{"attempts": 1, "completed_at": "2026-11-22T02:58:58Z", "cpu_seconds": 1.5013640379999984, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-22d283ae6b45", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "562164a3d73d3cd47fc0315dbd138073fa63b7671051d1dbc6cb7acfdafae550", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-22d283ae6b45", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 277976947, "treatment": "iw_temperature"}, "priority": 1.9851346262653344, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ddc2a2a91235530c5dfa6bafb147a94c6f1c48d16fbdf4fccfcd450feaaaf7d8", "run_seconds": 1.5022625923156738, "seed": 277976947, "started_at": "2026-11-22T02:58:56Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01743, 0.003516], "effect": -0.006957, "evaluated_at": "2026-11-22T02:58:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-22d283ae6b45", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.09026587, 0.1416694, 0.148683647, 0.148413405, 0.185658663, 0.168929158, 0.163354968, 0.252893333, 0.198625598, 0.185732862, 0.1687107, 0.151507063, 0.118383212, 0.130642095, 0.13988682, 0.188570045], "raw": [0.130849267, 0.13063202, 0.126365485, 0.154019768, 0.179512162, 0.11727279, 0.152867198, 0.23552699, 0.133136882, 0.150102148, 0.1687107, 0.151507063, 0.11350671, 0.103725763, 0.176135772, 0.177453328], "temperature": [0.130849267, 0.145129082, 0.143181713, 0.140760758, 0.183848365, 0.134501672, 0.15774532, 0.23552699, 0.169203648, 0.164291575, 0.13749332, 0.145330415, 0.11350671, 0.130642095, 0.176135772, 0.16246648]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0016, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.996", "prereg_hash": "562164a3d73d3cd47fc0315dbd138073fa63b7671051d1dbc6cb7acfdafae550", "primary_control": "temperature", "result_hash": "ddc2a2a91235530c5dfa6bafb147a94c6f1c48d16fbdf4fccfcd450feaaaf7d8", "signature": "eb77416cd91819469c590f69637f95d684499d0dcf0308e9aab24fb8cb8d0b1f625097892d79c107fef271fcbf1e1bcce5e0463e1c895c2662ebe4cdec8c2a0f", "state": "VERIFIED", "summary": {"iw_temperature": 0.16137, "raw": 0.150083, "temperature": 0.154413}, "ties": 1, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T02:58:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:21Z", "cpu_seconds": 1.6444157940000004, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-2340226e758a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "489149ab18cd5dab42440ed0ddfbb19ee502f904c757209cbdef3a7c08169fd8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-2340226e758a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 435439152, "treatment": "iw_temperature"}, "priority": 1.995501828958669, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "3a9738f4f227a12ace02f64ef2ba4b1ca16efbc1640d902ee76c60d494ee3ffa", "run_seconds": 1.7111554145812988, "seed": 435439152, "started_at": "2026-11-22T02:58:19Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.007325, 0.001495], "effect": -0.002915, "evaluated_at": "2026-11-22T02:58:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-2340226e758a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.069806385, 0.045011172, 0.05101673, 0.076144835, 0.02320776, 0.082103612, 0.055256792, 0.05546612, 0.0341611, 0.02737075, 0.089234332, 0.058755737, 0.059176375, 0.034689448, 0.04696446, 0.044813583], "raw": [0.069806385, 0.04626133, 0.062213825, 0.072312655, 0.05209445, 0.067201017, 0.078171002, 0.053800155, 0.068987055, 0.050933307, 0.063410085, 0.059953075, 0.072930837, 0.034689448, 0.051735043, 0.067466283], "temperature": [0.065589328, 0.045011172, 0.047883803, 0.060515407, 0.036180163, 0.067201017, 0.053825093, 0.05546612, 0.038428188, 0.02737075, 0.063410085, 0.058755737, 0.059176375, 0.034689448, 0.04696446, 0.046076105]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 6, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "489149ab18cd5dab42440ed0ddfbb19ee502f904c757209cbdef3a7c08169fd8", "primary_control": "temperature", "result_hash": "3a9738f4f227a12ace02f64ef2ba4b1ca16efbc1640d902ee76c60d494ee3ffa", "signature": "8d6ef74c74e1ad821fee2fcbebeb04565c613daf30faa02788b80234a6cfc8e2821659671a6891ef6cd1d0781c909df539392097ebf839072242105180b1eb00", "state": "VERIFIED", "summary": {"iw_temperature": 0.053324, "raw": 0.060748, "temperature": 0.050409}, "ties": 7, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T02:58:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:00Z", "cpu_seconds": 1.6060914220000004, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-31ab62c756eb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "8ad7fff0b17011dfc3cdcaf4978d39104852cbedfa5af82d2ee6a0f1bdd9f814", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-31ab62c756eb", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1286014294, "treatment": "iw_temperature"}, "priority": 2.267302283616136, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "e77ff81981aa67557382dcf31dc1fc76e093a2c824417a481dbd0bc82c1a4b2e", "run_seconds": 1.6117746829986572, "seed": 1286014294, "started_at": "2026-11-22T02:53:58Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019413, 0.011918], "effect": -0.003748, "evaluated_at": "2026-11-22T02:54:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-31ab62c756eb", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.29894608, 0.150670925, 0.3089463, 0.338398083, 0.209428453, 0.255905685, 0.241047465, 0.212834607, 0.21551307, 0.240597242, 0.255042657, 0.267264772, 0.201071472, 0.255837068, 0.159989997, 0.279552745], "raw": [0.291757362, 0.133142223, 0.289373748, 0.338398083, 0.158743972, 0.244768052, 0.211971917, 0.162598848, 0.21551307, 0.172150267, 0.23013493, 0.256927837, 0.224085292, 0.20810375, 0.159989997, 0.222137302], "temperature": [0.296009227, 0.171680622, 0.296678947, 0.323415305, 0.158743972, 0.26564675, 0.293064952, 0.17336052, 0.21551307, 0.190474195, 0.239974423, 0.264331395, 0.224085292, 0.242745143, 0.220984998, 0.25437271]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "8ad7fff0b17011dfc3cdcaf4978d39104852cbedfa5af82d2ee6a0f1bdd9f814", "primary_control": "temperature", "result_hash": "e77ff81981aa67557382dcf31dc1fc76e093a2c824417a481dbd0bc82c1a4b2e", "signature": "b004f40be5ad9cda0db5c7351264083c622f2bc96e475deae341c2b056a7a7156429bf058518ead4f027495925efe5a5b2a326beb958806ba84275093a23ef00", "state": "VERIFIED", "summary": {"iw_temperature": 0.24319, "raw": 0.219987, "temperature": 0.239443}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T02:54:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:54Z", "cpu_seconds": 1.5300833380000114, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-3bc21928824d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "c5a068a15460c8fabe0e631a36f26113ffd72d9a396362dfb403b554ebbde710", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-3bc21928824d", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1132356442, "treatment": "iw_temperature"}, "priority": 1.9861551009357306, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "384f7e1301f63be3b7144bf255ace7e9b620b7814417c5592527280018a63d03", "run_seconds": 1.6747097969055176, "seed": 1132356442, "started_at": "2026-11-22T02:58:52Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014548, 0.017352], "effect": 0.001402, "evaluated_at": "2026-11-22T02:58:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-3bc21928824d", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.137695223, 0.090386472, 0.172215655, 0.140790933, 0.137183293, 0.201671043, 0.130712305, 0.213685943, 0.18758617, 0.178356442, 0.181189507, 0.126843107, 0.081466745, 0.160370645, 0.082998977, 0.126098948], "raw": [0.134377827, 0.169798103, 0.16005475, 0.113974085, 0.126010383, 0.172452627, 0.116634188, 0.174032895, 0.141838648, 0.153738373, 0.18870882, 0.118033007, 0.143060123, 0.15165686, 0.08401637, 0.128207973], "temperature": [0.13826101, 0.169798103, 0.149380222, 0.125657578, 0.137183293, 0.201671043, 0.130712305, 0.213685943, 0.141838648, 0.171574938, 0.165552295, 0.097396675, 0.156377072, 0.160370645, 0.08401637, 0.128207973]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1393, "losses": 6, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.726", "prereg_hash": "c5a068a15460c8fabe0e631a36f26113ffd72d9a396362dfb403b554ebbde710", "primary_control": "temperature", "result_hash": "384f7e1301f63be3b7144bf255ace7e9b620b7814417c5592527280018a63d03", "signature": "3f2fdf7efa93f601cfc9d30b1a92250c324edd5a22dcd616c45f696c63519f62aacdd309e4b54d1ab0cedee14fdefedca9b6e6968e45cbcc402469159686d304", "state": "VERIFIED", "summary": {"iw_temperature": 0.146828, "raw": 0.142287, "temperature": 0.14823}, "ties": 5, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T02:58:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:12Z", "cpu_seconds": 1.694184442, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4473ceb9e096", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "5e8352e63724f770043e4ac517d7807d3167b31afa12b57a656792ec04ad2ccc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4473ceb9e096", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 795191391, "treatment": "iw_temperature"}, "priority": 2.006029500022733, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "672d9a4c93a7ba7eb05fa1fc6aa820b879c151e04052e43d183a113b67e4c04f", "run_seconds": 2.0590872764587402, "seed": 795191391, "started_at": "2026-11-22T02:54:10Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.00758, 0.034361], "effect": 0.013391, "evaluated_at": "2026-11-22T02:54:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4473ceb9e096", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.144937563, 0.161811417, 0.24901185, 0.241300293, 0.10037081, 0.217686283, 0.269666932, 0.19840951, 0.276925142, 0.20631341, 0.208944618, 0.161635515, 0.198522362, 0.153529833, 0.293520173, 0.219642632], "raw": [0.144937563, 0.183588125, 0.230800665, 0.225575573, 0.070606768, 0.169531798, 0.202119227, 0.189575375, 0.198435073, 0.20631341, 0.163498645, 0.224213812, 0.18305725, 0.19915203, 0.23079839, 0.219642632], "temperature": [0.187550943, 0.23918916, 0.271318935, 0.241300293, 0.137266373, 0.18349858, 0.259055643, 0.227065623, 0.246420905, 0.229652163, 0.177172912, 0.278889558, 0.18305725, 0.179626113, 0.2557728, 0.219642632]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.4031, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.395", "prereg_hash": "5e8352e63724f770043e4ac517d7807d3167b31afa12b57a656792ec04ad2ccc", "primary_control": "temperature", "result_hash": "672d9a4c93a7ba7eb05fa1fc6aa820b879c151e04052e43d183a113b67e4c04f", "signature": "252a2a3782f3bfe2bd8a4fba9d569608a181c1fa099633ba9de031ef4df258ef7873ff54fcdd04fbd74d1002bd772a53b94425498724ed96530caa5222458309", "state": "VERIFIED", "summary": {"iw_temperature": 0.206389, "raw": 0.190115, "temperature": 0.21978}, "ties": 2, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T02:54:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:21Z", "cpu_seconds": 1.3677180359999994, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4c9237a255bd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "43d09d7e49929301980ad25481ab3c661ce7a802a8cfe80981493092c5e5c5c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4c9237a255bd", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 236405283, "treatment": "iw_temperature"}, "priority": 2.0026603797229856, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "d87b56d08b5eb309e0d55b433cbe64986892db414e05fbea5d6fee30ce1007cd", "run_seconds": 1.4042994976043701, "seed": 236405283, "started_at": "2026-11-22T02:54:19Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.007002, 0.008448], "effect": 0.000723, "evaluated_at": "2026-11-22T02:54:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4c9237a255bd", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.171914115, 0.21676063, 0.20009049, 0.186920775, 0.13043462, 0.180535185, 0.239060793, 0.216224768, 0.20513697, 0.258042245, 0.264171752, 0.187698572, 0.18624547, 0.296649245, 0.161307097, 0.230044725], "raw": [0.209379865, 0.224434448, 0.161738675, 0.201265838, 0.16356837, 0.148946063, 0.187955388, 0.185420363, 0.171422915, 0.218988227, 0.229464985, 0.187698572, 0.168790342, 0.227627887, 0.185262325, 0.198744832], "temperature": [0.196043315, 0.21676063, 0.218812307, 0.186920775, 0.13043462, 0.180535185, 0.224668702, 0.18724242, 0.20513697, 0.274908932, 0.24791967, 0.17373845, 0.199529118, 0.2809157, 0.185262325, 0.233972792]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "43d09d7e49929301980ad25481ab3c661ce7a802a8cfe80981493092c5e5c5c3", "primary_control": "temperature", "result_hash": "d87b56d08b5eb309e0d55b433cbe64986892db414e05fbea5d6fee30ce1007cd", "signature": "2f458a796296911510c5762218b86dd0be4d9a5488eab7d94d878fcc4f3f74d0337e74ea1fa1af7cb600203b1261b6ba2eaccb70668bf8a569b8ca555d517e01", "state": "VERIFIED", "summary": {"iw_temperature": 0.208202, "raw": 0.191919, "temperature": 0.208925}, "ties": 5, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T02:54:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:47Z", "cpu_seconds": 1.6330440999999993, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-56b78c664c35", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "582aba8eba1db6546aedc322c65be435298e6797336982130260005587266c51", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-56b78c664c35", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 370541079, "treatment": "iw_temperature"}, "priority": 1.999627851820116, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "0298aad182c9036a489d237ba5a3d35ba1040a08089163915bdd3735b8e847b7", "run_seconds": 2.194295883178711, "seed": 370541079, "started_at": "2026-11-22T02:54:45Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.006838, 0.01223], "effect": 0.002696, "evaluated_at": "2026-11-22T02:54:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-56b78c664c35", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.105503038, 0.077480958, 0.11406269, 0.192256495, 0.135214285, 0.160055892, 0.116164177, 0.091077627, 0.094935058, 0.18235624, 0.033653565, 0.15583428, 0.144291915, 0.150204068, 0.08950117, 0.10958198], "raw": [0.14414371, 0.074817513, 0.127311108, 0.207632675, 0.104993785, 0.126037092, 0.084886165, 0.121783783, 0.141291808, 0.142769088, 0.1193571, 0.14322958, 0.119154845, 0.09991021, 0.105238375, 0.11790015], "temperature": [0.15063886, 0.087552455, 0.127311108, 0.196036765, 0.104993785, 0.160055892, 0.116164177, 0.108945495, 0.119114813, 0.164705525, 0.04574899, 0.143005473, 0.1203175, 0.135895393, 0.105238375, 0.10958198]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5368, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.291", "prereg_hash": "582aba8eba1db6546aedc322c65be435298e6797336982130260005587266c51", "primary_control": "temperature", "result_hash": "0298aad182c9036a489d237ba5a3d35ba1040a08089163915bdd3735b8e847b7", "signature": "dcf4f3226c978e715d4fcf5348bc564f08b9521a80b1ed7b633423f3ac24246d3216f2465bc600abf7cb5775597cd4b9143186757e7fa34c2e692ca45cf9e104", "state": "VERIFIED", "summary": {"iw_temperature": 0.122011, "raw": 0.123779, "temperature": 0.124707}, "ties": 3, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T02:54:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:53Z", "cpu_seconds": 1.6308068239999987, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-77c6ee597499", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "3f8f2452b18d852026181a69ac61b2572f80257e3ece1dbd75a4475ef5a25f5a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-77c6ee597499", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 2026546043, "treatment": "iw_temperature"}, "priority": 1.9994491514495156, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "03e0a04d686f82bb14546f3cdf1b51421bec6a887cad5f140aab975c731e897b", "run_seconds": 1.8783442974090576, "seed": 2026546043, "started_at": "2026-11-22T02:54:51Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021677, 0.018906], "effect": -0.001385, "evaluated_at": "2026-11-22T02:54:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-77c6ee597499", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.282208652, 0.170683957, 0.230768255, 0.258490653, 0.24110895, 0.296174312, 0.202726225, 0.123944943, 0.193941935, 0.292778653, 0.312338558, 0.199284687, 0.281016108, 0.22208431, 0.267580765, 0.207026622], "raw": [0.238140385, 0.136366133, 0.278917598, 0.236520123, 0.199775422, 0.234876513, 0.164204873, 0.19786183, 0.193941935, 0.27958764, 0.280839885, 0.199284687, 0.271111928, 0.205955227, 0.247972635, 0.230800153], "temperature": [0.2620006, 0.107529018, 0.294856917, 0.27011518, 0.218063117, 0.25618764, 0.183213302, 0.19786183, 0.22646061, 0.241469805, 0.26874709, 0.215654975, 0.287281408, 0.22208431, 0.247972635, 0.260492958]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1572, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.696", "prereg_hash": "3f8f2452b18d852026181a69ac61b2572f80257e3ece1dbd75a4475ef5a25f5a", "primary_control": "temperature", "result_hash": "03e0a04d686f82bb14546f3cdf1b51421bec6a887cad5f140aab975c731e897b", "signature": "79081d8ace9a9afd9fcb913dd9df375b8a86a7d9ea48c274945dea40aff87de3a876999f9e3aca217ff12cbc1aadd82844826eb61837aaa1883f7ae96e557904", "state": "VERIFIED", "summary": {"iw_temperature": 0.236385, "raw": 0.22476, "temperature": 0.234999}, "ties": 1, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T02:54:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:54Z", "cpu_seconds": 1.5704721620000015, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c09d620c4242", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "090a7f827ed3b07edffcb5832331b53942c633f0138d2b4c26d483c44aa8a724", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c09d620c4242", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1523298844, "treatment": "iw_temperature"}, "priority": 1.9990060495100215, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "e2b41e8f71cdb5a8c0c9b458284df3388a4786bacbc875719ab1240d9e3694a5", "run_seconds": 1.9613375663757324, "seed": 1523298844, "started_at": "2026-11-22T02:54:52Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010289, 0.001334], "effect": -0.004477, "evaluated_at": "2026-11-22T02:54:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c09d620c4242", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.108947978, 0.051599412, 0.074370748, 0.0937767, 0.102960925, 0.035336483, 0.052961248, 0.074012415, 0.085397655, 0.054443685, 0.052642305, 0.073984155, 0.089214685, 0.029698167, 0.103667473, 0.071625713], "raw": [0.068658285, 0.06729097, 0.074370748, 0.0937767, 0.08845184, 0.03314923, 0.052961248, 0.074012415, 0.10012796, 0.047184618, 0.04295037, 0.06964478, 0.087033953, 0.055104753, 0.106319732, 0.064426913], "temperature": [0.080069718, 0.0482429, 0.074370748, 0.079770513, 0.083335775, 0.03726117, 0.077577393, 0.07422087, 0.090535817, 0.04584613, 0.04295037, 0.073984155, 0.089214685, 0.029698167, 0.091496293, 0.064426913]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "090a7f827ed3b07edffcb5832331b53942c633f0138d2b4c26d483c44aa8a724", "primary_control": "temperature", "result_hash": "e2b41e8f71cdb5a8c0c9b458284df3388a4786bacbc875719ab1240d9e3694a5", "signature": "15b4f93d1ae35ed6d10fe62b1aea47458c18de20e039e5e307fcc93fdc9d50133e264688e1a594a95b959198d79aaa3cd32ca513e311c2b7e3d7e587f4c97607", "state": "VERIFIED", "summary": {"iw_temperature": 0.072165, "raw": 0.070342, "temperature": 0.067688}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T02:54:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:57Z", "cpu_seconds": 1.685947161, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c7cd6d2c2c2c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "1fb0423b318cfa2d633ef7db2b2387183b5a718ec10c8be6221dcb839ca79e6c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c7cd6d2c2c2c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1137530641, "treatment": "iw_temperature"}, "priority": 1.9977024914043384, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ada76c84ac6c1d5b9611f48b272ddd26708aa11d86b94f768c38af471dd1ec33", "run_seconds": 1.6879019737243652, "seed": 1137530641, "started_at": "2026-11-22T02:57:56Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017246, -0.002243], "effect": -0.009744, "evaluated_at": "2026-11-22T02:57:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c7cd6d2c2c2c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.036864008, 0.007247195, 0.096627413, 0.09389476, 0.085289215, 0.010539517, 0.10037193, 0.051816602, 0.038892525, 0.054296287, 0.017290485, 0.038773932, 0.098611402, 0.013246378, 0.044789205, 0.044086183], "raw": [0.030807683, 0.016243568, 0.093556187, 0.04604064, 0.051360648, 0.022958483, 0.068152585, 0.05561648, 0.032343735, 0.054296287, 0.049811227, 0.044066365, 0.099171162, 0.028083378, 0.07751583, 0.05941069], "temperature": [0.036864008, 0.01265437, 0.073946645, 0.050006158, 0.064776778, 0.012018852, 0.064988618, 0.035182345, 0.02440208, 0.053633578, 0.028323152, 0.024820595, 0.099171162, 0.01505217, 0.045843435, 0.035045745]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "1fb0423b318cfa2d633ef7db2b2387183b5a718ec10c8be6221dcb839ca79e6c", "primary_control": "temperature", "result_hash": "ada76c84ac6c1d5b9611f48b272ddd26708aa11d86b94f768c38af471dd1ec33", "signature": "2cc8b37e8df93f7a429041d6cd82af2b07ff0ab6372e68cc2a1a216d0dd70bcdba0dd1094d1c39120a4c9c87ea00ce2c28b0cbd57d2e11231280a03c628a4a07", "state": "VERIFIED", "summary": {"iw_temperature": 0.05204, "raw": 0.05184, "temperature": 0.042296}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T02:57:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:46Z", "cpu_seconds": 1.8257742179999994, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-d7b58620f522", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "7780bfea9bf30e9a0244033b11b6b6ac494d8f50467b1cf024f9d05855a5c9f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-d7b58620f522", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 2390797, "treatment": "iw_temperature"}, "priority": 2.0003299560268966, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "38f7bd300f9376dcf3763234e8fa1c50273ae0316ea5a7a61f83a7c0b47402fc", "run_seconds": 2.7457542419433594, "seed": 2390797, "started_at": "2026-11-22T02:54:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019418, -0.000224], "effect": -0.009821, "evaluated_at": "2026-11-22T02:54:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-d7b58620f522", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.013133317, 0.017499098, 0.092089377, 0.02673525, 0.174026407, 0.06162169, 0.053289362, 0.204537657, 0.02694794, 0.063508785, 0.088836555, 0.009265885, 0.02897706, 0.155297655, 0.003779185, 0.015501467], "raw": [0.01571509, 0.007487293, 0.092089377, 0.026184175, 0.173132, 0.050224662, 0.029771822, 0.192608042, 0.025509908, 0.03703496, 0.07308442, 0.018620798, 0.029606113, 0.078272698, 0.003779185, 0.033908395], "temperature": [0.033370848, 0.017499098, 0.05961188, 0.02883604, 0.172648118, 0.048123055, 0.02939061, 0.201656123, 0.035129615, 0.02941125, 0.066818383, 0.016102762, 0.020188387, 0.098582245, 0.002276, 0.018262967]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "7780bfea9bf30e9a0244033b11b6b6ac494d8f50467b1cf024f9d05855a5c9f4", "primary_control": "temperature", "result_hash": "38f7bd300f9376dcf3763234e8fa1c50273ae0316ea5a7a61f83a7c0b47402fc", "signature": "b6bc6fe8bc5e52ed92733275da85842c156f7d9bbaddb98963ae42cbc14edd9a4e4d68fd86d9513ca4330d9446b398aa0d03fc405ca9cf5a08477a1c1a0f8f01", "state": "VERIFIED", "summary": {"iw_temperature": 0.06469, "raw": 0.055439, "temperature": 0.054869}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T02:54:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:22Z", "cpu_seconds": 1.4764572030000096, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-fd2f5aeb7ff4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "d3011aac1db396bb5b8283866fb5716ee3bc63c4b3f69443a85dba6c16d5e2ad", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-fd2f5aeb7ff4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:53:48Z", "seed": 594193507, "treatment": "iw_temperature"}, "priority": 1.9950027388928093, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "01421c16deb4987e5daa8a3d01aafdaca8da9bf57d687825de5ba67716ed613f", "run_seconds": 1.5391700267791748, "seed": 594193507, "started_at": "2026-11-22T02:58:21Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.02145, 0.00751], "effect": -0.00697, "evaluated_at": "2026-11-22T02:58:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-fd2f5aeb7ff4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.241849033, 0.137816318, 0.27665043, 0.1426761, 0.218948115, 0.272258577, 0.285964192, 0.3240472, 0.30420702, 0.245746235, 0.336446187, 0.283323945, 0.269111735, 0.238206583, 0.282250437, 0.19903196], "raw": [0.27579888, 0.220673083, 0.261029835, 0.20798992, 0.193074675, 0.245715282, 0.209929022, 0.322859748, 0.20496631, 0.15431793, 0.233110298, 0.251992192, 0.263847423, 0.190815755, 0.264435695, 0.179181185], "temperature": [0.28276017, 0.186011958, 0.26524831, 0.176669988, 0.218948115, 0.274155145, 0.250261828, 0.3249523, 0.27566438, 0.19600594, 0.284836152, 0.279279522, 0.269670248, 0.208169475, 0.2752057, 0.179181185]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "d3011aac1db396bb5b8283866fb5716ee3bc63c4b3f69443a85dba6c16d5e2ad", "primary_control": "temperature", "result_hash": "01421c16deb4987e5daa8a3d01aafdaca8da9bf57d687825de5ba67716ed613f", "signature": "992d7ad7da40d7689e1621a6670b1b41727c042ed1df02681636980796238c06229691a617b06ca3f1a2b7b015b68ff0d19b3c4c670cc7a0f34d5cb1d8fa6609", "state": "VERIFIED", "summary": {"iw_temperature": 0.253658, "raw": 0.229984, "temperature": 0.246689}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T02:58:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:53:58Z", "cpu_seconds": 8.33686817, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-5baff1581977", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "9f832aa9dc401848c90605bf9f4315a487b176df5eadade4eee33db97f9058e8", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-5baff1581977", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1937424006, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "4880d0f73f00837b84fda288e0b42228bf45af684be6e3a5e946e29a84575219", "run_seconds": 9.245905876159668, "seed": 1937424006, "started_at": "2026-11-22T02:53:49Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:54:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-5baff1581977", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0], "root": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9f832aa9dc401848c90605bf9f4315a487b176df5eadade4eee33db97f9058e8", "primary_control": "reset", "result_hash": "4880d0f73f00837b84fda288e0b42228bf45af684be6e3a5e946e29a84575219", "signature": "7283caeb26605af2b4302cca186e0a69a0def115550f0ca0599fd1e759b2578e7eddc1163f8c6258d31c06918b10e5ed82629e1c9b01e5173737e869ea509d09", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.43333333333333335}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:54:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:53:58Z", "cpu_seconds": 8.053295315, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-ba492203544c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "73ea4296dd97376bdba6e4bb7ed193f75f5882f11742995ba6424ac5d0075558", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-ba492203544c", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1123367653, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "ae1d8fd39a9e7fb9101dfb89c3bb2d83942e859f506d112856f856a23e7a2cb4", "run_seconds": 8.930095434188843, "seed": 1123367653, "started_at": "2026-11-22T02:53:49Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:53:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-ba492203544c", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "73ea4296dd97376bdba6e4bb7ed193f75f5882f11742995ba6424ac5d0075558", "primary_control": "reset", "result_hash": "ae1d8fd39a9e7fb9101dfb89c3bb2d83942e859f506d112856f856a23e7a2cb4", "signature": "756bed01b5057d9aca287daad6047e319194d6d07d884f070168d71aba18c337e65fb5cb066690ffd0bd779728ac236f86bad1ac4f48ca8dac0386ff2caf2b05", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.2833333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:54:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:03Z", "cpu_seconds": 0.6444592280000006, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-1a1d6c56260e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "148eb5c6476d89a3f465576a33b5851af8e6c2a0428cd2b9af0d3358a325daaa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-1a1d6c56260e", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 318138816, "treatment": "frontier"}, "priority": 2.009661086144572, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "874b27d9a6abc29af17b6c0b0cfd5f81ab9d03b668b7e0e0895cb1b4ad443044", "run_seconds": 0.7125859260559082, "seed": 318138816, "started_at": "2026-11-22T02:54:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.064111, 0.087723], "effect": 0.075917, "evaluated_at": "2026-11-22T02:54:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-1a1d6c56260e", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.924603175, 0.854700855, 0.935546875, 0.918803419, 0.896694215, 0.905172414, 0.91796875, 0.897435897, 0.910416667, 0.913306452, 0.877118644, 0.92], "frontier": [0.992063492, 0.995726496, 0.998046875, 1.0, 0.993801653, 1.0, 0.98828125, 0.997863248, 0.991666667, 0.993951613, 0.995762712, 0.988], "random_walk": [0.910714286, 0.873931624, 0.947265625, 0.94017094, 0.898760331, 0.924568966, 0.9296875, 0.901709402, 0.91875, 0.933467742, 0.913135593, 0.932]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "148eb5c6476d89a3f465576a33b5851af8e6c2a0428cd2b9af0d3358a325daaa", "primary_control": "random_walk", "result_hash": "874b27d9a6abc29af17b6c0b0cfd5f81ab9d03b668b7e0e0895cb1b4ad443044", "signature": "f60a11959ff06d58a8b18b50b84eb85e32e5e0b30f2254ada2faa668ece4fb0fc0cfae8a8282526702c1b8a250727d744affcd5ffd879ba4e926016c9b447405", "state": "VERIFIED", "summary": {"curiosity": 0.905981, "frontier": 0.994597, "random_walk": 0.91868}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:54:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:40Z", "cpu_seconds": 0.45321278499999984, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-38290d04ea05", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "0c27ac3606fc2cc1e36866ac8cf14488f12fce4b0fb9ce77be01631c5430c4d0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-38290d04ea05", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 894073025, "treatment": "frontier"}, "priority": 1.9911037773438154, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "3d100d04848227c35d0bb7e01033961f8d99edc5f9e5e05d1ce5dfb4fef19a8e", "run_seconds": 0.506690263748169, "seed": 894073025, "started_at": "2026-11-22T02:58:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.066948, 0.147086], "effect": 0.107017, "evaluated_at": "2026-11-22T02:58:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-38290d04ea05", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.854166667, 0.86637931, 0.890625, 1.0, 0.818181818, 0.94, 0.845, 0.9125, 0.864035088, 0.883064516, 0.880681818, 1.0], "frontier": [0.945833333, 0.987068966, 1.0, 1.0, 1.0, 1.0, 1.0, 0.975, 1.0, 0.991935484, 0.926136364, 1.0], "random_walk": [0.8375, 0.797413793, 0.822916667, 1.0, 0.881818182, 0.87, 0.81, 0.975, 0.868421053, 0.85483871, 0.823863636, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "0c27ac3606fc2cc1e36866ac8cf14488f12fce4b0fb9ce77be01631c5430c4d0", "primary_control": "random_walk", "result_hash": "3d100d04848227c35d0bb7e01033961f8d99edc5f9e5e05d1ce5dfb4fef19a8e", "signature": "31e001089daa7b3f04c74bfd346119894fc23d8eebfa9e30987cf7fb367b611edef1081c4042d1980312703361d21adca83e37e299ceeb9572275001f15af901", "state": "VERIFIED", "summary": {"curiosity": 0.89622, "frontier": 0.985498, "random_walk": 0.878481}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T02:58:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:51Z", "cpu_seconds": 0.534731727999997, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-44c0b9bdc44d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "0333fa8915a6fa2e3eab0cf3b264fcfdf935331d1e3c9095ef3b9fb6bce5e15f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-44c0b9bdc44d", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 984215600, "treatment": "frontier"}, "priority": 1.9872736474220667, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "162125693121c03fa079147a48f014d649e39175db2875933cde588175056a84", "run_seconds": 0.5576002597808838, "seed": 984215600, "started_at": "2026-11-22T02:58:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.044068, 0.075959], "effect": 0.060014, "evaluated_at": "2026-11-22T02:58:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-44c0b9bdc44d", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.9, 0.868534483, 0.910714286, 0.94, 0.87605042, 0.853448276, 0.931640625, 0.919354839, 0.897916667, 0.908119658, 0.921875, 0.904471545], "frontier": [0.98, 0.896551724, 0.986111111, 0.988, 0.983193277, 0.993534483, 0.98046875, 0.981854839, 0.885416667, 0.995726496, 0.99609375, 0.989837398], "random_walk": [0.916, 0.877155172, 0.926587302, 0.926, 0.903361345, 0.903017241, 0.94921875, 0.911290323, 0.875, 0.888888889, 0.94140625, 0.918699187]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "0333fa8915a6fa2e3eab0cf3b264fcfdf935331d1e3c9095ef3b9fb6bce5e15f", "primary_control": "random_walk", "result_hash": "162125693121c03fa079147a48f014d649e39175db2875933cde588175056a84", "signature": "c7c32b1a68eab636c4b89fa420f3454ba54e72d13aa6f81a0bb7f80e7205363865b1ffca7107338090daabf33ec3ef78082be959bc132b2b7e3acf5cef6fab05", "state": "VERIFIED", "summary": {"curiosity": 0.902677, "frontier": 0.971399, "random_walk": 0.911385}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:58:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:57Z", "cpu_seconds": 0.3760132670000047, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-51c0030be1b2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "158b7994dbbb2b7564ca9cdd3e6d25ad3c3f8d071f0074854b470f0102ae6d23", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-51c0030be1b2", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 63763904, "treatment": "frontier"}, "priority": 1.9974882015781654, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "ac8fbec41b857eba85d19407a189d2cd6b3a1b7afd004609ab9c8b761cdd43d5", "run_seconds": 0.45860958099365234, "seed": 63763904, "started_at": "2026-11-22T02:57:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.061302, 0.080033], "effect": 0.070667, "evaluated_at": "2026-11-22T02:57:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-51c0030be1b2", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.90530303, 0.920454545, 0.967857143, 0.923913043, 0.924603175, 0.922794118, 0.928030303, 0.895833333, 0.933823529, 0.926470588, 0.915384615, 0.93115942], "frontier": [0.988636364, 0.992424242, 0.996428571, 0.981884058, 1.0, 0.981617647, 0.992424242, 1.0, 0.992647059, 0.977941176, 0.988461538, 0.985507246], "random_walk": [0.909090909, 0.924242424, 0.928571429, 0.938405797, 0.916666667, 0.922794118, 0.939393939, 0.895833333, 0.911764706, 0.915441176, 0.903846154, 0.923913043]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "158b7994dbbb2b7564ca9cdd3e6d25ad3c3f8d071f0074854b470f0102ae6d23", "primary_control": "random_walk", "result_hash": "ac8fbec41b857eba85d19407a189d2cd6b3a1b7afd004609ab9c8b761cdd43d5", "signature": "d6f92827a3648c0d05834c5c55be20883b3da3cce4a4da008dea2b6ba236bcfb7de1b2ff4e0559dc418c0ec4ebd0747f8697d941ca4aa77852d3c2133981690c", "state": "VERIFIED", "summary": {"curiosity": 0.924636, "frontier": 0.989831, "random_walk": 0.919164}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:57:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:19Z", "cpu_seconds": 4.694429608999997, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-6c983a8e8255", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "9390ee065eea90539eb8c214ed35223137ede9ea256d49f843d2f871806a31c1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-6c983a8e8255", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 976833245, "treatment": "frontier"}, "priority": 1.9987817643705983, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "213d6d7b81d70c184da927ef9681e85239b7d5752118ed5ca3c5e921e3194273", "run_seconds": 4.801470041275024, "seed": 976833245, "started_at": "2026-11-22T02:57:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.038259, 0.050085], "effect": 0.044172, "evaluated_at": "2026-11-22T02:57:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-6c983a8e8255", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.922619048, 0.913793103, 0.900826446, 0.939516129, 0.907786885, 0.919491525, 0.948076923, 0.915966387, 0.918803419, 0.9375, 0.91733871, 0.93125], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.94047619, 0.952586207, 0.950413223, 0.943548387, 0.954918033, 0.966101695, 0.969230769, 0.947478992, 0.961538462, 0.95703125, 0.951612903, 0.975]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "9390ee065eea90539eb8c214ed35223137ede9ea256d49f843d2f871806a31c1", "primary_control": "random_walk", "result_hash": "213d6d7b81d70c184da927ef9681e85239b7d5752118ed5ca3c5e921e3194273", "signature": "fce02421132a72ec30d3e9118793bf15b990570c6a31f1a631d105da596e2597ebe757fd10afa31ebb360e01ea3b6ff6b6249779ef203a647db26b9efb850f0b", "state": "VERIFIED", "summary": {"curiosity": 0.922747, "frontier": 1.0, "random_walk": 0.955828}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:57:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:47Z", "cpu_seconds": 0.32852271599999483, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-863e6aa82d86", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "9b95f7b06bbebaf5a101b9b8d46bbfff81fc5fbb114efd1030ee4140dd0e5554", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-863e6aa82d86", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 7253526, "treatment": "frontier"}, "priority": 1.9881410329006528, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "1f5ea61703279999ed0edfffe4d9107827773638c56563a684a2ce77935aff04", "run_seconds": 0.32975172996520996, "seed": 7253526, "started_at": "2026-11-22T02:58:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058275, 0.123785], "effect": 0.09103, "evaluated_at": "2026-11-22T02:58:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-863e6aa82d86", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.741847826, 0.771276596, 0.809782609, 0.786082474, 0.822727273, 0.770114943, 0.807291667, 0.762254902, 0.81, 0.793209877, 0.752808989], "frontier": [1.0, 0.877717391, 0.890957447, 0.880434783, 0.783505155, 1.0, 0.885057471, 0.8828125, 0.860294118, 0.8625, 0.916666667, 0.876404494], "random_walk": [1.0, 0.698369565, 0.792553191, 0.828804348, 0.742268041, 0.827272727, 0.761494253, 0.815104167, 0.803921569, 0.825, 0.762345679, 0.766853933]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "9b95f7b06bbebaf5a101b9b8d46bbfff81fc5fbb114efd1030ee4140dd0e5554", "primary_control": "random_walk", "result_hash": "1f5ea61703279999ed0edfffe4d9107827773638c56563a684a2ce77935aff04", "signature": "698d6689a8b5adbb42746ae6848fda9698497d89ec17b2144cd0b85740382696676b312512e15a33fd85a5698e614b3403b019620cfb7531e3956709e948000e", "state": "VERIFIED", "summary": {"curiosity": 0.802283, "frontier": 0.893029, "random_walk": 0.801999}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:58:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:41Z", "cpu_seconds": 1.4511720510000004, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-8a29dd005ea3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "30f11c9ac00db858d56c16aa2c70683a113945734e8fba8bd21a50d10ba4a221", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-8a29dd005ea3", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 132607823, "treatment": "frontier"}, "priority": 1.989517593174352, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "168922f8a936e4575e554d73ddd9ea560f77a7f5457b1b956d1ad82c75000872", "run_seconds": 1.4999241828918457, "seed": 132607823, "started_at": "2026-11-22T02:58:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.057858, 0.125537], "effect": 0.091698, "evaluated_at": "2026-11-22T02:58:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-8a29dd005ea3", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.889830508, 0.806818182, 0.921875, 0.857142857, 0.92, 0.87244898, 0.936440678, 0.881578947, 0.867346939, 1.0, 0.892857143, 0.861111111], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.902542373, 0.778409091, 0.90625, 0.821428571, 0.935, 0.918367347, 0.961864407, 0.899122807, 0.959183673, 1.0, 0.900793651, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "30f11c9ac00db858d56c16aa2c70683a113945734e8fba8bd21a50d10ba4a221", "primary_control": "random_walk", "result_hash": "168922f8a936e4575e554d73ddd9ea560f77a7f5457b1b956d1ad82c75000872", "signature": "c2255454679eeeb3ed57d59b2c58eb847ce5a72e12770590525cb5f0a7e7df733b629811d9a2c2200a7bd5a7ba9175bcb1b4468542eb4b6578aa05ac9a4c7e03", "state": "VERIFIED", "summary": {"curiosity": 0.892288, "frontier": 1.0, "random_walk": 0.908302}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:58:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:05Z", "cpu_seconds": 0.16709811499999994, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-94243d4a2a51", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "c40f4157704a95a56007e6e8e174cbc4cc5cba4a56ee89a3eaab3ff0b9ec15cb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-94243d4a2a51", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1033594074, "treatment": "frontier"}, "priority": 2.0075764409717065, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "65e422a77970388e6dc4453a98d75b1f33f814ef74a3e811f87fdabd19f86098", "run_seconds": 0.17339134216308594, "seed": 1033594074, "started_at": "2026-11-22T02:54:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019812, 0.053546], "effect": 0.036679, "evaluated_at": "2026-11-22T02:54:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-94243d4a2a51", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.827868852, 1.0, 0.842307692, 0.762711864, 0.919117647, 0.873134328, 0.890151515, 0.895522388, 0.841666667, 0.87109375, 0.818965517, 0.949324324], "frontier": [0.889344262, 1.0, 0.892307692, 0.88559322, 0.9375, 0.880597015, 0.916666667, 0.914179104, 0.895833333, 0.87890625, 0.922413793, 0.956081081], "random_walk": [0.81147541, 1.0, 0.838461538, 0.788135593, 0.908088235, 0.869402985, 0.893939394, 0.891791045, 0.854166667, 0.875, 0.86637931, 0.932432432]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "c40f4157704a95a56007e6e8e174cbc4cc5cba4a56ee89a3eaab3ff0b9ec15cb", "primary_control": "random_walk", "result_hash": "65e422a77970388e6dc4453a98d75b1f33f814ef74a3e811f87fdabd19f86098", "signature": "155323c79589ecfad13dc49f665aac9a58b14ad63e3be81412ea082c68820172f8805f1fb9a38c09ee85b94a229fa3a11328bb0ca26b8365f051b3af1790a306", "state": "VERIFIED", "summary": {"curiosity": 0.874322, "frontier": 0.914119, "random_walk": 0.877439}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:54:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:28Z", "cpu_seconds": 0.36973978199999635, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-b07a21a8e345", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "012959c10cefa49ab3fe326b132fdb8a86014cbeec9089e828e4ebd3fdd044bb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-b07a21a8e345", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 36616671, "treatment": "frontier"}, "priority": 1.9925014635739235, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "57509922e3425e0035fd7494e21175ac1ad6708c6400adb84d80a8944813db70", "run_seconds": 0.47602367401123047, "seed": 36616671, "started_at": "2026-11-22T02:58:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.018884, 0.0462], "effect": 0.032542, "evaluated_at": "2026-11-22T02:58:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-b07a21a8e345", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.842105263, 0.892561983, 0.909663866, 0.887362637, 0.822727273, 0.846982759, 0.891129032, 0.858974359, 0.897177419, 0.901209677, 0.888429752, 0.906], "frontier": [0.885964912, 0.92768595, 0.918067227, 0.92032967, 0.902272727, 0.894396552, 0.935483871, 0.89957265, 0.929435484, 0.927419355, 0.863636364, 0.93], "random_walk": [0.835526316, 0.871900826, 0.915966387, 0.892857143, 0.820454545, 0.838362069, 0.911290323, 0.865384615, 0.921370968, 0.901209677, 0.857438017, 0.912]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "012959c10cefa49ab3fe326b132fdb8a86014cbeec9089e828e4ebd3fdd044bb", "primary_control": "random_walk", "result_hash": "57509922e3425e0035fd7494e21175ac1ad6708c6400adb84d80a8944813db70", "signature": "7c859cd8182d3017a382683ae5aaf5235bcd8169deb6f9f4f20d75c84d1170f911e668428c6a0846ae8032bc2b34cfaa2bdf78ad3027e363f1e761a1124d7d09", "state": "VERIFIED", "summary": {"curiosity": 0.878694, "frontier": 0.911189, "random_walk": 0.878647}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:58:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:07Z", "cpu_seconds": 4.325789386, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-b392e83d7ef2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "d7d3d08a1ba58d38f0e5d749ddd3a70f378c19199423f828457dec9f782b96b9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-b392e83d7ef2", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 1216666235, "treatment": "frontier"}, "priority": 1.997054652574272, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "f68d79c661314500ef9b55584473d1a7a867a5d7a7022490d0086417669b7d4a", "run_seconds": 4.8054609298706055, "seed": 1216666235, "started_at": "2026-11-22T02:58:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.044718, 0.061347], "effect": 0.053033, "evaluated_at": "2026-11-22T02:58:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-b392e83d7ef2", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.916666667, 0.970238095, 0.906521739, 0.913135593, 0.930894309, 0.870535714, 0.890495868, 0.91875, 0.885321101, 0.931818182, 0.933884298, 0.922764228], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.947154472, 0.954365079, 0.958695652, 0.957627119, 0.965447154, 0.930803571, 0.948347107, 0.952083333, 0.917431193, 0.962809917, 0.933884298, 0.93495935]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "d7d3d08a1ba58d38f0e5d749ddd3a70f378c19199423f828457dec9f782b96b9", "primary_control": "random_walk", "result_hash": "f68d79c661314500ef9b55584473d1a7a867a5d7a7022490d0086417669b7d4a", "signature": "60a4d27ee841d6de9aa2f827eed674315b21d1146664f231ac22b6531c716f40519956f0e7a51286bffb08e9c88c7091d7c063d69c4b439d52c85904fffcd707", "state": "VERIFIED", "summary": {"curiosity": 0.915919, "frontier": 1.0, "random_walk": 0.946967}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:58:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:46Z", "cpu_seconds": 0.33496436000000074, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-f2b0bc37a614", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "5c1df6caa391dbde1aeb296978d5712040f9340039053c66f1a460f007bbac41", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-f2b0bc37a614", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 468622847, "treatment": "frontier"}, "priority": 1.9881637974639494, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "f0c7a8c167c68eaf481e324a5e825fc3087e50a67f6101367b0e4c59d223b204", "run_seconds": 0.34287452697753906, "seed": 468622847, "started_at": "2026-11-22T02:58:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.026916, 0.078015], "effect": 0.052465, "evaluated_at": "2026-11-22T02:58:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-f2b0bc37a614", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.762345679, 1.0, 0.733333333, 0.857142857, 0.77393617, 0.877232143, 1.0, 0.821052632, 0.801470588, 0.7975, 1.0, 1.0], "frontier": [0.885802469, 1.0, 0.886111111, 0.895089286, 0.882978723, 0.924107143, 1.0, 0.894736842, 0.884803922, 0.8825, 1.0, 1.0], "random_walk": [0.768518519, 1.0, 0.766666667, 0.845982143, 0.795212766, 0.879464286, 1.0, 0.839473684, 0.81372549, 0.7975, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "5c1df6caa391dbde1aeb296978d5712040f9340039053c66f1a460f007bbac41", "primary_control": "random_walk", "result_hash": "f0c7a8c167c68eaf481e324a5e825fc3087e50a67f6101367b0e4c59d223b204", "signature": "ab1624ba0241e22760dcb1a1b125f2f2a2974fb2f5a52930c56c9ce2ff7411e8a5e633f9cca2428bf2cfd5354da0eae326283656c94cef76c17d74aaaf804206", "state": "VERIFIED", "summary": {"curiosity": 0.868668, "frontier": 0.928011, "random_walk": 0.875545}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T02:58:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:53:53Z", "cpu_seconds": 3.623631782, "created_at": "2026-11-22T02:53:48Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-fe4294ea7c9b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "da1e191d9a8df7da70e958223afe6b3bc6ded14446251edb420d281bf7a70e1a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-fe4294ea7c9b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:53:48Z", "seed": 482031188, "treatment": "frontier"}, "priority": 2.2999003325238503, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "01bd670fb29174d64ad51fa76e07a5dbb78392f475317395e1d9cec373024cfa", "run_seconds": 4.230284690856934, "seed": 482031188, "started_at": "2026-11-22T02:53:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.023119, 0.094217], "effect": 0.058668, "evaluated_at": "2026-11-22T02:53:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-fe4294ea7c9b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.832317073, 0.875, 1.0, 0.823369565, 1.0, 0.78313253, 0.808139535, 0.794520548, 0.895348837, 0.854651163, 1.0, 0.837628866], "frontier": [0.990853659, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.819767442, 1.0, 1.0], "random_walk": [0.862804878, 0.950581395, 1.0, 0.89673913, 1.0, 0.903614458, 0.869186047, 0.924657534, 0.988372093, 0.85755814, 1.0, 0.853092784]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.7093, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0195", "prereg_hash": "da1e191d9a8df7da70e958223afe6b3bc6ded14446251edb420d281bf7a70e1a", "primary_control": "random_walk", "result_hash": "01bd670fb29174d64ad51fa76e07a5dbb78392f475317395e1d9cec373024cfa", "signature": "554cfd2b05d8912a5603114c0a2687d0128a513178a6126c9cdb4076be8e878784d66f48d115f5114aa7f5fc3e3b09310a824572fa0bcb79b5fab374839f630b", "state": "VERIFIED", "summary": {"curiosity": 0.875342, "frontier": 0.984218, "random_walk": 0.925551}, "ties": 3, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T02:53:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:09Z", "cpu_seconds": 0.3155377260000005, "created_at": "2026-11-22T02:54:04Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-1f9a846df59c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "f767027001462fbf4d7f6c74ec6eaf55403dd9f6361340e9c725f5354d63c834", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-1f9a846df59c", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:04Z", "seed": 1714454915, "treatment": "frontier"}, "priority": 2.00670228213061, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "8d643f8335b4abbe151da0f80364849eac3a91c1b0518298d8944a8bbfeef477", "run_seconds": 0.5001871585845947, "seed": 1714454915, "started_at": "2026-11-22T02:54:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.05487, 0.084611], "effect": 0.06974, "evaluated_at": "2026-11-22T02:54:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-1f9a846df59c", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.934027778, 0.907692308, 0.943661972, 0.935714286, 0.949324324, 0.928571429, 0.915441176, 0.906716418, 0.954861111, 0.880597015, 0.922794118, 0.940972222], "frontier": [0.989583333, 1.0, 0.996478873, 0.989285714, 0.993243243, 0.996428571, 0.988970588, 1.0, 1.0, 0.985074627, 0.988970588, 0.986111111], "random_walk": [0.913194444, 0.873076923, 0.940140845, 0.928571429, 0.945945946, 0.928571429, 0.919117647, 0.884328358, 0.958333333, 0.932835821, 0.919117647, 0.934027778]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "f767027001462fbf4d7f6c74ec6eaf55403dd9f6361340e9c725f5354d63c834", "primary_control": "random_walk", "result_hash": "8d643f8335b4abbe151da0f80364849eac3a91c1b0518298d8944a8bbfeef477", "signature": "3aea24086eac4ab2d005d249d1a235ff6f43fcc277637de80b3142fe1e0d1b26ac99521237abc3f23a17005841c0d66ee771a88a30f33d750493d197c0700901", "state": "VERIFIED", "summary": {"curiosity": 0.926698, "frontier": 0.992846, "random_walk": 0.923105}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:54:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:14Z", "cpu_seconds": 0.19892498399999958, "created_at": "2026-11-22T02:54:04Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-7318b98fe360", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "04e42bcc5c54d02c46e1823a48d1449102ad340513a78ca5190db05213af2d22", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-7318b98fe360", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:04Z", "seed": 195420938, "treatment": "frontier"}, "priority": 2.004683197533074, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "725c4189cef7b00206ff55ccbf53f14031500e8b871b5ff1bf4c4d7cd7394b04", "run_seconds": 0.2358400821685791, "seed": 195420938, "started_at": "2026-11-22T02:54:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.038386, 0.09105], "effect": 0.064718, "evaluated_at": "2026-11-22T02:54:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-7318b98fe360", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.785714286, 0.861538462, 0.856557377, 1.0, 1.0, 0.841836735, 0.802631579, 0.85, 0.773148148, 1.0, 0.823275862, 0.851851852], "frontier": [0.880952381, 0.896153846, 0.889344262, 1.0, 1.0, 0.897959184, 0.877192982, 1.0, 0.87962963, 1.0, 0.879310345, 0.902777778], "random_walk": [0.81547619, 0.846153846, 0.803278689, 1.0, 0.975, 0.790816327, 0.798245614, 0.833333333, 0.814814815, 1.0, 0.793103448, 0.856481481]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "04e42bcc5c54d02c46e1823a48d1449102ad340513a78ca5190db05213af2d22", "primary_control": "random_walk", "result_hash": "725c4189cef7b00206ff55ccbf53f14031500e8b871b5ff1bf4c4d7cd7394b04", "signature": "b73cd0b185eb0013655497e8ea0897946b3d788870ca7ddbceb945e5eef3cbe80ec3b6c03bf51e10a6d6adf0b4b52e450af144a62f5bf554bb61d5042d06a30a", "state": "VERIFIED", "summary": {"curiosity": 0.870546, "frontier": 0.925277, "random_walk": 0.860559}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T02:54:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:17Z", "cpu_seconds": 0.22781504900000105, "created_at": "2026-11-22T02:54:04Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-e24f6cd1a3bc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "40755878b5c3bbbcfe166e6245eabaabf242a1b39aef7340eeca978188c2777a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-e24f6cd1a3bc", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:04Z", "seed": 1750480904, "treatment": "frontier"}, "priority": 2.0033479660989513, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "cba966bb1ac108de2f7e6856d7460ef2572a78cbf5e26c0f29c1e626e78f6251", "run_seconds": 0.35645031929016113, "seed": 1750480904, "started_at": "2026-11-22T02:54:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.014289, 0.032814], "effect": 0.023552, "evaluated_at": "2026-11-22T02:54:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-e24f6cd1a3bc", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.930555556, 0.889705882, 0.87109375, 0.942028986, 1.0, 0.905797101, 0.894927536, 0.921232877, 0.862318841, 0.85546875, 0.908450704, 0.887681159], "frontier": [0.940972222, 0.930147059, 0.88671875, 0.93115942, 1.0, 0.913043478, 0.938405797, 0.948630137, 0.902173913, 0.890625, 0.936619718, 0.927536232], "random_walk": [0.909722222, 0.875, 0.87109375, 0.916666667, 1.0, 0.898550725, 0.887681159, 0.938356164, 0.873188406, 0.87890625, 0.908450704, 0.905797101]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "40755878b5c3bbbcfe166e6245eabaabf242a1b39aef7340eeca978188c2777a", "primary_control": "random_walk", "result_hash": "cba966bb1ac108de2f7e6856d7460ef2572a78cbf5e26c0f29c1e626e78f6251", "signature": "07bccb69fc3d7ecea37ba36d21c51592f518cd42b02827e7effceeda6aa1492ee0ed858b3825906d1951705cb0957465db2845f8b43efbf268b88cec1e35800f", "state": "VERIFIED", "summary": {"curiosity": 0.905772, "frontier": 0.928836, "random_walk": 0.905284}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:54:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:35Z", "cpu_seconds": 1.8612147820000011, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-6142c61c3138", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "57217afcf9a26b4156ef6ff8bd048e949b4d6f424de2b707fcc695243870a02e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-6142c61c3138", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:54:21Z", "seed": 2045620866, "treatment": "iw_temperature"}, "priority": 1.9913955731273025, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "12e5ca70a3fe36dddcddbb12db6c7073060403a522b4fc6cb0efee158756f811", "run_seconds": 2.2048351764678955, "seed": 2045620866, "started_at": "2026-11-22T02:58:33Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.00822, 0.003678], "effect": -0.002271, "evaluated_at": "2026-11-22T02:58:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-6142c61c3138", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.08163066, 0.10395512, 0.098453225, 0.144903837, 0.140598653, 0.163116987, 0.137347127, 0.128956983, 0.08997007, 0.181825638, 0.178739797, 0.128264457, 0.12906409, 0.09023387, 0.133440005, 0.13930802], "raw": [0.08163066, 0.10395512, 0.078738983, 0.093300485, 0.140598653, 0.114969892, 0.118007158, 0.116153993, 0.084550493, 0.12502458, 0.122863612, 0.09484641, 0.12906409, 0.070879867, 0.117419097, 0.118944547], "temperature": [0.088954943, 0.10421449, 0.098453225, 0.132440265, 0.140598653, 0.14828499, 0.137347127, 0.12520505, 0.07650928, 0.16698732, 0.151339228, 0.144725867, 0.138717323, 0.105480462, 0.14125295, 0.132961552]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "57217afcf9a26b4156ef6ff8bd048e949b4d6f424de2b707fcc695243870a02e", "primary_control": "temperature", "result_hash": "12e5ca70a3fe36dddcddbb12db6c7073060403a522b4fc6cb0efee158756f811", "signature": "26f24596273de01dd0aa3b3db2961b3f1d365dd089ab66fa7c5adccdbf366a29c505b490794aa9671e01f369663fc67a00101983c0b2423cf8e3c688b293b406", "state": "VERIFIED", "summary": {"iw_temperature": 0.129363, "raw": 0.106934, "temperature": 0.127092}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T02:58:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:28Z", "cpu_seconds": 1.6001057210000003, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f0f05cbc1fd0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "ccaac535b468d906774e81636460ad6548de1747199f82201ef649825d669f2d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f0f05cbc1fd0", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:54:21Z", "seed": 1958384736, "treatment": "iw_temperature"}, "priority": 1.9935030782392513, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "41b383dee6bbdbea836b35a2c8fd1d160dafae1fc43bd6ea6a7f3796181bcde6", "run_seconds": 1.6131377220153809, "seed": 1958384736, "started_at": "2026-11-22T02:58:26Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.013299, 0.000486], "effect": -0.006407, "evaluated_at": "2026-11-22T02:58:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f0f05cbc1fd0", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.149000315, 0.1452185, 0.133840835, 0.1266202, 0.103404787, 0.166245078, 0.162357125, 0.07882152, 0.161413957, 0.132020455, 0.16171014, 0.1628659, 0.117137635, 0.15172982, 0.197599015, 0.132108537], "raw": [0.129554435, 0.076130497, 0.11202911, 0.10816512, 0.122073015, 0.10381823, 0.114084385, 0.104219695, 0.145727087, 0.112883705, 0.143105165, 0.146830182, 0.13008808, 0.14297336, 0.176885767, 0.127444485], "temperature": [0.149000315, 0.107631913, 0.118403842, 0.10816512, 0.127914395, 0.154395217, 0.144736022, 0.091209783, 0.161413957, 0.124576178, 0.155490213, 0.146830182, 0.117137635, 0.14297336, 0.197599015, 0.132108537]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0026, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.994", "prereg_hash": "ccaac535b468d906774e81636460ad6548de1747199f82201ef649825d669f2d", "primary_control": "temperature", "result_hash": "41b383dee6bbdbea836b35a2c8fd1d160dafae1fc43bd6ea6a7f3796181bcde6", "signature": "143da4b77e16e059991647b6249348e12f91d74fe79c949226f079bfbdcc1f0a28a21919bdf316dad0838a4ac655d4784e36c3004bbc5aa8e89b259e6aa51b0e", "state": "VERIFIED", "summary": {"iw_temperature": 0.142631, "raw": 0.124751, "temperature": 0.136224}, "ties": 5, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T02:58:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:35Z", "cpu_seconds": 1.754921629000009, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f906d7399216", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "f868e59cc70af1e7c25f10a3fae2a354d9981f8693062b3653fcbf209f5ebbe5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f906d7399216", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:54:21Z", "seed": 553535404, "treatment": "iw_temperature"}, "priority": 1.9912470913562403, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "07d9e2dde332210150ab05772a3808de20f2b0d7a29af3187b88052d7f4a8a93", "run_seconds": 2.052396059036255, "seed": 553535404, "started_at": "2026-11-22T02:58:33Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.013462, 0.007397], "effect": -0.003032, "evaluated_at": "2026-11-22T02:58:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f906d7399216", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.205187407, 0.251531945, 0.186196692, 0.219952988, 0.24450026, 0.254076732, 0.13358087, 0.244637998, 0.218537583, 0.225971187, 0.25139279, 0.217439098, 0.171327887, 0.215573735, 0.232647265, 0.235332753], "raw": [0.179367583, 0.182375135, 0.22613298, 0.19003394, 0.211089508, 0.199579137, 0.147246928, 0.197427363, 0.194192573, 0.19718863, 0.188557827, 0.196138265, 0.171327887, 0.162805922, 0.18807198, 0.189646712], "temperature": [0.192025012, 0.213966685, 0.198879142, 0.21374851, 0.24450026, 0.254076732, 0.17783124, 0.248458818, 0.229431112, 0.211195203, 0.233375737, 0.251891463, 0.157322948, 0.203192913, 0.224500895, 0.20497099]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "f868e59cc70af1e7c25f10a3fae2a354d9981f8693062b3653fcbf209f5ebbe5", "primary_control": "temperature", "result_hash": "07d9e2dde332210150ab05772a3808de20f2b0d7a29af3187b88052d7f4a8a93", "signature": "c4be1a12b587d846c263f28aea32afea98ce0950f09e7ebf250cdff378da32090de113998ec0685de07e1c2f032972359b7edabc328a07a28a5eb1db75d8bd03", "state": "VERIFIED", "summary": {"iw_temperature": 0.219243, "raw": 0.188824, "temperature": 0.21621}, "ties": 2, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T02:58:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:35Z", "cpu_seconds": 7.686907335000001, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-66bd6d79adfb", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "07b24e0c8b6ffb7be92cf9e55a1cc5ec580d5d01e54786bc5b5be66c8d779b29", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-66bd6d79adfb", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:54:21Z", "seed": 1076650606, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "05e862b855787f89d2d6183f46863e3738de29847e262739e7ea7faa60ed5d54", "run_seconds": 8.894245624542236, "seed": 1076650606, "started_at": "2026-11-22T02:54:26Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:54:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-66bd6d79adfb", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.31227374880677183, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "07b24e0c8b6ffb7be92cf9e55a1cc5ec580d5d01e54786bc5b5be66c8d779b29", "primary_control": "reset", "result_hash": "05e862b855787f89d2d6183f46863e3738de29847e262739e7ea7faa60ed5d54", "signature": "5feac48099b2ba235d33b3abca7bf94bc7a58dae6bbd8eeff10a1a4ded3cd374d0e9fd1912a26078d9f629bb2939afb6b2b85b9fda5c4738484648c67a24f505", "state": "PARTIAL", "summary": {"inherit": 0.7333333333333333, "reset": 0.7333333333333333, "root": 0.4166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:54:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:29Z", "cpu_seconds": 7.735941787, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-8151ab0a0f8c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "663281598e24cc694ac0a5c5fa57a8426f00ed6f94df0c3f8c8162de30b4c47d", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-8151ab0a0f8c", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:54:21Z", "seed": 1613235009, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "4d7d4c32a4cb1c90950c1735a199d0757b749c697b558c37bcd446d31430f1df", "run_seconds": 8.250279188156128, "seed": 1613235009, "started_at": "2026-11-22T02:54:21Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.249189, 0.015855], "effect": -0.116667, "evaluated_at": "2026-11-22T02:54:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-8151ab0a0f8c", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0108, "losses": 12, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.975", "prereg_hash": "663281598e24cc694ac0a5c5fa57a8426f00ed6f94df0c3f8c8162de30b4c47d", "primary_control": "reset", "result_hash": "4d7d4c32a4cb1c90950c1735a199d0757b749c697b558c37bcd446d31430f1df", "signature": "99ccd961fd9822691b2eec3373110781792f1f43fd58d973101f7e6a4cd8a4f9f557686a4242037dcc18ff1ec6b382168abe55646f8f5c71683d47f66968220d", "state": "VERIFIED", "summary": {"inherit": 0.5166666666666667, "reset": 0.6333333333333333, "root": 0.25}, "ties": 43, "treatment": "inherit", "wins": 5}, "verified_at": "2026-11-22T02:54:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:32Z", "cpu_seconds": 1.133281223000001, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-1acdd2bba8e0", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "f637dac77628f22d13595c7cc90c7e0c2ec322f5fa25f4c50352f22edffb2d5f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-1acdd2bba8e0", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:21Z", "seed": 71763421, "treatment": "frontier"}, "priority": 2.002803736063905, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "ff0cd4c95c1524fded77051f9a4e2ab7900a3f60fc694023a3665bfcce4eb48c", "run_seconds": 1.6902003288269043, "seed": 71763421, "started_at": "2026-11-22T02:54:31Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T02:54:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-1acdd2bba8e0", "family": "world_model_frontier", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"curiosity": [1.0, 0.84125], "frontier": [1.0, 0.999242], "random_walk": [1.0, 0.874949]}, "prereg_hash": "f637dac77628f22d13595c7cc90c7e0c2ec322f5fa25f4c50352f22edffb2d5f", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "ff0cd4c95c1524fded77051f9a4e2ab7900a3f60fc694023a3665bfcce4eb48c", "signature": "e15a9601ec56ae5955364c889100b8e74bb84bdfc2921793bf3ec3afe366e7671842d96b08cf10d4a5e00087a9c77c1035362b88a547fd9b6504eb182b1d7a03", "state": "REJECTED", "summary": {"curiosity": 0.84125, "frontier": 0.999242, "random_walk": 0.874949}}, "verified_at": "2026-11-22T02:54:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:26Z", "cpu_seconds": 3.04051325, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-3bb17d270761", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "4756292d03dfedadc6a74db4896d0f7b06b035d01673abf5d8e8a19f36acadfe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-3bb17d270761", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:21Z", "seed": 1913479534, "treatment": "frontier"}, "priority": 2.0074843738104935, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "8d3fe35b4369793cc70a59b2d363051996e1dbbf828df1ea5b89d676b250b1c4", "run_seconds": 3.1268222332000732, "seed": 1913479534, "started_at": "2026-11-22T02:54:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.052061, 0.108965], "effect": 0.080513, "evaluated_at": "2026-11-22T02:54:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-3bb17d270761", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.818421053, 0.823979592, 0.828947368, 0.863207547, 1.0, 0.79047619, 0.811111111, 0.788690476, 0.867424242, 0.854166667, 1.0, 0.858247423], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.839473684, 0.897959184, 0.881578947, 0.903301887, 1.0, 0.85, 0.927777778, 0.9375, 0.912878788, 0.9375, 1.0, 0.945876289]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "4756292d03dfedadc6a74db4896d0f7b06b035d01673abf5d8e8a19f36acadfe", "primary_control": "random_walk", "result_hash": "8d3fe35b4369793cc70a59b2d363051996e1dbbf828df1ea5b89d676b250b1c4", "signature": "26133614802400723bd6cfdf0fac67588c65bc6f996c43a18ac7814a529524a2aa7d56d71c9d358fda1d35a151079c9b4dd30690966788aca6fc111cdcda1009", "state": "VERIFIED", "summary": {"curiosity": 0.858723, "frontier": 1.0, "random_walk": 0.919487}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T02:54:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:52Z", "cpu_seconds": 1.4969458990000106, "created_at": "2026-11-22T02:54:21Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-f5b84e8fff38", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "c199c36d1e8d1336ef00e51c9815a3b472b8f390fe2e110ea81b4b316cc42988", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-f5b84e8fff38", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:21Z", "seed": 1121325023, "treatment": "frontier"}, "priority": 1.998089636605581, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "c06128fb96d7052f0e38bc9d1f1e68908b07b84071aab528a706af17b7363f2a", "run_seconds": 1.649825096130371, "seed": 1121325023, "started_at": "2026-11-22T02:57:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.03231, 0.053439], "effect": 0.042874, "evaluated_at": "2026-11-22T02:57:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-f5b84e8fff38", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.901515152, 0.964788732, 0.983333333, 0.954861111, 0.954225352, 0.941176471, 0.952898551, 0.939285714, 0.939393939, 1.0, 0.911764706, 0.93115942], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.935606061, 0.954225352, 0.976666667, 0.961805556, 0.964788732, 0.963235294, 0.960144928, 0.95, 0.943181818, 1.0, 0.933823529, 0.942028986]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "c199c36d1e8d1336ef00e51c9815a3b472b8f390fe2e110ea81b4b316cc42988", "primary_control": "random_walk", "result_hash": "c06128fb96d7052f0e38bc9d1f1e68908b07b84071aab528a706af17b7363f2a", "signature": "7ac490de189e6ed9d67b03696b949d65a6d88975e1777cc80fab182d0d2670117b5770133005609ee14c0722800ee25fc6c4e63ccb0091d40d4b793a83c05305", "state": "VERIFIED", "summary": {"curiosity": 0.947867, "frontier": 1.0, "random_walk": 0.957126}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:57:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:40Z", "cpu_seconds": 0.20944806500000013, "created_at": "2026-11-22T02:54:37Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-a40f2274e39c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "e2067c2fb9bacd69a0e57c98116da2093a11ffb6a9d3f558ba143c167ac2de89", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-a40f2274e39c", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:37Z", "seed": 1133172755, "treatment": "frontier"}, "priority": 2.0049129543046713, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-e24f6cd1a3bc", "result_hash": "6b747ca67216091deb615acb0492b70eea30ccc0fc932d88764f6c6f7b2ee384", "run_seconds": 0.22944140434265137, "seed": 1133172755, "started_at": "2026-11-22T02:54:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019169, 0.038287], "effect": 0.028728, "evaluated_at": "2026-11-22T02:54:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-a40f2274e39c", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.882352941, 0.954861111, 0.875, 0.845238095, 0.927536232, 0.902173913, 0.83984375, 0.894927536, 0.914285714, 0.882352941, 0.928571429, 0.905797101], "frontier": [0.908088235, 0.96875, 0.90530303, 0.900793651, 0.960144928, 0.90942029, 0.87890625, 0.916666667, 0.917857143, 0.911764706, 0.942857143, 0.902173913], "random_walk": [0.856617647, 0.927083333, 0.878787879, 0.845238095, 0.93115942, 0.913043478, 0.8359375, 0.891304348, 0.903571429, 0.889705882, 0.917857143, 0.887681159]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.4984, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00317", "prereg_hash": "e2067c2fb9bacd69a0e57c98116da2093a11ffb6a9d3f558ba143c167ac2de89", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-e24f6cd1a3bc", "result_hash": "6b747ca67216091deb615acb0492b70eea30ccc0fc932d88764f6c6f7b2ee384", "signature": "98a90354998fe9aa41aa9070235db15cc2071236fe6aea4fba264e9f57e92fa4d44c4303a10a06430821cdd824999dedbc3248d033744682bff4905fe542ca0c", "state": "VERIFIED", "summary": {"curiosity": 0.896078, "frontier": 0.91856, "random_walk": 0.889832}, "ties": 0, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:54:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:54:38Z", "cpu_seconds": 0.4800418509999993, "created_at": "2026-11-22T02:54:37Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-d1e39d0ae411", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "9677fbe21108de1c30343ca3d81da486a726f282e5af19e6ccd57ddabf4e6e17", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-d1e39d0ae411", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:37Z", "seed": 166620614, "treatment": "frontier"}, "priority": 2.007652082817327, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "b281bb6c61eed87afdb64343a5b8906d85e216cfd6da080e743643846610ccad", "run_seconds": 0.5566296577453613, "seed": 166620614, "started_at": "2026-11-22T02:54:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.079833, 0.149362], "effect": 0.114597, "evaluated_at": "2026-11-22T02:54:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-d1e39d0ae411", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.855769231, 0.861111111, 0.930327869, 0.778846154, 0.824074074, 0.84375, 0.891304348, 1.0, 0.845454545, 0.816326531, 0.85], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.822115385, 0.875, 0.901639344, 0.807692308, 0.842592593, 0.870535714, 0.902173913, 1.0, 0.872727273, 0.892857143, 0.8375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "9677fbe21108de1c30343ca3d81da486a726f282e5af19e6ccd57ddabf4e6e17", "primary_control": "random_walk", "result_hash": "b281bb6c61eed87afdb64343a5b8906d85e216cfd6da080e743643846610ccad", "signature": "e38aaeeffaca011b8ee36d1b1d96c8bdaf99e1d75b19ec1a7eb37736593f007682c759500397b5cab03bd481fb651871fb75492da6c99f43fb541ae0e4c20b05", "state": "VERIFIED", "summary": {"curiosity": 0.874747, "frontier": 1.0, "random_walk": 0.885403}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T02:54:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:07Z", "cpu_seconds": 1.3618277000000063, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-0812fcead8af", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "dca7db62e25880b522f2096de6283e64012fff5c876e992ac5b9c6f1793eb388", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-0812fcead8af", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:54:53Z", "seed": 2101609709, "treatment": "iw_temperature"}, "priority": 1.9698977553960901, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "34f19d9214a854e8619040355e48aec1249c974de4ba6912cbef3f01d6768e36", "run_seconds": 1.3675868511199951, "seed": 2101609709, "started_at": "2026-11-22T02:59:05Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01726, -0.001959], "effect": -0.009609, "evaluated_at": "2026-11-22T02:59:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-0812fcead8af", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.032705575, 0.141284945, 0.032861015, 0.047220495, 0.021928755, 0.165783803, 0.055730242, 0.08014465, 0.03575166, 0.058715112, 0.03462442, 0.122319102, 0.065897195, 0.064262878, 0.069428183, 0.114165197], "raw": [0.023316993, 0.108901125, 0.043258198, 0.037492923, 0.02429866, 0.151301298, 0.039222787, 0.04824183, 0.064487188, 0.076924467, 0.04944194, 0.134189145, 0.039416635, 0.048065935, 0.06276077, 0.064643905], "temperature": [0.026561915, 0.141284945, 0.041157752, 0.031872035, 0.021928755, 0.135482432, 0.047066832, 0.04915851, 0.03575166, 0.05185868, 0.03462442, 0.117837303, 0.08348007, 0.048065935, 0.052533353, 0.070412497]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0014, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.997", "prereg_hash": "dca7db62e25880b522f2096de6283e64012fff5c876e992ac5b9c6f1793eb388", "primary_control": "temperature", "result_hash": "34f19d9214a854e8619040355e48aec1249c974de4ba6912cbef3f01d6768e36", "signature": "3de4876c68e08069d3df53398fc5ba907dab7868692da45a2675beda244a49d5a8988981c4866fc309d235fac7fc43d85438626e21b223d90f50599b30c5e702", "state": "VERIFIED", "summary": {"iw_temperature": 0.071426, "raw": 0.063498, "temperature": 0.061817}, "ties": 4, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T02:59:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:11Z", "cpu_seconds": 1.6285188300000009, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-14e80f988e5f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "3ff5d2287b803cd4b2c864c81226498ff3dbe0a79a789e461593fb0dc5be88fa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-14e80f988e5f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:54:53Z", "seed": 1005242428, "treatment": "iw_temperature"}, "priority": 1.9683494297762807, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a54b2f09c0758ae75001253bc62bcc70205364b1fb64b03b0b72edba1ad08963", "run_seconds": 1.637134075164795, "seed": 1005242428, "started_at": "2026-11-22T02:59:09Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015158, 0.010455], "effect": -0.002351, "evaluated_at": "2026-11-22T02:59:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-14e80f988e5f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.119440405, 0.17649574, 0.109064075, 0.128818475, 0.094102605, 0.10615041, 0.115145832, 0.149051097, 0.139298628, 0.145181325, 0.147865403, 0.092363175, 0.05882175, 0.141674645, 0.151721573, 0.084485882], "raw": [0.082045755, 0.111966828, 0.075637768, 0.093592797, 0.110813585, 0.089676567, 0.121547807, 0.119594378, 0.10101356, 0.07719961, 0.1170774, 0.087179845, 0.08856221, 0.060496735, 0.1047859, 0.084485882], "temperature": [0.096171965, 0.17649574, 0.089687005, 0.128818475, 0.094102605, 0.10615041, 0.16616065, 0.149051097, 0.1307762, 0.09438801, 0.123756433, 0.096808217, 0.102361015, 0.127013165, 0.126926465, 0.113390135]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0522, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.887", "prereg_hash": "3ff5d2287b803cd4b2c864c81226498ff3dbe0a79a789e461593fb0dc5be88fa", "primary_control": "temperature", "result_hash": "a54b2f09c0758ae75001253bc62bcc70205364b1fb64b03b0b72edba1ad08963", "signature": "01c3707415bed48fb09f8e1466d7008166053774a732a28d9fa6fd44dbe5372ad79f6e6f0ba94c1095ab8ec33ef05153f489eb07d003f11e6e3362e3d1d0c808", "state": "VERIFIED", "summary": {"iw_temperature": 0.12248, "raw": 0.095355, "temperature": 0.120129}, "ties": 5, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T02:59:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:00Z", "cpu_seconds": 1.493310907999998, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-62281f3a0b07", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "58d5b4e3d1fd18bbb4e898f9501fad502c58ad40bde60074b309bb26642cdaac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-62281f3a0b07", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:54:53Z", "seed": 765506328, "treatment": "iw_temperature"}, "priority": 1.9835656275280087, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-56b78c664c35", "result_hash": "5ae763710ff6c757a34c4db982733b5e35c249b3084193a9ddb889e0c62f3f2d", "run_seconds": 1.5503616333007812, "seed": 765506328, "started_at": "2026-11-22T02:58:59Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.005425, 0.012558], "effect": 0.003567, "evaluated_at": "2026-11-22T02:59:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-62281f3a0b07", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.163285853, 0.103123972, 0.15833617, 0.118512825, 0.17224452, 0.1078411, 0.15069833, 0.093648303, 0.151691147, 0.133934117, 0.15370572, 0.074521715, 0.176212795, 0.10507047, 0.098804662, 0.128151293], "raw": [0.126125552, 0.108044265, 0.118448823, 0.118512825, 0.118973295, 0.126097768, 0.159878007, 0.103805013, 0.158884385, 0.133934117, 0.073801455, 0.126789147, 0.169728022, 0.142004778, 0.101361385, 0.09836017], "temperature": [0.139089082, 0.1355585, 0.139924807, 0.139246488, 0.142895468, 0.126097768, 0.15069833, 0.118017948, 0.145269715, 0.138219267, 0.15370572, 0.1027733, 0.18916902, 0.10507047, 0.092960487, 0.128151293]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.4121, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.387", "prereg_hash": "58d5b4e3d1fd18bbb4e898f9501fad502c58ad40bde60074b309bb26642cdaac", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-56b78c664c35", "result_hash": "5ae763710ff6c757a34c4db982733b5e35c249b3084193a9ddb889e0c62f3f2d", "signature": "e6080206eb3119398243370ce96bb9bd1285d6d598d792e0b430f38e0f979a878349f45ad637a73a70a334a5df92def6dbc228e03d2d6f2bb82e1dbd43ebae04", "state": "VERIFIED", "summary": {"iw_temperature": 0.130611, "raw": 0.124047, "temperature": 0.134178}, "ties": 4, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T02:59:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:14Z", "cpu_seconds": 1.6078271789999992, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c3489ccfae44", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "411fa896427c4ff86af0f86b59f0316f46f5eed4cf47c99f38af2312fb09a7a1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c3489ccfae44", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:54:53Z", "seed": 1934645907, "treatment": "iw_temperature"}, "priority": 1.9968965726052421, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "66c7ec288a4d46e442093d2ce48696bb819f8fdd3a6e2897e49ccd38f4758644", "run_seconds": 1.6577610969543457, "seed": 1934645907, "started_at": "2026-11-22T02:58:12Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.028296, -0.00485], "effect": -0.016573, "evaluated_at": "2026-11-22T02:58:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c3489ccfae44", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.045759617, 0.037115773, 0.056035225, 0.069703038, 0.088834718, 0.031318733, 0.034396858, 0.0668845, 0.03572135, 0.050295965, 0.04694196, 0.061838285, 0.074195192, 0.06381138, 0.077295275, 0.138556607], "raw": [0.045759617, 0.037115773, 0.054753392, 0.047372685, 0.063069455, 0.068319105, 0.034396858, 0.038347745, 0.053142515, 0.052043728, 0.0291445, 0.0618048, 0.047645128, 0.052687652, 0.043159427, 0.066392515], "temperature": [0.021205505, 0.037211078, 0.07447364, 0.049898997, 0.06193013, 0.03562872, 0.029052255, 0.038347745, 0.032100242, 0.05270551, 0.0291445, 0.039636798, 0.060715968, 0.052687652, 0.050973355, 0.047819802]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0046, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "411fa896427c4ff86af0f86b59f0316f46f5eed4cf47c99f38af2312fb09a7a1", "primary_control": "temperature", "result_hash": "66c7ec288a4d46e442093d2ce48696bb819f8fdd3a6e2897e49ccd38f4758644", "signature": "5f5ab0ed8010f0b62c7b3bfa46cb8d2466d392633ac1400722901c8a7859f260198f5c74a19850d12a926c8fc7290522c73822899f22d3f0b6bdabc65a6a0500", "state": "VERIFIED", "summary": {"iw_temperature": 0.061169, "raw": 0.049697, "temperature": 0.044596}, "ties": 0, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T02:58:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:55:20Z", "cpu_seconds": 8.339701591, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-40afd1255429", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6f6ca1bc2ea99cf84b681bc8c1625f188ea913d834fbbf719ca6b0246f538c4f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-40afd1255429", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:54:53Z", "seed": 145527334, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "aeeb0c25504db8a0f1752b68bce1f8e6c3ceee1240a9f6518919c094520446ea", "run_seconds": 9.387705564498901, "seed": 145527334, "started_at": "2026-11-22T02:55:11Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.135366, 0.102032], "effect": -0.016667, "evaluated_at": "2026-11-22T02:55:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-40afd1255429", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "6f6ca1bc2ea99cf84b681bc8c1625f188ea913d834fbbf719ca6b0246f538c4f", "primary_control": "reset", "result_hash": "aeeb0c25504db8a0f1752b68bce1f8e6c3ceee1240a9f6518919c094520446ea", "signature": "1f3cdce4e6f9e39b0757b5bf44cc50afff5587a81d54f262b1466c10987b60e0ebd366d975ba350ed60071c351b1f4e2bbb723659b8cdd5613cd085b1f19c004", "state": "VERIFIED", "summary": {"inherit": 0.5833333333333334, "reset": 0.6, "root": 0.2833333333333333}, "ties": 47, "treatment": "inherit", "wins": 6}, "verified_at": "2026-11-22T02:55:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:55:06Z", "cpu_seconds": 6.400908331, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-7c01f6e43b6c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "3b275a7d861c4c80deb6bbeffdcb020abae2eb47bde3e79fc6794fd1221d80b5", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-7c01f6e43b6c", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:54:53Z", "seed": 1878147595, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "2d98ceb6e482a03ba6958ae13e1eccbeb60e82a13b451f64ce1844b601931721", "run_seconds": 6.63359522819519, "seed": 1878147595, "started_at": "2026-11-22T02:54:59Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:55:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-7c01f6e43b6c", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4258955870846385, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0], "root": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3b275a7d861c4c80deb6bbeffdcb020abae2eb47bde3e79fc6794fd1221d80b5", "primary_control": "reset", "result_hash": "2d98ceb6e482a03ba6958ae13e1eccbeb60e82a13b451f64ce1844b601931721", "signature": "2ac7a7e4b41ae6327982ab919b3a85685e30d440c198a5413d5be57365518e744d6d885a61f8c09c6b1f20999377cdaf38f441f218a599093d227b57f80c620e", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:55:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:56:28Z", "cpu_seconds": 0.19932126400000527, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-5b59b78fa401", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "61cd480781f905db4e4cae6f4e1a6b2ebb32a139274bd7c66104f52f69ccb18b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-5b59b78fa401", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:53Z", "seed": 10769311, "treatment": "frontier"}, "priority": 1.999282321907274, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "be5de3942a06ddf5d1e1b21c3e1f78c58f15f670a8a410134bd83b54ac2489a8", "run_seconds": 0.356365442276001, "seed": 10769311, "started_at": "2026-11-22T02:56:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.03322, 0.08362], "effect": 0.05842, "evaluated_at": "2026-11-22T02:56:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-5b59b78fa401", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.822115385, 0.828703704, 1.0, 0.824074074, 0.836363636, 1.0, 1.0, 0.866935484, 0.818181818, 0.827272727, 0.975, 0.813636364], "frontier": [0.875, 0.884259259, 1.0, 0.898148148, 0.909090909, 1.0, 1.0, 0.907258065, 0.904545455, 0.868181818, 1.0, 0.868181818], "random_walk": [0.764423077, 0.777777778, 1.0, 0.787037037, 0.8, 1.0, 1.0, 0.85483871, 0.840909091, 0.809090909, 0.975, 0.804545455]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "61cd480781f905db4e4cae6f4e1a6b2ebb32a139274bd7c66104f52f69ccb18b", "primary_control": "random_walk", "result_hash": "be5de3942a06ddf5d1e1b21c3e1f78c58f15f670a8a410134bd83b54ac2489a8", "signature": "d599a3acefc528b5495d0a409dd5838e56deb2a3bf78a0d0a32c61815fc3da3f1063d1c4fbb63388da4ba6231282d4b430d7013a6135e8cc55ff88e792bc8f0c", "state": "VERIFIED", "summary": {"curiosity": 0.884357, "frontier": 0.926222, "random_walk": 0.867802}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T02:56:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:55:27Z", "cpu_seconds": 1.387975941999997, "created_at": "2026-11-22T02:54:53Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-933d79bd1d90", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "11dd6e756d7742d2e3b989a6381c9389ed0e5f9207c95a323d4500e7a875c96c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-933d79bd1d90", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:54:53Z", "seed": 2084539509, "treatment": "frontier"}, "priority": 2.001133450286401, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "9d5eea6f3d64dd84e7e92c5500b4df0c8cf355471686af3e4ce2d491721310c5", "run_seconds": 1.6194052696228027, "seed": 2084539509, "started_at": "2026-11-22T02:55:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.060792, 0.111078], "effect": 0.085935, "evaluated_at": "2026-11-22T02:55:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-933d79bd1d90", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.976190476, 1.0, 0.823170732, 0.921568627, 0.908653846, 0.8, 0.842592593, 0.885245902, 0.845588235, 0.91, 0.846153846, 0.830508475], "frontier": [0.988095238, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.93452381, 1.0, 0.847560976, 0.931372549, 0.908653846, 0.859090909, 0.921296296, 0.93852459, 0.933823529, 0.85, 0.929487179, 0.902542373]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "11dd6e756d7742d2e3b989a6381c9389ed0e5f9207c95a323d4500e7a875c96c", "primary_control": "random_walk", "result_hash": "9d5eea6f3d64dd84e7e92c5500b4df0c8cf355471686af3e4ce2d491721310c5", "signature": "57035b37507bb38cf0d8feffbb35598621aeae66ceb2de6f610ffd651f4ab92d2de101e5343c20ed3eefda7899158dd9b0e9da52acc762d9f6ed7881cac2b00a", "state": "VERIFIED", "summary": {"curiosity": 0.882473, "frontier": 0.999008, "random_walk": 0.913073}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:55:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:55:39Z", "cpu_seconds": 6.491344461000004, "created_at": "2026-11-22T02:55:26Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1e90752200e5", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "1d1a94e176aca5861e0f4f978257c157eacba7a3a818d6c6090468ebf102b3ee", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1e90752200e5", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:55:26Z", "seed": 447561311, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "e877f8357d1eacbb33159b4f3c27d7458af7deee7e4e9e474b919d03e1f3cb6a", "run_seconds": 6.601968288421631, "seed": 447561311, "started_at": "2026-11-22T02:55:32Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:55:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1e90752200e5", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1d1a94e176aca5861e0f4f978257c157eacba7a3a818d6c6090468ebf102b3ee", "primary_control": "reset", "result_hash": "e877f8357d1eacbb33159b4f3c27d7458af7deee7e4e9e474b919d03e1f3cb6a", "signature": "dd5f905233623b0894bc295d3df30213d98b1da2ed2b40aee0dc695f8debab95a0c3a366d84cfdd8defe44fac2b531462ae2918d6844481e7f3eafbd90d9880d", "state": "PARTIAL", "summary": {"inherit": 0.6, "reset": 0.6, "root": 0.4666666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:55:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:55:57Z", "cpu_seconds": 0.8654750829999998, "created_at": "2026-11-22T02:55:26Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-d9b1a50449c7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "6629dc71ead226a777ba12ae8956f1e7931c137b17ffbdada90c5f6ba8e29652", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-d9b1a50449c7", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:55:26Z", "seed": 1384936185, "treatment": "frontier"}, "priority": 2.001135467568131, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "848797df366e6639623dfadfb87fadfdaa93efbe9bbf96f648d94a05e6c441fa", "run_seconds": 1.5238335132598877, "seed": 1384936185, "started_at": "2026-11-22T02:55:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.054679, 0.163237], "effect": 0.108958, "evaluated_at": "2026-11-22T02:56:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-d9b1a50449c7", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.794642857, 1.0, 0.779255319, 1.0, 0.832568807, 0.795673077, 0.7975, 0.789772727, 0.8, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.803571429, 1.0, 0.776595745, 0.977272727, 0.871559633, 0.824519231, 0.8025, 0.852272727, 0.784210526, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "6629dc71ead226a777ba12ae8956f1e7931c137b17ffbdada90c5f6ba8e29652", "primary_control": "random_walk", "result_hash": "848797df366e6639623dfadfb87fadfdaa93efbe9bbf96f648d94a05e6c441fa", "signature": "5e060ec4b5427e8d60a4c8f8e4f149d4d02572600f59d5357700d15b06217a4764177884e9264176e98c18a37714ff88b1d62d2e688f24c80cf29e3d5acd5709", "state": "VERIFIED", "summary": {"curiosity": 0.882451, "frontier": 1.0, "random_walk": 0.891042}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T02:56:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:55:51Z", "cpu_seconds": 6.683707122999998, "created_at": "2026-11-22T02:55:42Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-30b49108db31", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "3a5534f1f85be60b60ebf9e27b2f7cf7622ec2516529118abecb2445ce27f10f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-30b49108db31", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:55:42Z", "seed": 157818447, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "14e2772f98c57ede8227e8cb4e93c4183e1f7873faab38e851b84ea0d9e90ae8", "run_seconds": 7.163694143295288, "seed": 157818447, "started_at": "2026-11-22T02:55:44Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.078036, 0.421964], "effect": 0.25, "evaluated_at": "2026-11-22T02:55:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-30b49108db31", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 3}, "hypothesis_supported": true, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -2.2727, "losses": 8, "n": 60, "outcome": "SUPPORTED", "p_sign_display": "0.00534", "prereg_hash": "3a5534f1f85be60b60ebf9e27b2f7cf7622ec2516529118abecb2445ce27f10f", "primary_control": "reset", "result_hash": "14e2772f98c57ede8227e8cb4e93c4183e1f7873faab38e851b84ea0d9e90ae8", "signature": "413facb925150e5619a94df6defcc63ec29835982f497e2a70fd97c9822c86527f717b5a6d9642d1b261bf86e73a73b052bba1a57d1ad433065393d62e451706", "state": "VERIFIED", "summary": {"inherit": 0.75, "reset": 0.5, "root": 0.35}, "ties": 29, "treatment": "inherit", "wins": 23}, "verified_at": "2026-11-22T02:56:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:56:10Z", "cpu_seconds": 7.185594318, "created_at": "2026-11-22T02:55:58Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a438fe80ac20", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "4a883bc439ca66a89cf855a6ad9b687514179b752fb2f546d0887906ae388163", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a438fe80ac20", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:55:58Z", "seed": 1158592350, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "14c0ae640c30c96de633f748596144812ff0b1f5b2b6d44340abb44fd111633d", "run_seconds": 7.443701267242432, "seed": 1158592350, "started_at": "2026-11-22T02:56:02Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 4, "reset": 3}, "better": "higher", "ci95": [-0.023806, 0.257139], "effect": 0.116667, "evaluated_at": "2026-11-22T02:56:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a438fe80ac20", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.428985896009936, "rstep": 3}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0], "root": [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -1.0781, "losses": 6, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0835", "prereg_hash": "4a883bc439ca66a89cf855a6ad9b687514179b752fb2f546d0887906ae388163", "primary_control": "reset", "result_hash": "14c0ae640c30c96de633f748596144812ff0b1f5b2b6d44340abb44fd111633d", "signature": "79068fb87886c4506b0f18eb024e125a4526cced2cd0a098629747b086487403f3fbc891018c66e515302036f135a031064093d47f9050488b2ae58ac60f000d", "state": "VERIFIED", "summary": {"inherit": 0.7166666666666667, "reset": 0.6, "root": 0.2833333333333333}, "ties": 41, "treatment": "inherit", "wins": 13}, "verified_at": "2026-11-22T02:56:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:03Z", "cpu_seconds": 0.36491242399999635, "created_at": "2026-11-22T02:55:58Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-b35a5fb050a4", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "2401e402c31d8395dc0543ca344be134e60772e02202aae9fa1dc34b19bfc0e8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-b35a5fb050a4", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:55:58Z", "seed": 783946774, "treatment": "frontier"}, "priority": 1.996904653896162, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "ef168c2ff1b89787c89a6694f7db740bf079de5d735af3f8801178e1c762c15a", "run_seconds": 0.4768812656402588, "seed": 783946774, "started_at": "2026-11-22T02:58:02Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T02:58:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-b35a5fb050a4", "family": "world_model_frontier", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"curiosity": [1.0, 0.901308], "frontier": [1.0, 0.928756], "random_walk": [1.0, 0.902268]}, "prereg_hash": "2401e402c31d8395dc0543ca344be134e60772e02202aae9fa1dc34b19bfc0e8", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "ef168c2ff1b89787c89a6694f7db740bf079de5d735af3f8801178e1c762c15a", "signature": "60572e74f2cbb9483e6dbaa0fbabccc091bb089053d6709087986b8b22f8c4ea770f1734b05f34b4dc097896ec2d699c6da3dd9335f5ae3ce9e1a557e415510c", "state": "REJECTED", "summary": {"curiosity": 0.901308, "frontier": 0.928756, "random_walk": 0.902268}}, "verified_at": "2026-11-22T02:58:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:56:23Z", "cpu_seconds": 7.641831457000002, "created_at": "2026-11-22T02:56:14Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-6396e5e67f56", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "2041831571f0cc78bae1d53964280e9421be53a6b844e76e27e14c741c107435", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-6396e5e67f56", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:56:14Z", "seed": 1591647134, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f241eee953714690582e6206c54f7622dcdb5408c30ebf205433e2a69e44178c", "run_seconds": 7.941754102706909, "seed": 1591647134, "started_at": "2026-11-22T02:56:15Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 4, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:56:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-6396e5e67f56", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "2041831571f0cc78bae1d53964280e9421be53a6b844e76e27e14c741c107435", "primary_control": "reset", "result_hash": "f241eee953714690582e6206c54f7622dcdb5408c30ebf205433e2a69e44178c", "signature": "8e94a932ec06744278ab70b2b5d2753ec3cbf446aab19d480dd840da50a76febfd3901fe5bbfa0211fa355ad25ebaec740b9a67c1dda54b8f7c57a728b345c0b", "state": "PARTIAL", "summary": {"inherit": 0.7666666666666667, "reset": 0.7666666666666667, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:56:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:56:41Z", "cpu_seconds": 7.2110807480000005, "created_at": "2026-11-22T02:56:30Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-29acd67c03f6", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e5af4ecc09f4b1f7d10b95af0b07e77e73dfe99dcab4eac6cde21cc20e571d33", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-29acd67c03f6", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:56:30Z", "seed": 1318766203, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "93264719116bd9819f98f72d0ba6c6cec2351939fa22f65a55f707fa2209d69b", "run_seconds": 7.451743841171265, "seed": 1318766203, "started_at": "2026-11-22T02:56:33Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 1, "reset": 2}, "better": "higher", "ci95": [-0.130317, -0.003016], "effect": -0.066667, "evaluated_at": "2026-11-22T02:56:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-29acd67c03f6", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e5af4ecc09f4b1f7d10b95af0b07e77e73dfe99dcab4eac6cde21cc20e571d33", "primary_control": "reset", "result_hash": "93264719116bd9819f98f72d0ba6c6cec2351939fa22f65a55f707fa2209d69b", "signature": "42d87c0732c1202084215dc5669370be9dd7dd5725c9a6b7376406f168cdd02167c0b8530e5d3307c7215e7d610b50cd67bd7f235a62610acbdf7e8ccd9ec507", "state": "VERIFIED", "summary": {"inherit": 0.5666666666666667, "reset": 0.6333333333333333, "root": 0.4666666666666667}, "ties": 56, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:56:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:56:46Z", "cpu_seconds": 0.3607103239999958, "created_at": "2026-11-22T02:56:30Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-d9d5d5ce1fc3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "7111903ce22b46508bc6e2bbee0625bc364e3b142639589933a785a138e9d100", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-d9d5d5ce1fc3", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:56:30Z", "seed": 688626529, "treatment": "frontier"}, "priority": 2.007253913531651, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "58d7d543e6f79c1140fc45c74a77bc9e9dcf127aefaddce851f0382af08f1cc0", "run_seconds": 0.36215901374816895, "seed": 688626529, "started_at": "2026-11-22T02:56:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.016043, 0.027557], "effect": 0.0218, "evaluated_at": "2026-11-22T02:56:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-d9d5d5ce1fc3", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.905511811, 0.91015625, 0.899193548, 0.911290323, 0.896, 0.948863636, 0.890495868, 0.914, 0.923076923, 0.909448819, 0.894628099, 0.88125], "frontier": [0.921259843, 0.9296875, 0.921370968, 0.941532258, 0.93, 0.948863636, 0.909090909, 0.928, 0.944230769, 0.933070866, 0.923553719, 0.9125], "random_walk": [0.897637795, 0.904296875, 0.907258065, 0.909274194, 0.902, 0.946969697, 0.876033058, 0.916, 0.934615385, 0.907480315, 0.900826446, 0.879166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "7111903ce22b46508bc6e2bbee0625bc364e3b142639589933a785a138e9d100", "primary_control": "random_walk", "result_hash": "58d7d543e6f79c1140fc45c74a77bc9e9dcf127aefaddce851f0382af08f1cc0", "signature": "7257a77121cc887105389d4953ba73987cbbd29e81dc3a97b8c38fba6d34486f5f72acb69d6a4e6f34270d83dc20d2597fac79c398a7b19b61f397804680ae0f", "state": "VERIFIED", "summary": {"curiosity": 0.906993, "frontier": 0.928597, "random_walk": 0.906797}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T02:56:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:56:58Z", "cpu_seconds": 6.664384783999992, "created_at": "2026-11-22T02:56:46Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3c47fbc8f907", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "4e3328ff199f541d13ebcc9672e411d62fc0c32bc5221e413dd83e65bf008592", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3c47fbc8f907", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:56:46Z", "seed": 1363656841, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "34fbd4f831e8705a042dd8863f38b1c0a47754806f768fba00d6210f92c8111e", "run_seconds": 6.895127534866333, "seed": 1363656841, "started_at": "2026-11-22T02:56:51Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.045096, 0.245096], "effect": 0.1, "evaluated_at": "2026-11-22T02:56:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3c47fbc8f907", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.8808, "losses": 7, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.132", "prereg_hash": "4e3328ff199f541d13ebcc9672e411d62fc0c32bc5221e413dd83e65bf008592", "primary_control": "reset", "result_hash": "34fbd4f831e8705a042dd8863f38b1c0a47754806f768fba00d6210f92c8111e", "signature": "5b57d784bf0e90f17d52959b3b7e7c598fd98d364240b15d3f5ebb826bb801131b405f5626747f0acb58955d204bb7f7a52b7e136035b4f7efd016e07614c60f", "state": "VERIFIED", "summary": {"inherit": 0.8, "reset": 0.7, "root": 0.4166666666666667}, "ties": 40, "treatment": "inherit", "wins": 13}, "verified_at": "2026-11-22T02:57:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:04Z", "cpu_seconds": 0.881208756000003, "created_at": "2026-11-22T02:56:46Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-6b9c3d1b6c31", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "9c9e5d1e3a3d8395e84ba7672c4c1f7fbec715942dd47f37178b455255f94ae7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-6b9c3d1b6c31", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:56:46Z", "seed": 1349108667, "treatment": "frontier"}, "priority": 1.9770190031773744, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "c61cc874c846c7ae45554b68c271b2aa8f947301fa1c32be4a0926aeb83c1cac", "run_seconds": 0.8975770473480225, "seed": 1349108667, "started_at": "2026-11-22T02:59:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.099366, 0.187254], "effect": 0.14331, "evaluated_at": "2026-11-22T02:59:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-6b9c3d1b6c31", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.809895833, 0.798295455, 0.826315789, 0.798969072, 1.0, 0.803921569, 0.77688172, 0.827777778, 0.8, 0.829207921, 1.0, 0.842857143], "frontier": [1.0, 1.0, 0.981578947, 1.0, 1.0, 0.992647059, 1.0, 1.0, 1.0, 1.0, 1.0, 0.997619048], "random_walk": [0.776041667, 0.778409091, 0.894736842, 0.865979381, 1.0, 0.840686275, 0.819892473, 0.786111111, 0.85, 0.806930693, 1.0, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "9c9e5d1e3a3d8395e84ba7672c4c1f7fbec715942dd47f37178b455255f94ae7", "primary_control": "random_walk", "result_hash": "c61cc874c846c7ae45554b68c271b2aa8f947301fa1c32be4a0926aeb83c1cac", "signature": "717d2ef3a9ce2212417aaf0b3f502bcd480d28e1d4ab9aaf5dbb50884b86ba6e4e848db554bb6986e25906c843fd351331bada2cc4767a8219cb88c22279dd01", "state": "VERIFIED", "summary": {"curiosity": 0.842844, "frontier": 0.997654, "random_walk": 0.854343}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T02:59:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:09Z", "cpu_seconds": 6.183649322999997, "created_at": "2026-11-22T02:57:03Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-34566033e15d", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f2897ed7513d1944607e922470890f9a57c2567d643169cbb57ca03ce814d568", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-34566033e15d", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:57:03Z", "seed": 496266991, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "b67d82044dcfed2a0495b667200375b8ce3fbac4983fd120479636df8e312eed", "run_seconds": 6.276302814483643, "seed": 496266991, "started_at": "2026-11-22T02:57:03Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.045332, 0.278666], "effect": 0.116667, "evaluated_at": "2026-11-22T02:57:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-34566033e15d", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5683809493410882, "rstep": 3}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.9402, "losses": 9, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.115", "prereg_hash": "f2897ed7513d1944607e922470890f9a57c2567d643169cbb57ca03ce814d568", "primary_control": "reset", "result_hash": "b67d82044dcfed2a0495b667200375b8ce3fbac4983fd120479636df8e312eed", "signature": "ebc0504ff527cf328a50fc4836acf13578f84112f7a9dcf582e907d0b4daa6eeea05da5dfc044b8050819b6b428be658272aa7ef3ff8eca0c05ca210f318d707", "state": "VERIFIED", "summary": {"inherit": 0.7666666666666667, "reset": 0.65, "root": 0.38333333333333336}, "ties": 35, "treatment": "inherit", "wins": 16}, "verified_at": "2026-11-22T02:57:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:45Z", "cpu_seconds": 7.5412721670000025, "created_at": "2026-11-22T02:57:19Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3462d15ce719", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f07ad606fa47f115012e348cc1d0159af852cfbf8310d54b9f5e334895da64ac", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3462d15ce719", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:57:19Z", "seed": 1680865675, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "7c084aa21e1f4457d15218cf20423a4c75c89baf588f62db8b34efdb465f747a", "run_seconds": 7.627123117446899, "seed": 1680865675, "started_at": "2026-11-22T02:57:38Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.016, 0.049333], "effect": 0.016667, "evaluated_at": "2026-11-22T02:57:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3462d15ce719", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.36802521548565303, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0], "root": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f07ad606fa47f115012e348cc1d0159af852cfbf8310d54b9f5e334895da64ac", "primary_control": "reset", "result_hash": "7c084aa21e1f4457d15218cf20423a4c75c89baf588f62db8b34efdb465f747a", "signature": "9e71162b4cc924443428cc20b7d5b3de71a34ec343c74807626e5c7c7fb7da9ee55d4ff742068f26edc99f116b88fc7f00f3cfbbe478ba06c139a4bcdea0ab0d", "state": "VERIFIED", "summary": {"inherit": 0.5333333333333333, "reset": 0.5166666666666667, "root": 0.3333333333333333}, "ties": 59, "treatment": "inherit", "wins": 1}, "verified_at": "2026-11-22T02:57:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:33Z", "cpu_seconds": 8.417294225999996, "created_at": "2026-11-22T02:57:19Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-fa7264277822", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "39fdb5e99aa2552d186a8e73a7331799acb60266ee1a300b4acea612b39251b2", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-fa7264277822", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:57:19Z", "seed": 617253856, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "479fa92c5b4a9b71a9bf2bf0a413fb3de7caba4493c017cdcfe64a587ced2814", "run_seconds": 8.445898056030273, "seed": 617253856, "started_at": "2026-11-22T02:57:24Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:57:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-fa7264277822", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "39fdb5e99aa2552d186a8e73a7331799acb60266ee1a300b4acea612b39251b2", "primary_control": "reset", "result_hash": "479fa92c5b4a9b71a9bf2bf0a413fb3de7caba4493c017cdcfe64a587ced2814", "signature": "cab1e846e76acfa959c68348bebade6996870c5588c5eec5467c18dc27da21bd92baac98b8afa1d07e6c2889b4c339944ba146cb115dcc44d803980bb7692107", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:57:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:51Z", "cpu_seconds": 1.426142904, "created_at": "2026-11-22T02:57:19Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-57d0209f9fca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "adc692881546fce3520a2ae9c306af437166a92aef17e8e94ba5b2fa5970cc00", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-57d0209f9fca", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:57:19Z", "seed": 1027764802, "treatment": "frontier"}, "priority": 2.002375511078746, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "8f69ac50cdc061ce91887e10ba3dca9b47fbf99ee210c9890c68dcce14a66ec8", "run_seconds": 1.4948158264160156, "seed": 1027764802, "started_at": "2026-11-22T02:57:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.028799, 0.074421], "effect": 0.05161, "evaluated_at": "2026-11-22T02:57:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-57d0209f9fca", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.940140845, 0.840163934, 0.917910448, 0.926470588, 0.965753425, 0.955882353, 0.94140625, 0.97265625, 0.916666667, 0.967857143, 0.972222222, 0.958955224], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.875, 1.0, 1.0, 1.0], "random_walk": [0.929577465, 0.905737705, 0.917910448, 0.908088235, 0.97260274, 0.955882353, 0.953125, 0.9609375, 0.931818182, 0.928571429, 0.954861111, 0.936567164]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4984, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00317", "prereg_hash": "adc692881546fce3520a2ae9c306af437166a92aef17e8e94ba5b2fa5970cc00", "primary_control": "random_walk", "result_hash": "8f69ac50cdc061ce91887e10ba3dca9b47fbf99ee210c9890c68dcce14a66ec8", "signature": "20a395d37b29788133bb0f94f7a0d544c00fd4131586cf311e3ea77de496910d15a2031bcc3123673b5ad0cf802b809ba3a8bcb2464211e13661e5c130ef4808", "state": "VERIFIED", "summary": {"curiosity": 0.939674, "frontier": 0.989583, "random_walk": 0.937973}, "ties": 0, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T02:57:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:18Z", "cpu_seconds": 1.7138147270000026, "created_at": "2026-11-22T02:57:51Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5cf45a9608bf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "259697111114349736602a00730c8d2db54f3033a34b1f7a2784518a854e344a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5cf45a9608bf", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:57:51Z", "seed": 1401285192, "treatment": "iw_temperature"}, "priority": 0.7959669079825877, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "d1aefafdca768d8f8b38ffd57cd3e9a713c22a02451b234c0c05945a10dca0b3", "run_seconds": 1.721797227859497, "seed": 1401285192, "started_at": "2026-11-22T02:59:16Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015927, 0.005738], "effect": -0.005094, "evaluated_at": "2026-11-22T02:59:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5cf45a9608bf", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.317519855, 0.312357767, 0.298221197, 0.2769983, 0.227217955, 0.275011258, 0.314842133, 0.292309902, 0.365301277, 0.231305665, 0.280674582, 0.219848655, 0.2164962, 0.23551614, 0.272647808, 0.256582525], "raw": [0.284359085, 0.308032873, 0.256022568, 0.256985525, 0.202474427, 0.242539745, 0.272540603, 0.277968228, 0.307492328, 0.261138182, 0.2739544, 0.20458821, 0.214378725, 0.223237812, 0.206582175, 0.175919818], "temperature": [0.317519855, 0.31607972, 0.288713795, 0.223685192, 0.21864009, 0.275011258, 0.304106897, 0.289180095, 0.354366105, 0.261138182, 0.297198167, 0.219848655, 0.2164962, 0.267661995, 0.23737849, 0.22431877]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "259697111114349736602a00730c8d2db54f3033a34b1f7a2784518a854e344a", "primary_control": "temperature", "result_hash": "d1aefafdca768d8f8b38ffd57cd3e9a713c22a02451b234c0c05945a10dca0b3", "signature": "63e7fffc74f10416d24bd4e7b95e673ed5b5b9a7d9e2a054243a8b96e49293f8618440c89ad16224ec7cafc0e1af758d4cd33a433d5308263fb946985c99a50a", "state": "VERIFIED", "summary": {"iw_temperature": 0.274553, "raw": 0.248013, "temperature": 0.269459}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T02:59:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:14Z", "cpu_seconds": 1.4336445939999862, "created_at": "2026-11-22T02:57:51Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-6049517bb695", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "c8233334e257d098d1124fbf38da43e6e8f789d7f861a236cf5b463bc1097f57", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-6049517bb695", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:57:51Z", "seed": 184223245, "treatment": "iw_temperature"}, "priority": 0.5029820109939623, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-77c6ee597499", "result_hash": "d90b554d8492bbbd572a5f042b7e3c4b8313ce31a63a42e722c23e92ea924450", "run_seconds": 1.7584469318389893, "seed": 184223245, "started_at": "2026-11-22T03:00:12Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01681, 0.018825], "effect": 0.001008, "evaluated_at": "2026-11-22T03:00:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-6049517bb695", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.25510996, 0.295913298, 0.206441723, 0.23873469, 0.300725005, 0.105921218, 0.227659688, 0.21369633, 0.272378863, 0.160410153, 0.251298315, 0.210511017, 0.196902385, 0.270867053, 0.33219792, 0.238203535], "raw": [0.28756063, 0.30221076, 0.206441723, 0.228835493, 0.273716465, 0.174927, 0.200796402, 0.21369633, 0.248781937, 0.160410153, 0.228396908, 0.245360955, 0.208839995, 0.26384604, 0.273529545, 0.18477475], "temperature": [0.295385395, 0.295913298, 0.176760082, 0.20957766, 0.300725005, 0.195823673, 0.239095585, 0.219873477, 0.205484747, 0.17667686, 0.251298315, 0.233624607, 0.2214055, 0.264672458, 0.2879552, 0.218821318]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "c8233334e257d098d1124fbf38da43e6e8f789d7f861a236cf5b463bc1097f57", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-77c6ee597499", "result_hash": "d90b554d8492bbbd572a5f042b7e3c4b8313ce31a63a42e722c23e92ea924450", "signature": "11155d2bad8f2fd424be8ded2c49c3679fd3213fcea96c3c4063389efef69cbb340706b7c5c51b8c357f38cb028f3ae6aded047ae8901f033282efc7b9bad701", "state": "VERIFIED", "summary": {"iw_temperature": 0.236061, "raw": 0.231383, "temperature": 0.237068}, "ties": 3, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:00:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:57:59Z", "cpu_seconds": 6.540566087, "created_at": "2026-11-22T02:57:51Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-2aebdfa17174", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6122302f9d9e2458461149351cb3a143768cc11afdf3efc77f08e5fecd04bf88", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-2aebdfa17174", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:57:51Z", "seed": 1670171133, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a746b9f2bdc05a91867a99b60b6b04a46de1526928f63a29d471f8f1d3786651", "run_seconds": 6.606740951538086, "seed": 1670171133, "started_at": "2026-11-22T02:57:52Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.068088, 0.168088], "effect": 0.05, "evaluated_at": "2026-11-22T02:58:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-2aebdfa17174", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6112619746388116, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.5368, "losses": 5, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.291", "prereg_hash": "6122302f9d9e2458461149351cb3a143768cc11afdf3efc77f08e5fecd04bf88", "primary_control": "reset", "result_hash": "a746b9f2bdc05a91867a99b60b6b04a46de1526928f63a29d471f8f1d3786651", "signature": "aee19058c26f282689b40de0f487d82ba96df8659ec85ae6a8ef3f4e20c5fcbe83bc831e71b55842bfb062ecd27bf3efa78f42115472af1350e8409361d5a60c", "state": "VERIFIED", "summary": {"inherit": 0.9, "reset": 0.85, "root": 0.43333333333333335}, "ties": 47, "treatment": "inherit", "wins": 8}, "verified_at": "2026-11-22T02:58:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:25Z", "cpu_seconds": 1.7754655199999831, "created_at": "2026-11-22T02:58:07Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-8d6fed1767f7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "44832d75db9677355bc2e8e773a2b00dc6ab63c5c4a1c5aa269de3c4897fc78e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-8d6fed1767f7", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:07Z", "seed": 1335115820, "treatment": "iw_temperature"}, "priority": 0.5243083472482308, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a588aa99792e6c7c8071f3787ee491ae4b8ab3838e59d4e5bae07f81d93f6fb1", "run_seconds": 1.9802014827728271, "seed": 1335115820, "started_at": "2026-11-22T02:59:23Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.013586, 0.001474], "effect": -0.006056, "evaluated_at": "2026-11-22T02:59:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-8d6fed1767f7", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.089106797, 0.030818145, 0.01639025, 0.07121155, 0.03593763, 0.014131748, 0.08911249, 0.00735234, 0.01046064, 0.0731098, 0.00852147, 0.013833097, 0.040816673, 0.028718987, 0.08279331, 0.003833057], "raw": [0.042016735, 0.017458938, 0.02876212, 0.038053117, 0.05481724, 0.023834558, 0.112862115, 0.015786998, 0.026306087, 0.048807412, 0.015119657, 0.020671078, 0.057234992, 0.035552485, 0.037900242, 0.015986845], "temperature": [0.052968462, 0.030818145, 0.019588415, 0.043552625, 0.030510395, 0.016117535, 0.112862115, 0.011299947, 0.011491263, 0.044635247, 0.009221722, 0.013833097, 0.035644147, 0.022929898, 0.059099088, 0.004680872]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2184, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.605", "prereg_hash": "44832d75db9677355bc2e8e773a2b00dc6ab63c5c4a1c5aa269de3c4897fc78e", "primary_control": "temperature", "result_hash": "a588aa99792e6c7c8071f3787ee491ae4b8ab3838e59d4e5bae07f81d93f6fb1", "signature": "11d0a4c1698fd3c0cfd429deaa208483854f65904904663e3a5aaf9c77e576fec50761a729cd648f5ed7f51866f645b2976dbdc9f4cd836386fbee452139da02", "state": "VERIFIED", "summary": {"iw_temperature": 0.038509, "raw": 0.036948, "temperature": 0.032453}, "ties": 2, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T02:59:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:21Z", "cpu_seconds": 1.4933511469999985, "created_at": "2026-11-22T02:58:07Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-a8710b80565b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "541511dac5e91f576ea8380d85e384608dcd2c882c9581fa72d77a916df9d503", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-a8710b80565b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:07Z", "seed": 552518517, "treatment": "iw_temperature"}, "priority": 0.4973266107401937, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "59f77a3901e38260dbee6b0488cc111c49af12d53a798adad80b9e47c135f1c8", "run_seconds": 1.7817728519439697, "seed": 552518517, "started_at": "2026-11-22T03:00:19Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.008667, 0.002093], "effect": -0.003287, "evaluated_at": "2026-11-22T03:00:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-a8710b80565b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.041206158, 0.022238662, 0.05188948, 0.180670138, 0.048115895, 0.038445673, 0.094109715, 0.060838013, 0.014341365, 0.07854271, 0.05823917, 0.054792865, 0.079805905, 0.043333843, 0.010573347, 0.063987787], "raw": [0.024883595, 0.02609925, 0.0453561, 0.18892371, 0.056036858, 0.105895202, 0.066067123, 0.040715, 0.014341365, 0.09037701, 0.060978802, 0.0614934, 0.046480672, 0.022139585, 0.02326127, 0.055697465], "temperature": [0.041206158, 0.027861852, 0.040840897, 0.196977238, 0.039067537, 0.053661025, 0.066067123, 0.049818475, 0.01124291, 0.07854271, 0.05823917, 0.047872397, 0.063487197, 0.039261252, 0.010573347, 0.063812495]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0085, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.981", "prereg_hash": "541511dac5e91f576ea8380d85e384608dcd2c882c9581fa72d77a916df9d503", "primary_control": "temperature", "result_hash": "59f77a3901e38260dbee6b0488cc111c49af12d53a798adad80b9e47c135f1c8", "signature": "7aed90ecb0d78658de051b885f219cbc6b308e8515eb6cfa783ea271b5ca1f461f2909b327a903d327dae71d09a1264ef3a2b357fb524d40626c362c4ec1a207", "state": "VERIFIED", "summary": {"iw_temperature": 0.058821, "raw": 0.058047, "temperature": 0.055533}, "ties": 4, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:00:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:38Z", "cpu_seconds": 1.7104326189999988, "created_at": "2026-11-22T02:58:07Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b493e90e05f5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "1f2b6ab0897beeabb5679352f4eee06995b717663821985d589eb5dc1ba42a0f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b493e90e05f5", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:07Z", "seed": 11762395, "treatment": "iw_temperature"}, "priority": 0.5208387915162609, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "810250516da5a3d1d4d53c8fdf06aad5f38be0a7850a87b89222b88bbfee57a2", "run_seconds": 1.9076204299926758, "seed": 11762395, "started_at": "2026-11-22T02:59:36Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.009194, 0.004134], "effect": -0.00253, "evaluated_at": "2026-11-22T02:59:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b493e90e05f5", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.0421209, 0.060083083, 0.125484415, 0.058393372, 0.053084397, 0.06702845, 0.054524463, 0.031478063, 0.083659032, 0.06259851, 0.082267525, 0.065292652, 0.044404272, 0.041132585, 0.058698582, 0.063515093], "raw": [0.069783175, 0.05153379, 0.100790288, 0.049495648, 0.04327268, 0.063175063, 0.05194639, 0.052783552, 0.096698015, 0.06259851, 0.084900133, 0.065116175, 0.0642963, 0.054972973, 0.050310243, 0.051202227], "temperature": [0.047530563, 0.04835455, 0.087062775, 0.05289161, 0.04816066, 0.06702845, 0.045062695, 0.031478063, 0.083659032, 0.074142733, 0.089978105, 0.065116175, 0.044404272, 0.067063785, 0.050149805, 0.051202227]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "1f2b6ab0897beeabb5679352f4eee06995b717663821985d589eb5dc1ba42a0f", "primary_control": "temperature", "result_hash": "810250516da5a3d1d4d53c8fdf06aad5f38be0a7850a87b89222b88bbfee57a2", "signature": "401e766ed4172b83f385fb8946453531446644cd066ee90b42c373748500ad6c9cc3a99ebab98b6146438281b1b7742d43e905d43e2e87bcbaa1d17b38d15e0c", "state": "VERIFIED", "summary": {"iw_temperature": 0.06211, "raw": 0.063305, "temperature": 0.05958}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T02:59:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:15Z", "cpu_seconds": 5.96265175, "created_at": "2026-11-22T02:58:07Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-2b05637c3b11", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "8c430ee39c8ca53ee8dd0b95c322a1a8dfff9054f48d881bf24691d3015a3def", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-2b05637c3b11", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:58:07Z", "seed": 1821423919, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "e438d0abd360c84014eea47a9a7ebfe40236a67e6f2975accd46420125bb2a45", "run_seconds": 6.051678419113159, "seed": 1821423919, "started_at": "2026-11-22T02:58:09Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:58:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-2b05637c3b11", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.2890517613150465, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8c430ee39c8ca53ee8dd0b95c322a1a8dfff9054f48d881bf24691d3015a3def", "primary_control": "reset", "result_hash": "e438d0abd360c84014eea47a9a7ebfe40236a67e6f2975accd46420125bb2a45", "signature": "562697a580ae65b37ec318fc021076fcd285120c68ed925151d2322e5503f402209a70c3c85e8ddb563d47a14392bbd4a74a688d983ea2c553843eb0e76c6807", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.5166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:58:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:16Z", "cpu_seconds": 7.468228945999996, "created_at": "2026-11-22T02:58:07Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-9778e4560445", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f06182a6b4dcbcb875481682257e3fd889fdb5174a94de92d8cddd8adf92f61a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-9778e4560445", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:58:07Z", "seed": 2053602284, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "e331a923a4f3a8b4a8584804e81f63ba1f96144c2a05d2e68de6b48b71ba1056", "run_seconds": 7.6693103313446045, "seed": 2053602284, "started_at": "2026-11-22T02:58:08Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.016, 0.049333], "effect": 0.016667, "evaluated_at": "2026-11-22T02:58:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-9778e4560445", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.565167569807461, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f06182a6b4dcbcb875481682257e3fd889fdb5174a94de92d8cddd8adf92f61a", "primary_control": "reset", "result_hash": "e331a923a4f3a8b4a8584804e81f63ba1f96144c2a05d2e68de6b48b71ba1056", "signature": "439df64e5d0208a4fc529440412fa2af497529edb022bf6df9accbf8832d89613b3ca42ccaa21338090f947a1564d3bfd157b07021a9f1fc99483a2067300a01", "state": "VERIFIED", "summary": {"inherit": 0.5666666666666667, "reset": 0.55, "root": 0.25}, "ties": 59, "treatment": "inherit", "wins": 1}, "verified_at": "2026-11-22T02:58:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:48Z", "cpu_seconds": 2.730726943999997, "created_at": "2026-11-22T02:58:07Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-7256712810e8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "15e01369cbcb3843f8e62e09feda5eb4ad9a8cbcd0f42f89b91a3aa0336d4905", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-7256712810e8", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:07Z", "seed": 488035906, "treatment": "frontier"}, "priority": 0.464065788088916, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "468c3d662e5837d8a3bab4b8ae57e7c01e76ceb16cce79b420f8daee34189d10", "run_seconds": 2.967940330505371, "seed": 488035906, "started_at": "2026-11-22T03:00:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.042906, 0.126727], "effect": 0.084817, "evaluated_at": "2026-11-22T03:00:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-7256712810e8", "family": "world_model_frontier", "hypothesis_supported": false, "instance_scores": {"curiosity": [1.0, 0.846153846, 0.859756098, 1.0, 0.787234043, 0.841397849, 0.813157895, 0.943877551, 0.830645161, 0.779069767, 0.825268817, 0.801136364], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.827956989, 1.0, 0.989247312, 0.866477273], "random_walk": [1.0, 0.871794872, 0.792682927, 1.0, 0.875, 0.852150538, 0.868421053, 0.915816327, 0.833333333, 0.912790698, 0.857526882, 0.886363636]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2621, "losses": 2, "n": 12, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0547", "prereg_hash": "15e01369cbcb3843f8e62e09feda5eb4ad9a8cbcd0f42f89b91a3aa0336d4905", "primary_control": "random_walk", "result_hash": "468c3d662e5837d8a3bab4b8ae57e7c01e76ceb16cce79b420f8daee34189d10", "signature": "9c10bc39a2068a0f4d4ea7c8391e63e920ae085b560e400d5d2d6993a4d13fe5b3abafe96a4d5950f9a827d4e906fa23f2c6d55e8359d7621957d8758c3e5501", "state": "VERIFIED", "summary": {"curiosity": 0.860641, "frontier": 0.97364, "random_walk": 0.888823}, "ties": 2, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:00:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:24Z", "cpu_seconds": 1.6427348749999986, "created_at": "2026-11-22T02:58:23Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f27ac428ec79", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "2b06addaf347ad11f284d852379d12ae67a7da83c4e267e44671cdc02b78e56f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f27ac428ec79", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:23Z", "seed": 854499326, "treatment": "iw_temperature"}, "priority": 0.5281304625422502, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "646bc954850ccf6ce7442e2a6ce73f45e4fdaa5198dd1f2c94c945dd1b915a3f", "run_seconds": 1.7239353656768799, "seed": 854499326, "started_at": "2026-11-22T02:59:23Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.031908, -0.000501], "effect": -0.016205, "evaluated_at": "2026-11-22T02:59:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f27ac428ec79", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.167052893, 0.057003888, 0.088172623, 0.022916152, 0.051178037, 0.08924686, 0.046519962, 0.032643005, 0.02828514, 0.023743332, 0.042836087, 0.044645052, 0.012507798, 0.03136398, 0.030808355, 0.026050305], "raw": [0.040818345, 0.049035583, 0.104286022, 0.014381757, 0.051178037, 0.0665317, 0.046519962, 0.013273205, 0.032647805, 0.03033076, 0.046609335, 0.031706455, 0.017903322, 0.076467762, 0.04409677, 0.042076337], "temperature": [0.039914805, 0.05087257, 0.041212273, 0.015468348, 0.04584451, 0.06908078, 0.04535016, 0.016716365, 0.029354505, 0.023743332, 0.025277125, 0.036253237, 0.012742222, 0.036534472, 0.02732421, 0.02001105]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0016, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.996", "prereg_hash": "2b06addaf347ad11f284d852379d12ae67a7da83c4e267e44671cdc02b78e56f", "primary_control": "temperature", "result_hash": "646bc954850ccf6ce7442e2a6ce73f45e4fdaa5198dd1f2c94c945dd1b915a3f", "signature": "715d91abae2c802c22e1d7019e63dc37bf71e1dad3070af87e9379b7f48bd361981a1ac796c5cda3223fa114b538ae99c9c9f1ba90bab07e9aafec4097587e09", "state": "VERIFIED", "summary": {"iw_temperature": 0.049686, "raw": 0.044241, "temperature": 0.033481}, "ties": 1, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T02:59:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:33Z", "cpu_seconds": 7.874556953000001, "created_at": "2026-11-22T02:58:23Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-cc7db7372b18", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f98e633e7810e3771674dc9d73c254b341102f8b7656d26af848d232a61eb89b", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-cc7db7372b18", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:58:23Z", "seed": 1138989088, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9b3dbdc7b7ff0332bb09942e7122a821482ed80e4bdba429ef15669ad62576c5", "run_seconds": 8.053025245666504, "seed": 1138989088, "started_at": "2026-11-22T02:58:25Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:58:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-cc7db7372b18", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6473021792433731, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f98e633e7810e3771674dc9d73c254b341102f8b7656d26af848d232a61eb89b", "primary_control": "reset", "result_hash": "9b3dbdc7b7ff0332bb09942e7122a821482ed80e4bdba429ef15669ad62576c5", "signature": "73bfb11cb13e1e1cfb5e5cf1580d156d89515677827c4e5eb88a7ccaad0b93ab0deb2cc3d419be52312657af4127aa6f0de6072dd890568e1c1e04d2887f8905", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:58:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:21Z", "cpu_seconds": 1.6588058340000202, "created_at": "2026-11-22T02:58:23Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-14ee82711183", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "3fb3e5860d9363393e829673e029ada3483c26a009e4312744d25ad45ea2ece0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-14ee82711183", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:23Z", "seed": 513972495, "treatment": "frontier"}, "priority": 0.24976534507228607, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-f5b84e8fff38", "result_hash": "97ad6e34efbe6cb81c0e9a583b7e56eede6679a6c9622bef99af80985149c78c", "run_seconds": 1.788947582244873, "seed": 513972495, "started_at": "2026-11-22T03:06:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.03487, 0.053018], "effect": 0.043944, "evaluated_at": "2026-11-22T03:06:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-14ee82711183", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.931818182, 0.983333333, 0.928571429, 0.962837838, 0.947183099, 0.943181818, 0.964285714, 0.953846154, 0.8828125, 0.967857143, 0.974264706, 0.964788732], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.943181818, 0.97, 0.960714286, 0.97972973, 0.978873239, 0.935606061, 0.957142857, 0.946153846, 0.9375, 0.971428571, 0.952205882, 0.940140845]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "3fb3e5860d9363393e829673e029ada3483c26a009e4312744d25ad45ea2ece0", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-f5b84e8fff38", "result_hash": "97ad6e34efbe6cb81c0e9a583b7e56eede6679a6c9622bef99af80985149c78c", "signature": "8e4e906a1f23a5cc0af792325a86b996796a46cc0e8723aa2f1471c89547daa74ea0d28ac920c232c0500f27c92908e6d58fe277e7101f607e1f2384f7a24d0d", "state": "VERIFIED", "summary": {"curiosity": 0.950398, "frontier": 1.0, "random_walk": 0.956056}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:06:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:52Z", "cpu_seconds": 0.4048312609999982, "created_at": "2026-11-22T02:58:23Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-17939181ac4b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "21ff61ad4f7a73b87d97751f7ba114aafcfe60a2b40cbe5250915637cec7c59b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-17939181ac4b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:23Z", "seed": 324005809, "treatment": "frontier"}, "priority": 0.25378215886390426, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "c9ff1ce34d76709148c4ef385c344b4d912d842f19174bf148aaaeb8e3a0ddfe", "run_seconds": 0.41371631622314453, "seed": 324005809, "started_at": "2026-11-22T03:05:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058972, 0.131945], "effect": 0.095459, "evaluated_at": "2026-11-22T03:06:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-17939181ac4b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.884920635, 0.905172414, 0.847457627, 0.89516129, 0.823863636, 1.0, 0.787037037, 1.0, 0.855932203, 1.0, 0.8875, 0.850961538], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.900793651, 0.853448276, 0.847457627, 0.907258065, 0.835227273, 1.0, 0.824074074, 1.0, 0.88559322, 1.0, 0.920833333, 0.879807692]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "21ff61ad4f7a73b87d97751f7ba114aafcfe60a2b40cbe5250915637cec7c59b", "primary_control": "random_walk", "result_hash": "c9ff1ce34d76709148c4ef385c344b4d912d842f19174bf148aaaeb8e3a0ddfe", "signature": "d815327270c59d0b9a66a50aded39afd06db0e13a1d9e1b6843a343bba7d913eb3caa3c0a6c8a13750ce6dad2c333c12c60955e5018890c3e4ad4771399a2a00", "state": "VERIFIED", "summary": {"curiosity": 0.894834, "frontier": 1.0, "random_walk": 0.904541}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:06:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:12Z", "cpu_seconds": 1.4271875450000024, "created_at": "2026-11-22T02:58:39Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4e762789e23a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "8df0f4cd80f9c2ef1aacab872f8935e64f981e1b7f2eeea2d37edac697a8ff14", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4e762789e23a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:39Z", "seed": 495743436, "treatment": "iw_temperature"}, "priority": 0.5054805417696824, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "78203467c056ad5516009884784c56a86d06c45d9e3800a35833c0aea7102e4a", "run_seconds": 1.5905344486236572, "seed": 495743436, "started_at": "2026-11-22T03:00:11Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018806, 0.009766], "effect": -0.00452, "evaluated_at": "2026-11-22T03:00:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4e762789e23a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.250617235, 0.273955022, 0.300829383, 0.322137357, 0.275480618, 0.309003743, 0.259351773, 0.29212527, 0.164169152, 0.290533558, 0.288156377, 0.124375197, 0.286033607, 0.214165607, 0.26992654, 0.297311143], "raw": [0.220908935, 0.1830784, 0.29957956, 0.25673511, 0.279600172, 0.320411985, 0.24877754, 0.283487253, 0.164169152, 0.28901039, 0.28230017, 0.124375197, 0.212344862, 0.211640812, 0.232254023, 0.270005612], "temperature": [0.250617235, 0.232025358, 0.300829383, 0.29106416, 0.26024982, 0.320411985, 0.259351773, 0.283487253, 0.175389912, 0.2896562, 0.269222382, 0.20040697, 0.244894997, 0.244822765, 0.253413338, 0.270005612]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0205, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.954", "prereg_hash": "8df0f4cd80f9c2ef1aacab872f8935e64f981e1b7f2eeea2d37edac697a8ff14", "primary_control": "temperature", "result_hash": "78203467c056ad5516009884784c56a86d06c45d9e3800a35833c0aea7102e4a", "signature": "f7b7df478548ad35bc27c38ae9ab20e81d771f621ae0667d69242113950f52b5e0a1636bc814a7fec52670441f56bee19e5b9c415ec07583fd00c8864d408d0c", "state": "VERIFIED", "summary": {"iw_temperature": 0.263636, "raw": 0.242417, "temperature": 0.259116}, "ties": 3, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:00:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:00Z", "cpu_seconds": 1.5634427929999788, "created_at": "2026-11-22T02:58:39Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-7537704cf190", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "e70fbc15e40362d58520adf9dcdcbd094bc4b37ecb9c378bedebc24d10860f80", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-7537704cf190", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:39Z", "seed": 511036484, "treatment": "iw_temperature"}, "priority": 0.5076824924340464, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "1817c052242fe529801505d4cb714ff2505b5b0a8b55f3c6539b49685303680d", "run_seconds": 1.5652942657470703, "seed": 511036484, "started_at": "2026-11-22T02:59:58Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020063, 0.007089], "effect": -0.006487, "evaluated_at": "2026-11-22T03:00:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-7537704cf190", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.085469522, 0.131902483, 0.135581618, 0.192942078, 0.103258342, 0.159842285, 0.10839579, 0.161457783, 0.16544778, 0.213461035, 0.082737385, 0.14025089, 0.115658405, 0.15585161, 0.15438349, 0.155587465], "raw": [0.100917637, 0.101835, 0.129792125, 0.164403187, 0.10947847, 0.123959745, 0.10839579, 0.111086332, 0.10042193, 0.163838797, 0.117779815, 0.131551863, 0.083930783, 0.08318293, 0.099603283, 0.128597242], "temperature": [0.100917637, 0.131902483, 0.131063978, 0.18318233, 0.14275276, 0.146262235, 0.105813213, 0.13782954, 0.114324127, 0.163838797, 0.134278768, 0.131551863, 0.115658405, 0.114826692, 0.14863957, 0.155587465]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "e70fbc15e40362d58520adf9dcdcbd094bc4b37ecb9c378bedebc24d10860f80", "primary_control": "temperature", "result_hash": "1817c052242fe529801505d4cb714ff2505b5b0a8b55f3c6539b49685303680d", "signature": "015f1b875af9adafc4fb7961571c7fc5d5471c819d42f91ec293116bc8d610b2e6e8f4162aae7a60b8f078980d4a8446381a07dde62373fa176404fc49585f0f", "state": "VERIFIED", "summary": {"iw_temperature": 0.141389, "raw": 0.116173, "temperature": 0.134902}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:00:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:28Z", "cpu_seconds": 1.856968686999977, "created_at": "2026-11-22T02:58:39Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f10e6fccd035", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "5a3f08da44303b39c22f115da910156eabc0ca0606858d79cc6c371b262095f2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f10e6fccd035", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:39Z", "seed": 1020258910, "treatment": "iw_temperature"}, "priority": 0.4890883726403459, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a0faed7904161a2d97853f8d5b58dd552db37290e340abc18032d9af14439d22", "run_seconds": 2.041590929031372, "seed": 1020258910, "started_at": "2026-11-22T03:00:26Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011036, 0.000588], "effect": -0.005224, "evaluated_at": "2026-11-22T03:00:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f10e6fccd035", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.0250647, 0.04044694, 0.06108349, 0.059809415, 0.021914383, 0.0773857, 0.039608295, 0.036714645, 0.021232963, 0.070811845, 0.034592962, 0.030098255, 0.041845968, 0.050161425, 0.033293932, 0.039352135], "raw": [0.056778105, 0.04584644, 0.077199917, 0.050317948, 0.035816505, 0.049568822, 0.078482657, 0.050507437, 0.02979837, 0.06233486, 0.042351707, 0.044035925, 0.041845968, 0.08178944, 0.049721782, 0.04156522], "temperature": [0.0250647, 0.03667171, 0.031880532, 0.049177525, 0.025771152, 0.04368077, 0.045914125, 0.036714645, 0.02979837, 0.05845805, 0.034592962, 0.02430014, 0.042918145, 0.039656802, 0.033663993, 0.04156522]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "5a3f08da44303b39c22f115da910156eabc0ca0606858d79cc6c371b262095f2", "primary_control": "temperature", "result_hash": "a0faed7904161a2d97853f8d5b58dd552db37290e340abc18032d9af14439d22", "signature": "e08cdb4cde3cb4d141eab9835dea89e08e362144373c34346f3f9323c7bb827ed5f3684b720b06523bee1423421f85e662442e633db6fee645c0ced02740d60f", "state": "VERIFIED", "summary": {"iw_temperature": 0.042714, "raw": 0.052373, "temperature": 0.037489}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:00:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:58:47Z", "cpu_seconds": 7.819376552000001, "created_at": "2026-11-22T02:58:39Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-e9a2902760bd", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6d0eeb992b9f5cc115a1b96fb3823981f00331ce5ad30d2df69395653b18f5f0", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-e9a2902760bd", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:58:39Z", "seed": 273525517, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "cc93c67274613371c89f3bf81f7702814a5328cd0fa6c43a382a496f066dfb9d", "run_seconds": 7.919478893280029, "seed": 273525517, "started_at": "2026-11-22T02:58:39Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:58:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-e9a2902760bd", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4926865775845492, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6d0eeb992b9f5cc115a1b96fb3823981f00331ce5ad30d2df69395653b18f5f0", "primary_control": "reset", "result_hash": "cc93c67274613371c89f3bf81f7702814a5328cd0fa6c43a382a496f066dfb9d", "signature": "ea21f0e0f52a863b62c60a09777ad35dc971ec3653eacbe89a3eb7350711ac746ee017994ca0aa6031f77dae62e6337c973608d9140397b9cc36dc727a4c4109", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:58:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:11Z", "cpu_seconds": 0.1708177790000036, "created_at": "2026-11-22T02:58:39Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-c333f1bf81d4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "59c5ceedd4d4c977777fb095521de716028924a2f3bc3e02d7efde2a10657534", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-c333f1bf81d4", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:39Z", "seed": 861433568, "treatment": "frontier"}, "priority": 0.2620108560849221, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "539af08a8f33710a76e996b17335f28a31ef72bc68161a83e103d5e5b32f606f", "run_seconds": 0.17101645469665527, "seed": 861433568, "started_at": "2026-11-22T03:04:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.042372, 0.102784], "effect": 0.072578, "evaluated_at": "2026-11-22T03:04:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-c333f1bf81d4", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.916666667, 0.829787234, 0.778409091, 0.790178571, 0.909836066, 1.0, 0.8125, 0.773148148, 1.0, 0.878787879, 0.836206897], "frontier": [1.0, 1.0, 0.920212766, 0.931818182, 0.888392857, 0.885245902, 1.0, 0.864583333, 0.888888889, 1.0, 0.897727273, 0.900862069], "random_walk": [1.0, 0.851851852, 0.819148936, 0.761363636, 0.790178571, 0.856557377, 1.0, 0.807291667, 0.796296296, 0.928571429, 0.863636364, 0.831896552]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "59c5ceedd4d4c977777fb095521de716028924a2f3bc3e02d7efde2a10657534", "primary_control": "random_walk", "result_hash": "539af08a8f33710a76e996b17335f28a31ef72bc68161a83e103d5e5b32f606f", "signature": "5a848d405b6624bd1374ac1944b92abb73843980648e6c185fc6c4d1f753c2e7fe65f1fdf1681bc325bba029a1f169830f5899a5ad54eacce2d082facc6f530a", "state": "VERIFIED", "summary": {"curiosity": 0.877127, "frontier": 0.931478, "random_walk": 0.858899}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:04:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:33Z", "cpu_seconds": 1.5449478220000046, "created_at": "2026-11-22T02:58:55Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-344519108278", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "568e51f2a0535fa38e1111fe3176cc1088e705d0dae3df2475a9a1c7d5e7dd81", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-344519108278", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:55Z", "seed": 452846828, "treatment": "iw_temperature"}, "priority": 0.481997707930124, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "6bedb43a2291b645e16e6870984c0a6ba0ed4f1d71507e06c0f3da2f0749705f", "run_seconds": 1.5807313919067383, "seed": 452846828, "started_at": "2026-11-22T03:00:32Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.02261, 0.001229], "effect": -0.010691, "evaluated_at": "2026-11-22T03:00:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-344519108278", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.098788953, 0.168956402, 0.14227673, 0.069587265, 0.135793528, 0.13872905, 0.138773085, 0.198831732, 0.147989712, 0.117547713, 0.26463502, 0.167644165, 0.077654095, 0.113659338, 0.17525735, 0.124582742], "raw": [0.108808575, 0.11024717, 0.122263268, 0.069587265, 0.11123916, 0.10595286, 0.091623053, 0.116746375, 0.100787732, 0.117547713, 0.16835066, 0.109985237, 0.061713868, 0.105227585, 0.13555449, 0.114926557], "temperature": [0.1135315, 0.13229718, 0.150207767, 0.091925447, 0.123503045, 0.125143472, 0.121293832, 0.134305115, 0.142407452, 0.134084623, 0.2174293, 0.167644165, 0.077654095, 0.113659338, 0.139986748, 0.124582742]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "568e51f2a0535fa38e1111fe3176cc1088e705d0dae3df2475a9a1c7d5e7dd81", "primary_control": "temperature", "result_hash": "6bedb43a2291b645e16e6870984c0a6ba0ed4f1d71507e06c0f3da2f0749705f", "signature": "a0feb445fcf937e4b306c5bf46146a653421fba86060e29182b5250038ba2d09590c8a894b6d23f3cad2b7ba1be274c5f0f27293d2bd9f6fe50937bce7d9e501", "state": "VERIFIED", "summary": {"iw_temperature": 0.142544, "raw": 0.10941, "temperature": 0.131853}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:00:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:58Z", "cpu_seconds": 1.5703284079999982, "created_at": "2026-11-22T02:58:55Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-64f43541b177", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "b6a8d0f430a5405de61d42c87803abc151e0d1b9d5880f24e062c703fbe7f879", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-64f43541b177", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:58:55Z", "seed": 464707481, "treatment": "iw_temperature"}, "priority": 0.5097953448689929, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "c33f1bb0c2de378d70aa242b183a2835c4f33af593090c0ab07395254478ec68", "run_seconds": 1.5822179317474365, "seed": 464707481, "started_at": "2026-11-22T02:59:56Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012759, 0.01062], "effect": -0.001069, "evaluated_at": "2026-11-22T03:00:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-64f43541b177", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.191764495, 0.255620528, 0.253208265, 0.17294108, 0.240365382, 0.30126114, 0.165483887, 0.236768737, 0.160046913, 0.238880257, 0.331525782, 0.220073093, 0.22400304, 0.24390591, 0.15670711, 0.27374325], "raw": [0.20223509, 0.209264545, 0.20460594, 0.133776218, 0.227483078, 0.206092765, 0.122300475, 0.183801495, 0.177812712, 0.200208318, 0.276405728, 0.220073093, 0.19060078, 0.179244435, 0.146293973, 0.222117367], "temperature": [0.217854023, 0.242241352, 0.235821512, 0.193463937, 0.25398822, 0.30126114, 0.165483887, 0.217534065, 0.219443307, 0.223105013, 0.331525782, 0.22608254, 0.22400304, 0.218192478, 0.113347645, 0.26583901]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0936, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.806", "prereg_hash": "b6a8d0f430a5405de61d42c87803abc151e0d1b9d5880f24e062c703fbe7f879", "primary_control": "temperature", "result_hash": "c33f1bb0c2de378d70aa242b183a2835c4f33af593090c0ab07395254478ec68", "signature": "9661bab05e0a59c9fe42601bfeeda34cb558e8a326780b06e6a463ddf55b1d7f2bdd2c2ecac098ca787829c05707ec41c2de9af0a5b5402055bcf94696506c01", "state": "VERIFIED", "summary": {"iw_temperature": 0.229144, "raw": 0.193895, "temperature": 0.228074}, "ties": 4, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:00:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:06Z", "cpu_seconds": 8.364774653999998, "created_at": "2026-11-22T02:58:55Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-857d1a1113b7", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "24032087e7a8895a084151fcc1920db2dc4508f75addc7efc33ab38931ac7a92", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-857d1a1113b7", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:58:55Z", "seed": 1358517220, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "29846f99458e7db81559c68fd6b4cf75d73d8ebe3cd67b25c2680e41c164cada", "run_seconds": 8.416388750076294, "seed": 1358517220, "started_at": "2026-11-22T02:58:57Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:59:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-857d1a1113b7", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "24032087e7a8895a084151fcc1920db2dc4508f75addc7efc33ab38931ac7a92", "primary_control": "reset", "result_hash": "29846f99458e7db81559c68fd6b4cf75d73d8ebe3cd67b25c2680e41c164cada", "signature": "2be3042ff7f1e408e6e58e6e2d09b4fdc1c9c592cf2ec575916c40f57f68bb6367a307ef9e335e1ae671abb3a313c60f38de21e4300179ea388d2967ad0ad209", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.3333333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:59:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:40Z", "cpu_seconds": 0.2004980070000073, "created_at": "2026-11-22T02:58:55Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-9b048faf518b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "3d48fae1e7b630266ba66082e3fcf4c64219cfedf42850e22211157461756736", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-9b048faf518b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:55Z", "seed": 671211447, "treatment": "frontier"}, "priority": 0.25414610155749195, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "b06092f4df1d4c3f6f2b50fc3d0d30ad0c8bfe3de20db79c92f1c144c25d4c77", "run_seconds": 0.38686680793762207, "seed": 671211447, "started_at": "2026-11-22T03:05:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.017363, 0.044387], "effect": 0.030875, "evaluated_at": "2026-11-22T03:05:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-9b048faf518b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.90530303, 0.849206349, 0.956521739, 0.931506849, 0.863636364, 0.891666667, 0.907692308, 0.882352941, 0.834677419, 0.876923077, 0.915384615, 0.925], "frontier": [0.924242424, 0.900793651, 0.960144928, 0.934931507, 0.886363636, 0.933333333, 0.896153846, 0.897058824, 0.899193548, 0.896153846, 0.934615385, 0.942857143], "random_walk": [0.897727273, 0.829365079, 0.945652174, 0.917808219, 0.852272727, 0.904166667, 0.888461538, 0.878676471, 0.822580645, 0.846153846, 0.934615385, 0.917857143]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "3d48fae1e7b630266ba66082e3fcf4c64219cfedf42850e22211157461756736", "primary_control": "random_walk", "result_hash": "b06092f4df1d4c3f6f2b50fc3d0d30ad0c8bfe3de20db79c92f1c144c25d4c77", "signature": "51342942772c2c8d2e4ac40811f2e7afa5a81a624192a5f0a15b15611ca0f8ac58999eddba0c07a45a447f8cc1e29414b67bf66279030e63dc47c53209ed920f", "state": "VERIFIED", "summary": {"curiosity": 0.894989, "frontier": 0.917154, "random_walk": 0.886278}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:05:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:26Z", "cpu_seconds": 0.35835251199998197, "created_at": "2026-11-22T02:58:55Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-b1f360cb5ced", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "d7e120af15a49a6a727a755d640e95c15c8e6a4967e62eae565c56d0efb1c711", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-b1f360cb5ced", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:55Z", "seed": 1934290717, "treatment": "frontier"}, "priority": 0.2493613685454093, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-863e6aa82d86", "result_hash": "971e934c3e1dc4ddaad6cb88f61a3a6bb810f10b9dc418d53c68049868c91917", "run_seconds": 0.3639698028564453, "seed": 1934290717, "started_at": "2026-11-22T03:06:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.033187, 0.095164], "effect": 0.064176, "evaluated_at": "2026-11-22T03:06:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-b1f360cb5ced", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.8125, 1.0, 1.0, 0.785714286, 1.0, 1.0, 0.759259259, 0.797619048, 0.820588235, 0.775, 0.814516129, 0.806701031], "frontier": [0.89, 1.0, 1.0, 0.895604396, 1.0, 1.0, 0.898148148, 0.866666667, 0.923529412, 0.8775, 0.887096774, 0.894329897], "random_walk": [0.82, 1.0, 1.0, 0.777472527, 1.0, 1.0, 0.737654321, 0.816666667, 0.826470588, 0.8025, 0.801075269, 0.780927835]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "d7e120af15a49a6a727a755d640e95c15c8e6a4967e62eae565c56d0efb1c711", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-863e6aa82d86", "result_hash": "971e934c3e1dc4ddaad6cb88f61a3a6bb810f10b9dc418d53c68049868c91917", "signature": "37f9a16a596036829efc63cdf1f7e488e3713be34a6fc260e6ba1b1bfbad079b4bc4f629a2ea7717c821957d499e8015264529fea85858c520496b48d5784c09", "state": "VERIFIED", "summary": {"curiosity": 0.864325, "frontier": 0.92774, "random_walk": 0.863564}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:06:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:14Z", "cpu_seconds": 0.21282453999998552, "created_at": "2026-11-22T02:58:55Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-b6d6ed89e251", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "0e708875aa979f539e0d3bb9d6d82844f14fb08c3f49df0e108843eb9f6bc0d0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-b6d6ed89e251", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:55Z", "seed": 1062842521, "treatment": "frontier"}, "priority": 0.2499519364898938, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-e24f6cd1a3bc", "result_hash": "13c4c3832ff18a9e66ffa7ac97c6340cb3e04820700056017d874ee8d5c0fde1", "run_seconds": 0.22950983047485352, "seed": 1062842521, "started_at": "2026-11-22T03:06:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.027984, 0.053433], "effect": 0.040708, "evaluated_at": "2026-11-22T03:06:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-b6d6ed89e251", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.887681159, 0.902985075, 0.853174603, 0.934027778, 0.887681159, 0.817460317, 0.915441176, 0.945205479, 0.893382353, 0.919014085, 0.882142857, 0.888888889], "frontier": [0.934782609, 0.914179104, 0.908730159, 0.940972222, 0.920289855, 0.876984127, 0.966911765, 0.952054795, 0.911764706, 0.940140845, 0.921428571, 0.958333333], "random_walk": [0.862318841, 0.869402985, 0.857142857, 0.916666667, 0.90942029, 0.80952381, 0.911764706, 0.938356164, 0.893382353, 0.915492958, 0.885714286, 0.888888889]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "0e708875aa979f539e0d3bb9d6d82844f14fb08c3f49df0e108843eb9f6bc0d0", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-e24f6cd1a3bc", "result_hash": "13c4c3832ff18a9e66ffa7ac97c6340cb3e04820700056017d874ee8d5c0fde1", "signature": "690ac0af8853abcaa2db2db39c0a094c187e8de90b04c6d33634658e934aed0c151a31edeab0e2df53deda39de8222cbf9826c733b6e6689299429d156cb780a", "state": "VERIFIED", "summary": {"curiosity": 0.893924, "frontier": 0.928881, "random_walk": 0.888173}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:06:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:20Z", "cpu_seconds": 0.3886469090000162, "created_at": "2026-11-22T02:58:55Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-e41583d38779", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "d85a92993690a228ba052bb5b07a44e83ab966f2b641aeaa41f754e5af0d84dc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-e41583d38779", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:58:55Z", "seed": 524520757, "treatment": "frontier"}, "priority": 0.26006332547821526, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "2193a106573e69ddbd60f31ae4487c29e59617909af4de0fbb35fa8c16f1cd74", "run_seconds": 0.3885972499847412, "seed": 524520757, "started_at": "2026-11-22T03:04:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.013172, 0.05787], "effect": 0.035521, "evaluated_at": "2026-11-22T03:04:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-e41583d38779", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.794392523, 0.805825243, 1.0, 1.0, 0.807692308, 0.828431373, 1.0, 0.943181818, 0.809405941, 0.822072072, 1.0, 0.871621622], "frontier": [0.887850467, 0.800970874, 1.0, 1.0, 0.892857143, 0.894607843, 1.0, 1.0, 0.910891089, 0.896396396, 1.0, 0.882882883], "random_walk": [0.794392523, 0.817961165, 1.0, 1.0, 0.821428571, 0.830882353, 1.0, 0.943181818, 0.834158416, 0.826576577, 1.0, 0.871621622]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.454, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0352", "prereg_hash": "d85a92993690a228ba052bb5b07a44e83ab966f2b641aeaa41f754e5af0d84dc", "primary_control": "random_walk", "result_hash": "2193a106573e69ddbd60f31ae4487c29e59617909af4de0fbb35fa8c16f1cd74", "signature": "9a505acca74de2ebc519c4c1a36c77b02d5880c9e699751ac31989dd10a3a412d9a2aacf2da9d4230573dab73382416228ea57c7595c2edbb9bfee6aa7c1d004", "state": "VERIFIED", "summary": {"curiosity": 0.890219, "frontier": 0.930538, "random_walk": 0.895017}, "ties": 4, "treatment": "frontier", "wins": 7}, "verified_at": "2026-11-22T03:04:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:07Z", "cpu_seconds": 1.4987406569999848, "created_at": "2026-11-22T02:59:11Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-28c7aefd7b13", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "2533554847999235c5db1b7e6e5da287167fc536a4accf0de258fd854612114b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-28c7aefd7b13", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:11Z", "seed": 1155101470, "treatment": "iw_temperature"}, "priority": 0.5055239357508546, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "9a2a3b60ce6da34b6a7d99ceee5db622dc4f42ea9c593886bae7b2f4f7e2bc29", "run_seconds": 1.6302740573883057, "seed": 1155101470, "started_at": "2026-11-22T03:00:05Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.005946, 0.018763], "effect": 0.006408, "evaluated_at": "2026-11-22T03:00:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-28c7aefd7b13", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.129657055, 0.253638747, 0.13065486, 0.161209988, 0.16068159, 0.150162537, 0.118682645, 0.119890052, 0.11129126, 0.14978429, 0.143012688, 0.24440278, 0.104234152, 0.132758883, 0.17022493, 0.095916982], "raw": [0.13870611, 0.204372312, 0.11877407, 0.152079655, 0.15730213, 0.112778572, 0.118682645, 0.12914218, 0.137884715, 0.14978429, 0.116258692, 0.208776357, 0.104234152, 0.118430047, 0.148570742, 0.15128599], "temperature": [0.129657055, 0.23214198, 0.13065486, 0.19411689, 0.15730213, 0.115593115, 0.12756744, 0.157814147, 0.143345285, 0.14978429, 0.123118995, 0.224546403, 0.146134498, 0.132336858, 0.17022493, 0.14439783]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2127, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.613", "prereg_hash": "2533554847999235c5db1b7e6e5da287167fc536a4accf0de258fd854612114b", "primary_control": "temperature", "result_hash": "9a2a3b60ce6da34b6a7d99ceee5db622dc4f42ea9c593886bae7b2f4f7e2bc29", "signature": "fc97cbea4fe0eccfe7b05fd59631d97e9bd60d4bc16105711d089bf01fccd376c9f69cabd941f860346f1fcd0676ebba090777ec09f5fb1d0ad438ab885f2307", "state": "VERIFIED", "summary": {"iw_temperature": 0.148513, "raw": 0.141691, "temperature": 0.154921}, "ties": 4, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:00:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:53Z", "cpu_seconds": 1.5331192310000006, "created_at": "2026-11-22T02:59:11Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-7bb3b38b0597", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "c8ceb151380dfd524def3e542efa3fe4512a2d4177f2fba5433ca2afd4ed80ee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-7bb3b38b0597", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:11Z", "seed": 1188731486, "treatment": "iw_temperature"}, "priority": 0.5112190348820533, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "68c1e8c6d023719f9dc17f9377eb32d88068d57093a38fc6f0ed2ada7da99204", "run_seconds": 1.5829856395721436, "seed": 1188731486, "started_at": "2026-11-22T02:59:52Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.025274, 0.009395], "effect": -0.007939, "evaluated_at": "2026-11-22T02:59:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-7bb3b38b0597", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.09393627, 0.203819525, 0.177292153, 0.082271443, 0.135792093, 0.212868968, 0.128428522, 0.181161692, 0.176194798, 0.090414335, 0.170621125, 0.115757898, 0.150150208, 0.121259515, 0.214292258, 0.157388698], "raw": [0.067102935, 0.18903783, 0.094447997, 0.18664125, 0.143088157, 0.135885608, 0.128428522, 0.120623855, 0.117545895, 0.125357615, 0.15298239, 0.105886825, 0.09422349, 0.085096037, 0.1660215, 0.139075613], "temperature": [0.09393627, 0.229591685, 0.111089703, 0.156694537, 0.142636318, 0.190668345, 0.120052038, 0.14863161, 0.14336165, 0.130765483, 0.15381795, 0.105886825, 0.150150208, 0.076002465, 0.1660215, 0.165313407]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "c8ceb151380dfd524def3e542efa3fe4512a2d4177f2fba5433ca2afd4ed80ee", "primary_control": "temperature", "result_hash": "68c1e8c6d023719f9dc17f9377eb32d88068d57093a38fc6f0ed2ada7da99204", "signature": "c5cc3210481cd3632eb271ff4982546d1f226ca30d488ce3646cf6b1ce0977ec5d65b7a91454154931db1cd9355a722216fb48efe3cb89e9b1146013313f970d", "state": "VERIFIED", "summary": {"iw_temperature": 0.150728, "raw": 0.128215, "temperature": 0.142789}, "ties": 2, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T02:59:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:18Z", "cpu_seconds": 6.260599045000021, "created_at": "2026-11-22T02:59:11Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-6de2436fc35e", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "0f93283d1595e58740facc40fcf3212e65d3277867df3e8968d15082d2575d53", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-6de2436fc35e", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:59:11Z", "seed": 314453852, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "1ec3821bf25d3144ecca3887af2007e16d90c42f0e2829e76cfaa7337f83e55b", "run_seconds": 6.530988454818726, "seed": 314453852, "started_at": "2026-11-22T02:59:12Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:59:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-6de2436fc35e", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.1, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0f93283d1595e58740facc40fcf3212e65d3277867df3e8968d15082d2575d53", "primary_control": "reset", "result_hash": "1ec3821bf25d3144ecca3887af2007e16d90c42f0e2829e76cfaa7337f83e55b", "signature": "967cabcbe0f9db10fcffb42eaf0296bbb900fbbc125b02b37bc5a4f0637d8b347410686c3bda659784f1676390e01bdb7410bb2bae4cc07b43c7f0b49b530608", "state": "PARTIAL", "summary": {"inherit": 0.6333333333333333, "reset": 0.6333333333333333, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:59:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:51Z", "cpu_seconds": 0.3534389170000054, "created_at": "2026-11-22T02:59:11Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-3c0979d7fa58", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "81e147a0d0bc6208555f7f4786cf930aab5a5c99b4963e6cfee26ff21f08b75b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-3c0979d7fa58", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:59:11Z", "seed": 424243852, "treatment": "frontier"}, "priority": 0.2585157796190289, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "7da3149d454d689706927818dc4a5ed63710a13276f2c00ca985eab75e5d93db", "run_seconds": 0.3549642562866211, "seed": 424243852, "started_at": "2026-11-22T03:04:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.065458, 0.099815], "effect": 0.082636, "evaluated_at": "2026-11-22T03:04:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-3c0979d7fa58", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.900735294, 0.915322581, 0.93115942, 0.958333333, 0.919354839, 0.9140625, 0.943661972, 0.885964912, 0.9375, 0.890151515, 0.917910448, 0.948275862], "frontier": [0.985294118, 1.0, 0.974637681, 0.996527778, 1.0, 0.984375, 0.982394366, 1.0, 0.986111111, 0.992424242, 0.996268657, 0.99137931], "random_walk": [0.900735294, 0.883064516, 0.913043478, 0.954861111, 0.915322581, 0.875, 0.929577465, 0.864035088, 0.940972222, 0.882575758, 0.929104478, 0.909482759]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "81e147a0d0bc6208555f7f4786cf930aab5a5c99b4963e6cfee26ff21f08b75b", "primary_control": "random_walk", "result_hash": "7da3149d454d689706927818dc4a5ed63710a13276f2c00ca985eab75e5d93db", "signature": "41ed9c17cd8301d1eb855e695f9b64cd752e46e7708b9b5de5fdc30bc7e9eb8eb75903981370b1c68423f62504c9bd14db20742e81e1e3792a3e75598c6f4802", "state": "VERIFIED", "summary": {"curiosity": 0.921869, "frontier": 0.990784, "random_walk": 0.908148}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:04:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:17Z", "cpu_seconds": 0.6101412079999875, "created_at": "2026-11-22T02:59:12Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-bdcbd97fdb37", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "1bebc889765a7d43d7838f7dea9916339b1ab04dd83822e205ace4de697c1737", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-bdcbd97fdb37", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:59:12Z", "seed": 565379493, "treatment": "frontier"}, "priority": 0.25539293173624833, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "2386198b92c04debae84622385a88757b3dcc1a78d4c7e48a64b608e4e7f403a", "run_seconds": 0.8851261138916016, "seed": 565379493, "started_at": "2026-11-22T03:05:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.072381, 0.109925], "effect": 0.091153, "evaluated_at": "2026-11-22T03:05:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-bdcbd97fdb37", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.869369369, 0.85, 0.889830508, 0.916666667, 0.870535714, 0.926470588, 0.941532258, 0.892561983, 0.913385827, 0.894067797, 0.904471545, 0.916], "frontier": [1.0, 0.997727273, 0.995762712, 0.986111111, 1.0, 0.987394958, 0.989919355, 0.987603306, 0.986220472, 0.991525424, 0.987804878, 0.988], "random_walk": [0.853603604, 0.843181818, 0.879237288, 0.912698413, 0.886160714, 0.911764706, 0.91733871, 0.921487603, 0.92519685, 0.908898305, 0.916666667, 0.928]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "1bebc889765a7d43d7838f7dea9916339b1ab04dd83822e205ace4de697c1737", "primary_control": "random_walk", "result_hash": "2386198b92c04debae84622385a88757b3dcc1a78d4c7e48a64b608e4e7f403a", "signature": "79e5c44eed1c1abab674a0e17a69c739c47a758bddf17df8186952fd12f30b414cd1d030c0d915481b0e2ce12d3c44d6e625a77bdf60311b89172fd428af3408", "state": "VERIFIED", "summary": {"curiosity": 0.898741, "frontier": 0.991506, "random_walk": 0.900353}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:05:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:58Z", "cpu_seconds": 0.4279466679999899, "created_at": "2026-11-22T02:59:12Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-e513ad9c71bc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "cf286582d3ec0bc694c02dd76bb03752b7dffbdd2c43dfe6860e531bffb2efe6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-e513ad9c71bc", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T02:59:12Z", "seed": 596737443, "treatment": "frontier"}, "priority": 0.2520146854039069, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "7271b855bbb3d8c747b6d9372fa38ffd8dffd5d096b86c0cdacb784a0ad61211", "run_seconds": 0.523496150970459, "seed": 596737443, "started_at": "2026-11-22T03:05:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.024291, 0.041636], "effect": 0.032964, "evaluated_at": "2026-11-22T03:06:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-e513ad9c71bc", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.894308943, 0.898760331, 0.922619048, 0.894308943, 0.891666667, 1.0, 0.862288136, 0.898760331, 0.898373984, 0.877118644, 0.932, 0.859649123], "frontier": [0.924796748, 0.938016529, 0.96031746, 0.914634146, 0.922916667, 1.0, 0.891949153, 0.911157025, 0.941056911, 0.913135593, 0.938, 0.912280702], "random_walk": [0.892276423, 0.882231405, 0.924603175, 0.888211382, 0.902083333, 1.0, 0.849576271, 0.880165289, 0.894308943, 0.88559322, 0.914, 0.859649123]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "cf286582d3ec0bc694c02dd76bb03752b7dffbdd2c43dfe6860e531bffb2efe6", "primary_control": "random_walk", "result_hash": "7271b855bbb3d8c747b6d9372fa38ffd8dffd5d096b86c0cdacb784a0ad61211", "signature": "51af61f511089c3feafe0b50ae9098dbfecb020ad2020aa23643196698450afb8af6276225b18a2baa7ce0b5eaefe0fc51453a3650f111d99a32de3684b4680a", "state": "VERIFIED", "summary": {"curiosity": 0.902488, "frontier": 0.930688, "random_walk": 0.897725}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:06:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:45Z", "cpu_seconds": 1.4692787610000018, "created_at": "2026-11-22T02:59:28Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-58d5d43c4ea9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "3f9530472f5a4d1fd62542dbb2aed11b758f98bb864d26408be60f26fa8bee47", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-58d5d43c4ea9", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:28Z", "seed": 1874438162, "treatment": "iw_temperature"}, "priority": 0.5191679258294118, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a8dabda75bfb5b65f50153822d2b629fbe0f147c6a32d946aa741a74ca2fa830", "run_seconds": 1.5802993774414062, "seed": 1874438162, "started_at": "2026-11-22T02:59:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012485, 0.001442], "effect": -0.005521, "evaluated_at": "2026-11-22T02:59:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-58d5d43c4ea9", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.04084393, 0.016010185, 0.03149129, 0.026349395, 0.018326027, 0.01195344, 0.076298743, 0.178975708, 0.061968838, 0.151332358, 0.074493462, 0.04772667, 0.036803553, 0.029943532, 0.027403043, 0.052662195], "raw": [0.028494027, 0.02850368, 0.048908505, 0.047458513, 0.025455173, 0.008865075, 0.056494227, 0.166293023, 0.061968838, 0.113292518, 0.075799978, 0.047864178, 0.071041155, 0.023944785, 0.032018322, 0.070844935], "temperature": [0.038726215, 0.021567317, 0.033593797, 0.033071082, 0.01925532, 0.00912289, 0.052965415, 0.166293023, 0.034510485, 0.111055283, 0.089505085, 0.044902258, 0.036803553, 0.019914707, 0.027403043, 0.055549277]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "3f9530472f5a4d1fd62542dbb2aed11b758f98bb864d26408be60f26fa8bee47", "primary_control": "temperature", "result_hash": "a8dabda75bfb5b65f50153822d2b629fbe0f147c6a32d946aa741a74ca2fa830", "signature": "1c3abd84a1e2f20a1a3a6e9bad6921c2505c8f90899073761df17793230ba827eb9dd71d1d5dc9265636f565e84849a42961ec6d0416bb8b62b6996b4567a00f", "state": "VERIFIED", "summary": {"iw_temperature": 0.055161, "raw": 0.056703, "temperature": 0.04964}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T02:59:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:31Z", "cpu_seconds": 1.548893580999998, "created_at": "2026-11-22T02:59:28Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-674aba139eea", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "f833475a08a0ba538dc1e72aee02049a47eebd4e4fd69bc2e106e8a922c78903", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-674aba139eea", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:28Z", "seed": 1419223976, "treatment": "iw_temperature"}, "priority": 0.6939929506182143, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "3457ec3f2f69889a65dea0951d803f3f58accf2a00fb5a085415cdf470b0cec3", "run_seconds": 1.586672067642212, "seed": 1419223976, "started_at": "2026-11-22T02:59:29Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018874, 0.0183], "effect": -0.000287, "evaluated_at": "2026-11-22T02:59:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-674aba139eea", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.012738152, 0.257248262, 0.264625397, 0.260236923, 0.318356648, 0.29224554, 0.32019789, 0.223840197, 0.31023627, 0.269058603, 0.28999113, 0.194724847, 0.23271189, 0.30353755, 0.19538929, 0.255713212], "raw": [0.17576397, 0.250340768, 0.252924875, 0.213447532, 0.29261588, 0.225912567, 0.212556858, 0.223840197, 0.304141097, 0.230768793, 0.288774228, 0.165332248, 0.168896135, 0.31086836, 0.167318355, 0.218497342], "temperature": [0.124974833, 0.257248262, 0.264625397, 0.245884587, 0.29261588, 0.241124827, 0.271195838, 0.26431366, 0.308636353, 0.250018603, 0.291027592, 0.213826413, 0.23271189, 0.317761478, 0.17460551, 0.24568853]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "f833475a08a0ba538dc1e72aee02049a47eebd4e4fd69bc2e106e8a922c78903", "primary_control": "temperature", "result_hash": "3457ec3f2f69889a65dea0951d803f3f58accf2a00fb5a085415cdf470b0cec3", "signature": "5573d7c010ce6c49160ddf745b88cb7f066100f2d4125be667f26d98f051b2dcdb67174e7dcae1cbe80bcfb4a95cd1f943cfedab5876de3059c5e54a97559c0b", "state": "VERIFIED", "summary": {"iw_temperature": 0.250053, "raw": 0.231375, "temperature": 0.249766}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T02:59:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:27Z", "cpu_seconds": 1.6719424030000027, "created_at": "2026-11-22T02:59:28Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b273ea9fa848", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "2a2ee196c3dbaeb90abb67ec46162377d10db720e6b3e6e3fb6b98ecfb3ae3df", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b273ea9fa848", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:28Z", "seed": 452435753, "treatment": "iw_temperature"}, "priority": 0.4902691402482059, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-4473ceb9e096", "result_hash": "2e4a87607c7b3467ba7a3255246288c3670d1cc15148cd27961874283187dbb4", "run_seconds": 1.9289419651031494, "seed": 452435753, "started_at": "2026-11-22T03:00:24Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018103, 0.025307], "effect": 0.003602, "evaluated_at": "2026-11-22T03:00:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b273ea9fa848", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.194326362, 0.17357756, 0.171494748, 0.2083799, 0.177163203, 0.281097317, 0.200803152, 0.219584493, 0.26496784, 0.215780943, 0.30938595, 0.203836713, 0.08929188, 0.29979679, 0.230275205, 0.24501185], "raw": [0.194326362, 0.189874618, 0.206851362, 0.154806447, 0.154867647, 0.225901617, 0.187444868, 0.15831482, 0.17345171, 0.154647248, 0.277042867, 0.186911308, 0.198411282, 0.222905325, 0.230275205, 0.229274695], "temperature": [0.194326362, 0.233512658, 0.217753335, 0.178818735, 0.187138797, 0.281097317, 0.200803152, 0.165617855, 0.209347403, 0.19327937, 0.318625685, 0.186911308, 0.213513092, 0.27680929, 0.239837378, 0.24501185]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.2127, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.613", "prereg_hash": "2a2ee196c3dbaeb90abb67ec46162377d10db720e6b3e6e3fb6b98ecfb3ae3df", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-4473ceb9e096", "result_hash": "2e4a87607c7b3467ba7a3255246288c3670d1cc15148cd27961874283187dbb4", "signature": "1cd9c04ba2b981f0fb391aa4d4ae6e5444f8c46f246c5c943209a82d27c3d4278e500aaed4d8d8cc8fd4ba1dcf7e7ace7bdc4775b747bbc9b5992776db7fb70c", "state": "VERIFIED", "summary": {"iw_temperature": 0.217798, "raw": 0.196582, "temperature": 0.2214}, "ties": 4, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:00:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:34Z", "cpu_seconds": 5.289661095, "created_at": "2026-11-22T02:59:28Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-87742f58ab7a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "c89e6a45713f401997edc81612df5d11d4ddf6e44dc7dbf8834697ed1ec1ecd4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-87742f58ab7a", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:59:28Z", "seed": 1940991630, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "8fa2714499f9ef82c285d1177995ee4f1a207317aedba62ebebd6674f1ccdd19", "run_seconds": 5.337064027786255, "seed": 1940991630, "started_at": "2026-11-22T02:59:29Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 4, "reset": 4}, "better": "higher", "ci95": [-0.005613, 0.105613], "effect": 0.05, "evaluated_at": "2026-11-22T02:59:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-87742f58ab7a", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.656305108464547, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c89e6a45713f401997edc81612df5d11d4ddf6e44dc7dbf8834697ed1ec1ecd4", "primary_control": "reset", "result_hash": "8fa2714499f9ef82c285d1177995ee4f1a207317aedba62ebebd6674f1ccdd19", "signature": "60ac952486a1dee6af3f912f7bf37350eb3788db4a736ce45843e3de4ef03d02fa2ba3c31b8c46695cbd98e111d3dc0a0564631b6a1c5ab588bcee4328d2b906", "state": "VERIFIED", "summary": {"inherit": 0.9333333333333333, "reset": 0.8833333333333333, "root": 0.4}, "ties": 57, "treatment": "inherit", "wins": 3}, "verified_at": "2026-11-22T02:59:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:40Z", "cpu_seconds": 9.363274859, "created_at": "2026-11-22T02:59:28Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-cbb436f0b329", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "874be627a5c73216c402b5213b95b7b6b38c65c181847b1c2529fc50b64de751", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-cbb436f0b329", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:59:28Z", "seed": 1990879748, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "2a902670a38208814f76fff69d426cccc4b2d439c34934db9ab3ae3b722dae8c", "run_seconds": 9.650429010391235, "seed": 1990879748, "started_at": "2026-11-22T02:59:30Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T02:59:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-cbb436f0b329", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "874be627a5c73216c402b5213b95b7b6b38c65c181847b1c2529fc50b64de751", "primary_control": "reset", "result_hash": "2a902670a38208814f76fff69d426cccc4b2d439c34934db9ab3ae3b722dae8c", "signature": "3ec90c58385352a6ec3dd4179e927f548d125428b2209f8893aad28ee9816b5af23a10c4b97ac8a3cc3b0468eb717ff5a53b9dee95444fa0f0bd3497ebfaba06", "state": "PARTIAL", "summary": {"inherit": 0.6833333333333333, "reset": 0.6833333333333333, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T02:59:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:21Z", "cpu_seconds": 1.5853003419999965, "created_at": "2026-11-22T02:59:44Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-0990932b6af3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "42401aa8c7b28d614a91cb8efdd259bc6b1bd991ec609301505191b51a24bc1d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-0990932b6af3", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:44Z", "seed": 1167948271, "treatment": "iw_temperature"}, "priority": 0.3784688184475684, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "3e63f7328941b85b25e7f913a4567ef79c84612d1d69844421775a5d6fa05852", "run_seconds": 1.6276297569274902, "seed": 1167948271, "started_at": "2026-11-22T03:02:20Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018719, -0.002021], "effect": -0.01037, "evaluated_at": "2026-11-22T03:02:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-0990932b6af3", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.023204457, 0.01560113, 0.009842788, 0.058994612, 0.070345618, 0.021215567, 0.037933165, 0.06200457, 0.008359647, 0.03770022, 0.064753855, 0.063805315, 0.0235799, 0.112594412, 0.018196238, 0.005165022], "raw": [0.030520818, 0.032237475, 0.0114608, 0.06554875, 0.059388217, 0.016021823, 0.066874597, 0.06200457, 0.01370257, 0.038158512, 0.044817808, 0.055978613, 0.00546149, 0.06617283, 0.023966048, 0.012559752], "temperature": [0.039850953, 0.01560113, 0.005378777, 0.057737228, 0.057035645, 0.009240085, 0.035940115, 0.035985275, 0.01370257, 0.028877372, 0.044817808, 0.037624765, 0.00724273, 0.05498275, 0.018196238, 0.005165022]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0007, "losses": 11, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.998", "prereg_hash": "42401aa8c7b28d614a91cb8efdd259bc6b1bd991ec609301505191b51a24bc1d", "primary_control": "temperature", "result_hash": "3e63f7328941b85b25e7f913a4567ef79c84612d1d69844421775a5d6fa05852", "signature": "5a1c8fe1c643189b339d00cecf8bbc38a56cb846c18a089763ea6eb1a1e7177c8474b85009f305d582af36b9288919c2e41c5f02eddf00af502d9f8fbbaac80c", "state": "VERIFIED", "summary": {"iw_temperature": 0.039581, "raw": 0.037805, "temperature": 0.029211}, "ties": 3, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:02:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:47Z", "cpu_seconds": 1.6020082420000108, "created_at": "2026-11-22T02:59:44Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1e4a76bab22b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "1d2324505c042b61896e9aad7a4544eae0865b1ad47ee690f56f0ebedc998521", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1e4a76bab22b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:44Z", "seed": 752888999, "treatment": "iw_temperature"}, "priority": 0.6114840571436146, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "c72a7df651dfcedb0c1e4aad09e374580aaad0003bbc7a4415c31fa0f8069967", "run_seconds": 1.6723549365997314, "seed": 752888999, "started_at": "2026-11-22T02:59:45Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.005937, 0.024007], "effect": 0.009035, "evaluated_at": "2026-11-22T02:59:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1e4a76bab22b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.31990237, 0.102924897, 0.264458778, 0.321641723, 0.27383342, 0.312686788, 0.214485145, 0.214582347, 0.11230399, 0.24517533, 0.307546013, 0.161540312, 0.306617183, 0.287647725, 0.32181254, 0.294635837], "raw": [0.31872468, 0.180658293, 0.249065815, 0.300733563, 0.27383342, 0.293667955, 0.196738743, 0.260031842, 0.119338222, 0.212292757, 0.29521034, 0.184143365, 0.244494373, 0.25960825, 0.297000717, 0.283521745], "temperature": [0.319483078, 0.177761008, 0.262712053, 0.310485995, 0.27634526, 0.3060112, 0.224437242, 0.297621132, 0.119338222, 0.24517533, 0.305583755, 0.198504253, 0.280096177, 0.267319512, 0.32181254, 0.293666563]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "1d2324505c042b61896e9aad7a4544eae0865b1ad47ee690f56f0ebedc998521", "primary_control": "temperature", "result_hash": "c72a7df651dfcedb0c1e4aad09e374580aaad0003bbc7a4415c31fa0f8069967", "signature": "e045267111ffbc3d80c313d044f97dfff0fd9d859e0b9ad37f5af017e7cdc35877bf5c91a1cac4d65e11de9f1bf43ffe7caaa1fc7d508aa1658ac6bd4fde310f", "state": "VERIFIED", "summary": {"iw_temperature": 0.253862, "raw": 0.248067, "temperature": 0.262897}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T02:59:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:51Z", "cpu_seconds": 1.5534627190000023, "created_at": "2026-11-22T02:59:44Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-64dacb30eb0e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "816547a5b283c70b7228ce45e87663008b1a11e2436200c16e1b272aec83a3b8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-64dacb30eb0e", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T02:59:44Z", "seed": 941325778, "treatment": "iw_temperature"}, "priority": 0.5345752713153248, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "2b8849f91213ae1b9920e1fef5a0e4bab981b995cbae9ffa457e183be7908a04", "run_seconds": 1.5780794620513916, "seed": 941325778, "started_at": "2026-11-22T02:59:50Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.009989, 0.001929], "effect": -0.00403, "evaluated_at": "2026-11-22T02:59:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-64dacb30eb0e", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.13988271, 0.20396839, 0.123136075, 0.113470368, 0.09349862, 0.15703713, 0.121388258, 0.16342314, 0.049519125, 0.128797025, 0.126841215, 0.103814515, 0.078437535, 0.08924483, 0.13396962, 0.115865835], "raw": [0.08070688, 0.12464271, 0.10422532, 0.105422265, 0.082796665, 0.11151252, 0.09543879, 0.133911363, 0.06907912, 0.080557797, 0.08689557, 0.103814515, 0.081521585, 0.091138652, 0.08755477, 0.06208457], "temperature": [0.13988271, 0.18759261, 0.123136075, 0.113470368, 0.082796665, 0.14866572, 0.121388258, 0.153580422, 0.070757855, 0.097604465, 0.142237057, 0.103814515, 0.0737221, 0.091138652, 0.121638075, 0.10639299]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "816547a5b283c70b7228ce45e87663008b1a11e2436200c16e1b272aec83a3b8", "primary_control": "temperature", "result_hash": "2b8849f91213ae1b9920e1fef5a0e4bab981b995cbae9ffa457e183be7908a04", "signature": "d388693e7a782c7ecf41225584a9a62d542d81f1c919dadfadc61676d8bf1728bf27e57cfc19c6058555b68450a39f9d9ab5dc50610f251920c14cd9256a2503", "state": "VERIFIED", "summary": {"iw_temperature": 0.121393, "raw": 0.093831, "temperature": 0.117364}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T02:59:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T02:59:53Z", "cpu_seconds": 8.614228066999999, "created_at": "2026-11-22T02:59:44Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0fd4ba636714", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ebfb8c74cc98bf864fcce5147274b355bd6f2078333aba92ca99cedc761e0796", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0fd4ba636714", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T02:59:44Z", "seed": 671072619, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "63f3bbd71edec2018141a8c18eb3ace114a49ca21ee412b441bbcb00a90d6ec5", "run_seconds": 8.890531301498413, "seed": 671072619, "started_at": "2026-11-22T02:59:44Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 4, "reset": 3}, "better": "higher", "ci95": [0.047666, 0.252334], "effect": 0.15, "evaluated_at": "2026-11-22T02:59:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0fd4ba636714", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6468300569023522, "rstep": 1}, "hypothesis_supported": true, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -2.2321, "losses": 1, "n": 60, "outcome": "SUPPORTED", "p_sign_display": "0.00586", "prereg_hash": "ebfb8c74cc98bf864fcce5147274b355bd6f2078333aba92ca99cedc761e0796", "primary_control": "reset", "result_hash": "63f3bbd71edec2018141a8c18eb3ace114a49ca21ee412b441bbcb00a90d6ec5", "signature": "801680a3ee11977dbe6058edd2a66bfaf3287048bbfd8cafeeb6ebf882e202ab4213b24633d6c522fd61dec1298ed34059b9a320236a655a8a475da7ef82d300", "state": "VERIFIED", "summary": {"inherit": 0.7, "reset": 0.55, "root": 0.35}, "ties": 49, "treatment": "inherit", "wins": 10}, "verified_at": "2026-11-22T03:00:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:31Z", "cpu_seconds": 1.4629184809999742, "created_at": "2026-11-22T03:00:00Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-35ec266702f4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "c26b34235e2db0dd32b8ff9a9b92b00297f5e73525b32f7fd3d015122ac1f762", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-35ec266702f4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:00Z", "seed": 658649823, "treatment": "iw_temperature"}, "priority": 0.3677579160081484, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a86dbd96989b38050f8f771619b8ea8957d56f90601fa524e597253eb30560e8", "run_seconds": 1.5553836822509766, "seed": 658649823, "started_at": "2026-11-22T03:02:29Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01384, 0.004931], "effect": -0.004454, "evaluated_at": "2026-11-22T03:02:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-35ec266702f4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.12708719, 0.151937475, 0.144375077, 0.203344673, 0.100189032, 0.150285515, 0.17503937, 0.144705965, 0.14453569, 0.205184055, 0.177680295, 0.126017465, 0.116055158, 0.179712857, 0.156791368, 0.197187975], "raw": [0.137581482, 0.168778208, 0.12613746, 0.14552415, 0.100189032, 0.153207408, 0.161282312, 0.09840558, 0.10798524, 0.26671571, 0.13218367, 0.13362938, 0.101599317, 0.132507267, 0.142964842, 0.157467455], "temperature": [0.126054698, 0.156239298, 0.123089262, 0.175887828, 0.115299135, 0.145841297, 0.182674258, 0.144705965, 0.10798524, 0.2498535, 0.16623904, 0.126017465, 0.116055158, 0.15970151, 0.15665051, 0.176568668]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0205, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.954", "prereg_hash": "c26b34235e2db0dd32b8ff9a9b92b00297f5e73525b32f7fd3d015122ac1f762", "primary_control": "temperature", "result_hash": "a86dbd96989b38050f8f771619b8ea8957d56f90601fa524e597253eb30560e8", "signature": "baf98b9aa21845597ac95d203fed4feaaa2e14e26893219c81b8330e843065f47bd8493e3a96988c5be02012948153ee7989213aee8beef00f0efa2cfcaaab0e", "state": "VERIFIED", "summary": {"iw_temperature": 0.156258, "raw": 0.141635, "temperature": 0.151804}, "ties": 3, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:02:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:06Z", "cpu_seconds": 1.6769078159999964, "created_at": "2026-11-22T03:00:00Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-3b8aef663f1e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "496dd41eb12a828e6e8954b4f1f3d852e3ab0d8da60d61c55185df11a2c45cb2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-3b8aef663f1e", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:00Z", "seed": 790454159, "treatment": "iw_temperature"}, "priority": 0.5376304137568738, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ed771e17f485396c5ea95a41229b4bb92bcf0a56243cd93e1b92a855265a2cc4", "run_seconds": 2.3028621673583984, "seed": 790454159, "started_at": "2026-11-22T03:00:03Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.006134, 0.036282], "effect": 0.015074, "evaluated_at": "2026-11-22T03:00:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-3b8aef663f1e", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.258509163, 0.27075901, 0.261554908, 0.07105882, 0.09888955, 0.297665077, 0.314006618, 0.203632522, 0.267635083, 0.157020077, 0.313774182, 0.277468915, 0.199506032, 0.271461518, 0.334246432, 0.159716417], "raw": [0.228829035, 0.25913098, 0.20569268, 0.216851472, 0.138999012, 0.289122548, 0.266168072, 0.226349053, 0.2034402, 0.18472441, 0.303829038, 0.256039803, 0.16524962, 0.271461518, 0.312331045, 0.18158229], "temperature": [0.25159527, 0.27715728, 0.251739878, 0.216851472, 0.138999012, 0.297665077, 0.326620968, 0.237360557, 0.232745955, 0.207784275, 0.31227039, 0.269399663, 0.16524962, 0.300895223, 0.317364555, 0.194386415]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "496dd41eb12a828e6e8954b4f1f3d852e3ab0d8da60d61c55185df11a2c45cb2", "primary_control": "temperature", "result_hash": "ed771e17f485396c5ea95a41229b4bb92bcf0a56243cd93e1b92a855265a2cc4", "signature": "9c78af8c302bc8467212583565d0ab565b34aae86627d3ffb5e627b75e22fc90ed9bb17369b6365e7c8a7ec032a6775d0b29fa79f63f5e7ef7485ebcf4f12a03", "state": "VERIFIED", "summary": {"iw_temperature": 0.234807, "raw": 0.231863, "temperature": 0.24988}, "ties": 1, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:00:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:19Z", "cpu_seconds": 1.5415867030000072, "created_at": "2026-11-22T03:00:00Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4b180e00c6a4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "afe3244f6c94367871d81caf3348d35fa9591129504a99a3167dcf7351fb4826", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4b180e00c6a4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:00Z", "seed": 63130968, "treatment": "iw_temperature"}, "priority": 0.3942078944427099, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "1e1c1d63eb1cf3d23396f5cb5007600f2a1ddea47a51b7913fafa07139b73748", "run_seconds": 1.722245454788208, "seed": 63130968, "started_at": "2026-11-22T03:01:17Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.009567, 0.005805], "effect": -0.001881, "evaluated_at": "2026-11-22T03:01:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4b180e00c6a4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.057778387, 0.152273365, 0.08088748, 0.130920387, 0.1103555, 0.128283495, 0.092917627, 0.165366008, 0.119719413, 0.149114258, 0.152233135, 0.18829071, 0.130995212, 0.118224153, 0.09547166, 0.118666995], "raw": [0.07833739, 0.113119837, 0.112160485, 0.08404316, 0.072531883, 0.06500529, 0.083883395, 0.11755315, 0.094173948, 0.10472956, 0.092110025, 0.095878398, 0.10036795, 0.111796582, 0.076289842, 0.103883537], "temperature": [0.057778387, 0.152273365, 0.125991567, 0.135297255, 0.1103555, 0.100967962, 0.083883395, 0.165366008, 0.119719413, 0.149114258, 0.152233135, 0.170883275, 0.12213364, 0.111796582, 0.076289842, 0.127316548]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 6, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "afe3244f6c94367871d81caf3348d35fa9591129504a99a3167dcf7351fb4826", "primary_control": "temperature", "result_hash": "1e1c1d63eb1cf3d23396f5cb5007600f2a1ddea47a51b7913fafa07139b73748", "signature": "0c9ae72062ad59fa45e4cfd1655f0563d3cc907d4c66ba824eddf721bb998e8238b57c4d561e1343b0677d0da24f0657125767e3291695d67e74890956258c0b", "state": "VERIFIED", "summary": {"iw_temperature": 0.124469, "raw": 0.094117, "temperature": 0.122588}, "ties": 7, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:01:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:33Z", "cpu_seconds": 1.5324804729999997, "created_at": "2026-11-22T03:00:00Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c6c284f5e783", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "c8f72e4b6b35d93d58f450afadf9e0eb508839dbc4abfe359fbb8d05350093e6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c6c284f5e783", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:00Z", "seed": 981701988, "treatment": "iw_temperature"}, "priority": 0.3919667785003842, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "e413d31379e3d83915bfc13ea2438adb9822ed0e35d55bb2a8ff8c302b58cd04", "run_seconds": 1.7892284393310547, "seed": 981701988, "started_at": "2026-11-22T03:01:31Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010749, 0.016836], "effect": 0.003043, "evaluated_at": "2026-11-22T03:01:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c6c284f5e783", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.077914643, 0.06839218, 0.034789535, 0.015870133, 0.081915025, 0.03601626, 0.051903138, 0.050771995, 0.048234563, 0.043549837, 0.052080015, 0.08226628, 0.01687065, 0.068959443, 0.04522652, 0.020511205], "raw": [0.098054387, 0.047536373, 0.04051563, 0.018761728, 0.061646065, 0.030925697, 0.063790735, 0.039221848, 0.064354465, 0.042188383, 0.04502036, 0.06994484, 0.03294288, 0.029446152, 0.06059064, 0.028580443], "temperature": [0.173416558, 0.064446377, 0.061880918, 0.015870133, 0.073970072, 0.03601626, 0.063424635, 0.050771995, 0.036296175, 0.040841492, 0.049153402, 0.06994484, 0.016816832, 0.028373335, 0.04522652, 0.017515837]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0085, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.981", "prereg_hash": "c8f72e4b6b35d93d58f450afadf9e0eb508839dbc4abfe359fbb8d05350093e6", "primary_control": "temperature", "result_hash": "e413d31379e3d83915bfc13ea2438adb9822ed0e35d55bb2a8ff8c302b58cd04", "signature": "fec4fec744796b29d73e4f9a89af9d2c80c3720cd9e109120696fd8ddb05776711aeaa6cc127c4cd3bdd0c4ed678e9e1140d3aa31e4aac42d9f74a683d170407", "state": "VERIFIED", "summary": {"iw_temperature": 0.049704, "raw": 0.048345, "temperature": 0.052748}, "ties": 4, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:01:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:09Z", "cpu_seconds": 7.492304987999994, "created_at": "2026-11-22T03:00:00Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3f7de97253d1", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d14c59f26fde82d1f18f7fb73b0d34d029c19ae6c22283dcb67debcb879c476b", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3f7de97253d1", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:00:00Z", "seed": 635650893, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "607bfed6eb6e5f30d66af0320825c1b882088d2f16b48e67653a981dee928232", "run_seconds": 8.366901636123657, "seed": 635650893, "started_at": "2026-11-22T03:00:01Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:00:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3f7de97253d1", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "reset": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "root": [0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d14c59f26fde82d1f18f7fb73b0d34d029c19ae6c22283dcb67debcb879c476b", "primary_control": "reset", "result_hash": "607bfed6eb6e5f30d66af0320825c1b882088d2f16b48e67653a981dee928232", "signature": "3d11423827e63d701d4dfd4d1f97fee705eecf9339847df1d175888337cb29427eab16f6b6887183eb191636148cede0e40c6e6dfda4565932bbcd19abdaed00", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:00:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:52Z", "cpu_seconds": 0.5669783399999986, "created_at": "2026-11-22T03:00:00Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-8406a9900225", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "e51442da2cac5ed13031ac53138bc95556c3c33c8e25b964945e59784e4150b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-8406a9900225", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:00:00Z", "seed": 1611005215, "treatment": "frontier"}, "priority": 0.3838925807520741, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-d1e39d0ae411", "result_hash": "af5ced07d5f045d71e46a36b952f26a7bf8519eced540bdd2d45953c35ce490c", "run_seconds": 0.6339161396026611, "seed": 1611005215, "started_at": "2026-11-22T03:01:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.082343, 0.18408], "effect": 0.133212, "evaluated_at": "2026-11-22T03:01:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-8406a9900225", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.791666667, 0.813636364, 1.0, 0.784482759, 0.835106383, 1.0, 0.875, 1.0, 0.776785714, 0.760638298, 0.896825397, 0.834745763], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.827380952, 0.809090909, 1.0, 0.806034483, 0.877659574, 1.0, 0.845833333, 1.0, 0.785714286, 0.739361702, 0.896825397, 0.813559322]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "e51442da2cac5ed13031ac53138bc95556c3c33c8e25b964945e59784e4150b6", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-d1e39d0ae411", "result_hash": "af5ced07d5f045d71e46a36b952f26a7bf8519eced540bdd2d45953c35ce490c", "signature": "db528c2a1a7f8c9a950fdcb16996143ccee4a949ffab745ad21310447051d39305ef5af31c2e3f217ef7f492c41c83b2ea5dae05591ee1e3c5867d86def03c0a", "state": "VERIFIED", "summary": {"curiosity": 0.864074, "frontier": 1.0, "random_walk": 0.866788}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:01:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:47Z", "cpu_seconds": 1.400839194999996, "created_at": "2026-11-22T03:00:16Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b50d516aaf5c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "d8554aca14bfbecfb71955be0f21f55603b343c350a5d55cc389317b58168d04", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b50d516aaf5c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:16Z", "seed": 1785037741, "treatment": "iw_temperature"}, "priority": 0.3868822515489911, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "46361e03c8ca49ceb1cbd72d0ac30000343284bb2f34c4108ccc1399b9bf8fb3", "run_seconds": 1.5106208324432373, "seed": 1785037741, "started_at": "2026-11-22T03:01:45Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.022465, 0.0078], "effect": -0.007333, "evaluated_at": "2026-11-22T03:01:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b50d516aaf5c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.154961815, 0.163753245, 0.113118935, 0.14067124, 0.143034255, 0.144585898, 0.123381083, 0.18755275, 0.216098208, 0.152438547, 0.19425974, 0.138568798, 0.155797368, 0.122142115, 0.170565632, 0.100787377], "raw": [0.128613937, 0.171868327, 0.112703637, 0.101637513, 0.13064075, 0.123787415, 0.113403693, 0.13198272, 0.122327235, 0.165419468, 0.176651272, 0.138568798, 0.219292802, 0.110506617, 0.150854222, 0.101378362], "temperature": [0.146738987, 0.14742864, 0.113118935, 0.120623658, 0.13064075, 0.125183015, 0.113321782, 0.148638172, 0.155941025, 0.152438547, 0.176651272, 0.1459068, 0.24622154, 0.121398297, 0.15935814, 0.100787377]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0007, "losses": 11, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.998", "prereg_hash": "d8554aca14bfbecfb71955be0f21f55603b343c350a5d55cc389317b58168d04", "primary_control": "temperature", "result_hash": "46361e03c8ca49ceb1cbd72d0ac30000343284bb2f34c4108ccc1399b9bf8fb3", "signature": "7f1f2bc412785fde693fff5ab4a99e41258a1f18ed2a4ba9c80e77c3ac1381fcfb7e5bc352c458ec4e0b0f4b72f49fd363a4ff991a26bfe1067ba00eb937c202", "state": "VERIFIED", "summary": {"iw_temperature": 0.151357, "raw": 0.137477, "temperature": 0.144025}, "ties": 3, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:01:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:19Z", "cpu_seconds": 1.6584526299999993, "created_at": "2026-11-22T03:00:16Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c5307ea0853f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "d2577b5f1b0bf1437c0eec3a1857764be015a8ca237cff3d98d23e057d2d107d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c5307ea0853f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:16Z", "seed": 1494074474, "treatment": "iw_temperature"}, "priority": 0.5124300432749596, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "08b0ca9bfad1281fb0b46930301565d2146ec91dc3d2313f1c4c06d647d63617", "run_seconds": 2.1934797763824463, "seed": 1494074474, "started_at": "2026-11-22T03:00:17Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.025221, 0.009116], "effect": -0.008052, "evaluated_at": "2026-11-22T03:00:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c5307ea0853f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.130373332, 0.270278005, 0.269125983, 0.241465825, 0.290325588, 0.197411602, 0.227390723, 0.295080565, 0.318154215, 0.225088895, 0.28463783, 0.295364027, 0.281594012, 0.229590945, 0.27997213, 0.269401295], "raw": [0.227057075, 0.264157635, 0.171538262, 0.179562095, 0.290185225, 0.164322082, 0.193532755, 0.279879312, 0.268985095, 0.235395545, 0.24906465, 0.256427652, 0.281594012, 0.229590945, 0.217206692, 0.230049233], "temperature": [0.187276555, 0.270278005, 0.187160043, 0.196246407, 0.309836505, 0.197411602, 0.227390723, 0.291179285, 0.290889128, 0.264381672, 0.278330167, 0.23862049, 0.29389446, 0.242442525, 0.252259245, 0.248818823]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "d2577b5f1b0bf1437c0eec3a1857764be015a8ca237cff3d98d23e057d2d107d", "primary_control": "temperature", "result_hash": "08b0ca9bfad1281fb0b46930301565d2146ec91dc3d2313f1c4c06d647d63617", "signature": "b08e10773c29a2fbdf52d986f161441ce3edca7502fe4d254e7f3a5f887469666ddad94b6138b895d64e6c74e00cd681725d89159e4b2c4113a0a5cb07192f05", "state": "VERIFIED", "summary": {"iw_temperature": 0.256578, "raw": 0.233659, "temperature": 0.248526}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:00:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:55Z", "cpu_seconds": 1.9340023299999984, "created_at": "2026-11-22T03:00:16Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c58428873e63", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "42f3ea57f554eb447085afbdda15cdb8a87edb520440a94aed0780c74df4d0b3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c58428873e63", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:16Z", "seed": 1044103558, "treatment": "iw_temperature"}, "priority": 0.4033996325033293, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "aed36081c1e3d3ef5a09310d7a2794dda538e51179684773a9646a4c35a4dd3e", "run_seconds": 2.124546766281128, "seed": 1044103558, "started_at": "2026-11-22T03:00:53Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011398, 0.019405], "effect": 0.004003, "evaluated_at": "2026-11-22T03:00:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c58428873e63", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.074785775, 0.09478407, 0.096179455, 0.058262762, 0.10571713, 0.148390678, 0.081598955, 0.171613863, 0.13716662, 0.06739293, 0.154391462, 0.096049375, 0.14679883, 0.112119595, 0.224005795, 0.153105185], "raw": [0.094875515, 0.13598508, 0.122735397, 0.058262762, 0.097077035, 0.09323403, 0.06537573, 0.135999337, 0.11894054, 0.06739293, 0.145936718, 0.071171607, 0.121473165, 0.124521457, 0.113302103, 0.11061346], "temperature": [0.117702797, 0.144892978, 0.138341098, 0.065124055, 0.140938463, 0.153544478, 0.056855502, 0.150920652, 0.11894054, 0.048880458, 0.149671475, 0.13243396, 0.14679883, 0.12511039, 0.159779842, 0.1364809]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "42f3ea57f554eb447085afbdda15cdb8a87edb520440a94aed0780c74df4d0b3", "primary_control": "temperature", "result_hash": "aed36081c1e3d3ef5a09310d7a2794dda538e51179684773a9646a4c35a4dd3e", "signature": "75ad2cea73e186c6e67af0f73528c30bd6f5c3f26ca2c49912cd7b4504616be1bf946fe92b95ef7686cc81f9f50aba9fac5e033756cf17a62f977189eb62920a", "state": "VERIFIED", "summary": {"iw_temperature": 0.120148, "raw": 0.104806, "temperature": 0.124151}, "ties": 1, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:00:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:00Z", "cpu_seconds": 1.6849066600000242, "created_at": "2026-11-22T03:00:16Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-e0ad50ff1c86", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "0647b1d72564f2f3035df12ef71581cbcf481b7cc2866fb6816dbfb8dfd2603e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-e0ad50ff1c86", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:16Z", "seed": 1804930454, "treatment": "iw_temperature"}, "priority": 0.381953831710315, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-7537704cf190", "result_hash": "d84eca06693133171ac7d31674f5593227600c0a965d433a26248a18bfd08766", "run_seconds": 1.7782773971557617, "seed": 1804930454, "started_at": "2026-11-22T03:01:58Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010197, 0.020712], "effect": 0.005258, "evaluated_at": "2026-11-22T03:02:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-e0ad50ff1c86", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.134630478, 0.140174525, 0.121911655, 0.110534043, 0.143158288, 0.171046247, 0.065222785, 0.126598265, 0.151402865, 0.135091032, 0.06399368, 0.18872408, 0.19111843, 0.147165925, 0.126361195, 0.152789872], "raw": [0.134630478, 0.160002177, 0.078502365, 0.142385802, 0.07797936, 0.132066483, 0.102476348, 0.191697005, 0.144710375, 0.117252365, 0.12429126, 0.100558062, 0.155485542, 0.12462635, 0.131659785, 0.118231035], "temperature": [0.15823944, 0.153580097, 0.132095868, 0.073126507, 0.120397847, 0.163527253, 0.090345615, 0.191697005, 0.14751541, 0.135091032, 0.129386247, 0.138100152, 0.177265987, 0.134326685, 0.144016262, 0.165334828]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "0647b1d72564f2f3035df12ef71581cbcf481b7cc2866fb6816dbfb8dfd2603e", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-7537704cf190", "result_hash": "d84eca06693133171ac7d31674f5593227600c0a965d433a26248a18bfd08766", "signature": "1f8404aa0758a151688e066b085b2f840c0db2c6df712121aaf5bc81ace752d52070d765c634ba76e50314da80ad738974d9779486c8de770ec58b6489e38a03", "state": "VERIFIED", "summary": {"iw_temperature": 0.13562, "raw": 0.127285, "temperature": 0.140878}, "ties": 1, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:02:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:25Z", "cpu_seconds": 7.246813482000007, "created_at": "2026-11-22T03:00:16Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-8fe1caf6c7ff", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7d5e97b93871429aa9129203d0e35ef0029b19962958ad678cc1d06109af4ee4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-8fe1caf6c7ff", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:00:16Z", "seed": 687587391, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "ad016ba73124c669b8e392e2526e1f1c7486954abcfeb34c98d7eaa54db41ae6", "run_seconds": 7.791571855545044, "seed": 687587391, "started_at": "2026-11-22T03:00:17Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.137158, 0.070492], "effect": -0.033333, "evaluated_at": "2026-11-22T03:00:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-8fe1caf6c7ff", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6827516717382824, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0819, "losses": 6, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.828", "prereg_hash": "7d5e97b93871429aa9129203d0e35ef0029b19962958ad678cc1d06109af4ee4", "primary_control": "reset", "result_hash": "ad016ba73124c669b8e392e2526e1f1c7486954abcfeb34c98d7eaa54db41ae6", "signature": "308ef34a15cc9905dfbcc5df4975c3880bca312207da1713cfa9360d2bf121bc132680aebf2dac042343aa753620522d6d7dfea404075296d0305d6074a36d09", "state": "VERIFIED", "summary": {"inherit": 0.6833333333333333, "reset": 0.7166666666666667, "root": 0.4}, "ties": 50, "treatment": "inherit", "wins": 4}, "verified_at": "2026-11-22T03:00:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:24Z", "cpu_seconds": 1.4318644210000002, "created_at": "2026-11-22T03:00:33Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-03a1cb12fe67", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "5b8891f378cd9895d3bb80938142d6c4d3e53d98279947d82f4ca8555904dabf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-03a1cb12fe67", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:33Z", "seed": 1218151650, "treatment": "iw_temperature"}, "priority": 0.369712867901093, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ddbe44340a130117df809e3074ef6feafd4b7a645502d33a10fd4226dbfc61a9", "run_seconds": 1.493241786956787, "seed": 1218151650, "started_at": "2026-11-22T03:02:23Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.029517, 0.001032], "effect": -0.014242, "evaluated_at": "2026-11-22T03:02:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-03a1cb12fe67", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.213627162, 0.267443505, 0.218725353, 0.296113355, 0.310021048, 0.246601265, 0.274172785, 0.250231778, 0.182577915, 0.231611253, 0.15867657, 0.263186288, 0.273291503, 0.227000252, 0.305624922, 0.250598402], "raw": [0.174229935, 0.267443505, 0.21097956, 0.209913552, 0.310021048, 0.205182355, 0.23597604, 0.23940612, 0.198695443, 0.200089905, 0.180401458, 0.251763148, 0.193511598, 0.18408295, 0.239413813, 0.25765524], "temperature": [0.235273947, 0.243022697, 0.250401868, 0.25013361, 0.302369308, 0.225479232, 0.24698674, 0.234112082, 0.205997945, 0.21016899, 0.180401458, 0.248504663, 0.203638902, 0.197255563, 0.239413813, 0.268466548]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.017, "losses": 11, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.962", "prereg_hash": "5b8891f378cd9895d3bb80938142d6c4d3e53d98279947d82f4ca8555904dabf", "primary_control": "temperature", "result_hash": "ddbe44340a130117df809e3074ef6feafd4b7a645502d33a10fd4226dbfc61a9", "signature": "de7b2063a99d864629dca40d92027cf9b0d39605de7be836249aec49137f3f77105d37cbf3dc4119878a3cc5f95d0886cc9ac09d43e29cdf25fbfecbf107e101", "state": "VERIFIED", "summary": {"iw_temperature": 0.248094, "raw": 0.222423, "temperature": 0.233852}, "ties": 0, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:02:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:48Z", "cpu_seconds": 1.728507602999997, "created_at": "2026-11-22T03:00:33Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5517e17c136a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "e5e1e6fd3d61dd47c0bda4ca769def7ea2183d886ae6df30f29b851391d00608", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5517e17c136a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:33Z", "seed": 656389341, "treatment": "iw_temperature"}, "priority": 0.40370888848554776, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "536a567da388776966e1c364a26c9a35048fd06abdb1bc1ad8ee4aad9b3a1b93", "run_seconds": 1.8275814056396484, "seed": 656389341, "started_at": "2026-11-22T03:00:46Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011603, 0.006201], "effect": -0.002701, "evaluated_at": "2026-11-22T03:00:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5517e17c136a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.046772735, 0.080439858, 0.053802178, 0.075499643, 0.03811209, 0.051164773, 0.017947177, 0.047280495, 0.038385833, 0.04380539, 0.04468275, 0.023452343, 0.037744587, 0.044455505, 0.080294535, 0.062465885], "raw": [0.078585108, 0.052137355, 0.053802178, 0.063347295, 0.03811209, 0.049020263, 0.02338272, 0.07423128, 0.059076608, 0.04096769, 0.074152093, 0.0327984, 0.053320922, 0.049450203, 0.060244778, 0.045555017], "temperature": [0.078585108, 0.052137355, 0.070004333, 0.075499643, 0.062736928, 0.040089218, 0.012818307, 0.047280495, 0.051251595, 0.048033195, 0.042653458, 0.023452343, 0.02654675, 0.021705337, 0.050181213, 0.040118127]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "e5e1e6fd3d61dd47c0bda4ca769def7ea2183d886ae6df30f29b851391d00608", "primary_control": "temperature", "result_hash": "536a567da388776966e1c364a26c9a35048fd06abdb1bc1ad8ee4aad9b3a1b93", "signature": "08fa593192e9bf2dc6c9589635fce6958fa52437756888444327134bc83c1da010f0b5f5e6579f28ba985a88ade1c3c0d4158325f9f8d023ad0a2714afa8b702", "state": "VERIFIED", "summary": {"iw_temperature": 0.049144, "raw": 0.053011, "temperature": 0.046443}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:00:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:02Z", "cpu_seconds": 1.7534404029999848, "created_at": "2026-11-22T03:00:33Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-7ab266962501", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "a00af7d7f435a6deec4d8cdd745f8e3d0c5fc7e508fd3410b8d1f4278565eabc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-7ab266962501", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:33Z", "seed": 1032591765, "treatment": "iw_temperature"}, "priority": 0.3992287215977097, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "29cf10d0034a8d29b32d51562f78e6b7bf21692740844be979f89c75c65da1b5", "run_seconds": 1.8700222969055176, "seed": 1032591765, "started_at": "2026-11-22T03:01:00Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.007678, 0.011595], "effect": 0.001959, "evaluated_at": "2026-11-22T03:01:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-7ab266962501", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.215738065, 0.184602075, 0.234565095, 0.228596867, 0.23048102, 0.18718854, 0.270665612, 0.24983754, 0.214702417, 0.232800115, 0.180259613, 0.171640352, 0.237950125, 0.1796004, 0.25703361, 0.269209947], "raw": [0.230198857, 0.148117715, 0.234565095, 0.203318055, 0.187916922, 0.200239058, 0.22284555, 0.200988595, 0.21012752, 0.192952947, 0.13018056, 0.171640352, 0.210001355, 0.149245237, 0.200246533, 0.17378525], "temperature": [0.262970723, 0.184602075, 0.26668361, 0.228596867, 0.23048102, 0.200239058, 0.258207545, 0.235829232, 0.214702417, 0.232800115, 0.143550015, 0.191782572, 0.237950125, 0.1796004, 0.239003032, 0.269209947]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1961, "losses": 4, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.637", "prereg_hash": "a00af7d7f435a6deec4d8cdd745f8e3d0c5fc7e508fd3410b8d1f4278565eabc", "primary_control": "temperature", "result_hash": "29cf10d0034a8d29b32d51562f78e6b7bf21692740844be979f89c75c65da1b5", "signature": "128698dfc1ce71644b2b948c9b39838a78260dc73a035dee58963ae0806388e6fb0205adc86de2da19c3e30c74644949e26d4f3e1c3834295ea47324359a6301", "state": "VERIFIED", "summary": {"iw_temperature": 0.221554, "raw": 0.191648, "temperature": 0.223513}, "ties": 8, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:01:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:40Z", "cpu_seconds": 1.5541934190000006, "created_at": "2026-11-22T03:00:33Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-80422121f721", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "4b4d2a9fd32ddca56d44dbfbe51cf33c94aa36c8785af4459f70c415f9802e1a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-80422121f721", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:33Z", "seed": 621970619, "treatment": "iw_temperature"}, "priority": 0.49323971620078516, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "55b056bfbc4518a276d767487df41e7e4194c754fdd616991ad28a2bfe6ffa93", "run_seconds": 1.5711989402770996, "seed": 621970619, "started_at": "2026-11-22T03:00:38Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.024631, 0.00441], "effect": -0.01011, "evaluated_at": "2026-11-22T03:00:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-80422121f721", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.170364422, 0.279911617, 0.261998703, 0.285503377, 0.225685198, 0.295840472, 0.22517624, 0.223179365, 0.288515647, 0.252787347, 0.193358995, 0.23766551, 0.30405136, 0.24409448, 0.316973617, 0.215979513], "raw": [0.18461293, 0.262562947, 0.179945713, 0.234767298, 0.15923197, 0.224132088, 0.19862081, 0.259609508, 0.27431609, 0.16992247, 0.17241133, 0.21344131, 0.302905575, 0.237289165, 0.269124198, 0.179537323], "temperature": [0.145670415, 0.279911617, 0.210560665, 0.251713297, 0.184425985, 0.268859325, 0.24243425, 0.271315442, 0.279969382, 0.202573105, 0.239911563, 0.23766551, 0.304586662, 0.241172415, 0.29859254, 0.199955983]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "4b4d2a9fd32ddca56d44dbfbe51cf33c94aa36c8785af4459f70c415f9802e1a", "primary_control": "temperature", "result_hash": "55b056bfbc4518a276d767487df41e7e4194c754fdd616991ad28a2bfe6ffa93", "signature": "2bbb47185f8a06c25e18727d130c5b5403f00acdbc37c7b867dd73323cd44b7754cb1242dfc82e1be0a27336602697608464a4fc5b384571286981eeed801605", "state": "VERIFIED", "summary": {"iw_temperature": 0.251318, "raw": 0.220152, "temperature": 0.241207}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:00:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:02Z", "cpu_seconds": 1.626837866999999, "created_at": "2026-11-22T03:00:33Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-84b736447c3e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "5f9fc713ae0ea23666b966b67abce41ea5951ae3dc571058d26fb4ead6eb594e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-84b736447c3e", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:33Z", "seed": 360967244, "treatment": "iw_temperature"}, "priority": 0.39881424256616116, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "7c8862b8879597048b1e462e2122a38732ca846d2743ca2e8a3002cc44c928cf", "run_seconds": 1.7074251174926758, "seed": 360967244, "started_at": "2026-11-22T03:01:00Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.030722, 0.018345], "effect": -0.006188, "evaluated_at": "2026-11-22T03:01:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-84b736447c3e", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.291010928, 0.240788497, 0.29843606, 0.132136788, 0.254321757, 0.284259468, 0.279203382, 0.321462355, 0.26544001, 0.290840415, 0.285541615, 0.322961132, 0.305947677, 0.248679075, 0.044915473, 0.266787832], "raw": [0.282519857, 0.216600902, 0.31649667, 0.228785215, 0.238469168, 0.233260117, 0.178124718, 0.248066407, 0.26869144, 0.25643209, 0.241440088, 0.322852128, 0.26578908, 0.221048163, 0.20198212, 0.258203102], "temperature": [0.291010928, 0.216600902, 0.308652667, 0.181961708, 0.238469168, 0.233260117, 0.245535207, 0.293376385, 0.260192827, 0.273659713, 0.22866167, 0.310172173, 0.29174999, 0.21525212, 0.20198212, 0.243180053]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0016, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.996", "prereg_hash": "5f9fc713ae0ea23666b966b67abce41ea5951ae3dc571058d26fb4ead6eb594e", "primary_control": "temperature", "result_hash": "7c8862b8879597048b1e462e2122a38732ca846d2743ca2e8a3002cc44c928cf", "signature": "234e13d874be8ed17f703bf108e3467fb7ce013093644e16c95b992e70cbf17abca23b088e9608606374093bf1544d0d59cd65ea9e7e592387433bbd5597650e", "state": "VERIFIED", "summary": {"iw_temperature": 0.258296, "raw": 0.248673, "temperature": 0.252107}, "ties": 1, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:01:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:41Z", "cpu_seconds": 7.897804391999983, "created_at": "2026-11-22T03:00:33Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-12a52988db09", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d14bbb6bd7363515e11fd5bbba79d51adb12cd2830b6d9d7a56daed8b03a6795", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-12a52988db09", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:00:33Z", "seed": 1680230027, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a43b6c21830e857a4a43ea238896e02a63e0d64de3c0a15a5615aa7eb753b06d", "run_seconds": 8.036461353302002, "seed": 1680230027, "started_at": "2026-11-22T03:00:33Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:00:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-12a52988db09", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d14bbb6bd7363515e11fd5bbba79d51adb12cd2830b6d9d7a56daed8b03a6795", "primary_control": "reset", "result_hash": "a43b6c21830e857a4a43ea238896e02a63e0d64de3c0a15a5615aa7eb753b06d", "signature": "61763c3532ab6fff4289bf8e62ee0c571f0f653bde6d13bb432e75193f6070c5873784beaa9047e5d257f52349f9e6ae7440606db00c5343b89fa0217f912b0c", "state": "PARTIAL", "summary": {"inherit": 0.5333333333333333, "reset": 0.5333333333333333, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:00:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:06Z", "cpu_seconds": 1.4175856210000006, "created_at": "2026-11-22T03:00:49Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-011ed4b03c4b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "1ce6bb73f8deca03e7a85dedfbe7f514f1cb1632d88fe0bce30333bc296d6f66", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-011ed4b03c4b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:49Z", "seed": 1277176200, "treatment": "iw_temperature"}, "priority": 0.37933443872485134, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "6e9e63187f671740874f1bf5e905dfe03c92f554ff952b9b86f306c2aa69af66", "run_seconds": 1.4212782382965088, "seed": 1277176200, "started_at": "2026-11-22T03:02:05Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018936, -0.001827], "effect": -0.010382, "evaluated_at": "2026-11-22T03:02:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-011ed4b03c4b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.092245588, 0.098405197, 0.008834703, 0.14697462, 0.013897207, 0.048786293, 0.017498395, 0.119685747, 0.055182065, 0.05062184, 0.017543098, 0.06167385, 0.179031015, 0.091125615, 0.011343565, 0.07658782], "raw": [0.080050257, 0.060067985, 0.032144092, 0.093681618, 0.02844764, 0.05173916, 0.007021383, 0.105174223, 0.06009807, 0.03573703, 0.042216905, 0.06167385, 0.119213648, 0.065849027, 0.03044417, 0.070160975], "temperature": [0.092245588, 0.067282133, 0.014450722, 0.115328448, 0.019032645, 0.04522145, 0.007021383, 0.112384945, 0.056061615, 0.030506098, 0.019554953, 0.06167385, 0.121878848, 0.091125615, 0.013296077, 0.056267907]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "1ce6bb73f8deca03e7a85dedfbe7f514f1cb1632d88fe0bce30333bc296d6f66", "primary_control": "temperature", "result_hash": "6e9e63187f671740874f1bf5e905dfe03c92f554ff952b9b86f306c2aa69af66", "signature": "2770bd1d4eb6f7674cbecda19ab8239a527cb0b327a8d651dfc410ca8dc708e75b802548e14ac69a3ec955b9335b09891ced3b218f6b57226c18e48c926f9602", "state": "VERIFIED", "summary": {"iw_temperature": 0.06809, "raw": 0.058983, "temperature": 0.057708}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:02:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:55Z", "cpu_seconds": 1.7687507980000134, "created_at": "2026-11-22T03:00:49Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-49fa070948f2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "60c7df2a60acd4d9e1ae47cf7a3bb5761a5852ffe93b42fb3fe4001ebc8aac27", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-49fa070948f2", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:49Z", "seed": 503723161, "treatment": "iw_temperature"}, "priority": 0.46155088787025833, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "193fb6a8047be5e59a96ea34725e98d6a38946abf321bdd9e6eb419132e2b6f1", "run_seconds": 1.8413314819335938, "seed": 503723161, "started_at": "2026-11-22T03:00:53Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.013897, 0.001491], "effect": -0.006203, "evaluated_at": "2026-11-22T03:00:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-49fa070948f2", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.27857443, 0.252935775, 0.330462435, 0.276149093, 0.185893263, 0.250733643, 0.097071857, 0.245653265, 0.312545313, 0.32318046, 0.296795987, 0.316017585, 0.339367682, 0.287516155, 0.293861187, 0.219947415], "raw": [0.220540992, 0.181292043, 0.322953357, 0.263450358, 0.185893263, 0.20611775, 0.117142807, 0.202866218, 0.28847297, 0.313752623, 0.290947863, 0.308692912, 0.333786118, 0.22077686, 0.280373733, 0.219947415], "temperature": [0.26253689, 0.213883722, 0.33052505, 0.281469882, 0.182305525, 0.250733643, 0.12745911, 0.230467178, 0.308239225, 0.320360135, 0.294078005, 0.300480787, 0.340029657, 0.255212687, 0.289731995, 0.219947415]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "60c7df2a60acd4d9e1ae47cf7a3bb5761a5852ffe93b42fb3fe4001ebc8aac27", "primary_control": "temperature", "result_hash": "193fb6a8047be5e59a96ea34725e98d6a38946abf321bdd9e6eb419132e2b6f1", "signature": "5f7bca867d01cfab1410260f6c77df060d82947b79ecf5fe0417ab18978cd708f888d7184266e2ac30f71e9c688b96443d3bca08ddc1946b1ee3a0a677001f0d", "state": "VERIFIED", "summary": {"iw_temperature": 0.269169, "raw": 0.247313, "temperature": 0.262966}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:00:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:39Z", "cpu_seconds": 1.8996058170000083, "created_at": "2026-11-22T03:00:49Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5ffac869b386", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "2ab43d81613755054feb35d23fe9866497de874e4a1a7a80ce8272c40a7d8ca7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5ffac869b386", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:00:49Z", "seed": 1248781440, "treatment": "iw_temperature"}, "priority": 0.3917673474822859, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "096dfbc3d60e003fb0346a16604b3b3af231b3c813f4421e9deb54ff47184791", "run_seconds": 2.004950761795044, "seed": 1248781440, "started_at": "2026-11-22T03:01:36Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012251, 0.002622], "effect": -0.004814, "evaluated_at": "2026-11-22T03:01:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5ffac869b386", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.08238321, 0.048730745, 0.044961917, 0.03310372, 0.070309198, 0.030232732, 0.138195458, 0.064151452, 0.073566397, 0.076150922, 0.067173677, 0.04005088, 0.09686647, 0.036848432, 0.084324405, 0.092435387], "raw": [0.091197815, 0.050962442, 0.050098102, 0.0825329, 0.081675657, 0.044618512, 0.142883707, 0.06715537, 0.067117702, 0.066264175, 0.067173677, 0.04005088, 0.066340583, 0.042385655, 0.095636783, 0.0414612], "temperature": [0.091197815, 0.049330542, 0.036212023, 0.03310372, 0.090235433, 0.030232732, 0.141831603, 0.064151452, 0.067117702, 0.068517328, 0.067173677, 0.031816398, 0.075397055, 0.03106358, 0.083613922, 0.0414612]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "2ab43d81613755054feb35d23fe9866497de874e4a1a7a80ce8272c40a7d8ca7", "primary_control": "temperature", "result_hash": "096dfbc3d60e003fb0346a16604b3b3af231b3c813f4421e9deb54ff47184791", "signature": "12027ec51694750f9077366e9ca72dd7f41a492f1bcdd3fb5936c72156b5e9b45c46632690f196ceb96e4aa73841fd43fb0740c59bb92f21a792ecbf70a84c06", "state": "VERIFIED", "summary": {"iw_temperature": 0.067468, "raw": 0.068597, "temperature": 0.062654}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:01:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:00:57Z", "cpu_seconds": 4.4839135110000115, "created_at": "2026-11-22T03:00:49Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-17237dcf8ef8", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7776e8bf43b67ce56ce40f6cc3f8c21f810a2a54358afc2d5037a6124166a1b6", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-17237dcf8ef8", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:00:49Z", "seed": 174834533, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "67098dc838a12fb99add3b669e1698f5ee559ad72daf4a25bc595f640811e528", "run_seconds": 4.57878565788269, "seed": 174834533, "started_at": "2026-11-22T03:00:53Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:00:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-17237dcf8ef8", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "7776e8bf43b67ce56ce40f6cc3f8c21f810a2a54358afc2d5037a6124166a1b6", "primary_control": "reset", "result_hash": "67098dc838a12fb99add3b669e1698f5ee559ad72daf4a25bc595f640811e528", "signature": "0ad8e3215716c9d0184786d50f9cc0c15e4b7638d473afee8e7b1d902683b77e4fb2f48675d1f926791cea80cd9f1317134f3c1d93b32982940fc494a042a502", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:01:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:18Z", "cpu_seconds": 1.7873082040000057, "created_at": "2026-11-22T03:01:05Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-dc62a965cd6b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "2a4fc1a26fffdee8fe01edb1a03a52284554ca9d60ea14bf4ca8fbdb9c0b2e90", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-dc62a965cd6b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:05Z", "seed": 533467457, "treatment": "iw_temperature"}, "priority": 0.4241518759797689, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "eba2bd2585f603eb7deefce237df9229ee819aec4842c80c0471af95a7c88862", "run_seconds": 2.4632132053375244, "seed": 533467457, "started_at": "2026-11-22T03:01:15Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01314, 0.004043], "effect": -0.004548, "evaluated_at": "2026-11-22T03:01:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-dc62a965cd6b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.149259592, 0.25763695, 0.24104498, 0.233587433, 0.23182764, 0.282124942, 0.238062518, 0.253399185, 0.287241155, 0.260573053, 0.288595485, 0.300415405, 0.239637227, 0.33146396, 0.245361313, 0.254781278], "raw": [0.194175702, 0.242432343, 0.26381872, 0.195380783, 0.214613667, 0.253929658, 0.177824688, 0.24420698, 0.219605078, 0.21391476, 0.24422755, 0.277474705, 0.21639218, 0.323834587, 0.211730918, 0.254781278], "temperature": [0.186311062, 0.242432343, 0.269435977, 0.213922632, 0.23182764, 0.273917762, 0.224038143, 0.24420698, 0.272021925, 0.226226575, 0.269071975, 0.298300705, 0.239637227, 0.33074293, 0.245361313, 0.254781278]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0014, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.997", "prereg_hash": "2a4fc1a26fffdee8fe01edb1a03a52284554ca9d60ea14bf4ca8fbdb9c0b2e90", "primary_control": "temperature", "result_hash": "eba2bd2585f603eb7deefce237df9229ee819aec4842c80c0471af95a7c88862", "signature": "f5412755c31776771165d500c67b318303bb51d843ffa3c9a2603fd99ccdf6138873702fe353ff7106ea910307e68661bd704bd1bce4d33a28c4d3dbafa24206", "state": "VERIFIED", "summary": {"iw_temperature": 0.255938, "raw": 0.234271, "temperature": 0.25139}, "ties": 4, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:01:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:35Z", "cpu_seconds": 1.7289228449999996, "created_at": "2026-11-22T03:01:05Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f1adea782567", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "dd85c35a019fe440f4510a8250bef994f6502720d0449d4040d668f9e9f64e61", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f1adea782567", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:05Z", "seed": 1743540289, "treatment": "iw_temperature"}, "priority": 0.35833508561776034, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "66a5f7eaac21336769f8c650a98939e1097d4c505bdd944e2d86b91ea53cd061", "run_seconds": 1.9450793266296387, "seed": 1743540289, "started_at": "2026-11-22T03:02:33Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017327, -0.000743], "effect": -0.009035, "evaluated_at": "2026-11-22T03:02:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f1adea782567", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.02818745, 0.04898774, 0.135321463, 0.068362403, 0.048885655, 0.056618295, 0.03716864, 0.049668167, 0.043779388, 0.038059003, 0.06979962, 0.035258073, 0.059595118, 0.059021263, 0.041720287, 0.059519038], "raw": [0.040831573, 0.052939963, 0.056475287, 0.047631572, 0.047606015, 0.067308438, 0.06325748, 0.056368662, 0.04725254, 0.053222605, 0.034880622, 0.052297978, 0.066652982, 0.05359796, 0.0480679, 0.051031737], "temperature": [0.02818745, 0.04898774, 0.07120714, 0.06047778, 0.03443507, 0.058789722, 0.038066735, 0.039881877, 0.04725254, 0.038059003, 0.041123558, 0.03459239, 0.061840083, 0.05359796, 0.030358215, 0.04853424]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0205, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.954", "prereg_hash": "dd85c35a019fe440f4510a8250bef994f6502720d0449d4040d668f9e9f64e61", "primary_control": "temperature", "result_hash": "66a5f7eaac21336769f8c650a98939e1097d4c505bdd944e2d86b91ea53cd061", "signature": "aec85c04afcd96c1bc28ecb7798e4aed2a0118fce4468eb354be29689cb0b137a10577b76fd6d85204c3067517f4f23fc89150b1d4f3c0a705a67192a893ec04", "state": "VERIFIED", "summary": {"iw_temperature": 0.054997, "raw": 0.052464, "temperature": 0.045962}, "ties": 3, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:02:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:12Z", "cpu_seconds": 5.396694319000005, "created_at": "2026-11-22T03:01:05Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-96e6c3c42607", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "520280af78b0bbe456afcb91d1aeb278cc6886bf6e2234b4bff7d60c6ee35b83", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-96e6c3c42607", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:01:05Z", "seed": 1490537637, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f59239c5e2a8fd6b85cf6ee65db0e703ae29cb90d1f4c3824eff46e7f8faad22", "run_seconds": 5.642853021621704, "seed": 1490537637, "started_at": "2026-11-22T03:01:07Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:01:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-96e6c3c42607", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "520280af78b0bbe456afcb91d1aeb278cc6886bf6e2234b4bff7d60c6ee35b83", "primary_control": "reset", "result_hash": "f59239c5e2a8fd6b85cf6ee65db0e703ae29cb90d1f4c3824eff46e7f8faad22", "signature": "3110a5f49e604bc2c6691d295de38933bc7148ebadfda8eed575f0f21fea182922ef90a693690a1c371dff9f0fd2b605373f233987dc5ef45b530e5e89089f0b", "state": "PARTIAL", "summary": {"inherit": 0.6, "reset": 0.6, "root": 0.21666666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:01:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:10Z", "cpu_seconds": 4.952766769999997, "created_at": "2026-11-22T03:01:05Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d00d64635b99", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "9975031b0fd51812813df0a745534b7e9db602f4d6ebeff3dc27093c75aa17dd", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d00d64635b99", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:01:05Z", "seed": 1506862921, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "837b74d4fdd72cf10e385733b50e35926a1f232847c6a20f580c6c0de58e1b30", "run_seconds": 5.279294013977051, "seed": 1506862921, "started_at": "2026-11-22T03:01:05Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.20744, 0.140774], "effect": -0.033333, "evaluated_at": "2026-11-22T03:01:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d00d64635b99", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0], "reset": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0], "root": [0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1462, "losses": 15, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.714", "prereg_hash": "9975031b0fd51812813df0a745534b7e9db602f4d6ebeff3dc27093c75aa17dd", "primary_control": "reset", "result_hash": "837b74d4fdd72cf10e385733b50e35926a1f232847c6a20f580c6c0de58e1b30", "signature": "e9e63a1bc984531a99a406a84458058aef487591acc394493cdef22ce52f56b8694aec466d81ecb8bf106e7ee8a3f8f417cc8110275957536a82b258df0b6b05", "state": "VERIFIED", "summary": {"inherit": 0.4666666666666667, "reset": 0.5, "root": 0.2833333333333333}, "ties": 32, "treatment": "inherit", "wins": 13}, "verified_at": "2026-11-22T03:01:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:10Z", "cpu_seconds": 2.9317570070000016, "created_at": "2026-11-22T03:01:05Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-148645d4b419", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "9a11bd4591ffd661d349b0eb76e75a2053d99562175ff9f6c2f8e24b231b819d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-148645d4b419", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:01:05Z", "seed": 819381173, "treatment": "frontier"}, "priority": 0.43590347877843, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "9ffd535d225a25014ef0ce3ae65b61deea116ef686c4e400c8e76dcc736c65b1", "run_seconds": 3.281560182571411, "seed": 819381173, "started_at": "2026-11-22T03:01:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.04094, 0.111371], "effect": 0.076155, "evaluated_at": "2026-11-22T03:01:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-148645d4b419", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.875, 0.746835443, 1.0, 0.781914894, 0.784574468, 0.829326923, 0.837662338, 0.836633663, 1.0, 0.824519231, 1.0, 0.840206186], "frontier": [1.0, 1.0, 1.0, 1.0, 0.869680851, 1.0, 1.0, 1.0, 1.0, 0.831730769, 1.0, 1.0], "random_walk": [0.875, 0.819620253, 1.0, 0.92287234, 0.792553191, 0.829326923, 0.944805195, 0.913366337, 1.0, 0.795673077, 1.0, 0.894329897]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "9a11bd4591ffd661d349b0eb76e75a2053d99562175ff9f6c2f8e24b231b819d", "primary_control": "random_walk", "result_hash": "9ffd535d225a25014ef0ce3ae65b61deea116ef686c4e400c8e76dcc736c65b1", "signature": "1bc64a72fb5590ef7db2e8185046c107478d856629929c25a0e9c01c9a04b9657b5f356d175aba71a93f7c9c6ba1b465545089e52d8bc57d672e1669b0a40d09", "state": "VERIFIED", "summary": {"curiosity": 0.863056, "frontier": 0.975118, "random_walk": 0.898962}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:01:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:31Z", "cpu_seconds": 0.31430669100001296, "created_at": "2026-11-22T03:01:05Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-1cc59e0bc5a0", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "a8dc8c85c33c1b7b6857fd34d0e37f92a3f088bd6ab6e9944c3a22a6f49c9f35", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-1cc59e0bc5a0", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:01:05Z", "seed": 1066302865, "treatment": "frontier"}, "priority": 0.39198189601454064, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-b07a21a8e345", "result_hash": "22a27cc12ebe73094b9a0dee892c327dd9824b5e5c7747598eef8d7d83d16e94", "run_seconds": 0.35367631912231445, "seed": 1066302865, "started_at": "2026-11-22T03:01:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.020653, 0.038], "effect": 0.029327, "evaluated_at": "2026-11-22T03:01:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-1cc59e0bc5a0", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.909274194, 0.863839286, 0.863247863, 0.898305085, 0.860869565, 0.872881356, 0.926923077, 0.88034188, 0.889344262, 0.904471545, 0.839912281, 0.86875], "frontier": [0.941532258, 0.90625, 0.903846154, 0.921610169, 0.904347826, 0.898305085, 0.930769231, 0.908119658, 0.924180328, 0.93902439, 0.901315789, 0.914583333], "random_walk": [0.909274194, 0.861607143, 0.865384615, 0.896186441, 0.860869565, 0.86440678, 0.934615385, 0.905982906, 0.893442623, 0.904471545, 0.872807018, 0.872916667]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.4984, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00317", "prereg_hash": "a8dc8c85c33c1b7b6857fd34d0e37f92a3f088bd6ab6e9944c3a22a6f49c9f35", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-b07a21a8e345", "result_hash": "22a27cc12ebe73094b9a0dee892c327dd9824b5e5c7747598eef8d7d83d16e94", "signature": "83a2063d69b7971e1baa24f8b609deda7a095dab7ef965eaf3afc81344963b34e92fbd08af0de242311d3de16bd669b22e79f59d1a2c361842551e1682b6790d", "state": "VERIFIED", "summary": {"curiosity": 0.881513, "frontier": 0.916157, "random_walk": 0.88683}, "ties": 0, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:01:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:23Z", "cpu_seconds": 1.8500043039999952, "created_at": "2026-11-22T03:01:21Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-805eaad352f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "76cda358244d979f12a573d418fe1f9246985a467ac10f0bb8c8edf6e8fb3bb3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-805eaad352f0", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:21Z", "seed": 915034572, "treatment": "iw_temperature"}, "priority": 0.3123035210020101, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "9cd0305daffbe4da82efb163f4f064c48db6cdd9d9f6f49bff6013ab6102d5d9", "run_seconds": 1.9951281547546387, "seed": 915034572, "started_at": "2026-11-22T03:03:21Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.036079, -0.007302], "effect": -0.021691, "evaluated_at": "2026-11-22T03:03:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-805eaad352f0", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.02081209, 0.046419835, 0.05638872, 0.01666382, 0.065022455, 0.05702516, 0.034303835, 0.038050847, 0.027671585, 0.033917955, 0.112872765, 0.122536492, 0.115586792, 0.074012517, 0.04020361, 0.093651313], "raw": [0.026136235, 0.075654107, 0.04764662, 0.015282915, 0.044378325, 0.05574916, 0.04738336, 0.026408345, 0.027671585, 0.09273597, 0.04550687, 0.069117212, 0.047032127, 0.058952483, 0.052380642, 0.04676326], "temperature": [0.01986706, 0.06003832, 0.033327668, 0.015282915, 0.039367097, 0.05702516, 0.02894218, 0.021688653, 0.041929697, 0.028016045, 0.039181095, 0.069117212, 0.047032127, 0.04425908, 0.03772323, 0.025289685]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0002, "losses": 13, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "76cda358244d979f12a573d418fe1f9246985a467ac10f0bb8c8edf6e8fb3bb3", "primary_control": "temperature", "result_hash": "9cd0305daffbe4da82efb163f4f064c48db6cdd9d9f6f49bff6013ab6102d5d9", "signature": "10023b5173c7e71b32c26808ca4b3fba1ff1339773f3fa91d831687d3530bf2819db800c311a330134cf1c17a6b44b0e8cf43bb35b81e84b2c10177fb52f180c", "state": "VERIFIED", "summary": {"iw_temperature": 0.059696, "raw": 0.048675, "temperature": 0.038005}, "ties": 1, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:03:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:26Z", "cpu_seconds": 1.6661781949999863, "created_at": "2026-11-22T03:01:21Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c79ee1e13ff4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "6abad33a456b031d09020ca3aeb01875d2ea28aa166bac1e26daeae50596865f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c79ee1e13ff4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:21Z", "seed": 1012861281, "treatment": "iw_temperature"}, "priority": 0.3990157988388038, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a251d705dacbecd9e33172b10ca61909495e655dbfab3df5fe3c8b0cb87100d4", "run_seconds": 1.6931240558624268, "seed": 1012861281, "started_at": "2026-11-22T03:01:24Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012136, 0.00758], "effect": -0.002278, "evaluated_at": "2026-11-22T03:01:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c79ee1e13ff4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.167336932, 0.162762975, 0.117468303, 0.10685429, 0.157196053, 0.11727572, 0.17326955, 0.088104745, 0.130472977, 0.113148275, 0.177270737, 0.116305408, 0.105393735, 0.108200048, 0.136629543, 0.135714618], "raw": [0.119766657, 0.135870022, 0.105784787, 0.066829395, 0.139610275, 0.10544273, 0.15007219, 0.067006917, 0.06619685, 0.095930317, 0.169770185, 0.092150875, 0.115474085, 0.069079903, 0.111759327, 0.103147923], "temperature": [0.142496272, 0.143662345, 0.135713947, 0.073785143, 0.14388105, 0.13446309, 0.165190243, 0.109169578, 0.101455412, 0.12395676, 0.196330148, 0.099877985, 0.115474085, 0.108200048, 0.16458873, 0.118716455]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1572, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.696", "prereg_hash": "6abad33a456b031d09020ca3aeb01875d2ea28aa166bac1e26daeae50596865f", "primary_control": "temperature", "result_hash": "a251d705dacbecd9e33172b10ca61909495e655dbfab3df5fe3c8b0cb87100d4", "signature": "2fb52294085d948ba13a52721f71784b1ec1253da9dc1770c659bb0020bb511a8a3b91cd1e8f1ca6caaadf65da143dd631e5baf03db8aba369fd8b508dfb2b0f", "state": "VERIFIED", "summary": {"iw_temperature": 0.132088, "raw": 0.107118, "temperature": 0.12981}, "ties": 1, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:01:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:29Z", "cpu_seconds": 5.343815727000006, "created_at": "2026-11-22T03:01:21Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-632702f2d789", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "8b69309bcc4a46c56fc976992f7c2cefb3ff77af8506ee9beea80ddc7d5dcfd5", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-632702f2d789", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:01:21Z", "seed": 1432869704, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "6ddeab4ec68c063815b1e42f9c837c00e899f34733ddfe857706d2f4c4e05550", "run_seconds": 5.680409669876099, "seed": 1432869704, "started_at": "2026-11-22T03:01:23Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.10617, 0.172837], "effect": 0.033333, "evaluated_at": "2026-11-22T03:01:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-632702f2d789", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.29384112948631064, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0], "reset": [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.3901, "losses": 8, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.407", "prereg_hash": "8b69309bcc4a46c56fc976992f7c2cefb3ff77af8506ee9beea80ddc7d5dcfd5", "primary_control": "reset", "result_hash": "6ddeab4ec68c063815b1e42f9c837c00e899f34733ddfe857706d2f4c4e05550", "signature": "ff58aee20066b4cdc20080ea777e2f91497ffcfd99652dd6a90c480717365fcc3ff75d28ac0b4e90e3dde0555b3a17b2de75507fac257354505a73a8cab5420c", "state": "VERIFIED", "summary": {"inherit": 0.6666666666666666, "reset": 0.6333333333333333, "root": 0.43333333333333335}, "ties": 42, "treatment": "inherit", "wins": 10}, "verified_at": "2026-11-22T03:01:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:26Z", "cpu_seconds": 3.2210550149999975, "created_at": "2026-11-22T03:01:21Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-6711b80dbbf7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "4a0cdce09ea6869341742160fc7d30c02fba5bd2502c4b8856f6fcd231317f7c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-6711b80dbbf7", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:01:21Z", "seed": 717372504, "treatment": "frontier"}, "priority": 0.3995953455826136, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "380d3fa29f614003b8d8dc611566d4f9b483c9cfbf1834cfa727c1a1b186e0dd", "run_seconds": 3.293821096420288, "seed": 717372504, "started_at": "2026-11-22T03:01:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.053751, 0.126103], "effect": 0.089927, "evaluated_at": "2026-11-22T03:01:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-6711b80dbbf7", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.841584158, 0.791666667, 0.873376623, 0.806818182, 1.0, 1.0, 0.77020202, 0.866666667, 0.813333333, 1.0, 0.849514563, 0.8125], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.987373737, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.923267327, 0.8125, 0.886363636, 0.869318182, 1.0, 1.0, 0.878787879, 0.944444444, 0.843333333, 1.0, 0.861650485, 0.888586957]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "4a0cdce09ea6869341742160fc7d30c02fba5bd2502c4b8856f6fcd231317f7c", "primary_control": "random_walk", "result_hash": "380d3fa29f614003b8d8dc611566d4f9b483c9cfbf1834cfa727c1a1b186e0dd", "signature": "e7e0bcfed89d6fe38fc20f21e93b7dbc57f260a4baad0947fc44f05c755539bb4e54d86cd967343978abcc703450334246cf1adca386ad9c3e6d0403aeb62f0b", "state": "VERIFIED", "summary": {"curiosity": 0.868805, "frontier": 0.998948, "random_walk": 0.909021}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:01:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:43Z", "cpu_seconds": 1.8961246620000054, "created_at": "2026-11-22T03:01:38Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-35a72cf0a0fb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "17cc8535cb34c2f948d70764f66fd620def7c1244ca75bd67e26eefead7420f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-35a72cf0a0fb", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:38Z", "seed": 247719449, "treatment": "iw_temperature"}, "priority": 0.3057190941549976, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "d47191aa3f443db638bceecfe03cf835f1a0098659a3ef9188c5b478807f01f4", "run_seconds": 1.9383485317230225, "seed": 247719449, "started_at": "2026-11-22T03:03:41Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014188, 0.000437], "effect": -0.006876, "evaluated_at": "2026-11-22T03:03:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-35a72cf0a0fb", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.211220435, 0.260897825, 0.261416897, 0.227408237, 0.240343488, 0.225285035, 0.202056912, 0.22579034, 0.173593947, 0.215344373, 0.207192745, 0.309087752, 0.22028846, 0.215204935, 0.263592667, 0.153044563], "raw": [0.179635135, 0.239154677, 0.223152323, 0.180294845, 0.192441863, 0.198182655, 0.194751363, 0.180241528, 0.16269489, 0.19567364, 0.174429843, 0.22884498, 0.172523915, 0.215204935, 0.2041223, 0.160520808], "temperature": [0.211220435, 0.220016092, 0.261416897, 0.203737178, 0.22168386, 0.225285035, 0.202056912, 0.22579034, 0.178292857, 0.19567364, 0.218322115, 0.309087752, 0.194146645, 0.228960435, 0.253019412, 0.153044563]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 6, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "17cc8535cb34c2f948d70764f66fd620def7c1244ca75bd67e26eefead7420f4", "primary_control": "temperature", "result_hash": "d47191aa3f443db638bceecfe03cf835f1a0098659a3ef9188c5b478807f01f4", "signature": "898c5542adaa98e6a8bcc29155dd72b028045be52eee158d9056cfc6f2536cfdea4cb0888420acf5016d3d3c360c55cf63249d4c69d7ca5ea575a6a479aaee02", "state": "VERIFIED", "summary": {"iw_temperature": 0.225736, "raw": 0.193867, "temperature": 0.21886}, "ties": 7, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:03:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:29Z", "cpu_seconds": 1.761519695000004, "created_at": "2026-11-22T03:01:38Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4454a1fb6f8e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "bafbac21207bac923ffee2b6af24488eea9388aba3f930470d66dfb65f349c3d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4454a1fb6f8e", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:38Z", "seed": 1180823042, "treatment": "iw_temperature"}, "priority": 0.30996837394671917, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "9c3a5a7e708a3507d1957b1eeba9f18f26ed652c9e45975c782d8da1bc798c37", "run_seconds": 1.8242299556732178, "seed": 1180823042, "started_at": "2026-11-22T03:03:27Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018539, 0.01489], "effect": -0.001824, "evaluated_at": "2026-11-22T03:03:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4454a1fb6f8e", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.21677421, 0.148232128, 0.117914045, 0.181067202, 0.12550393, 0.105103788, 0.188745548, 0.15801932, 0.177155648, 0.112740878, 0.09728801, 0.146770157, 0.072245922, 0.146111075, 0.168608085, 0.129970003], "raw": [0.116456227, 0.076673387, 0.151000547, 0.118139765, 0.087915392, 0.113931553, 0.101000047, 0.144264875, 0.110351702, 0.071368313, 0.09728801, 0.139760403, 0.112612058, 0.092096815, 0.140359305, 0.109977267], "temperature": [0.158611675, 0.127457757, 0.151000547, 0.164674325, 0.106292783, 0.129940807, 0.169685338, 0.144264875, 0.159474152, 0.112740878, 0.1303691, 0.159219902, 0.150487458, 0.092096815, 0.163871928, 0.142871072]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "bafbac21207bac923ffee2b6af24488eea9388aba3f930470d66dfb65f349c3d", "primary_control": "temperature", "result_hash": "9c3a5a7e708a3507d1957b1eeba9f18f26ed652c9e45975c782d8da1bc798c37", "signature": "72a766945c782a2698640b1bb5f80719bc818d85589f1f31933e4733dc4ff432d811dbe3e383e5aed4f8fd1210362fa149544c960c2aa5686fde79160e8dbb06", "state": "VERIFIED", "summary": {"iw_temperature": 0.143266, "raw": 0.11145, "temperature": 0.141441}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:03:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:40Z", "cpu_seconds": 1.6390121579999999, "created_at": "2026-11-22T03:01:38Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5197b72a5372", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "1c4b250026bd074556b6d223acb818308d0dbdc01c91cd95412cf5657b369991", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5197b72a5372", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:38Z", "seed": 1343148747, "treatment": "iw_temperature"}, "priority": 0.41030655801897026, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "c23f41e6ecd92da2b54b18b70762715ca6ced5f2566ca443cac6631dee0cf3c7", "run_seconds": 1.6877248287200928, "seed": 1343148747, "started_at": "2026-11-22T03:01:38Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019824, 0.008565], "effect": -0.00563, "evaluated_at": "2026-11-22T03:01:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5197b72a5372", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.264631048, 0.336677868, 0.306639807, 0.299878927, 0.286506592, 0.29103816, 0.225384093, 0.27392144, 0.265307542, 0.279049037, 0.224428098, 0.266941697, 0.289237737, 0.286974715, 0.158350805, 0.281562327], "raw": [0.18003413, 0.321656747, 0.300644372, 0.235555008, 0.23713523, 0.202361093, 0.219704638, 0.236476423, 0.203373637, 0.233444513, 0.26307861, 0.255077923, 0.20713456, 0.273428713, 0.226783542, 0.243432], "temperature": [0.234875962, 0.328245452, 0.306639807, 0.293884902, 0.276892045, 0.23644111, 0.22723022, 0.26624805, 0.231871833, 0.279049037, 0.272544962, 0.260534897, 0.238634993, 0.28578373, 0.21008035, 0.297498883]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "1c4b250026bd074556b6d223acb818308d0dbdc01c91cd95412cf5657b369991", "primary_control": "temperature", "result_hash": "c23f41e6ecd92da2b54b18b70762715ca6ced5f2566ca443cac6631dee0cf3c7", "signature": "e1bb8f50a0ebdef5403ad52cc6c1edb7d9397faf5ca3b6e2e63fc1f8e468da5b3e54090e52ab0b3e4557d2f80ed0e89986314a289c88f89986c605421e471203", "state": "VERIFIED", "summary": {"iw_temperature": 0.271033, "raw": 0.239958, "temperature": 0.265404}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:01:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:29Z", "cpu_seconds": 1.6197255950000056, "created_at": "2026-11-22T03:01:38Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-89f2c03bce14", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "3da8bbc12a2522235014cfda8b105c8f5a342309010d32ab53dcb00621d80072", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-89f2c03bce14", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:38Z", "seed": 999209015, "treatment": "iw_temperature"}, "priority": 0.30948006220862057, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "6673070850d8d7061883d164334a090d1a769fa0c0124f2fb5d45308a0a10f31", "run_seconds": 1.6699204444885254, "seed": 999209015, "started_at": "2026-11-22T03:03:28Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014204, 0.020417], "effect": 0.003107, "evaluated_at": "2026-11-22T03:03:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-89f2c03bce14", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.265370357, 0.242373975, 0.266802445, 0.302064895, 0.207430562, 0.255691032, 0.24628961, 0.194528035, 0.26661165, 0.2787104, 0.19253164, 0.276297898, 0.324352825, 0.252692103, 0.228576825, 0.253351298], "raw": [0.248846435, 0.268590045, 0.238329782, 0.2917299, 0.303428727, 0.227751718, 0.241878093, 0.208760275, 0.258293603, 0.272911565, 0.235651827, 0.270965108, 0.227865503, 0.23888243, 0.257502453, 0.207016235], "temperature": [0.227141497, 0.268590045, 0.230955133, 0.299148478, 0.31087677, 0.26077495, 0.234931675, 0.223978487, 0.265254323, 0.276744883, 0.217409275, 0.274281388, 0.272424912, 0.252692103, 0.218198175, 0.269980715]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "3da8bbc12a2522235014cfda8b105c8f5a342309010d32ab53dcb00621d80072", "primary_control": "temperature", "result_hash": "6673070850d8d7061883d164334a090d1a769fa0c0124f2fb5d45308a0a10f31", "signature": "863e3e591462a6ff840338a8ebc50a6a4a39883fbca29a695007bb8b0eedb0e3dc272c0410e7f5047aaad032fe75a9e06849cd29a88fd54679f5846500d7bf06", "state": "VERIFIED", "summary": {"iw_temperature": 0.253355, "raw": 0.2499, "temperature": 0.256461}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:03:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:53Z", "cpu_seconds": 8.925557611000016, "created_at": "2026-11-22T03:01:38Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-4c69bff39124", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e8c44045aa3fa84ba66094798564627817c091df881d6ac1fc08532b1a76f4d3", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-4c69bff39124", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:01:38Z", "seed": 1377704456, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "8501acdd10bcc4568f897f75a006d63d7e6838702535ccd2ac6f13840a5aa2d2", "run_seconds": 9.304456949234009, "seed": 1377704456, "started_at": "2026-11-22T03:01:44Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:01:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-4c69bff39124", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e8c44045aa3fa84ba66094798564627817c091df881d6ac1fc08532b1a76f4d3", "primary_control": "reset", "result_hash": "8501acdd10bcc4568f897f75a006d63d7e6838702535ccd2ac6f13840a5aa2d2", "signature": "6a6621963b68130cd32b9e743d374a5f654a3de96209d5e65a74eca23b4d5f350ba7cc87a5faaeb75d60064ba82765531504726e30f8c1ea45cc109dab651605", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.4166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:02:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:01:45Z", "cpu_seconds": 6.299228631000005, "created_at": "2026-11-22T03:01:38Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-f5fb3d11911f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "fb81abe064dbe6b62b228c2c29f70d9400e3c111534b62dcd46022fe06f581a7", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-f5fb3d11911f", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:01:38Z", "seed": 829037444, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "b2a388e00dd092fe242f1db0f826064f65ae53aee85eddd875463eb119a329ca", "run_seconds": 6.401867628097534, "seed": 829037444, "started_at": "2026-11-22T03:01:39Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:01:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-f5fb3d11911f", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0], "root": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fb81abe064dbe6b62b228c2c29f70d9400e3c111534b62dcd46022fe06f581a7", "primary_control": "reset", "result_hash": "b2a388e00dd092fe242f1db0f826064f65ae53aee85eddd875463eb119a329ca", "signature": "2d68ddb33d15b75b59ff2a485d60b4d0ddac2d269332afe33f0929f4723409fbe8c34555357051d5b6557de5d346e7695379cc65f5353ba3b7adabc19aae6a0c", "state": "PARTIAL", "summary": {"inherit": 0.7166666666666667, "reset": 0.7166666666666667, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:01:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:08Z", "cpu_seconds": 2.816685007999993, "created_at": "2026-11-22T03:01:38Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-c1086f932966", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "83430da3ab2f7016f1bc0288c276a42f962ef2f80914f064f6c1112377539736", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-c1086f932966", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:01:38Z", "seed": 1254333982, "treatment": "frontier"}, "priority": 0.3787405611217199, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "3ee180831de640dc754dcd6bb02802610f4f749dfccd4d48f684041202bcfee7", "run_seconds": 3.070934534072876, "seed": 1254333982, "started_at": "2026-11-22T03:02:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.030466, 0.088012], "effect": 0.059239, "evaluated_at": "2026-11-22T03:02:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-c1086f932966", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.833333333, 0.816326531, 0.87962963, 0.864583333, 0.788659794, 1.0, 1.0, 0.9, 1.0, 0.899082569, 0.897727273, 0.879746835], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.997706422, 0.958333333, 1.0], "random_walk": [0.873809524, 0.905612245, 1.0, 0.895833333, 0.871134021, 1.0, 1.0, 0.925, 1.0, 0.931192661, 0.928030303, 0.914556962]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "83430da3ab2f7016f1bc0288c276a42f962ef2f80914f064f6c1112377539736", "primary_control": "random_walk", "result_hash": "3ee180831de640dc754dcd6bb02802610f4f749dfccd4d48f684041202bcfee7", "signature": "dfc5a5b5df34b94d330f75b3dd60c3017c783722af5dd36487376f7101ccf954ca97e49e07aade69f9638db0c13f6c13540f302dddeed928c028e42aff13410a", "state": "VERIFIED", "summary": {"curiosity": 0.896591, "frontier": 0.996337, "random_walk": 0.937097}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:02:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:16Z", "cpu_seconds": 1.6823609480000101, "created_at": "2026-11-22T03:01:54Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-a17074327d38", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "74f6c85d1c6178293585a4b5746bdaa289a8ce3443a8ed038677f74ba7e3bfc1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-a17074327d38", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:01:54Z", "seed": 1665785971, "treatment": "iw_temperature"}, "priority": 0.3149580967821837, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "5c5f8f2a225bf06ff9228545b46d166669f1e35cb505abf6aaefce0e29a718c6", "run_seconds": 1.7333574295043945, "seed": 1665785971, "started_at": "2026-11-22T03:03:14Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.003806, 0.008567], "effect": 0.002381, "evaluated_at": "2026-11-22T03:03:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-a17074327d38", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.06804403, 0.05041391, 0.040188595, 0.034489988, 0.021022827, 0.033098713, 0.137493293, 0.03205416, 0.019944553, 0.068527997, 0.117490013, 0.062297778, 0.216349378, 0.047832578, 0.056050765, 0.055098997], "raw": [0.09526922, 0.05041391, 0.055796898, 0.041937355, 0.047924492, 0.018779457, 0.137493293, 0.03831434, 0.023503065, 0.037978165, 0.095316253, 0.090702807, 0.236565615, 0.047832578, 0.054944555, 0.049233132], "temperature": [0.08129328, 0.055187987, 0.040188595, 0.055866502, 0.047924492, 0.018779457, 0.140854113, 0.047791527, 0.020089463, 0.073474557, 0.103877285, 0.050124993, 0.22443826, 0.039571723, 0.056050765, 0.042973797]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6737, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.212", "prereg_hash": "74f6c85d1c6178293585a4b5746bdaa289a8ce3443a8ed038677f74ba7e3bfc1", "primary_control": "temperature", "result_hash": "5c5f8f2a225bf06ff9228545b46d166669f1e35cb505abf6aaefce0e29a718c6", "signature": "a707f00454d727850197b7bcaebf88425c9caa3c207d8c6f90eb1f6ae6bffedd0d9837898f3fb5049d24543ee5312699b431ba242c0a2d6c3c7735cdbafe5802", "state": "VERIFIED", "summary": {"iw_temperature": 0.066275, "raw": 0.070125, "temperature": 0.068655}, "ties": 2, "treatment": "iw_temperature", "wins": 9}, "verified_at": "2026-11-22T03:03:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:00Z", "cpu_seconds": 2.543486821000002, "created_at": "2026-11-22T03:01:54Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-7bdee73f9866", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "2990633f12d733b2006c761f34745ae3cad3ce228bfe227941ab46f2b9f478e5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-7bdee73f9866", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:01:54Z", "seed": 933436349, "treatment": "frontier"}, "priority": 0.41959841803500664, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-fe4294ea7c9b", "result_hash": "b9eaeedf52fbbaf3db78f0beb44ea8c2ab2a9c110a5d478a453835b1901a600a", "run_seconds": 2.6272876262664795, "seed": 933436349, "started_at": "2026-11-22T03:01:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.023247, 0.098049], "effect": 0.060648, "evaluated_at": "2026-11-22T03:02:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-7bdee73f9866", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.75, 1.0, 0.794444444, 0.77393617, 1.0, 0.823979592, 1.0, 0.834558824, 1.0, 0.886842105, 1.0, 0.843434343], "frontier": [1.0, 1.0, 0.994444444, 0.965425532, 1.0, 1.0, 1.0, 1.0, 1.0, 0.978947368, 1.0, 1.0], "random_walk": [0.838815789, 1.0, 0.819444444, 0.917553191, 1.0, 0.941326531, 1.0, 0.863970588, 1.0, 0.915789474, 1.0, 0.914141414]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.1072, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00781", "prereg_hash": "2990633f12d733b2006c761f34745ae3cad3ce228bfe227941ab46f2b9f478e5", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-fe4294ea7c9b", "result_hash": "b9eaeedf52fbbaf3db78f0beb44ea8c2ab2a9c110a5d478a453835b1901a600a", "signature": "5caae2a23de256497dbe5253e5da7dfba7a80e4c05b585bda903cef879685e14b3daad1ff53bd358187824edd23b100696d75108af7833f853b2cb808c265809", "state": "VERIFIED", "summary": {"curiosity": 0.892266, "frontier": 0.994901, "random_walk": 0.934253}, "ties": 5, "treatment": "frontier", "wins": 7}, "verified_at": "2026-11-22T03:02:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:48Z", "cpu_seconds": 1.1270795070000048, "created_at": "2026-11-22T03:01:54Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-b06c9081c77a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "9d5757847bd8853c371aa5bf7c7bea59ccf9ae181f82b87380f963b3e6801721", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-b06c9081c77a", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:01:54Z", "seed": 2041050514, "treatment": "frontier"}, "priority": 0.32066505570385584, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-933d79bd1d90", "result_hash": "7a78980bf85f0ee252a29f9b43e0fbae82964b4ad5d8824663d8b77624d3596a", "run_seconds": 1.1857521533966064, "seed": 2041050514, "started_at": "2026-11-22T03:02:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.048193, 0.109637], "effect": 0.078915, "evaluated_at": "2026-11-22T03:02:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-b06c9081c77a", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.852040816, 0.911016949, 0.966666667, 1.0, 0.836956522, 1.0, 0.854166667, 0.94, 0.931818182, 0.877118644, 0.955555556, 0.8625], "frontier": [1.0, 1.0, 0.983333333, 1.0, 1.0, 1.0, 0.979166667, 1.0, 1.0, 1.0, 0.988888889, 1.0], "random_walk": [0.857142857, 0.898305085, 0.908333333, 1.0, 0.913043478, 1.0, 0.833333333, 0.945, 0.954545455, 0.93220339, 0.933333333, 0.829166667]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "9d5757847bd8853c371aa5bf7c7bea59ccf9ae181f82b87380f963b3e6801721", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-933d79bd1d90", "result_hash": "7a78980bf85f0ee252a29f9b43e0fbae82964b4ad5d8824663d8b77624d3596a", "signature": "7ad05ad6d4b702cc2ac7075aecea1fab34e46fcb557360eff28e3bb627215dfd73d47a004df6850099a6d2754a26b1e780f95136a5d45b08c6ed999a051ec809", "state": "VERIFIED", "summary": {"curiosity": 0.915653, "frontier": 0.995949, "random_walk": 0.917034}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:02:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:15Z", "cpu_seconds": 1.4064717360000003, "created_at": "2026-11-22T03:02:10Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-05fac24c813d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "3aac2ae4e46cf31262cfa420fd7afbf427322e402a9fcb8df60598f2fcb7b584", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-05fac24c813d", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:10Z", "seed": 326858097, "treatment": "iw_temperature"}, "priority": 0.40430756358375175, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "e89a70b9c91dde91f901553b3fec947de8a6de74eec3996a259d54ba0bda532e", "run_seconds": 1.4519915580749512, "seed": 326858097, "started_at": "2026-11-22T03:02:13Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020155, 0.025348], "effect": 0.002596, "evaluated_at": "2026-11-22T03:02:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-05fac24c813d", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.28192063, 0.322614705, 0.195482225, 0.226242532, 0.208414333, 0.325489542, 0.257325943, 0.18705, 0.0779436, 0.257403045, 0.302328042, 0.254049335, 0.268791387, 0.359776967, 0.198639453, 0.158241205], "raw": [0.250135747, 0.304267862, 0.183616353, 0.154865975, 0.22112857, 0.301709877, 0.210509162, 0.235066215, 0.184938322, 0.240554285, 0.204202357, 0.176228543, 0.22802824, 0.352822015, 0.222126743, 0.14173382], "temperature": [0.28192063, 0.31165349, 0.20135352, 0.177523705, 0.256173188, 0.291393348, 0.27574722, 0.235066215, 0.198275545, 0.240554285, 0.240096935, 0.202613592, 0.248551118, 0.359215485, 0.244873473, 0.158241205]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "3aac2ae4e46cf31262cfa420fd7afbf427322e402a9fcb8df60598f2fcb7b584", "primary_control": "temperature", "result_hash": "e89a70b9c91dde91f901553b3fec947de8a6de74eec3996a259d54ba0bda532e", "signature": "f37c16d59afb08b0e9076d908ebe4998ea53c56ed124e4e194c33a839deb905e15837d4756f230ee221bb824535c0c215a311bd7b1c962ce1efcd9ce7a70ef04", "state": "VERIFIED", "summary": {"iw_temperature": 0.242607, "raw": 0.225746, "temperature": 0.245203}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:02:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:09Z", "cpu_seconds": 1.4025393149999985, "created_at": "2026-11-22T03:02:10Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-6c51266cc711", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "1b52fe2a615b61e388ba76c505813278c28d13d4773cb85b2b684021b6352f0b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-6c51266cc711", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:10Z", "seed": 1609342890, "treatment": "iw_temperature"}, "priority": 0.31626727228445034, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "6b0f82db5b7f317e1e2f6c32f4b5c7b94b6bad1018ed583f90d28522466dd342", "run_seconds": 1.5919971466064453, "seed": 1609342890, "started_at": "2026-11-22T03:03:07Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012796, 0.002202], "effect": -0.005297, "evaluated_at": "2026-11-22T03:03:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-6c51266cc711", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.10416905, 0.03512906, 0.062226072, 0.02804237, 0.077174415, 0.028123298, 0.11825966, 0.12745627, 0.056460102, 0.039739512, 0.083311318, 0.034634685, 0.050267825, 0.075067942, 0.08295122, 0.046964457], "raw": [0.107705173, 0.057377937, 0.033456532, 0.05142173, 0.066184285, 0.05618996, 0.104514835, 0.1219615, 0.056460102, 0.039739512, 0.068416945, 0.05052274, 0.050640075, 0.075245775, 0.075264393, 0.04981252], "temperature": [0.100557455, 0.042774315, 0.014430885, 0.02804237, 0.066055468, 0.05226741, 0.107591762, 0.118041748, 0.043173705, 0.023878288, 0.083311318, 0.034634685, 0.050267825, 0.075067942, 0.091905992, 0.033220145]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "1b52fe2a615b61e388ba76c505813278c28d13d4773cb85b2b684021b6352f0b", "primary_control": "temperature", "result_hash": "6b0f82db5b7f317e1e2f6c32f4b5c7b94b6bad1018ed583f90d28522466dd342", "signature": "3934aec2e6dc3f32e5bcc167215ae03586efcc1edf235b9f7c829e9212aaf16152a8ac5e1aa7609cdca3fab55a4317cd59f86d40d4ce1e94fa44f4495b72f40c", "state": "VERIFIED", "summary": {"iw_temperature": 0.065624, "raw": 0.066557, "temperature": 0.060326}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:03:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:53Z", "cpu_seconds": 1.5046976040000004, "created_at": "2026-11-22T03:02:10Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-7494da0f4c37", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "e0434ed1b2fc39138241c3e362aae810c7ea22355400c6db9b061984ea6e4737", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-7494da0f4c37", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:10Z", "seed": 1754012222, "treatment": "iw_temperature"}, "priority": 0.31790365552249095, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-b493e90e05f5", "result_hash": "f02da25a1d02e5b0222d519e01a9929f8f97cc75b2e5021246d88aa594207648", "run_seconds": 1.5136878490447998, "seed": 1754012222, "started_at": "2026-11-22T03:02:52Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011118, 0.000487], "effect": -0.005316, "evaluated_at": "2026-11-22T03:03:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-7494da0f4c37", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.13990008, 0.093392305, 0.06991755, 0.039948925, 0.07966047, 0.050304852, 0.03922087, 0.131150948, 0.048702385, 0.052887458, 0.067886138, 0.043109115, 0.055355205, 0.08965776, 0.027324077, 0.051706205], "raw": [0.10796213, 0.076925778, 0.07335879, 0.039948925, 0.069978953, 0.071236383, 0.061589785, 0.108247375, 0.05081752, 0.041198923, 0.05988033, 0.042319457, 0.068813685, 0.094026275, 0.05251981, 0.057690145], "temperature": [0.10796213, 0.078674648, 0.07335879, 0.046254125, 0.069978953, 0.044085678, 0.03922087, 0.108247375, 0.05081752, 0.036851643, 0.05988033, 0.043109115, 0.051068528, 0.09891814, 0.039841498, 0.046805285]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "e0434ed1b2fc39138241c3e362aae810c7ea22355400c6db9b061984ea6e4737", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-b493e90e05f5", "result_hash": "f02da25a1d02e5b0222d519e01a9929f8f97cc75b2e5021246d88aa594207648", "signature": "2c6c6a5d46d5dc0b439bccddd4acae3f5d009526260e84083fe62a30b0aa86f48318f72da63cb2f8d45eb384b119511d9ef19906081ab18c918af6ce919d370d", "state": "VERIFIED", "summary": {"iw_temperature": 0.067508, "raw": 0.067282, "temperature": 0.062192}, "ties": 2, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:03:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:18Z", "cpu_seconds": 6.3599159609999845, "created_at": "2026-11-22T03:02:10Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a579a99f6446", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "1d78f387b1255fc83bae29a076da361e5a69ec97e24cb164abe0380094247e7e", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a579a99f6446", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:02:10Z", "seed": 1985775236, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "8a40f65c0afda91532095bf4e6f52792c4f51bd2329f8ef08fcfa47d084bc007", "run_seconds": 6.5509514808654785, "seed": 1985775236, "started_at": "2026-11-22T03:02:11Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 4}, "better": "higher", "ci95": [-0.302032, -0.064634], "effect": -0.183333, "evaluated_at": "2026-11-22T03:02:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a579a99f6446", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4302671633356294, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0002, "losses": 13, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "1d78f387b1255fc83bae29a076da361e5a69ec97e24cb164abe0380094247e7e", "primary_control": "reset", "result_hash": "8a40f65c0afda91532095bf4e6f52792c4f51bd2329f8ef08fcfa47d084bc007", "signature": "d8f99b54bac50e019bb3356368e01cbf55df911ecd27edab38cd5cfcc13d14c38175c5e9723e7171062a1bda36814d1d1f0d2fe7151167ab45b179e18fcd1a04", "state": "VERIFIED", "summary": {"inherit": 0.6666666666666666, "reset": 0.85, "root": 0.36666666666666664}, "ties": 45, "treatment": "inherit", "wins": 2}, "verified_at": "2026-11-22T03:02:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:19Z", "cpu_seconds": 6.372003556999999, "created_at": "2026-11-22T03:02:10Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-f682cb8e1e02", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "4ba9e16facc8f4c228307dd1af9c0101a4edd1d0e806eee8f231c348e15ef258", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-f682cb8e1e02", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:02:10Z", "seed": 761059060, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "7d1dbb52060699a852dcd8b747e9279e38d69e999c4d5ebb4420ab9ccac922e2", "run_seconds": 6.614106178283691, "seed": 761059060, "started_at": "2026-11-22T03:02:13Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.05851, 0.15851], "effect": 0.05, "evaluated_at": "2026-11-22T03:02:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-f682cb8e1e02", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.7749414479786718, "rstep": 3}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0], "reset": [1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0], "root": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.5616, "losses": 4, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.274", "prereg_hash": "4ba9e16facc8f4c228307dd1af9c0101a4edd1d0e806eee8f231c348e15ef258", "primary_control": "reset", "result_hash": "7d1dbb52060699a852dcd8b747e9279e38d69e999c4d5ebb4420ab9ccac922e2", "signature": "b89ee4466c6a76da6ed0e32a6e3c46204eb83a5b29a7f1d932a94d4427e1a4de51a1586c91eded833f850a6d482b331c7a6b2459f3e3800beff09e66a7a8ff02", "state": "VERIFIED", "summary": {"inherit": 0.48333333333333334, "reset": 0.43333333333333335, "root": 0.3}, "ties": 49, "treatment": "inherit", "wins": 7}, "verified_at": "2026-11-22T03:02:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:40Z", "cpu_seconds": 3.2054668650000053, "created_at": "2026-11-22T03:02:10Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-073a626dac3e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "e5172d03c64c4b3ee93456405ba9b3fe021ae483a1dc141e944538d76b27e51c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-073a626dac3e", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:02:10Z", "seed": 238609168, "treatment": "frontier"}, "priority": 0.340062830440561, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "aeb979548799a6c2c1d5a0dd69a3539dc210e8adefed7613396d6de2fced30e4", "run_seconds": 3.651916265487671, "seed": 238609168, "started_at": "2026-11-22T03:02:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.030778, 0.097402], "effect": 0.06409, "evaluated_at": "2026-11-22T03:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-073a626dac3e", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 1.0, 0.868421053, 0.832352941, 0.799479167, 0.830555556, 0.870192308, 1.0, 0.817010309, 0.850490196, 1.0, 0.777777778], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.976804124, 1.0, 1.0, 1.0], "random_walk": [1.0, 1.0, 0.942105263, 0.847058824, 0.950520833, 0.927777778, 0.918269231, 1.0, 0.891752577, 0.901960784, 1.0, 0.828282828]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "e5172d03c64c4b3ee93456405ba9b3fe021ae483a1dc141e944538d76b27e51c", "primary_control": "random_walk", "result_hash": "aeb979548799a6c2c1d5a0dd69a3539dc210e8adefed7613396d6de2fced30e4", "signature": "1b5fa1b1d749c69ffd42d575e1720fcde60f6f3c3ef063e3e229b6544d4eb9946cec5a44ed7024d595684940533accd762df78b634c82f226b1faca0f7299902", "state": "VERIFIED", "summary": {"curiosity": 0.88719, "frontier": 0.998067, "random_walk": 0.933977}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:00Z", "cpu_seconds": 1.4857565769999894, "created_at": "2026-11-22T03:02:26Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-0a020d9b028a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "15d619e7671f43796c7ed9553b3fd849ef6b95be5975aa875595b64d264d7b29", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-0a020d9b028a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:26Z", "seed": 987745849, "treatment": "iw_temperature"}, "priority": 0.3164639742021019, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "21124a21cee19ead86689cd55e3aa1a660f0ab9bfeae65e94957ee0eff6941c2", "run_seconds": 1.5224401950836182, "seed": 987745849, "started_at": "2026-11-22T03:02:58Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018379, -0.001997], "effect": -0.010188, "evaluated_at": "2026-11-22T03:03:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-0a020d9b028a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.20450809, 0.023774595, 0.037906935, 0.088078065, 0.06597178, 0.009112693, 0.03970504, 0.061358347, 0.066077995, 0.045051355, 0.048934042, 0.070964977, 0.05172934, 0.063406115, 0.093555138, 0.115473322], "raw": [0.20496678, 0.01831188, 0.037906935, 0.068018948, 0.050097152, 0.013502595, 0.03970504, 0.028588488, 0.050939697, 0.027231977, 0.067587865, 0.012765452, 0.05172934, 0.079147668, 0.069451118, 0.104009992], "temperature": [0.204115798, 0.02416567, 0.034551742, 0.062367715, 0.057669605, 0.008291887, 0.03970504, 0.01821679, 0.050560607, 0.027231977, 0.056066298, 0.019997868, 0.05172934, 0.065139985, 0.093555138, 0.109237632]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "15d619e7671f43796c7ed9553b3fd849ef6b95be5975aa875595b64d264d7b29", "primary_control": "temperature", "result_hash": "21124a21cee19ead86689cd55e3aa1a660f0ab9bfeae65e94957ee0eff6941c2", "signature": "16b8a77d511b62cb1531d7b11205e6352f22e404540ce1d2dc85c621f0eb129974f16e35f735f09dcebc55b0afa2c3ada9065c74fbde53f6224445f2d78d0404", "state": "VERIFIED", "summary": {"iw_temperature": 0.06785, "raw": 0.057748, "temperature": 0.057663}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:03:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:28Z", "cpu_seconds": 1.6432344480000012, "created_at": "2026-11-22T03:02:26Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-51490405f9cf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "3421e7f749c7d6f902850e1de30af4f145efb5415bcfca136997c3c7e5908179", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-51490405f9cf", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:26Z", "seed": 1984127782, "treatment": "iw_temperature"}, "priority": 0.37172929183590103, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ca7d6445958eee47a2b5fcbee24f9c54a3e71bd6717335ba70af124d3d24d34d", "run_seconds": 1.904891014099121, "seed": 1984127782, "started_at": "2026-11-22T03:02:26Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017114, 0.01465], "effect": -0.001232, "evaluated_at": "2026-11-22T03:02:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-51490405f9cf", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.28399351, 0.225774482, 0.266250313, 0.317511263, 0.228726172, 0.285139028, 0.333404965, 0.32305823, 0.206118517, 0.152638798, 0.302025515, 0.274381945, 0.23210495, 0.276846875, 0.293723388, 0.18885991], "raw": [0.193544112, 0.218999667, 0.258142653, 0.280471625, 0.20085669, 0.281965223, 0.317832658, 0.295948625, 0.215494998, 0.195437143, 0.28046136, 0.225809502, 0.176918425, 0.276511335, 0.272921092, 0.21438898], "temperature": [0.208732162, 0.234955123, 0.289662168, 0.280471625, 0.23926455, 0.285139028, 0.331520927, 0.31202541, 0.272300605, 0.152638798, 0.302025515, 0.250754932, 0.212769332, 0.276630247, 0.283491725, 0.238463677]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "3421e7f749c7d6f902850e1de30af4f145efb5415bcfca136997c3c7e5908179", "primary_control": "temperature", "result_hash": "ca7d6445958eee47a2b5fcbee24f9c54a3e71bd6717335ba70af124d3d24d34d", "signature": "620a4003926079045a0c7a2c9880df725a9a65e9a6a201dab7f3c38ebfc2adc639069647cd1594b5ef1642e79e655b32c6da353b468bbb6c67d0c1481dfa670e", "state": "VERIFIED", "summary": {"iw_temperature": 0.26191, "raw": 0.244107, "temperature": 0.260678}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:02:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:42Z", "cpu_seconds": 1.4197987910000052, "created_at": "2026-11-22T03:02:26Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c487ee48da85", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "f82028c5298bbb585b8df4a5df8a30df5253114663ed78a822a992666b44deff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c487ee48da85", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:26Z", "seed": 1674806291, "treatment": "iw_temperature"}, "priority": 0.3228486232662568, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "07c90df69f6db9193872f9f3b03935da6d08754a2c258228d8f43d8f0d903e0a", "run_seconds": 1.4668214321136475, "seed": 1674806291, "started_at": "2026-11-22T03:02:40Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012322, 0.004517], "effect": -0.003902, "evaluated_at": "2026-11-22T03:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c487ee48da85", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.127232585, 0.12684863, 0.172188527, 0.203026452, 0.131270778, 0.154487698, 0.147827597, 0.163964647, 0.136360585, 0.12673811, 0.17115392, 0.173902398, 0.128141202, 0.136110115, 0.127274635, 0.054976945], "raw": [0.139047358, 0.11538982, 0.164109455, 0.154487678, 0.11041576, 0.12155746, 0.147827597, 0.20336924, 0.121409523, 0.117574882, 0.155497393, 0.173902398, 0.130390112, 0.109082437, 0.09828437, 0.144255665], "temperature": [0.139047358, 0.116699515, 0.170217087, 0.165376618, 0.11424058, 0.129155908, 0.143858642, 0.182838645, 0.127603802, 0.136415162, 0.16588847, 0.15568013, 0.132690975, 0.121066118, 0.133498088, 0.08479198]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0482, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.895", "prereg_hash": "f82028c5298bbb585b8df4a5df8a30df5253114663ed78a822a992666b44deff", "primary_control": "temperature", "result_hash": "07c90df69f6db9193872f9f3b03935da6d08754a2c258228d8f43d8f0d903e0a", "signature": "fcf054bdb636031fa1fedf1f2bc76166699526213161effa0b2f2749eb9ef41497ebdafb625ffe22fb500654ff4a0db9080630ac77c960d6daffcd007db6de06", "state": "VERIFIED", "summary": {"iw_temperature": 0.142594, "raw": 0.137913, "temperature": 0.138692}, "ties": 0, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:36Z", "cpu_seconds": 7.828717679999997, "created_at": "2026-11-22T03:02:26Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-917d333c3df4", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "49e171fab50ed0c052ca2036a9eb03537df8191d16e4b46da192c0837da83184", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-917d333c3df4", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:02:26Z", "seed": 1004659192, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "ce1f36a94c27265b400c884974cbcf06fbaa88b4870f70baded86c29462f0617", "run_seconds": 8.505467414855957, "seed": 1004659192, "started_at": "2026-11-22T03:02:27Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:02:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-917d333c3df4", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "49e171fab50ed0c052ca2036a9eb03537df8191d16e4b46da192c0837da83184", "primary_control": "reset", "result_hash": "ce1f36a94c27265b400c884974cbcf06fbaa88b4870f70baded86c29462f0617", "signature": "bb9ba81a6717344faf9f9c39eae8eb39fecefaa55718d8cf00f6f4609561a8a32677306e405919febf3a702ffc8fd00d34aca57d37aea35e8468f343c5d1b604", "state": "PARTIAL", "summary": {"inherit": 0.5666666666666667, "reset": 0.5666666666666667, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:47Z", "cpu_seconds": 1.619880476999981, "created_at": "2026-11-22T03:02:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-190890b683d7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "f51a7878b0d39adb288637ba76a8444b86c1aeeb4242aaccc28d6cfb37170cf9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-190890b683d7", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:42Z", "seed": 1841372149, "treatment": "iw_temperature"}, "priority": 0.38010926336437695, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "f60c46b491ed7a432b70eabb2fa599b58b7910444d3ed1113a1a27f55ac50d1c", "run_seconds": 1.6212437152862549, "seed": 1841372149, "started_at": "2026-11-22T03:02:45Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.022391, -0.005534], "effect": -0.013962, "evaluated_at": "2026-11-22T03:02:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-190890b683d7", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.248204607, 0.293893735, 0.279004672, 0.285856873, 0.266051483, 0.260368698, 0.257498255, 0.303341502, 0.294701677, 0.072289535, 0.272255685, 0.263691943, 0.31794602, 0.290038798, 0.2444058, 0.245627115], "raw": [0.235880623, 0.288761022, 0.250344293, 0.272189962, 0.234969307, 0.17078152, 0.253882463, 0.252944828, 0.275252813, 0.094359985, 0.207823073, 0.188062215, 0.311944703, 0.257265283, 0.206675225, 0.182670562], "temperature": [0.248204607, 0.293893735, 0.269046532, 0.285856873, 0.234969307, 0.25322083, 0.221028717, 0.284420823, 0.294701677, 0.07629306, 0.232986443, 0.24181842, 0.317558797, 0.278593527, 0.2444058, 0.194779058]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0002, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "f51a7878b0d39adb288637ba76a8444b86c1aeeb4242aaccc28d6cfb37170cf9", "primary_control": "temperature", "result_hash": "f60c46b491ed7a432b70eabb2fa599b58b7910444d3ed1113a1a27f55ac50d1c", "signature": "f8b730136901d2fd4e55ffbb4e0d6efcb77479a75f375215a8417764fc039bb044db36224f77c1b80dd933f84c5a4086df53acdcd6a92990075066c945c74c0d", "state": "VERIFIED", "summary": {"iw_temperature": 0.262199, "raw": 0.230238, "temperature": 0.248236}, "ties": 5, "treatment": "iw_temperature", "wins": 1}, "verified_at": "2026-11-22T03:02:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:59Z", "cpu_seconds": 1.445795094999994, "created_at": "2026-11-22T03:02:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1a5d527dc7d1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "aba546f5560948bb06bb6877609b28cc332bd6ba7b24776549a915b5e8472b5c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1a5d527dc7d1", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:42Z", "seed": 1858121383, "treatment": "iw_temperature"}, "priority": 0.2797556578708844, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "df064a5b7dc518e8aeb40e2a7e6fcaf69537280e7f07a0a21414dc811bd1d34f", "run_seconds": 1.4594604969024658, "seed": 1858121383, "started_at": "2026-11-22T03:03:58Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.038792, 0.002715], "effect": -0.018039, "evaluated_at": "2026-11-22T03:04:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1a5d527dc7d1", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.146041822, 0.23725001, 0.200178642, 0.20731674, 0.284498728, 0.266411322, 0.280280045, 0.27574731, 0.23290919, 0.302871585, 0.236602483, 0.224132558, 0.298045292, 0.174975827, 0.129502968, 0.178904138], "raw": [0.194302313, 0.171440632, 0.146447955, 0.20731674, 0.25920703, 0.22798214, 0.288289795, 0.252361465, 0.110941558, 0.195981388, 0.196012722, 0.180495072, 0.21281234, 0.18274535, 0.170102898, 0.223903977], "temperature": [0.146041822, 0.23725001, 0.146447955, 0.19333474, 0.268199955, 0.241008178, 0.254904393, 0.252212237, 0.158574783, 0.223705098, 0.196012722, 0.180495072, 0.275549232, 0.174975827, 0.158199805, 0.28013804]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0007, "losses": 11, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.998", "prereg_hash": "aba546f5560948bb06bb6877609b28cc332bd6ba7b24776549a915b5e8472b5c", "primary_control": "temperature", "result_hash": "df064a5b7dc518e8aeb40e2a7e6fcaf69537280e7f07a0a21414dc811bd1d34f", "signature": "a38fca5632a05c3d5707cf6db92311fff1c5faa2cce361aab02c0ccbf754badf3e897369ed14d530a91bd36eda86ecda16eac173d1c42c9e71f262903af9e105", "state": "VERIFIED", "summary": {"iw_temperature": 0.229729, "raw": 0.201271, "temperature": 0.211691}, "ties": 3, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:04:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:57Z", "cpu_seconds": 1.4420121630000153, "created_at": "2026-11-22T03:02:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-696fe84cc7f7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "274c7e8f492b4917d00a80afcf3e9173e12f9a0da669c42b6daa0548411d88e4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-696fe84cc7f7", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:42Z", "seed": 1031416759, "treatment": "iw_temperature"}, "priority": 0.2962132209778627, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "6bb16d1ad9dc98558402f769ab2f11ced0e2ab1b849d130b2025730e42500868", "run_seconds": 1.4482362270355225, "seed": 1031416759, "started_at": "2026-11-22T03:03:55Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014413, 0.008691], "effect": -0.002861, "evaluated_at": "2026-11-22T03:03:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-696fe84cc7f7", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.198970558, 0.149216313, 0.174657465, 0.139681932, 0.135871278, 0.118372937, 0.066329293, 0.16532168, 0.19366607, 0.166436965, 0.144030953, 0.135399595, 0.081191645, 0.16360948, 0.084465775, 0.126585548], "raw": [0.174827203, 0.120332655, 0.149095952, 0.148367297, 0.165302035, 0.146728872, 0.085712685, 0.149127633, 0.13576046, 0.127248695, 0.144030953, 0.14356541, 0.136779463, 0.145245687, 0.132569038, 0.13067623], "temperature": [0.191052797, 0.149216313, 0.15672593, 0.127258695, 0.182498078, 0.165532892, 0.0572051, 0.17624078, 0.156935027, 0.136194273, 0.13652266, 0.135399595, 0.081021373, 0.146439595, 0.096867518, 0.102917615]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "274c7e8f492b4917d00a80afcf3e9173e12f9a0da669c42b6daa0548411d88e4", "primary_control": "temperature", "result_hash": "6bb16d1ad9dc98558402f769ab2f11ced0e2ab1b849d130b2025730e42500868", "signature": "25d880debad3fdc3d6f0210372130d7996fae83519bc872aa85cc034caa7dd847ba6ddee2f6f101faca1a7b0f1b2a3260b2d46bd96c565958500a7c3d3766a01", "state": "VERIFIED", "summary": {"iw_temperature": 0.140238, "raw": 0.139711, "temperature": 0.137377}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:03:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:55Z", "cpu_seconds": 1.6284338109999936, "created_at": "2026-11-22T03:02:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-8562032d1f71", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "1aa2856d2e6a51a6d8525d816eb7f2015cc440f33541d4ab5bdbf7ba1cf10994", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-8562032d1f71", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:42Z", "seed": 954119775, "treatment": "iw_temperature"}, "priority": 0.3174240092504691, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "5d1658e048afe52ae259add5da4ca2c70cd2377a3b1a28361d86de95b33bd9f4", "run_seconds": 1.705127239227295, "seed": 954119775, "started_at": "2026-11-22T03:02:53Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011237, -0.001555], "effect": -0.006396, "evaluated_at": "2026-11-22T03:03:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-8562032d1f71", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.144960097, 0.17359968, 0.111533088, 0.142559172, 0.134933008, 0.149183783, 0.141611668, 0.090979832, 0.094671647, 0.161942345, 0.137720728, 0.15522793, 0.11195896, 0.10678963, 0.145126672, 0.155086098], "raw": [0.11008371, 0.143355522, 0.097220388, 0.086520195, 0.08897507, 0.085110958, 0.10532554, 0.069521098, 0.10235409, 0.081639743, 0.103689473, 0.096629377, 0.09886941, 0.097085412, 0.115569435, 0.098617925], "temperature": [0.13851663, 0.184946715, 0.099720627, 0.142559172, 0.117097615, 0.1255015, 0.129280725, 0.07717364, 0.094671647, 0.16115246, 0.137720728, 0.143875267, 0.09886941, 0.11742111, 0.13195855, 0.155086098]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0014, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.997", "prereg_hash": "1aa2856d2e6a51a6d8525d816eb7f2015cc440f33541d4ab5bdbf7ba1cf10994", "primary_control": "temperature", "result_hash": "5d1658e048afe52ae259add5da4ca2c70cd2377a3b1a28361d86de95b33bd9f4", "signature": "400618bfe9149caa0ba9f81de731a686761739e14db1f60236c465f99d5636d722f90191722aa837ed72659b99509e83e729819fb74d6f404e0f23cadc2f6d02", "state": "VERIFIED", "summary": {"iw_temperature": 0.134868, "raw": 0.098785, "temperature": 0.128472}, "ties": 4, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:03:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:15Z", "cpu_seconds": 1.7746341760000064, "created_at": "2026-11-22T03:02:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-a871546bb57f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "51a712e397dcc6d5d24f952fb1abf2833a5fc4d3abdabfa2b391d792e78542a4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-a871546bb57f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:42Z", "seed": 884427326, "treatment": "iw_temperature"}, "priority": 0.3149812385230112, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ee6b196fa2dd4178faec7205502147fd86308db7b3f406e94a266963e70913e3", "run_seconds": 1.7942907810211182, "seed": 884427326, "started_at": "2026-11-22T03:03:14Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020152, 0.001544], "effect": -0.009304, "evaluated_at": "2026-11-22T03:03:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-a871546bb57f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.0421613, 0.067210225, 0.070834412, 0.059950507, 0.136150607, 0.059703398, 0.03476691, 0.067355338, 0.02229413, 0.04315429, 0.066866883, 0.056711352, 0.036545105, 0.039659667, 0.055929972, 0.048320002], "raw": [0.0421613, 0.04497492, 0.084314363, 0.074549295, 0.038805665, 0.055086823, 0.048547445, 0.052648255, 0.061469468, 0.054638052, 0.058400118, 0.054715835, 0.05506312, 0.038901623, 0.062867078, 0.036713048], "temperature": [0.039977867, 0.051797542, 0.070834412, 0.050564378, 0.049007967, 0.068887393, 0.04246878, 0.053161298, 0.02229413, 0.04315429, 0.066866883, 0.044393935, 0.036545105, 0.039659667, 0.043792907, 0.035340695]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0047, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "51a712e397dcc6d5d24f952fb1abf2833a5fc4d3abdabfa2b391d792e78542a4", "primary_control": "temperature", "result_hash": "ee6b196fa2dd4178faec7205502147fd86308db7b3f406e94a266963e70913e3", "signature": "b4d4bcb5c07fc2af9a2800b8fc5a1d74b287aba1f9109e3e78073fe49c455243d6fbe6fd5928795981257cfae75056c530395719556fac6a258d6afa0ec7b10a", "state": "VERIFIED", "summary": {"iw_temperature": 0.056726, "raw": 0.053991, "temperature": 0.047422}, "ties": 6, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:03:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:02:52Z", "cpu_seconds": 7.767536019999994, "created_at": "2026-11-22T03:02:42Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-137d8d6237ba", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ab3fb8fff1ed67c330b60aec8e0507f38b8605597d3f89f0aed23c358c646962", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-137d8d6237ba", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:02:42Z", "seed": 1218131646, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a643c713194859493a78ed2ea11f59d0a0b064fbe7fb1eef4cb06542cb23ae93", "run_seconds": 7.8234992027282715, "seed": 1218131646, "started_at": "2026-11-22T03:02:44Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:02:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-137d8d6237ba", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6665526564721131, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0], "root": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ab3fb8fff1ed67c330b60aec8e0507f38b8605597d3f89f0aed23c358c646962", "primary_control": "reset", "result_hash": "a643c713194859493a78ed2ea11f59d0a0b064fbe7fb1eef4cb06542cb23ae93", "signature": "061015581bcf8a29002eb48f0755e86a6e271900ce5e176bda11f14f15db2f86ac8bf602608d108110cd1f892b283f6422e1b0d6335931050c8002e641a4c901", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:03:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:11Z", "cpu_seconds": 1.7352107140000044, "created_at": "2026-11-22T03:02:59Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-100228897355", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "29712382c6c3093b5bf8081c44281dc7015f6669e4e801d696399aaee27166e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-100228897355", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:59Z", "seed": 193031323, "treatment": "iw_temperature"}, "priority": 0.2563804884351745, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "e98064aa30e1e399640ddd850dbaf5b19800b42d015cd733af9aa3d125419ac8", "run_seconds": 2.4243862628936768, "seed": 193031323, "started_at": "2026-11-22T03:05:08Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015844, 0.012432], "effect": -0.001706, "evaluated_at": "2026-11-22T03:05:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-100228897355", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.14090922, 0.16427613, 0.137670987, 0.120121583, 0.053195885, 0.111089507, 0.15192484, 0.07368513, 0.147378553, 0.128748982, 0.178922732, 0.159126783, 0.148784795, 0.16173125, 0.04645307, 0.080537395], "raw": [0.099385583, 0.155888647, 0.119600083, 0.105333287, 0.053078813, 0.098381542, 0.15192484, 0.08906008, 0.138585485, 0.120833418, 0.100608587, 0.144949482, 0.136489728, 0.151501733, 0.09643951, 0.080537395], "temperature": [0.113613357, 0.155888647, 0.137670987, 0.105333287, 0.03746438, 0.111089507, 0.166278715, 0.118163205, 0.147380672, 0.128748982, 0.117535722, 0.147619397, 0.136489728, 0.151501733, 0.116310985, 0.086174225]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "29712382c6c3093b5bf8081c44281dc7015f6669e4e801d696399aaee27166e2", "primary_control": "temperature", "result_hash": "e98064aa30e1e399640ddd850dbaf5b19800b42d015cd733af9aa3d125419ac8", "signature": "6a15642a87fdba3fccec85b5af2e52a1b6608f31f499768d5fe9e8435edc27b470bdc4b9abbbfa2a62720f3ed7bf5f96a5d89bd83738cc4f5a7be40748555703", "state": "VERIFIED", "summary": {"iw_temperature": 0.125285, "raw": 0.115162, "temperature": 0.123579}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:05:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:45Z", "cpu_seconds": 1.6898358069999944, "created_at": "2026-11-22T03:02:59Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-2f6d33b1680f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "8a5e61b4f189688f0e973dfc4b0bfb80956d647afe7d3e2e8aaec4c575ca6ec3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-2f6d33b1680f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:59Z", "seed": 1724761457, "treatment": "iw_temperature"}, "priority": 0.29701862603262846, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "66c956f06af793ef5acb14cd0eb744ee75a501dbe591dfeb3b4efddf47bbc845", "run_seconds": 1.7590925693511963, "seed": 1724761457, "started_at": "2026-11-22T03:03:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015009, 0.006148], "effect": -0.00443, "evaluated_at": "2026-11-22T03:03:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-2f6d33b1680f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.082048137, 0.065627825, 0.01515335, 0.057086795, 0.058538912, 0.010037287, 0.1587618, 0.041938213, 0.05107641, 0.02538074, 0.094772533, 0.032297055, 0.17294323, 0.094604702, 0.057747295, 0.042745665], "raw": [0.10552025, 0.057186605, 0.01515335, 0.046437915, 0.07447717, 0.031040398, 0.156283363, 0.052865405, 0.035091788, 0.02935114, 0.054392582, 0.050782252, 0.1766293, 0.104309917, 0.078656932, 0.047587845], "temperature": [0.126708365, 0.029974207, 0.01186487, 0.0526576, 0.054503413, 0.010018352, 0.156283363, 0.041938213, 0.018214387, 0.014246143, 0.05647304, 0.032297055, 0.17294323, 0.127194765, 0.041814625, 0.042745665]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0014, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.997", "prereg_hash": "8a5e61b4f189688f0e973dfc4b0bfb80956d647afe7d3e2e8aaec4c575ca6ec3", "primary_control": "temperature", "result_hash": "66c956f06af793ef5acb14cd0eb744ee75a501dbe591dfeb3b4efddf47bbc845", "signature": "b57c50ad42f28473f773338d21df0fc2b40a91e6949dff39dad00123000344af30472c9921dece2d81586281e02f1ebbee52e6f67bd622c1a4ef8574546ac401", "state": "VERIFIED", "summary": {"iw_temperature": 0.066297, "raw": 0.069735, "temperature": 0.061867}, "ties": 4, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:03:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:02Z", "cpu_seconds": 1.7841487919999963, "created_at": "2026-11-22T03:02:59Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5a2a086f729b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "e9f06f067f293bf85e31b57789f8d71c6f07646febbad44230d89e2957caf11b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5a2a086f729b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:02:59Z", "seed": 1935042244, "treatment": "iw_temperature"}, "priority": 0.3686993570018071, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "8aac7ce4c6f4c8df079d9fae02fcd7767583ca049b09077c60df5fa18d40b1db", "run_seconds": 1.8999931812286377, "seed": 1935042244, "started_at": "2026-11-22T03:03:00Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.023506, 0.022322], "effect": -0.000592, "evaluated_at": "2026-11-22T03:03:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5a2a086f729b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.223134068, 0.048797352, 0.182862283, 0.247734643, 0.220301377, 0.318148572, 0.269791497, 0.289212023, 0.293713487, 0.218027723, 0.288170525, 0.279861252, 0.261442693, 0.300049083, 0.27575537, 0.293616342], "raw": [0.223134068, 0.145452953, 0.222692238, 0.16951622, 0.26890241, 0.28185706, 0.255469675, 0.284215675, 0.263702882, 0.193953597, 0.25614463, 0.210363593, 0.24354264, 0.182161127, 0.27575537, 0.26867004], "temperature": [0.223134068, 0.182271017, 0.234352368, 0.203451595, 0.243329958, 0.300277577, 0.277663248, 0.289212023, 0.284111703, 0.206167505, 0.264200737, 0.25903831, 0.259354435, 0.209472092, 0.27575537, 0.28935422]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0205, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.954", "prereg_hash": "e9f06f067f293bf85e31b57789f8d71c6f07646febbad44230d89e2957caf11b", "primary_control": "temperature", "result_hash": "8aac7ce4c6f4c8df079d9fae02fcd7767583ca049b09077c60df5fa18d40b1db", "signature": "fba8ab3c8b6d580c594d6002409086d876c04b4ebb842dae1b55fc9f4e7431f3b0a714587f1b3a7ec2c3ac8655f581d63427173857b2c3ff633e28ef1a3dd308", "state": "VERIFIED", "summary": {"iw_temperature": 0.250664, "raw": 0.234096, "temperature": 0.250072}, "ties": 3, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:03:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:07Z", "cpu_seconds": 7.545835197999992, "created_at": "2026-11-22T03:02:59Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-c25f7685ae21", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "eb2243b294318da8b47aa68431cc3adb5f05642d37e8ad9d651beba472cc7291", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-c25f7685ae21", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:02:59Z", "seed": 1731096896, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "c1e1f0bd4542f0d6a5e2c23e1f9fa1d656976187bafdc205117308478b3c53f8", "run_seconds": 7.632633209228516, "seed": 1731096896, "started_at": "2026-11-22T03:03:00Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:03:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-c25f7685ae21", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "eb2243b294318da8b47aa68431cc3adb5f05642d37e8ad9d651beba472cc7291", "primary_control": "reset", "result_hash": "c1e1f0bd4542f0d6a5e2c23e1f9fa1d656976187bafdc205117308478b3c53f8", "signature": "dc583abcb314847817f76471b0adfbfad5359f18fcc03e73af4c46f86761c4942a80e1b3d2ca17deebcce11d3fdf5908c36bcba6c550d4d9b567e57fd334160b", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.3}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:03:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:03Z", "cpu_seconds": 0.42493037699998126, "created_at": "2026-11-22T03:02:59Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-63def4f482c9", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "36e62d6aa5b29b07af22a3da3d4c230d3c2e4ec3259e472db1994464d13627f5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-63def4f482c9", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:02:59Z", "seed": 220527980, "treatment": "frontier"}, "priority": 0.25784907259146933, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "05c137b9e044ad5727a0aa5c5d18eaa7d7fc56e2e7227df9c4cd57ac58deabad", "run_seconds": 0.4739227294921875, "seed": 220527980, "started_at": "2026-11-22T03:05:03Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:05:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-63def4f482c9", "family": "world_model_frontier", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"curiosity": [1.0, 0.925356], "frontier": [1.0, 0.970139], "random_walk": [1.0, 0.895429]}, "prereg_hash": "36e62d6aa5b29b07af22a3da3d4c230d3c2e4ec3259e472db1994464d13627f5", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "05c137b9e044ad5727a0aa5c5d18eaa7d7fc56e2e7227df9c4cd57ac58deabad", "signature": "931e266edea06036aa46b0e67657b43f96d5b27cd242c40686077c99fdf89682697b79b88e2b1736817e591e75501ef0c4425d9b71fba871698726612397d207", "state": "REJECTED", "summary": {"curiosity": 0.925356, "frontier": 0.970139, "random_walk": 0.895429}}, "verified_at": "2026-11-22T03:05:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:09Z", "cpu_seconds": 3.5575781770000106, "created_at": "2026-11-22T03:02:59Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-6b3938d40f05", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "9ad4bf44b4e9a70e064b4d8f5719f7dcf1043ec3146e371f770fb229850fe58a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-6b3938d40f05", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:02:59Z", "seed": 185417151, "treatment": "frontier"}, "priority": 0.34137140268385646, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "417aac99908abcf9a68aa8d92dd89386baa5823c3780d59ef0983e5f150ca21e", "run_seconds": 3.788395404815674, "seed": 185417151, "started_at": "2026-11-22T03:03:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.055077, 0.115315], "effect": 0.085196, "evaluated_at": "2026-11-22T03:03:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-6b3938d40f05", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.856796117, 0.798969072, 0.838068182, 1.0, 0.889175258, 0.891304348, 0.798295455, 0.846153846, 0.927884615, 0.775568182, 0.821022727, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 0.966494845, 1.0, 1.0, 1.0, 0.997596154, 1.0, 0.977272727, 1.0], "random_walk": [0.847087379, 0.904639175, 0.852272727, 1.0, 0.917525773, 0.95923913, 0.90625, 0.881868132, 0.908653846, 0.855113636, 0.886363636, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "9ad4bf44b4e9a70e064b4d8f5719f7dcf1043ec3146e371f770fb229850fe58a", "primary_control": "random_walk", "result_hash": "417aac99908abcf9a68aa8d92dd89386baa5823c3780d59ef0983e5f150ca21e", "signature": "11005a8ae5ad42e2edfda8a838fc7d481f5a82a7a53a870a5b2be5ddb551d1000aee7eb390c7f4ed34704deb61141044d43b0e6fb621b5a8b77a71ebfef76501", "state": "VERIFIED", "summary": {"curiosity": 0.87027, "frontier": 0.995114, "random_walk": 0.909918}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:03:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:22Z", "cpu_seconds": 1.7054374370000005, "created_at": "2026-11-22T03:03:15Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1ef7f67cb8de", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "40e0973c545aa7b74e2c50fbf4195b3247e15c9b4c029d880a359ffdf6f520a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1ef7f67cb8de", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:03:15Z", "seed": 39440470, "treatment": "iw_temperature"}, "priority": 0.37015189717452335, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "39cb998510d92adb3bdfa558cf2a426de811d88ffca8d016bdc9f49634210512", "run_seconds": 1.7626380920410156, "seed": 39440470, "started_at": "2026-11-22T03:03:20Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.030336, 0.002729], "effect": -0.013804, "evaluated_at": "2026-11-22T03:03:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1ef7f67cb8de", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.234460973, 0.256174137, 0.280625287, 0.238886428, 0.335413908, 0.233617983, 0.153513055, 0.196901273, 0.280575393, 0.242650738, 0.178059113, 0.228859488, 0.33277295, 0.246400532, 0.253897768, 0.25904669], "raw": [0.120542205, 0.27738684, 0.241170535, 0.17436368, 0.325905482, 0.236468115, 0.13466543, 0.22676995, 0.26871599, 0.195735032, 0.171721738, 0.228859488, 0.307630927, 0.221834075, 0.18392474, 0.199896848], "temperature": [0.168344978, 0.29902981, 0.280625287, 0.18702911, 0.330085083, 0.245061347, 0.111088445, 0.196901273, 0.280575393, 0.211552662, 0.178059113, 0.254593512, 0.321029772, 0.255554028, 0.169483867, 0.2419846]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "40e0973c545aa7b74e2c50fbf4195b3247e15c9b4c029d880a359ffdf6f520a2", "primary_control": "temperature", "result_hash": "39cb998510d92adb3bdfa558cf2a426de811d88ffca8d016bdc9f49634210512", "signature": "d009eb3e8a5ffc651d57ba0568dbda64665534ba64aaf5345c5fad01df4ae4c09a5ef5d7017fbc4347863c666666e09902dd6527d04956078ea7f7f23cc7e008", "state": "VERIFIED", "summary": {"iw_temperature": 0.246991, "raw": 0.219724, "temperature": 0.233187}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:03:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:41Z", "cpu_seconds": 1.4791496269999982, "created_at": "2026-11-22T03:03:15Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f2aeb584def9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "a3cdfbb9a6eb28e8eecab96de10787da466e8bf73a0d1bb4e8020a67c3e61ba2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f2aeb584def9", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:03:15Z", "seed": 35923419, "treatment": "iw_temperature"}, "priority": 0.24636361365076967, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "8669e921ebfaba649c4c16b52e6bf85671af374ee4623aabb9e51faec5267982", "run_seconds": 1.8087267875671387, "seed": 35923419, "started_at": "2026-11-22T03:06:39Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011858, 0.010137], "effect": -0.00086, "evaluated_at": "2026-11-22T03:06:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f2aeb584def9", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.160562965, 0.153531072, 0.102338955, 0.17854863, 0.136299255, 0.186133, 0.133656125, 0.095299755, 0.161206362, 0.19558303, 0.149798055, 0.100808323, 0.142741345, 0.20442329, 0.15311686, 0.128491983], "raw": [0.150769155, 0.136719445, 0.085553575, 0.16582277, 0.111783532, 0.174044617, 0.103854265, 0.133780038, 0.15340794, 0.11025389, 0.149798055, 0.13027595, 0.090239585, 0.15576479, 0.146406722, 0.123006867], "temperature": [0.154893857, 0.146158543, 0.093765067, 0.16997499, 0.136299255, 0.174044617, 0.103854265, 0.145688355, 0.156383405, 0.167147088, 0.170227242, 0.146862242, 0.142741345, 0.185834532, 0.146406722, 0.128491983]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "a3cdfbb9a6eb28e8eecab96de10787da466e8bf73a0d1bb4e8020a67c3e61ba2", "primary_control": "temperature", "result_hash": "8669e921ebfaba649c4c16b52e6bf85671af374ee4623aabb9e51faec5267982", "signature": "957dce7d38baef8939a6369a8be03cd1c24b1dd909a0b89dbfaac893bc8be78b425f41c1421bfe5e7b354d8b0bd729be512aab177b79016abf136f9df646d50c", "state": "VERIFIED", "summary": {"iw_temperature": 0.148909, "raw": 0.132593, "temperature": 0.148048}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:06:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:23Z", "cpu_seconds": 7.292563611999995, "created_at": "2026-11-22T03:03:15Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-eeb7dfcf8bf8", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "52e0765575984409575fab4591e5a87240e22e223601ace788bcc9e699ba049f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-eeb7dfcf8bf8", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:03:15Z", "seed": 443787246, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "8e15490b5cce8d016e6bf4320cec01bd1841733355d3fdd29b0865813df3d83f", "run_seconds": 7.356141567230225, "seed": 443787246, "started_at": "2026-11-22T03:03:15Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.080465, 0.147132], "effect": 0.033333, "evaluated_at": "2026-11-22T03:03:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-eeb7dfcf8bf8", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.39936788819561164, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.4121, "losses": 5, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.387", "prereg_hash": "52e0765575984409575fab4591e5a87240e22e223601ace788bcc9e699ba049f", "primary_control": "reset", "result_hash": "8e15490b5cce8d016e6bf4320cec01bd1841733355d3fdd29b0865813df3d83f", "signature": "66a7ab060afc80b63ad917a4bdbaf8da9f068eafdca7ba1db65f53fcc437a583950a5c3ebfb26403a67282ac0b4bf52a5372f2b30f279e25bba2cf3c13e7580f", "state": "VERIFIED", "summary": {"inherit": 0.5666666666666667, "reset": 0.5333333333333333, "root": 0.31666666666666665}, "ties": 48, "treatment": "inherit", "wins": 7}, "verified_at": "2026-11-22T03:03:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:09Z", "cpu_seconds": 0.40469599099998277, "created_at": "2026-11-22T03:03:15Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-3d5a303fb6ea", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "8873e5d692dff143101c6746e5eb437ca8aae9d04513d7c96fac66e0bfcd4f5b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-3d5a303fb6ea", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:03:15Z", "seed": 1562857178, "treatment": "frontier"}, "priority": 0.2519730957657058, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "146b646730a1f8d19565fa79e4e0daa2469edf6538df12b6672e6d2b48e3f27c", "run_seconds": 0.560150146484375, "seed": 1562857178, "started_at": "2026-11-22T03:06:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.014145, 0.032316], "effect": 0.02323, "evaluated_at": "2026-11-22T03:06:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-3d5a303fb6ea", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.910714286, 1.0, 0.849557522, 0.904471545, 0.884615385, 0.915354331, 0.90625, 0.912601626, 0.904166667, 0.95112782, 0.909448819, 0.895491803], "frontier": [0.93452381, 1.0, 0.911504425, 0.953252033, 0.918803419, 0.927165354, 0.94140625, 0.93699187, 0.925, 0.95112782, 0.94488189, 0.905737705], "random_walk": [0.906746032, 1.0, 0.85840708, 0.906504065, 0.895299145, 0.913385827, 0.91015625, 0.910569106, 0.908333333, 0.947368421, 0.917322835, 0.897540984]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "8873e5d692dff143101c6746e5eb437ca8aae9d04513d7c96fac66e0bfcd4f5b", "primary_control": "random_walk", "result_hash": "146b646730a1f8d19565fa79e4e0daa2469edf6538df12b6672e6d2b48e3f27c", "signature": "91a2c21fc6dd859cd9b0e9fd64582866daa9019c1b54ca8ed7b92c1b88ea05d211b8b92a4095e711a1d91bcd133eb04f997f5170ced736e97ed36db2d8bc080a", "state": "VERIFIED", "summary": {"curiosity": 0.911983, "frontier": 0.937533, "random_walk": 0.914303}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:06:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:02Z", "cpu_seconds": 0.38133711999998354, "created_at": "2026-11-22T03:03:15Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-92ea3ced8150", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "43e421a3d3ff0ae20a3202e23a169ac72ef9913fa29965400e564715433134ec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-92ea3ced8150", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:03:15Z", "seed": 1347974550, "treatment": "frontier"}, "priority": 0.26416948990566796, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "2f360f5db1665b5cd4966adc6a23e02d5ff8b1d6ae104696935f2088b564fca6", "run_seconds": 0.3829832077026367, "seed": 1347974550, "started_at": "2026-11-22T03:04:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.025228, 0.080605], "effect": 0.052917, "evaluated_at": "2026-11-22T03:04:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-92ea3ced8150", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.808333333, 1.0, 0.817757009, 1.0, 0.8, 0.75, 1.0, 0.773195876, 0.810679612, 1.0, 0.738764045, 1.0], "frontier": [0.911111111, 1.0, 0.901869159, 1.0, 0.885, 0.86407767, 1.0, 0.87628866, 0.895631068, 1.0, 0.882022472, 1.0], "random_walk": [0.816666667, 1.0, 0.83411215, 1.0, 0.79, 0.759708738, 1.0, 0.804123711, 0.817961165, 1.0, 0.758426966, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.1072, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00781", "prereg_hash": "43e421a3d3ff0ae20a3202e23a169ac72ef9913fa29965400e564715433134ec", "primary_control": "random_walk", "result_hash": "2f360f5db1665b5cd4966adc6a23e02d5ff8b1d6ae104696935f2088b564fca6", "signature": "f255fa024095b0f8afc6a2247ae066b0dfa52c9316fda31b3fa893ddf94cb8b54768d7c4a8b2299acf9fd0853a694aff593a45ab774da109eab35adb00bbd002", "state": "VERIFIED", "summary": {"curiosity": 0.874894, "frontier": 0.934667, "random_walk": 0.88175}, "ties": 5, "treatment": "frontier", "wins": 7}, "verified_at": "2026-11-22T03:04:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:46Z", "cpu_seconds": 0.34923708300001977, "created_at": "2026-11-22T03:03:15Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-e169bdfc5bb3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "e4c72db96e110d11e5e348ce8a3d13e549704dce7871dc2b080ea04eb67467d4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-e169bdfc5bb3", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:03:15Z", "seed": 1385725989, "treatment": "frontier"}, "priority": 0.24418779640896138, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "6b275134130ba3af779cd3b7948c0a3c4888481abdc71c18f17df3d3d7b1c0aa", "run_seconds": 0.35442399978637695, "seed": 1385725989, "started_at": "2026-11-22T03:06:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.059737, 0.090174], "effect": 0.074955, "evaluated_at": "2026-11-22T03:06:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-e169bdfc5bb3", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.890625, 0.928571429, 0.952054795, 0.9, 0.904761905, 0.913043478, 0.941780822, 0.895522388, 0.926056338, 0.913934426, 0.922535211, 0.914285714], "frontier": [1.0, 0.992857143, 1.0, 0.984615385, 1.0, 0.996376812, 0.982876712, 0.992537313, 0.985915493, 1.0, 0.982394366, 0.982142857], "random_walk": [0.89453125, 0.892857143, 0.952054795, 0.907692308, 0.880952381, 0.920289855, 0.917808219, 0.891791045, 0.929577465, 0.926229508, 0.947183099, 0.939285714]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "e4c72db96e110d11e5e348ce8a3d13e549704dce7871dc2b080ea04eb67467d4", "primary_control": "random_walk", "result_hash": "6b275134130ba3af779cd3b7948c0a3c4888481abdc71c18f17df3d3d7b1c0aa", "signature": "d866a8c98103136d2562e11f034d7232ca9becc9bdcdc9c148103246b24d6af69dd245923e2d2eb5b13232825204bdc297957a8ab0ebbb3c46af0ba134946407", "state": "VERIFIED", "summary": {"curiosity": 0.916931, "frontier": 0.991643, "random_walk": 0.916688}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:06:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:36Z", "cpu_seconds": 1.6579973559999814, "created_at": "2026-11-22T03:03:31Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-2d482c40235b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "4e28cc8b662c81e4da15d840b9aaf77dc52915e88df50a89bbe73b7d5d0e6dec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-2d482c40235b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:03:31Z", "seed": 1623663279, "treatment": "iw_temperature"}, "priority": 0.35105293010613475, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "339779b978cdd49cced8c3e38d455bf3690d4d0176c5f9dee98525ec52aa81eb", "run_seconds": 1.6724965572357178, "seed": 1623663279, "started_at": "2026-11-22T03:03:34Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.023112, 0.014137], "effect": -0.004487, "evaluated_at": "2026-11-22T03:03:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-2d482c40235b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.29734046, 0.23037345, 0.26086993, 0.288339552, 0.30788135, 0.15950333, 0.287654358, 0.300893313, 0.2518901, 0.101962, 0.282094825, 0.232499272, 0.218482265, 0.291097665, 0.3121017, 0.27907405], "raw": [0.294508975, 0.24032167, 0.243444585, 0.2647486, 0.240816707, 0.196391697, 0.272718985, 0.294804597, 0.227063217, 0.18871406, 0.21634536, 0.221352952, 0.161275522, 0.241964538, 0.304223217, 0.247920018], "temperature": [0.299803543, 0.276416748, 0.243444585, 0.282154365, 0.240816707, 0.15950333, 0.28332808, 0.287344313, 0.248855902, 0.200456415, 0.245404505, 0.24397644, 0.161275522, 0.27885788, 0.31403193, 0.264590318]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "4e28cc8b662c81e4da15d840b9aaf77dc52915e88df50a89bbe73b7d5d0e6dec", "primary_control": "temperature", "result_hash": "339779b978cdd49cced8c3e38d455bf3690d4d0176c5f9dee98525ec52aa81eb", "signature": "ed330b54575c16f81420ad54138d023289e6d51356c0002fbd8a1a0926c1fad0ba1d38a16306f23ebadf3997f7aa72fa41fd8ddb2e5fb07c5049c946c37bcf07", "state": "VERIFIED", "summary": {"iw_temperature": 0.256379, "raw": 0.241038, "temperature": 0.251891}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:03:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:25Z", "cpu_seconds": 1.6611001050000027, "created_at": "2026-11-22T03:03:31Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-7f328ec3a5ba", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "eda579d01e23d2b2e2a22e89a0a42013919730de23cf1aa1ff101568e7b9ea1f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-7f328ec3a5ba", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:03:31Z", "seed": 386075318, "treatment": "iw_temperature"}, "priority": 0.25530731302502724, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "fe6150fb8f6a2a6ffcf43a9e7f9ef3f06d078322d90071f914d6f3ea0dd0b0c9", "run_seconds": 2.5631637573242188, "seed": 386075318, "started_at": "2026-11-22T03:05:22Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012108, 0.005113], "effect": -0.003497, "evaluated_at": "2026-11-22T03:05:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-7f328ec3a5ba", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.042768852, 0.03767321, 0.013058743, 0.0368433, 0.041426443, 0.053614697, 0.113773515, 0.022588897, 0.076500985, 0.042673938, 0.017967312, 0.02653625, 0.052164705, 0.019293383, 0.070440503, 0.029155383], "raw": [0.03689942, 0.073343663, 0.022311325, 0.0368433, 0.052787862, 0.036463805, 0.056407803, 0.022159802, 0.05812701, 0.042192618, 0.030343468, 0.028215553, 0.04025058, 0.050658778, 0.074683235, 0.047418467], "temperature": [0.02969343, 0.04923089, 0.012677888, 0.040866103, 0.029856075, 0.068940578, 0.056407803, 0.015152275, 0.06189111, 0.06260827, 0.018754687, 0.02653625, 0.04025058, 0.019293383, 0.074683235, 0.033678822]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2184, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.605", "prereg_hash": "eda579d01e23d2b2e2a22e89a0a42013919730de23cf1aa1ff101568e7b9ea1f", "primary_control": "temperature", "result_hash": "fe6150fb8f6a2a6ffcf43a9e7f9ef3f06d078322d90071f914d6f3ea0dd0b0c9", "signature": "082c698f6820b7dff0d3d9b2ade4b3993666bfecdb440df8471c30fdf46bc8292508859b17d38b25f7990c94c6cd6940264fb1c20f69564e96cea775b8a94408", "state": "VERIFIED", "summary": {"iw_temperature": 0.04353, "raw": 0.044319, "temperature": 0.040033}, "ties": 2, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:05:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:40Z", "cpu_seconds": 7.382873836999991, "created_at": "2026-11-22T03:03:31Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1085ecf3dce2", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "8d27d36654bbf509687b46b02dc8e27142639acc60a267757d0c71ab3116f434", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1085ecf3dce2", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:03:31Z", "seed": 1381070388, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "1b14d0a85be411bccd412139f0553adab75cf1c7a1c4d8b87c3734f27599b7af", "run_seconds": 7.437522649765015, "seed": 1381070388, "started_at": "2026-11-22T03:03:33Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:03:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1085ecf3dce2", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0], "root": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8d27d36654bbf509687b46b02dc8e27142639acc60a267757d0c71ab3116f434", "primary_control": "reset", "result_hash": "1b14d0a85be411bccd412139f0553adab75cf1c7a1c4d8b87c3734f27599b7af", "signature": "f59aad71a1934a785e924ca7e0a7028c4dde99461b6604800c7c6915f7b900c42b3d3faed955d61a1f979132958cb39d2b11c3f5b1fd435bd58e3b18ef33bb00", "state": "PARTIAL", "summary": {"inherit": 0.6, "reset": 0.6, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:03:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:38Z", "cpu_seconds": 3.0298756889999936, "created_at": "2026-11-22T03:03:31Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-575189ffbcb0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "2f7edde981680c019fcce903faf25cc11cd3aedf5b8b4886a3a8ff8ad3fc2549", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-575189ffbcb0", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:03:31Z", "seed": 2121712495, "treatment": "frontier"}, "priority": 0.3388878600106436, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "fe6f679897601641265118bd7b8b212962bae08b480fab3850f318b2f82a6b1d", "run_seconds": 3.143961191177368, "seed": 2121712495, "started_at": "2026-11-22T03:03:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.016528, 0.089295], "effect": 0.052912, "evaluated_at": "2026-11-22T03:03:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-575189ffbcb0", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.833333333, 1.0, 0.818627451, 1.0, 1.0, 0.844387755, 0.779069767, 1.0, 0.810526316, 0.84, 1.0], "frontier": [1.0, 1.0, 1.0, 0.990196078, 1.0, 1.0, 1.0, 0.979651163, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.8, 1.0, 0.892156863, 1.0, 1.0, 0.897959184, 0.918604651, 1.0, 0.923684211, 0.9025, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.8062, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0156", "prereg_hash": "2f7edde981680c019fcce903faf25cc11cd3aedf5b8b4886a3a8ff8ad3fc2549", "primary_control": "random_walk", "result_hash": "fe6f679897601641265118bd7b8b212962bae08b480fab3850f318b2f82a6b1d", "signature": "b3491a0ab5a392d8429c6b7d6f78f4d19805af0bb3e1ad60cc7a5b246d2bef547bdc8a34cc6325b5f0e54b26b967d7f2a77e412daa46e4fc6b31e5617d5d0508", "state": "VERIFIED", "summary": {"curiosity": 0.910495, "frontier": 0.997487, "random_walk": 0.944575}, "ties": 6, "treatment": "frontier", "wins": 6}, "verified_at": "2026-11-22T03:03:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:41Z", "cpu_seconds": 0.21741928099999086, "created_at": "2026-11-22T03:03:31Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-9cf061193eef", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "3e7e47f68c0561af8d99d5e536c070d3e2b2573b36bba29be06cc5b40ac7cd25", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-9cf061193eef", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:03:31Z", "seed": 1048724981, "treatment": "frontier"}, "priority": 0.2539230864921169, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "25af543f6f8a39ad7038c742f10b03bbf91d4c1abd204715fd779c156f5a9688", "run_seconds": 0.21761298179626465, "seed": 1048724981, "started_at": "2026-11-22T03:05:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.046027, 0.110504], "effect": 0.078265, "evaluated_at": "2026-11-22T03:05:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-9cf061193eef", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 1.0, 0.82, 1.0, 0.875, 0.846938776, 0.839622642, 0.808333333, 0.825, 0.827272727, 0.871621622, 0.855263158], "frontier": [1.0, 1.0, 0.925, 1.0, 1.0, 0.908163265, 0.905660377, 0.904166667, 0.870833333, 0.913636364, 0.952702703, 1.0], "random_walk": [1.0, 1.0, 0.815, 1.0, 0.852272727, 0.821428571, 0.825471698, 0.820833333, 0.795833333, 0.822727273, 0.871621622, 0.815789474]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "3e7e47f68c0561af8d99d5e536c070d3e2b2573b36bba29be06cc5b40ac7cd25", "primary_control": "random_walk", "result_hash": "25af543f6f8a39ad7038c742f10b03bbf91d4c1abd204715fd779c156f5a9688", "signature": "e0ceb8791168a013d43e0434b3cc9e7b2d8df2e658fe66ce59ce589ff4df9c80eef508cf8a6089fa7677e3f9bd6c52bdfdb36d7869e83a81326cf1bcf5c1bc04", "state": "VERIFIED", "summary": {"curiosity": 0.880754, "frontier": 0.948347, "random_walk": 0.870082}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:05:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:50Z", "cpu_seconds": 1.8174816389999933, "created_at": "2026-11-22T03:03:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5e6ff550a11d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "9b25f9687f3b40c3b997a4609cfb40ec5bb301226925e50ac5b93f4f0bbbddc4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5e6ff550a11d", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:03:47Z", "seed": 743141944, "treatment": "iw_temperature"}, "priority": 0.35666081243337316, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "aadff2fe449e8f48964ed3437ebb6a58b0dc18f3f99814b9428513e226bf5501", "run_seconds": 2.0379927158355713, "seed": 743141944, "started_at": "2026-11-22T03:03:48Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.031322, 0.002927], "effect": -0.014198, "evaluated_at": "2026-11-22T03:03:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5e6ff550a11d", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.294566415, 0.31967682, 0.292785735, 0.318383863, 0.319416957, 0.318362315, 0.26887129, 0.31053464, 0.275826752, 0.32023095, 0.29406093, 0.194683007, 0.31090239, 0.240421478, 0.223073755, 0.22076542], "raw": [0.281028753, 0.301809668, 0.257248432, 0.212418413, 0.284473865, 0.270621945, 0.220781517, 0.296708477, 0.203629818, 0.309386728, 0.199044167, 0.204955633, 0.27630594, 0.201705975, 0.223073755, 0.251122678], "temperature": [0.284497502, 0.328722202, 0.27536919, 0.285116492, 0.298980372, 0.298464248, 0.220781517, 0.303030168, 0.223574245, 0.318686203, 0.248012395, 0.251082648, 0.27630594, 0.17385139, 0.23337364, 0.27554915]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0046, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "9b25f9687f3b40c3b997a4609cfb40ec5bb301226925e50ac5b93f4f0bbbddc4", "primary_control": "temperature", "result_hash": "aadff2fe449e8f48964ed3437ebb6a58b0dc18f3f99814b9428513e226bf5501", "signature": "febbf8ace919fa28a5430be9cd9eec58a5686c8d4b9a4a9e624a123385c0fd86ffc60f8f86484b1b34bedcbb8862e6ac9eedb8b4314d5834cb2d4761063de30f", "state": "VERIFIED", "summary": {"iw_temperature": 0.28266, "raw": 0.249645, "temperature": 0.268462}, "ties": 0, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:03:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:11Z", "cpu_seconds": 1.5727596819999974, "created_at": "2026-11-22T03:03:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-686297da69e1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "7e9ba7a64ad13c107a17d177c5596aa58c716e05f1f491681c1908358af0a527", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-686297da69e1", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:03:47Z", "seed": 1489767549, "treatment": "iw_temperature"}, "priority": 0.2574818895952258, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "cfbe1b3b3168e896f0087ffe107338ef506c3f31453dc23cf5c90feebf68f639", "run_seconds": 2.222257375717163, "seed": 1489767549, "started_at": "2026-11-22T03:05:08Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017988, 0.004627], "effect": -0.006681, "evaluated_at": "2026-11-22T03:05:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-686297da69e1", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.060985507, 0.043964118, 0.098276188, 0.0034289, 0.054857632, 0.07605105, 0.054841185, 0.04260935, 0.00328047, 0.116817845, 0.071957325, 0.183885792, 0.165847612, 0.087923103, 0.138626128, 0.004219987], "raw": [0.03333715, 0.036762122, 0.030700272, 0.00861384, 0.05299752, 0.088872588, 0.074435115, 0.041520225, 0.00328047, 0.100887165, 0.0403376, 0.183885792, 0.165847612, 0.113876608, 0.12981145, 0.01301615], "temperature": [0.05509046, 0.03861415, 0.03212966, 0.007004973, 0.043760978, 0.069923157, 0.030664925, 0.039066662, 0.00328047, 0.116817845, 0.042400582, 0.183885792, 0.165847612, 0.138334472, 0.12981145, 0.004049283]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0014, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.997", "prereg_hash": "7e9ba7a64ad13c107a17d177c5596aa58c716e05f1f491681c1908358af0a527", "primary_control": "temperature", "result_hash": "cfbe1b3b3168e896f0087ffe107338ef506c3f31453dc23cf5c90feebf68f639", "signature": "903708fb24b3849c12431eb881b97a196c8837c4c13a069563a26902f993f0513368848b9fdf702f38ebbc04500b04a42097cef296d3cddbe9670a63e1a9f503", "state": "VERIFIED", "summary": {"iw_temperature": 0.075473, "raw": 0.069886, "temperature": 0.068793}, "ties": 4, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:05:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:48Z", "cpu_seconds": 1.536038813999994, "created_at": "2026-11-22T03:03:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-bb4c95da4737", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "d553934cf8e143ef721b8a339ce0539218eb6da265eb721fd1cdf0897ade1457", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-bb4c95da4737", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:03:47Z", "seed": 1402060925, "treatment": "iw_temperature"}, "priority": 0.2591519828836697, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a9de21e2b00a4feb60ad6a4688fbbddd0ccd1cc25fb2366ff69829354275407b", "run_seconds": 1.5405826568603516, "seed": 1402060925, "started_at": "2026-11-22T03:04:46Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.016718, 0.016606], "effect": -5.6e-05, "evaluated_at": "2026-11-22T03:04:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-bb4c95da4737", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.224212233, 0.10984261, 0.24598313, 0.232665025, 0.266617293, 0.260201635, 0.242292877, 0.299235172, 0.254120825, 0.107539257, 0.262879688, 0.128114765, 0.13249427, 0.30301754, 0.213047698, 0.32431106], "raw": [0.21688445, 0.214192638, 0.267821593, 0.242866665, 0.233255872, 0.193249945, 0.232129868, 0.288332812, 0.221156015, 0.191121537, 0.26907949, 0.229758923, 0.191586103, 0.293540565, 0.257994187, 0.326967505], "temperature": [0.224212233, 0.155964638, 0.24598313, 0.18636158, 0.233255872, 0.175825735, 0.262933845, 0.288332812, 0.251583105, 0.138797073, 0.253498305, 0.134525068, 0.165116343, 0.298258488, 0.257994187, 0.33303968]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2184, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.605", "prereg_hash": "d553934cf8e143ef721b8a339ce0539218eb6da265eb721fd1cdf0897ade1457", "primary_control": "temperature", "result_hash": "a9de21e2b00a4feb60ad6a4688fbbddd0ccd1cc25fb2366ff69829354275407b", "signature": "91e88e08b155b25b1643dee1a159f58ec1c273dde021a8f194d6fe18e95d073d236f2f94861ec1ab24b5eebb2157f2310d9a16dc0bfc703de9e1edf036505b05", "state": "VERIFIED", "summary": {"iw_temperature": 0.225411, "raw": 0.241871, "temperature": 0.225355}, "ties": 2, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:04:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:57Z", "cpu_seconds": 9.079971755000003, "created_at": "2026-11-22T03:03:47Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-b56aed50c9b4", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "0ba53ea16cefc08e65f04b0a717ef180648f3231d844be73d34190406c19ea96", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-b56aed50c9b4", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:03:47Z", "seed": 1846614362, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "63d961d2fd4eacfbf13a1962ddaf373298187e16ef0069c8c0b2b1df05a96336", "run_seconds": 9.340940237045288, "seed": 1846614362, "started_at": "2026-11-22T03:03:48Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:03:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-b56aed50c9b4", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4367628229934713, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0], "root": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0ba53ea16cefc08e65f04b0a717ef180648f3231d844be73d34190406c19ea96", "primary_control": "reset", "result_hash": "63d961d2fd4eacfbf13a1962ddaf373298187e16ef0069c8c0b2b1df05a96336", "signature": "70ef5edfbad94034eafd1577f71fab382544f839145af933c8899874ba39ec9e90090a4bbbc000d77517a48e93931f6e48097b58d937fa42d13e948f40a15000", "state": "PARTIAL", "summary": {"inherit": 0.7, "reset": 0.7, "root": 0.4666666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:04:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:03:53Z", "cpu_seconds": 2.789357486, "created_at": "2026-11-22T03:03:47Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-e2e0aa7a2687", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "a9f26c57001f059e3d7e24d25ce4fa5e86793438f57f54636605992514361dfa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-e2e0aa7a2687", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:03:47Z", "seed": 877457563, "treatment": "frontier"}, "priority": 0.3064802078004987, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "fc34379091d8ceeee07dc54a42d13781aa615f6b74b75ee20853dbac559f275f", "run_seconds": 2.946535348892212, "seed": 877457563, "started_at": "2026-11-22T03:03:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.032164, 0.106938], "effect": 0.069551, "evaluated_at": "2026-11-22T03:03:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-e2e0aa7a2687", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 1.0, 1.0, 0.785714286, 0.816091954, 1.0, 0.862980769, 0.888235294, 0.795918367, 0.822916667, 0.815217391, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 1.0, 1.0, 0.845238095, 0.893678161, 1.0, 0.870192308, 0.914705882, 0.905612245, 0.90625, 0.829710145, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.1072, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00781", "prereg_hash": "a9f26c57001f059e3d7e24d25ce4fa5e86793438f57f54636605992514361dfa", "primary_control": "random_walk", "result_hash": "fc34379091d8ceeee07dc54a42d13781aa615f6b74b75ee20853dbac559f275f", "signature": "3b76f9ecdb9de585fc3126f5dd948f3bf25972d8923584951b23444e3fe4a3f8bd9d948e0fa03b1e0bef5f22d5c929ed76b873391c7124f06952dfc078186c07", "state": "VERIFIED", "summary": {"curiosity": 0.898923, "frontier": 1.0, "random_walk": 0.930449}, "ties": 5, "treatment": "frontier", "wins": 7}, "verified_at": "2026-11-22T03:03:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:18Z", "cpu_seconds": 1.5502049029999938, "created_at": "2026-11-22T03:04:04Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4f7e0cdbfe02", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "81a6c4f4b45438435cbcbcfccc6bf3ff0adbf45571a4cd2e3fa17cc27f8f85f0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4f7e0cdbfe02", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:04Z", "seed": 1240017953, "treatment": "iw_temperature"}, "priority": 0.26015056651112045, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "0d23ee47e9dc096d358854618a7e2092d9932622567db621a792e592b7cb4811", "run_seconds": 1.5909478664398193, "seed": 1240017953, "started_at": "2026-11-22T03:04:16Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.016375, -0.000523], "effect": -0.008449, "evaluated_at": "2026-11-22T03:04:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4f7e0cdbfe02", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.073374278, 0.04139507, 0.081976287, 0.073043157, 0.044892725, 0.062733315, 0.086627403, 0.048077523, 0.051857647, 0.018793483, 0.041402895, 0.076878423, 0.055336565, 0.044071915, 0.105179597, 0.047765138], "raw": [0.047384485, 0.052097662, 0.087351107, 0.066618355, 0.07243202, 0.03821272, 0.069761685, 0.036801635, 0.085757605, 0.078539222, 0.049236272, 0.052258673, 0.0602844, 0.03811794, 0.04242126, 0.059647835], "temperature": [0.060507997, 0.04139507, 0.081976287, 0.072660062, 0.044892725, 0.03044285, 0.056076253, 0.048077523, 0.051857647, 0.025291203, 0.041402895, 0.063204737, 0.055336565, 0.044071915, 0.053257865, 0.047765138]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0034, "losses": 6, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.992", "prereg_hash": "81a6c4f4b45438435cbcbcfccc6bf3ff0adbf45571a4cd2e3fa17cc27f8f85f0", "primary_control": "temperature", "result_hash": "0d23ee47e9dc096d358854618a7e2092d9932622567db621a792e592b7cb4811", "signature": "c979059802be09ecfe781109ee7e9ff81a70925f23e56d30c0b2993a10a3181524df2cdc363aec9d18854c340bc4a62b25a1120a839dc091ef7dc60039231602", "state": "VERIFIED", "summary": {"iw_temperature": 0.059588, "raw": 0.058558, "temperature": 0.051139}, "ties": 9, "treatment": "iw_temperature", "wins": 1}, "verified_at": "2026-11-22T03:04:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:06Z", "cpu_seconds": 1.6469527100000079, "created_at": "2026-11-22T03:04:04Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b6600f5be285", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "22c746e65045223ad81a0ada3b23b08dffa6a1b8f2eb289bae5407240c792a00", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b6600f5be285", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:04Z", "seed": 1610712743, "treatment": "iw_temperature"}, "priority": 0.37759002953260223, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "1448440eeca60ebe111a82e5172c24d002772bc9d5b6344cfa6574f036d96d60", "run_seconds": 1.6582159996032715, "seed": 1610712743, "started_at": "2026-11-22T03:04:04Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010637, 0.016836], "effect": 0.0031, "evaluated_at": "2026-11-22T03:04:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b6600f5be285", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.299022855, 0.153183895, 0.34971362, 0.269298115, 0.302982283, 0.3294377, 0.259630225, 0.24573097, 0.16786658, 0.217264985, 0.296218272, 0.26008142, 0.265055457, 0.171176772, 0.26166516, 0.23162393], "raw": [0.27110909, 0.136239015, 0.342054517, 0.298356297, 0.228302715, 0.282300532, 0.248813003, 0.235069575, 0.19869689, 0.18093432, 0.281125663, 0.22450114, 0.230589122, 0.179696948, 0.18450315, 0.191366995], "temperature": [0.299022855, 0.1833588, 0.344936638, 0.298356297, 0.248925893, 0.31344453, 0.285768172, 0.267672227, 0.23444469, 0.204995637, 0.281125663, 0.26008142, 0.265055457, 0.179696948, 0.231037202, 0.23162393]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2127, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.613", "prereg_hash": "22c746e65045223ad81a0ada3b23b08dffa6a1b8f2eb289bae5407240c792a00", "primary_control": "temperature", "result_hash": "1448440eeca60ebe111a82e5172c24d002772bc9d5b6344cfa6574f036d96d60", "signature": "12cede4bee6c3243d96e2934fac356d12a35f1c659dc9121aaecdfb380ac67ee03e9b58988895b395d37977c107ab827e55742e2b32d969fe5cca52f76253b0b", "state": "VERIFIED", "summary": {"iw_temperature": 0.254997, "raw": 0.232104, "temperature": 0.258097}, "ties": 4, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:04:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:14Z", "cpu_seconds": 1.5057683279999878, "created_at": "2026-11-22T03:04:04Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-e95bf36b6831", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "c19f5e58a6fd678b1be0f5c2da48482ec158f21f0af1e9124669728455739f2b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-e95bf36b6831", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:04Z", "seed": 626657098, "treatment": "iw_temperature"}, "priority": 0.26049224663253157, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "0e9451b6362357065e48f3b9bb0b7b0f9a08dcf3e1dad2d724f8ca22f51aacf5", "run_seconds": 1.5972607135772705, "seed": 626657098, "started_at": "2026-11-22T03:04:13Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012015, 0.016205], "effect": 0.002095, "evaluated_at": "2026-11-22T03:04:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-e95bf36b6831", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.148377307, 0.16160546, 0.148088055, 0.086541268, 0.064180528, 0.098156227, 0.13868133, 0.150485858, 0.089227005, 0.128932445, 0.162718763, 0.160710125, 0.018379018, 0.130979993, 0.153882112, 0.10448079], "raw": [0.071089455, 0.105214335, 0.13632086, 0.099797855, 0.064180528, 0.103957713, 0.132799757, 0.179859523, 0.07275409, 0.128932445, 0.135581563, 0.19130974, 0.167596888, 0.100087465, 0.107715485, 0.10448079], "temperature": [0.091976638, 0.16160546, 0.148088055, 0.104911195, 0.109289698, 0.114131045, 0.118874513, 0.189176285, 0.102673518, 0.143606802, 0.153493943, 0.200019375, 0.03505031, 0.100087465, 0.118213532, 0.087750853]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.4031, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.395", "prereg_hash": "c19f5e58a6fd678b1be0f5c2da48482ec158f21f0af1e9124669728455739f2b", "primary_control": "temperature", "result_hash": "0e9451b6362357065e48f3b9bb0b7b0f9a08dcf3e1dad2d724f8ca22f51aacf5", "signature": "31e639ef5623cbe2447cbff362a05a637fa1c4895f988164300feaff7e78282c552a2ea885df213f21faf41ecbb22885287f67c4bb62cbdaed5c8c77e6b7ad0a", "state": "VERIFIED", "summary": {"iw_temperature": 0.121589, "raw": 0.118855, "temperature": 0.123684}, "ties": 2, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:04:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:11Z", "cpu_seconds": 6.607178750000003, "created_at": "2026-11-22T03:04:04Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a5cbf4c14d6c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "2f62ceabf4c6f5f2132ab0f273695186bd114a086e7a7b28d6bf6e006e20394b", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a5cbf4c14d6c", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:04:04Z", "seed": 306252592, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "d871bac770f97bef3f9a0b528e1cedd3c0a9743419d4e5604c479634ef470078", "run_seconds": 6.663407325744629, "seed": 306252592, "started_at": "2026-11-22T03:04:04Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 4, "reset": 2}, "better": "higher", "ci95": [-0.271822, 0.038489], "effect": -0.116667, "evaluated_at": "2026-11-22T03:04:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a5cbf4c14d6c", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.33076587804125634, "rstep": 3}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0], "reset": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0207, "losses": 15, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.953", "prereg_hash": "2f62ceabf4c6f5f2132ab0f273695186bd114a086e7a7b28d6bf6e006e20394b", "primary_control": "reset", "result_hash": "d871bac770f97bef3f9a0b528e1cedd3c0a9743419d4e5604c479634ef470078", "signature": "a119f985b28eecf4dd63a9680cd925569413631525149920011cd88d175c6da1c9013b68b3978f27935869651db0b3cb4fe5df87b8add899dc0e3f5c2bc86e05", "state": "VERIFIED", "summary": {"inherit": 0.5833333333333334, "reset": 0.7, "root": 0.4166666666666667}, "ties": 37, "treatment": "inherit", "wins": 8}, "verified_at": "2026-11-22T03:04:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:26Z", "cpu_seconds": 1.8570057219999967, "created_at": "2026-11-22T03:04:04Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-70f30e6e9f4b", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "e3eed314d740d40bb860b219ceed1fda59ad34998a407d36a9432196ac0c4ce5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-70f30e6e9f4b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:04:04Z", "seed": 1155340677, "treatment": "frontier"}, "priority": 0.25503166242763076, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-f5b84e8fff38", "result_hash": "cc302cc6481084714ff46bf4cc7ddb9898aede3c16d61bad9877b578be229052", "run_seconds": 3.1708743572235107, "seed": 1155340677, "started_at": "2026-11-22T03:05:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.032618, 0.071229], "effect": 0.051924, "evaluated_at": "2026-11-22T03:05:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-70f30e6e9f4b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.91796875, 0.8828125, 0.964788732, 0.97972973, 0.875, 0.975694444, 0.928030303, 0.975352113, 0.920454545, 0.982876712, 0.9375, 0.953571429], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.921875, 0.90625, 0.978873239, 0.989864865, 0.879032258, 0.972222222, 0.943181818, 0.964788732, 0.943181818, 0.97260274, 0.926470588, 0.978571429]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "e3eed314d740d40bb860b219ceed1fda59ad34998a407d36a9432196ac0c4ce5", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-f5b84e8fff38", "result_hash": "cc302cc6481084714ff46bf4cc7ddb9898aede3c16d61bad9877b578be229052", "signature": "d98d04fee87048d2c485669f04481d2fe1891baf4a33fe9ca471c70f3675beeba15e51180e77f4872b709e24464a8ccf4a6e942684a0bf8cc085613e76129907", "state": "VERIFIED", "summary": {"curiosity": 0.941148, "frontier": 1.0, "random_walk": 0.948076}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:05:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:08Z", "cpu_seconds": 0.32628275799999074, "created_at": "2026-11-22T03:04:04Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-ce94b411f359", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "afeddfd6e5420f4004dc1f1cea9a413cd220c4f3a416173f4b32bd1e4c228402", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-ce94b411f359", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:04:04Z", "seed": 368404844, "treatment": "frontier"}, "priority": 0.31353686774446404, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-d9d5d5ce1fc3", "result_hash": "9256e1ae39895fc584d61af1fbd0b7c664c1fc4bbe59c5ff34ff2aa08f6c9824", "run_seconds": 0.37779974937438965, "seed": 368404844, "started_at": "2026-11-22T03:04:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.026786, 0.042207], "effect": 0.034496, "evaluated_at": "2026-11-22T03:04:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-ce94b411f359", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.915354331, 0.926923077, 0.8875, 0.849137931, 0.845982143, 0.902542373, 0.892, 0.896694215, 0.899590164, 0.897058824, 0.892857143, 0.902892562], "frontier": [0.938976378, 0.946153846, 0.929166667, 0.905172414, 0.879464286, 0.917372881, 0.93, 0.944214876, 0.922131148, 0.922268908, 0.922619048, 0.944214876], "random_walk": [0.907480315, 0.926923077, 0.883333333, 0.885775862, 0.839285714, 0.870762712, 0.888, 0.890495868, 0.891393443, 0.894957983, 0.912698413, 0.896694215]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "afeddfd6e5420f4004dc1f1cea9a413cd220c4f3a416173f4b32bd1e4c228402", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-d9d5d5ce1fc3", "result_hash": "9256e1ae39895fc584d61af1fbd0b7c664c1fc4bbe59c5ff34ff2aa08f6c9824", "signature": "c84c33c9b0efccbdfb166f262f8e23dddc98727492f108442da9a0e1bc03e65b0064e03f9a5dbe5056b01d7f7d9862e2be3fa8f9ec8912c6c5ae9a01cfc9d602", "state": "VERIFIED", "summary": {"curiosity": 0.892378, "frontier": 0.925146, "random_walk": 0.89065}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:04:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:46Z", "cpu_seconds": 1.5830061059999991, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-29800edf289d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "a7d45fe0dd87c2ca710440d9864d1f2447a5ed5534c378baae7055cfa5f5c7bf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-29800edf289d", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:20Z", "seed": 927461199, "treatment": "iw_temperature"}, "priority": 0.2631635040720448, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "e01e6c61926d59dbb2e061288a4b1319703919c878b513bc2315ed8ff3c2907b", "run_seconds": 1.6083958148956299, "seed": 927461199, "started_at": "2026-11-22T03:04:44Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012984, 0.005539], "effect": -0.003723, "evaluated_at": "2026-11-22T03:04:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-29800edf289d", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.064675435, 0.040468693, 0.003362123, 0.00939535, 0.012481685, 0.055778315, 0.03966457, 0.01056899, 0.063673118, 0.08829015, 0.048126158, 0.02943448, 0.00979707, 0.04085563, 0.052997777, 0.062021875], "raw": [0.035524942, 0.053448123, 0.005891635, 0.023132477, 0.012481685, 0.03543126, 0.058458788, 0.037342495, 0.07782015, 0.091128747, 0.061069608, 0.031356593, 0.00722062, 0.016628178, 0.05740315, 0.042514772], "temperature": [0.028004455, 0.036307247, 0.002805675, 0.013897472, 0.015421222, 0.0234448, 0.052850885, 0.054132457, 0.063673118, 0.096607035, 0.045838395, 0.027180215, 0.002204265, 0.016628178, 0.049308495, 0.043725618]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "a7d45fe0dd87c2ca710440d9864d1f2447a5ed5534c378baae7055cfa5f5c7bf", "primary_control": "temperature", "result_hash": "e01e6c61926d59dbb2e061288a4b1319703919c878b513bc2315ed8ff3c2907b", "signature": "c73e09ef44ec8695ef7c2efcab4824ace863dd6e6cdd95236ff763dbacff649c835191f5a8f80f11d72b0720c17de972a9aac18db5963bf1ea5fe5c9ef2d2400", "state": "VERIFIED", "summary": {"iw_temperature": 0.039474, "raw": 0.040428, "temperature": 0.035752}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:04:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:24Z", "cpu_seconds": 1.4584752469999955, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-9be250fbe0d1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "84f6634465fc1723d9c822d1d9462f53aafe75985b1eced1dd3eddcc9e49b8fb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-9be250fbe0d1", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:20Z", "seed": 224944622, "treatment": "iw_temperature"}, "priority": 0.3781595905596689, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "7e3cac06abac525af1dddead51d348ceb9409d6807201ca75857de470081c0ee", "run_seconds": 1.46891188621521, "seed": 224944622, "started_at": "2026-11-22T03:04:23Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018526, 0.019313], "effect": 0.000393, "evaluated_at": "2026-11-22T03:04:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-9be250fbe0d1", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.191654437, 0.240481867, 0.254329038, 0.26385448, 0.214314767, 0.325371263, 0.328814237, 0.297350973, 0.29948442, 0.193416857, 0.291518198, 0.284817492, 0.218031633, 0.28693494, 0.23597488, 0.194849098], "raw": [0.131205035, 0.20997486, 0.210000055, 0.207375487, 0.21030875, 0.248927532, 0.308987348, 0.293925105, 0.292280428, 0.216765275, 0.275851945, 0.296491495, 0.230990837, 0.24464312, 0.216731967, 0.22196256], "temperature": [0.214816885, 0.22946306, 0.210000055, 0.207375487, 0.234942092, 0.284225318, 0.315415625, 0.293925105, 0.296383485, 0.279775037, 0.275851945, 0.296491495, 0.271232287, 0.24464312, 0.23597488, 0.236971792]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "84f6634465fc1723d9c822d1d9462f53aafe75985b1eced1dd3eddcc9e49b8fb", "primary_control": "temperature", "result_hash": "7e3cac06abac525af1dddead51d348ceb9409d6807201ca75857de470081c0ee", "signature": "ffb4c229cb19262b1dbec50c3fb730844ff793ad3b327805d6392d391df77b1ae7d1a1c8d4e584cd62b01e1ff5f6d2f590d732375200e8963f41dc2f74a1a00f", "state": "VERIFIED", "summary": {"iw_temperature": 0.257575, "raw": 0.238526, "temperature": 0.257968}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:04:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:56Z", "cpu_seconds": 1.6465864150000016, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b00bd6c19bda", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "90791d753af677b72d41be3be7e5586779125eda342e3dd40a644bf496706093", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b00bd6c19bda", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:20Z", "seed": 1128969006, "treatment": "iw_temperature"}, "priority": 0.25364351942317886, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a1af86f600ed5340f654d5b5fd153821d17f1730f161decb1e8e90acaefe7344", "run_seconds": 1.7484583854675293, "seed": 1128969006, "started_at": "2026-11-22T03:05:55Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021929, 0.000472], "effect": -0.010729, "evaluated_at": "2026-11-22T03:06:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b00bd6c19bda", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.203729395, 0.21797414, 0.177391743, 0.24670239, 0.119429622, 0.238466585, 0.2408614, 0.179732443, 0.206509028, 0.264127243, 0.251594555, 0.233668798, 0.228370545, 0.264615545, 0.267300402, 0.239093147], "raw": [0.151456162, 0.17237251, 0.15927075, 0.197879295, 0.17540361, 0.219694313, 0.203582223, 0.190390923, 0.235095467, 0.188532788, 0.188963342, 0.1707332, 0.189581575, 0.209884377, 0.23346711, 0.170054568], "temperature": [0.203729395, 0.21797414, 0.13634312, 0.24670239, 0.090882895, 0.230238852, 0.2408614, 0.179732443, 0.24576562, 0.280632625, 0.245587903, 0.1836556, 0.189581575, 0.233294002, 0.255653158, 0.22727288]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0026, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.994", "prereg_hash": "90791d753af677b72d41be3be7e5586779125eda342e3dd40a644bf496706093", "primary_control": "temperature", "result_hash": "a1af86f600ed5340f654d5b5fd153821d17f1730f161decb1e8e90acaefe7344", "signature": "cb133f35e1767734e7c97e9a6e9ffc4650d1da8e75bf176d6388836f86e548ca134ba8ae4310a9779375f513b93eb5e3b59f3dcaaa7bb7bac16914c9b05f1b07", "state": "VERIFIED", "summary": {"iw_temperature": 0.223723, "raw": 0.191023, "temperature": 0.212994}, "ties": 5, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:06:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:32Z", "cpu_seconds": 6.786132000000009, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-7a505fa7d4bc", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "71b98d88a4965898fc0694da44863807736fdb5f7926f1df8200aecf0209f94a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-7a505fa7d4bc", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:04:20Z", "seed": 884469287, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "8e8444805de5eecdb0a81c1fef50475a1e52451ea6e6483e034999eb4fe79bfe", "run_seconds": 7.133978605270386, "seed": 884469287, "started_at": "2026-11-22T03:04:25Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:04:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-7a505fa7d4bc", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6005837231445188, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "reset": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "71b98d88a4965898fc0694da44863807736fdb5f7926f1df8200aecf0209f94a", "primary_control": "reset", "result_hash": "8e8444805de5eecdb0a81c1fef50475a1e52451ea6e6483e034999eb4fe79bfe", "signature": "385650e8aded7d8bf51300da01c4ca3c3255a2a10281cb1b5db815a89c781e97c7a2b52601f1b3ac051cbe9018f495b6163d7ad60b05b29b1bfc4d345223b704", "state": "PARTIAL", "summary": {"inherit": 0.5333333333333333, "reset": 0.5333333333333333, "root": 0.2833333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:04:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:26Z", "cpu_seconds": 5.903142298000006, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-852bfaba337f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f59dd6815e473070e30d36401e0a71ea3e454d8c97063e4039776400ad1d27af", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-852bfaba337f", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:04:20Z", "seed": 1075250880, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9a294e003618ae082e230c63d35147e42c071e29a8f432dd5b7b5391070c20be", "run_seconds": 5.945267200469971, "seed": 1075250880, "started_at": "2026-11-22T03:04:21Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:04:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-852bfaba337f", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f59dd6815e473070e30d36401e0a71ea3e454d8c97063e4039776400ad1d27af", "primary_control": "reset", "result_hash": "9a294e003618ae082e230c63d35147e42c071e29a8f432dd5b7b5391070c20be", "signature": "d37b5cd62f83d794b967f310e016ecfa403ce6acf022d09089458141ec30c6a5cc83ba84cc087aa36efb94dee6593c5c839bdb1afcc34dd6af875526b2be980f", "state": "PARTIAL", "summary": {"inherit": 0.6333333333333333, "reset": 0.6333333333333333, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:04:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:27Z", "cpu_seconds": 0.4277585810000346, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-1f302df359f8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "5c3d7388659ac1e99952a6db6fb717df3372b11c220c9acf9a1b67f7afe1ac24", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-1f302df359f8", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:04:20Z", "seed": 1175250431, "treatment": "frontier"}, "priority": 0.2477439641357973, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "23811dbf64c22de7ce27dfb1609f683eab6321c12f1b7deba48e68ec85f1b5c2", "run_seconds": 0.43038439750671387, "seed": 1175250431, "started_at": "2026-11-22T03:06:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.05981, 0.149005], "effect": 0.104407, "evaluated_at": "2026-11-22T03:06:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-1f302df359f8", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.877777778, 0.882075472, 1.0, 0.886792453, 0.762820513, 0.783333333, 0.830357143, 1.0, 0.875, 0.790909091, 0.8125, 0.797169811], "frontier": [1.0, 0.825471698, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.895833333, 1.0, 1.0, 0.811320755], "random_walk": [0.85, 0.787735849, 1.0, 0.820754717, 0.852564103, 0.855555556, 0.772321429, 1.0, 0.847222222, 0.845454545, 0.850961538, 0.797169811]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "5c3d7388659ac1e99952a6db6fb717df3372b11c220c9acf9a1b67f7afe1ac24", "primary_control": "random_walk", "result_hash": "23811dbf64c22de7ce27dfb1609f683eab6321c12f1b7deba48e68ec85f1b5c2", "signature": "4e6f27671eac0e4af6a13a42467546c76c2b395336cfaceefe8a8d058754e22f05b8c68f318240f538a30f0a250ce6358c97ffbd9f60c16475b8c29ac022f009", "state": "VERIFIED", "summary": {"curiosity": 0.858228, "frontier": 0.961052, "random_walk": 0.856645}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:06:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:14Z", "cpu_seconds": 0.9194161659999907, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-ae204ab1eb94", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "0e2d795f6a83fe1d78ca2bf2272f47c419556839686de45004cca90430cfad4a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-ae204ab1eb94", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:04:20Z", "seed": 383073900, "treatment": "frontier"}, "priority": 0.2508233317198415, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "ef0f2dd2ace3377f27f1d30658e0193803c8e02ba2b441cb0cad592e4eec1745", "run_seconds": 1.0793216228485107, "seed": 383073900, "started_at": "2026-11-22T03:06:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.065547, 0.176979], "effect": 0.121263, "evaluated_at": "2026-11-22T03:06:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-ae204ab1eb94", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.77688172, 0.798507463, 0.855555556, 1.0, 0.830188679, 0.768987342, 0.800531915, 0.766129032, 0.853773585, 0.834951456, 0.838095238], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.747340426, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.752688172, 0.805970149, 0.830555556, 1.0, 0.853773585, 0.829113924, 0.829787234, 0.811827957, 0.882075472, 0.810679612, 0.885714286]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.9689, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0107", "prereg_hash": "0e2d795f6a83fe1d78ca2bf2272f47c419556839686de45004cca90430cfad4a", "primary_control": "random_walk", "result_hash": "ef0f2dd2ace3377f27f1d30658e0193803c8e02ba2b441cb0cad592e4eec1745", "signature": "f103ebb00a13ee3b02c9b5db25fa2dfd3293591efb1cee6cac42e0f533f3e8e0f4a7c4b834a683231895fc9b376b7229b6e89d72765ecc063ec7d0f1674de80f", "state": "VERIFIED", "summary": {"curiosity": 0.843633, "frontier": 0.978945, "random_walk": 0.857682}, "ties": 2, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:06:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:32Z", "cpu_seconds": 2.705228665000007, "created_at": "2026-11-22T03:04:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-ce5dc369a23b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "233af80c5b4e672910379815916656031dc3fe898f3e907036902cf0ac8ee3b9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-ce5dc369a23b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:04:20Z", "seed": 258495141, "treatment": "frontier"}, "priority": 0.3196204731617716, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "c5fe2642d428b7916385e4aeaac1ead8b916ca6b3915e87855e95633468fc511", "run_seconds": 3.0192275047302246, "seed": 258495141, "started_at": "2026-11-22T03:04:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.037787, 0.12821], "effect": 0.082998, "evaluated_at": "2026-11-22T03:04:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-ce5dc369a23b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.864864865, 0.79375, 1.0, 0.798469388, 0.782967033, 1.0, 0.826732673, 0.795454545, 0.81443299, 1.0, 0.830729167, 0.886363636], "frontier": [1.0, 1.0, 1.0, 0.862244898, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.890625, 0.991477273], "random_walk": [0.903153153, 0.865625, 1.0, 0.864795918, 0.78021978, 1.0, 0.861386139, 0.792929293, 0.922680412, 1.0, 0.854166667, 0.903409091]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.7093, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0195", "prereg_hash": "233af80c5b4e672910379815916656031dc3fe898f3e907036902cf0ac8ee3b9", "primary_control": "random_walk", "result_hash": "c5fe2642d428b7916385e4aeaac1ead8b916ca6b3915e87855e95633468fc511", "signature": "4ea5441dc23016f8f98d585464a124e82ff0942a5e0c0e407232edfcd13092775d07f7ff654c49735f0aed01a0472576ae09bd95f359431b208e7c7d861dc101", "state": "VERIFIED", "summary": {"curiosity": 0.866147, "frontier": 0.978696, "random_walk": 0.895697}, "ties": 3, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:04:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:39Z", "cpu_seconds": 1.6171420280000177, "created_at": "2026-11-22T03:04:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-dd183dae9f61", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "a5b926474588f1f2cd33dcd446790ff24098b34483571e6326c2be148821cc9d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-dd183dae9f61", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:36Z", "seed": 324494720, "treatment": "iw_temperature"}, "priority": 0.35328949140747967, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "d0a68cc8126e5a982055605e12f2eb18d76ef064715a159dca90546ee6c8dd13", "run_seconds": 1.8749115467071533, "seed": 324494720, "started_at": "2026-11-22T03:04:37Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.016325, 0.029362], "effect": 0.006518, "evaluated_at": "2026-11-22T03:04:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-dd183dae9f61", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.357964435, 0.2416389, 0.258107677, 0.315362237, 0.2863746, 0.247485228, 0.297457012, 0.258836923, 0.179538745, 0.164271895, 0.251159132, 0.272249298, 0.099449935, 0.087036858, 0.29436042, 0.090087765], "raw": [0.342579265, 0.264925638, 0.262331157, 0.291514843, 0.194837677, 0.221957695, 0.28172641, 0.230928082, 0.215901102, 0.164037693, 0.175671775, 0.242206945, 0.201586315, 0.11187041, 0.222963555, 0.172897382], "temperature": [0.359362923, 0.273350845, 0.273818198, 0.291514843, 0.22481777, 0.231527047, 0.296277647, 0.239055593, 0.215901102, 0.177067643, 0.218769302, 0.252225498, 0.195652635, 0.12914655, 0.237809825, 0.189376917]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2232, "losses": 8, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.598", "prereg_hash": "a5b926474588f1f2cd33dcd446790ff24098b34483571e6326c2be148821cc9d", "primary_control": "temperature", "result_hash": "d0a68cc8126e5a982055605e12f2eb18d76ef064715a159dca90546ee6c8dd13", "signature": "8329248f886926cf5b64e34a8853a103f169dd0df2a9d28eb15a324a196329e439b1aac0c33f704b798e92350d80ecddb2c9e74158b692413d7d6b067751660b", "state": "VERIFIED", "summary": {"iw_temperature": 0.231336, "raw": 0.224871, "temperature": 0.237855}, "ties": 0, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:04:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:46Z", "cpu_seconds": 8.426639995000016, "created_at": "2026-11-22T03:04:36Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-ca7a73a86dbf", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ed73907b26d6c3da18a9d7b9c7b39b09ee5758ee110d2bf1915a408da7ba04ee", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-ca7a73a86dbf", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:04:36Z", "seed": 114337025, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "54e5f6505f54f7b12ab6c6c6de78a0f38a1ca8a561a508821d19521608cea715", "run_seconds": 9.306931972503662, "seed": 114337025, "started_at": "2026-11-22T03:04:36Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:04:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-ca7a73a86dbf", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ed73907b26d6c3da18a9d7b9c7b39b09ee5758ee110d2bf1915a408da7ba04ee", "primary_control": "reset", "result_hash": "54e5f6505f54f7b12ab6c6c6de78a0f38a1ca8a561a508821d19521608cea715", "signature": "977511a8a716b8aab9c249a8dc1375328f3cb6478bfab89bae284522650b1cd0c2dfcdfdaa5b1eeb64821ca969137588d28262eec69e2448ea7005cfb628ae0f", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:04:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:41Z", "cpu_seconds": 2.7769835749999885, "created_at": "2026-11-22T03:04:36Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-bdb680f38eca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "d4ed1f75b6a0bc736782161ccd3ca15080655f728c65156028a59cde98ee772b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-bdb680f38eca", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:04:36Z", "seed": 1976074222, "treatment": "frontier"}, "priority": 0.28348004049472314, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "49351d229da7aff696be9c5b7557503f8865f01e768fe600c0547fb9c3de4742", "run_seconds": 3.424715518951416, "seed": 1976074222, "started_at": "2026-11-22T03:04:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.044152, 0.121527], "effect": 0.082839, "evaluated_at": "2026-11-22T03:04:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-bdb680f38eca", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.818181818, 1.0, 0.875, 0.849537037, 0.881410256, 0.885964912, 0.91875, 1.0, 1.0, 0.841463415, 0.808823529], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.975, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.800505051, 1.0, 0.929166667, 0.877314815, 0.900641026, 0.89254386, 0.8625, 1.0, 1.0, 0.850609756, 0.867647059]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "d4ed1f75b6a0bc736782161ccd3ca15080655f728c65156028a59cde98ee772b", "primary_control": "random_walk", "result_hash": "49351d229da7aff696be9c5b7557503f8865f01e768fe600c0547fb9c3de4742", "signature": "aba45d13cc458495c6ad7ecd00c9b3cb23af9cdf1d9f214b45d1b54c1bd4820472369a487606f577929ef1376721d5c31d226cc796a26875eb3faef9da3d1903", "state": "VERIFIED", "summary": {"curiosity": 0.906594, "frontier": 0.997917, "random_walk": 0.915077}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:04:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:59Z", "cpu_seconds": 1.5057376569999974, "created_at": "2026-11-22T03:04:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4e12438970cb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "e0e1289ba2637f58e7d93ebab817d6f9064db3837eb15ffe0aebceb8dfc2edae", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4e12438970cb", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:52Z", "seed": 1408440194, "treatment": "iw_temperature"}, "priority": 0.2387648760689801, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "118e1192fdbc68cc05d9de03b28e97a9a0332c1d758e98ecc984d960439b6218", "run_seconds": 1.718231201171875, "seed": 1408440194, "started_at": "2026-11-22T03:06:58Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.00962, 0.01162], "effect": 0.001, "evaluated_at": "2026-11-22T03:07:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4e12438970cb", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.2742096, 0.187725902, 0.14249202, 0.224163932, 0.160773115, 0.16543031, 0.167013568, 0.218176865, 0.258497897, 0.205125757, 0.186828755, 0.211336105, 0.193275853, 0.237473297, 0.199286338, 0.1998957], "raw": [0.21346838, 0.21633504, 0.186050862, 0.171693792, 0.170271332, 0.238608955, 0.203839465, 0.19463138, 0.20864047, 0.17322183, 0.186828755, 0.211336105, 0.193275853, 0.22864596, 0.223162745, 0.206139495], "temperature": [0.258510477, 0.21633504, 0.13122447, 0.187680508, 0.190915343, 0.176159877, 0.179982185, 0.218176865, 0.23680808, 0.200769545, 0.229125672, 0.215705015, 0.212692428, 0.22864596, 0.184330965, 0.180638387]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1572, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.696", "prereg_hash": "e0e1289ba2637f58e7d93ebab817d6f9064db3837eb15ffe0aebceb8dfc2edae", "primary_control": "temperature", "result_hash": "118e1192fdbc68cc05d9de03b28e97a9a0332c1d758e98ecc984d960439b6218", "signature": "2658d9958b1551344e84ff4a698a729ac941fc603d51a2238804091a4cf5a13a1fcf50a72a3b0cfa8a356c4076d00996f40452bdd12c1e06155dbc8cb35e0300", "state": "VERIFIED", "summary": {"iw_temperature": 0.201982, "raw": 0.201634, "temperature": 0.202981}, "ties": 1, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:58Z", "cpu_seconds": 1.6550021320000212, "created_at": "2026-11-22T03:04:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-89ba50873173", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "1d1b56e97c371dc295500c86a16e212a6e71b9457e2934c0cb128bef0844a098", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-89ba50873173", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:52Z", "seed": 1258198705, "treatment": "iw_temperature"}, "priority": 0.2674466636932463, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-8562032d1f71", "result_hash": "76f83c815c2c4fd7fcea0e19bb3918ea87f06e1cdf78905e9602e39c2f0d0bd3", "run_seconds": 1.753704309463501, "seed": 1258198705, "started_at": "2026-11-22T03:04:56Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.013673, 0.00848], "effect": -0.002597, "evaluated_at": "2026-11-22T03:05:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-89ba50873173", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.147032703, 0.13987222, 0.097233258, 0.139510587, 0.179210668, 0.128737915, 0.095244378, 0.14425149, 0.139056808, 0.199880295, 0.09031494, 0.150422082, 0.132273373, 0.10840937, 0.149004875, 0.14103898], "raw": [0.11091938, 0.089984683, 0.104608575, 0.099929527, 0.1224836, 0.080064505, 0.093999537, 0.090523525, 0.111005443, 0.10848304, 0.105621278, 0.10943396, 0.119565885, 0.086956718, 0.118033675, 0.127312917], "temperature": [0.147032703, 0.122725137, 0.132473645, 0.13538385, 0.179210668, 0.113430748, 0.119093335, 0.111208982, 0.139056808, 0.16641894, 0.12472606, 0.137646623, 0.122170395, 0.07175698, 0.166825522, 0.150787895]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "1d1b56e97c371dc295500c86a16e212a6e71b9457e2934c0cb128bef0844a098", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-8562032d1f71", "result_hash": "76f83c815c2c4fd7fcea0e19bb3918ea87f06e1cdf78905e9602e39c2f0d0bd3", "signature": "76e8c4f2e7654ffe3e3f182ffa9e32e05e520651f084fe71072b95752199e57cc8ed1a3303c4938df69601232ff64aef3b98c41552601bfca184620450a57d0e", "state": "VERIFIED", "summary": {"iw_temperature": 0.136343, "raw": 0.104933, "temperature": 0.133747}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:05:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:34Z", "cpu_seconds": 1.5414494380000008, "created_at": "2026-11-22T03:04:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-9e2c364df563", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "1fa5d8a4eb64fe27cc35daabc16f2b3dfcf6a728f4f97e670f3da1897e924e3a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-9e2c364df563", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:52Z", "seed": 794170708, "treatment": "iw_temperature"}, "priority": 0.25416197530128376, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "3ee2d99ff39cfc89580bd1055937480301aa51475ab13e4cd1523932d9c39200", "run_seconds": 3.0241949558258057, "seed": 794170708, "started_at": "2026-11-22T03:05:31Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010547, 0.026702], "effect": 0.008077, "evaluated_at": "2026-11-22T03:05:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-9e2c364df563", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.249056035, 0.211360857, 0.231968563, 0.22952049, 0.274624398, 0.210892137, 0.137335438, 0.215906337, 0.243872175, 0.248935715, 0.2263927, 0.133842777, 0.19873611, 0.316279945, 0.246030658, 0.191002917], "raw": [0.187375805, 0.22935235, 0.18639993, 0.243090473, 0.223722058, 0.221991298, 0.163635087, 0.245692012, 0.214611195, 0.133128792, 0.2263927, 0.147033543, 0.2358906, 0.292795125, 0.263578918, 0.21782053], "temperature": [0.238354325, 0.2143632, 0.24770616, 0.218709315, 0.223722058, 0.221991298, 0.186218973, 0.245692012, 0.24654328, 0.184366495, 0.187550838, 0.16154159, 0.262177323, 0.312136213, 0.29572257, 0.248198765]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6435, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.227", "prereg_hash": "1fa5d8a4eb64fe27cc35daabc16f2b3dfcf6a728f4f97e670f3da1897e924e3a", "primary_control": "temperature", "result_hash": "3ee2d99ff39cfc89580bd1055937480301aa51475ab13e4cd1523932d9c39200", "signature": "4af49f36e9a35fff1d32555efc1b21f3398d0652b3fd5a38b0e875a1e58fd722d884d7e216f49f44c234f9b7bd43f979b06cefe23fc66508b00a66f610ff5c08", "state": "VERIFIED", "summary": {"iw_temperature": 0.22286, "raw": 0.214532, "temperature": 0.230937}, "ties": 0, "treatment": "iw_temperature", "wins": 10}, "verified_at": "2026-11-22T03:05:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:04:55Z", "cpu_seconds": 1.743909862999999, "created_at": "2026-11-22T03:04:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c9d93809c76c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "dcadc751aaa103481b0f33ca31aa24f4ffa86c429c14d683760e76d5b79dc9aa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c9d93809c76c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:04:52Z", "seed": 998966197, "treatment": "iw_temperature"}, "priority": 0.34472921324392886, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "7d498cb765ca00d638a683af6598682ddbc92dc89954e39e5a19296490d972d2", "run_seconds": 1.9474701881408691, "seed": 998966197, "started_at": "2026-11-22T03:04:53Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.032996, 0.006186], "effect": -0.013405, "evaluated_at": "2026-11-22T03:04:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c9d93809c76c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.24315122, 0.324621407, 0.328944302, 0.295584135, 0.313738335, 0.207766437, 0.3122779, 0.185547445, 0.16267369, 0.308531915, 0.311396995, 0.275733803, 0.25608822, 0.22974333, 0.195229397, 0.22069064], "raw": [0.179209895, 0.30072505, 0.243182977, 0.258190612, 0.262663645, 0.278993935, 0.306473515, 0.17515311, 0.109540178, 0.239185407, 0.185973752, 0.18375458, 0.178714465, 0.187641345, 0.159499178, 0.253470693], "temperature": [0.201922928, 0.322273333, 0.295202008, 0.265116018, 0.262663645, 0.288741003, 0.313429563, 0.20621364, 0.16267369, 0.275090427, 0.234430365, 0.242785822, 0.21789916, 0.187641345, 0.227688133, 0.253470693]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "dcadc751aaa103481b0f33ca31aa24f4ffa86c429c14d683760e76d5b79dc9aa", "primary_control": "temperature", "result_hash": "7d498cb765ca00d638a683af6598682ddbc92dc89954e39e5a19296490d972d2", "signature": "35562d8c1b20d09b092c43bc584bb76d2b05ebb0030e34acf9da1b69e494381ed02b12c77c918f17c0c386eae43c067f76bd982f3369936e004ac7363d418d06", "state": "VERIFIED", "summary": {"iw_temperature": 0.260732, "raw": 0.218898, "temperature": 0.247328}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:04:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:02Z", "cpu_seconds": 7.508590830999992, "created_at": "2026-11-22T03:04:52Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-07bb86520814", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "96a5e765ff028ba1935610edf49343558146f00a3d43ddc68c7002847fcdad97", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-07bb86520814", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:04:52Z", "seed": 1561264472, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f2be0a50c823ea0396085b2cbfe421a9999f8ee815b474a6ae44a54c9b67a115", "run_seconds": 8.02938437461853, "seed": 1561264472, "started_at": "2026-11-22T03:04:54Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 4, "reset": 4}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:05:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-07bb86520814", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "96a5e765ff028ba1935610edf49343558146f00a3d43ddc68c7002847fcdad97", "primary_control": "reset", "result_hash": "f2be0a50c823ea0396085b2cbfe421a9999f8ee815b474a6ae44a54c9b67a115", "signature": "9e485fe9a48ca9f961791f6f1c3717514c7ca70b0e61a72da2c3205ad4a63ee1f4bb0d2fff18661879c4bc094e8d90b881cc05475ca66b668ffc2548b4a08403", "state": "PARTIAL", "summary": {"inherit": 0.9, "reset": 0.9, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:05:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:03Z", "cpu_seconds": 3.4914026459999974, "created_at": "2026-11-22T03:04:52Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-bdaf69650c7d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "71ab1f7076a54771ae18b80415c2fbebeddff5c42429dff649d891d1dfbcee35", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-bdaf69650c7d", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:04:52Z", "seed": 558550447, "treatment": "frontier"}, "priority": 0.26732227381663914, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "89297fd267c5acf8049545d4fd06c21f20f46640642bdf9d5f098eedd78fed7f", "run_seconds": 3.6935713291168213, "seed": 558550447, "started_at": "2026-11-22T03:05:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058923, 0.124027], "effect": 0.091475, "evaluated_at": "2026-11-22T03:05:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-bdaf69650c7d", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.923076923, 0.785714286, 0.851941748, 0.816455696, 0.747126437, 0.8175, 0.813291139, 0.816037736, 0.818681319, 1.0, 0.891176471, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 0.995, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.973076923, 0.849489796, 0.932038835, 0.901898734, 0.864942529, 0.885, 0.857594937, 0.830188679, 0.914835165, 1.0, 0.888235294, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "71ab1f7076a54771ae18b80415c2fbebeddff5c42429dff649d891d1dfbcee35", "primary_control": "random_walk", "result_hash": "89297fd267c5acf8049545d4fd06c21f20f46640642bdf9d5f098eedd78fed7f", "signature": "5c4aa42dd944f266859cb2f3d75ecbcbe01517791c49e620de26aaa3199adfc096dbc4c0055336fca6d39c041bb7b72917869e6d3537fc626cd8efc72254e908", "state": "VERIFIED", "summary": {"curiosity": 0.85675, "frontier": 0.999583, "random_walk": 0.908108}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:05:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:18Z", "cpu_seconds": 1.6637509440000002, "created_at": "2026-11-22T03:05:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5ff69429c3d3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "d9c793511dcffc7e3cfeb2c55c2cf99cb6e387ad556b3597021da12d1fc13455", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5ff69429c3d3", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:09Z", "seed": 1190736464, "treatment": "iw_temperature"}, "priority": 0.367581675533988, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "7ceef2f2f4c8bbb9efe358c951a787019e10c29364c42700338c1f97fd32a11d", "run_seconds": 2.0006659030914307, "seed": 1190736464, "started_at": "2026-11-22T03:05:16Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019985, 0.002732], "effect": -0.008626, "evaluated_at": "2026-11-22T03:05:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5ff69429c3d3", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.272335797, 0.3040315, 0.233040803, 0.257776625, 0.18472463, 0.28449503, 0.191203408, 0.160165628, 0.293909415, 0.239478643, 0.271348857, 0.178855948, 0.314052875, 0.322677727, 0.288197235, 0.284786348], "raw": [0.248204272, 0.268021798, 0.194560217, 0.261747045, 0.18472463, 0.26974466, 0.191203408, 0.192470235, 0.25100397, 0.20914211, 0.235983913, 0.158979773, 0.299435692, 0.314522578, 0.276866795, 0.279989402], "temperature": [0.26317928, 0.278167318, 0.21912004, 0.257776625, 0.156879307, 0.28449503, 0.176059192, 0.204968392, 0.22996678, 0.20914211, 0.271348857, 0.190612348, 0.318054163, 0.321968987, 0.276866795, 0.284453253]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "d9c793511dcffc7e3cfeb2c55c2cf99cb6e387ad556b3597021da12d1fc13455", "primary_control": "temperature", "result_hash": "7ceef2f2f4c8bbb9efe358c951a787019e10c29364c42700338c1f97fd32a11d", "signature": "88e4bca394b142702e569148ae527fd344a89db6778a0f918a06331feec6d908a68247d84e8cbbe7b8c12dfd64b7f303deaa8b4a698a1d0b9206ec8de62bb109", "state": "VERIFIED", "summary": {"iw_temperature": 0.255068, "raw": 0.239788, "temperature": 0.246441}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:05:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:42Z", "cpu_seconds": 1.475096972000017, "created_at": "2026-11-22T03:05:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-aa09c6250174", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "1790998a5ee0dc2dfb8a3d702e0fa5c7657e673b09171345cfdba6f2b245705d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-aa09c6250174", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:09Z", "seed": 590691379, "treatment": "iw_temperature"}, "priority": 0.2453472096580432, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "66bdb4e5fa383e5ab2374de2014fb87facc5e1af8617245037fd2276b5738652", "run_seconds": 1.906667947769165, "seed": 590691379, "started_at": "2026-11-22T03:06:40Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.005074, 0.006528], "effect": 0.000727, "evaluated_at": "2026-11-22T03:06:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-aa09c6250174", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.14674321, 0.063022832, 0.091832732, 0.045881972, 0.061866205, 0.020305703, 0.016460225, 0.03511202, 0.08421972, 0.031269118, 0.026570627, 0.055064105, 0.06219851, 0.04989697, 0.055029355, 0.132212572], "raw": [0.162357885, 0.07662879, 0.129032097, 0.060532687, 0.04690008, 0.024313712, 0.019034355, 0.046868708, 0.087188445, 0.027606265, 0.024468085, 0.043526245, 0.054109007, 0.055150015, 0.0562711, 0.14408732], "temperature": [0.156491533, 0.063022832, 0.123177475, 0.044538033, 0.04690008, 0.020305703, 0.019034355, 0.046868708, 0.087188445, 0.031269118, 0.02247391, 0.03627911, 0.054109007, 0.046778972, 0.046786248, 0.14408732]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "1790998a5ee0dc2dfb8a3d702e0fa5c7657e673b09171345cfdba6f2b245705d", "primary_control": "temperature", "result_hash": "66bdb4e5fa383e5ab2374de2014fb87facc5e1af8617245037fd2276b5738652", "signature": "2ee6ac3dbb5f9533f20a1442e38a4e98fd5aed761e8c4e5e9fe250a5ab3758536e2cf9421b0ff947478e0d5a3dd54d97e939a6623d9984f21cf2797ab51fd80a", "state": "VERIFIED", "summary": {"iw_temperature": 0.061105, "raw": 0.06613, "temperature": 0.061832}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:06:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:17Z", "cpu_seconds": 1.7593246710000017, "created_at": "2026-11-22T03:05:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-bf4237dbcef0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "f95d52fd0077c0afef7d679f1e033f1f9734962799f36fb0696850234b00bc58", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-bf4237dbcef0", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:09Z", "seed": 2094178908, "treatment": "iw_temperature"}, "priority": 0.22538011207488717, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "7db18d58dcddbfad77c3d6e9321ceca91741ef8608fb3215b76f30b1c64c5f71", "run_seconds": 2.4235968589782715, "seed": 2094178908, "started_at": "2026-11-22T03:07:15Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021654, -0.005008], "effect": -0.013331, "evaluated_at": "2026-11-22T03:07:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-bf4237dbcef0", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.111155278, 0.086962797, 0.134135108, 0.15862297, 0.195137063, 0.115321485, 0.18768178, 0.173174135, 0.164503502, 0.16027298, 0.132820333, 0.150642895, 0.150596157, 0.147168217, 0.121662193, 0.14890496], "raw": [0.111155278, 0.05135292, 0.124518902, 0.14242488, 0.182673687, 0.066389392, 0.110968445, 0.159444275, 0.092664905, 0.155275763, 0.129255385, 0.077887482, 0.104953665, 0.11003707, 0.113424145, 0.081343733], "temperature": [0.111155278, 0.05135292, 0.134135108, 0.14778537, 0.219354685, 0.111859607, 0.16438523, 0.159444275, 0.120558305, 0.163749292, 0.129255385, 0.12317848, 0.12533325, 0.123741295, 0.113424145, 0.126750523]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0004, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.999", "prereg_hash": "f95d52fd0077c0afef7d679f1e033f1f9734962799f36fb0696850234b00bc58", "primary_control": "temperature", "result_hash": "7db18d58dcddbfad77c3d6e9321ceca91741ef8608fb3215b76f30b1c64c5f71", "signature": "366d3d6e15a696f893deaf51c3cc6a322add663fb0bd57c49bf1a9482195e9598b661b7025741784a006f5517dd99a3ef73f710a19d11aa53cdc0e0f16d1e508", "state": "VERIFIED", "summary": {"iw_temperature": 0.146173, "raw": 0.113361, "temperature": 0.132841}, "ties": 2, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:07:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:17Z", "cpu_seconds": 5.999077569000008, "created_at": "2026-11-22T03:05:09Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-6e5687154703", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d36ba3b3aab462ad17f3fc3e0ccd8310565404e073169ec9183a6b669eea2700", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-6e5687154703", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:05:09Z", "seed": 424371430, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "3492528d3790e83b8c0035515a8989f088a9e9f465c4fe94cf20c9d0bb5375aa", "run_seconds": 7.133452892303467, "seed": 424371430, "started_at": "2026-11-22T03:05:10Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:05:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-6e5687154703", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0], "root": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d36ba3b3aab462ad17f3fc3e0ccd8310565404e073169ec9183a6b669eea2700", "primary_control": "reset", "result_hash": "3492528d3790e83b8c0035515a8989f088a9e9f465c4fe94cf20c9d0bb5375aa", "signature": "4c05c85377966d57800e17332a3de10b9abfae4991aa1abef2def45b598343b00c9d1bf5db0ae4cef31b2746d15e9dcf98c6b8d484e496e855080213c571ea05", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:05:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:01Z", "cpu_seconds": 0.7763126799999895, "created_at": "2026-11-22T03:05:09Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-0313289054b9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "4934d7b18054a38c93f5d59607c6830f7cfca32a04158f99e0d2a20a40c91d79", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-0313289054b9", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:05:09Z", "seed": 806540522, "treatment": "frontier"}, "priority": 0.23821177135019464, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "c4cd28f0d7a0288256b1b46c220ab13cacac08265ef019c257cd44892e586009", "run_seconds": 0.9920034408569336, "seed": 806540522, "started_at": "2026-11-22T03:07:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.081733, 0.185558], "effect": 0.133645, "evaluated_at": "2026-11-22T03:07:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-0313289054b9", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.763157895, 0.758152174, 1.0, 0.85, 0.779411765, 1.0, 0.770833333, 0.8075, 1.0, 0.78960396, 0.806930693, 0.821078431], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.771052632, 0.741847826, 1.0, 0.861111111, 0.818627451, 1.0, 0.796875, 0.87, 1.0, 0.849009901, 0.898514851, 0.789215686]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "4934d7b18054a38c93f5d59607c6830f7cfca32a04158f99e0d2a20a40c91d79", "primary_control": "random_walk", "result_hash": "c4cd28f0d7a0288256b1b46c220ab13cacac08265ef019c257cd44892e586009", "signature": "93a5ab745ca8004c0bac3bf67480fa3a6c252bf36a0a37e577687d5a6288bb7aa3bf6f50b84174c19813f2cb6e75f576c3ce5c4c39ad92c611f361ee52bd1001", "state": "VERIFIED", "summary": {"curiosity": 0.845556, "frontier": 1.0, "random_walk": 0.866355}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:18Z", "cpu_seconds": 0.6771885800000064, "created_at": "2026-11-22T03:05:09Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-0dfd4b65e4d4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "2dd547a8030c1914ab95db5af53bd6485274a0df52d6c25491a86619a960c75d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-0dfd4b65e4d4", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:05:09Z", "seed": 671010378, "treatment": "frontier"}, "priority": 0.22423729295319345, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "39e8bc002c988f50a3ef940deaa6613ee0f4f7619298529d7717f5c058db9210", "run_seconds": 0.9659926891326904, "seed": 671010378, "started_at": "2026-11-22T03:07:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.060987, 0.07826], "effect": 0.069623, "evaluated_at": "2026-11-22T03:07:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-0dfd4b65e4d4", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.898373984, 0.900406504, 0.900406504, 0.902, 0.921370968, 0.920634921, 0.894957983, 0.896341463, 0.936, 0.924180328, 0.893442623, 0.91468254], "frontier": [0.985772358, 0.995934959, 0.993902439, 0.99, 0.995967742, 0.992063492, 0.99789916, 0.995934959, 0.992, 0.993852459, 0.993852459, 0.992063492], "random_walk": [0.912601626, 0.900406504, 0.922764228, 0.9, 0.933467742, 0.94047619, 0.922268908, 0.945121951, 0.936, 0.932377049, 0.905737705, 0.932539683]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "2dd547a8030c1914ab95db5af53bd6485274a0df52d6c25491a86619a960c75d", "primary_control": "random_walk", "result_hash": "39e8bc002c988f50a3ef940deaa6613ee0f4f7619298529d7717f5c058db9210", "signature": "23528f227b5983c909fee7aa2b0f491533e3d23dc0e87288331d46ea550a50a4b37115b3c79bf28058f47f75d68001333ad08431a96c18ca087aefe5690f7306", "state": "VERIFIED", "summary": {"curiosity": 0.908566, "frontier": 0.99327, "random_walk": 0.923647}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:07:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:12Z", "cpu_seconds": 1.4951003879999973, "created_at": "2026-11-22T03:05:09Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-3502ceb5736b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "a4286fd3a443610e771ab31af7365025c256cb4ecf36cee5d93e3f71767b03c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-3502ceb5736b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:05:09Z", "seed": 479094144, "treatment": "frontier"}, "priority": 0.2335603352600129, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "0cd7d6fe85a9a775cc23b882a5a7d646f73bbeedcc27417e36485c57b2362351", "run_seconds": 2.049220323562622, "seed": 479094144, "started_at": "2026-11-22T03:07:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.025194, 0.075811], "effect": 0.050502, "evaluated_at": "2026-11-22T03:07:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-3502ceb5736b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 1.0, 1.0, 0.919230769, 0.813636364, 0.87295082, 1.0, 0.877118644, 0.927272727, 0.899038462, 0.94, 0.87704918], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 1.0, 1.0, 0.888461538, 0.954545455, 0.881147541, 1.0, 0.953389831, 0.959090909, 0.9375, 0.91, 0.909836066]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "a4286fd3a443610e771ab31af7365025c256cb4ecf36cee5d93e3f71767b03c6", "primary_control": "random_walk", "result_hash": "0cd7d6fe85a9a775cc23b882a5a7d646f73bbeedcc27417e36485c57b2362351", "signature": "c05381920887e89b664c42531f59e522691b9fee2b905610093cf4c84e8fdac48c60a4d24a1b7551baf577175fcbc1d524a96f3935631f5d1d7804b200a5cf0d", "state": "VERIFIED", "summary": {"curiosity": 0.927191, "frontier": 1.0, "random_walk": 0.949498}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:26Z", "cpu_seconds": 1.5428582609999921, "created_at": "2026-11-22T03:05:25Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-6a981899439c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "4e39e1caefe587e46e2e7d9724eb39d7cf078a5c71049f47ee117ed1254c289a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-6a981899439c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:25Z", "seed": 458376069, "treatment": "iw_temperature"}, "priority": 0.21363071995708377, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "73858335bbb8841aac03ea2eb1830a5a175def585d3047a8166a4ad93cdaaf47", "run_seconds": 1.609104871749878, "seed": 458376069, "started_at": "2026-11-22T03:08:24Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.005524, 0.014654], "effect": 0.004565, "evaluated_at": "2026-11-22T03:08:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-6a981899439c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.029293535, 0.202478085, 0.113946703, 0.118699987, 0.116663398, 0.17250922, 0.108849832, 0.11519853, 0.1533819, 0.17066499, 0.114557163, 0.163972848, 0.1235937, 0.15428594, 0.12048205, 0.163318872], "raw": [0.05849638, 0.123978145, 0.09118948, 0.09758388, 0.108725577, 0.113590348, 0.096034572, 0.11519853, 0.135251155, 0.142009867, 0.110964333, 0.136162605, 0.144752818, 0.130355935, 0.173713765, 0.153397682], "temperature": [0.054576, 0.218197725, 0.113946703, 0.099004417, 0.117131512, 0.126939238, 0.133997087, 0.098092338, 0.152096695, 0.17066499, 0.12518631, 0.162799597, 0.144752818, 0.15428594, 0.156736258, 0.186526247]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5368, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.291", "prereg_hash": "4e39e1caefe587e46e2e7d9724eb39d7cf078a5c71049f47ee117ed1254c289a", "primary_control": "temperature", "result_hash": "73858335bbb8841aac03ea2eb1830a5a175def585d3047a8166a4ad93cdaaf47", "signature": "f6df7dc982fa14a36290c70ad9de03ebdbe73f510bf2cbf5136fc77dfa33fc4831cdc74540bbcd90c75ad7b61a4e962a78973b9b3ee70e272df6abff8058a40f", "state": "VERIFIED", "summary": {"iw_temperature": 0.133869, "raw": 0.120713, "temperature": 0.138433}, "ties": 3, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:08:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:59Z", "cpu_seconds": 1.4367924849999838, "created_at": "2026-11-22T03:05:25Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f8b4b52e0711", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "6c333f1d7806d7156be0055d6f24ebbad1d200821c1b44ce81d7e099cb56cb73", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f8b4b52e0711", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:25Z", "seed": 106618179, "treatment": "iw_temperature"}, "priority": 0.21107159937916187, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "7148d67601054e8731e97e60e083c6e5b5241857b1aa3b11470a7f9b93366c21", "run_seconds": 1.4432623386383057, "seed": 106618179, "started_at": "2026-11-22T03:08:57Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.004539, 0.007575], "effect": 0.001518, "evaluated_at": "2026-11-22T03:09:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f8b4b52e0711", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.069499775, 0.042651125, 0.082497117, 0.006920185, 0.164504938, 0.079509462, 0.009762865, 0.08697373, 0.079259853, 0.037722735, 0.01187565, 0.02964694, 0.086856525, 0.04204894, 0.065708165, 0.055213882], "raw": [0.054067677, 0.042651125, 0.054760828, 0.00877773, 0.145971037, 0.079509462, 0.021451585, 0.10502252, 0.096824565, 0.042863297, 0.036646497, 0.021257713, 0.076566783, 0.046239815, 0.050840602, 0.065037185], "temperature": [0.071144555, 0.065599615, 0.054760828, 0.011263495, 0.16000157, 0.069263207, 0.014946135, 0.096589905, 0.081624785, 0.046446965, 0.014710337, 0.018698053, 0.106727957, 0.03413192, 0.063996167, 0.065037185]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6435, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.227", "prereg_hash": "6c333f1d7806d7156be0055d6f24ebbad1d200821c1b44ce81d7e099cb56cb73", "primary_control": "temperature", "result_hash": "7148d67601054e8731e97e60e083c6e5b5241857b1aa3b11470a7f9b93366c21", "signature": "f448b733db21015dc59a058f7e0d7e25e2a56009afbb231a30372a56901d4c8bf5aaa899bbb26cb0d817baaf37d9088bfe6d93ba1a0c3b9cd11c01d2d1f1e308", "state": "VERIFIED", "summary": {"iw_temperature": 0.059416, "raw": 0.059281, "temperature": 0.060934}, "ties": 0, "treatment": "iw_temperature", "wins": 10}, "verified_at": "2026-11-22T03:09:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:36Z", "cpu_seconds": 6.5235118849999765, "created_at": "2026-11-22T03:05:25Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-5513e275d878", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "97f7ba1d31ca992af38cc58da493616268cccf696d71f69311471854f921ead4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-5513e275d878", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:05:25Z", "seed": 1297513401, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "dd26e8f1a07627868d0f0453ebbeb50d1cb39303e983275de2da69c35e3b682f", "run_seconds": 9.506322383880615, "seed": 1297513401, "started_at": "2026-11-22T03:05:27Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:05:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-5513e275d878", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.8081926602980765, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "97f7ba1d31ca992af38cc58da493616268cccf696d71f69311471854f921ead4", "primary_control": "reset", "result_hash": "dd26e8f1a07627868d0f0453ebbeb50d1cb39303e983275de2da69c35e3b682f", "signature": "91be4b2183a3fa8766940cacb50f74237e74d9d23a133b40b05656499ec8d3c84984fadb19b02c3dcf97002b6f15a0c776ef77bba355002529a0dc3e0cd21809", "state": "PARTIAL", "summary": {"inherit": 0.7333333333333333, "reset": 0.7333333333333333, "root": 0.45}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:05:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:36Z", "cpu_seconds": 3.5192143579999993, "created_at": "2026-11-22T03:05:25Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-a338d408ab25", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "1ed54426194e585bf74c97e95b9cab0176ee37b6c3b8e4ff5b2cf5722fdd92d1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-a338d408ab25", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:05:25Z", "seed": 589323569, "treatment": "frontier"}, "priority": 0.2884317262726835, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "bcb141714975f112eb6d902db2377dada7d00a56f528a5f1d2e0f04cb6b664bb", "run_seconds": 6.20316743850708, "seed": 589323569, "started_at": "2026-11-22T03:05:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.041071, 0.115957], "effect": 0.078514, "evaluated_at": "2026-11-22T03:05:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-a338d408ab25", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.851010101, 0.835106383, 0.849009901, 0.772988506, 0.865591398, 1.0, 1.0, 0.805555556, 0.775, 1.0, 0.818681319, 1.0], "frontier": [0.994949495, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.911616162, 0.845744681, 0.930693069, 0.836206897, 0.930107527, 1.0, 1.0, 0.847222222, 0.858333333, 1.0, 0.892857143, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "1ed54426194e585bf74c97e95b9cab0176ee37b6c3b8e4ff5b2cf5722fdd92d1", "primary_control": "random_walk", "result_hash": "bcb141714975f112eb6d902db2377dada7d00a56f528a5f1d2e0f04cb6b664bb", "signature": "0404b9018a50226d62128136e6aea49d833b256cd6247fac875a817ca68a631a60ac9d25742c58ff1dca934f2c84ec8d6d46da0beebee0069003361202a1ef05", "state": "VERIFIED", "summary": {"curiosity": 0.881079, "frontier": 0.999579, "random_walk": 0.921065}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:05:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:42Z", "cpu_seconds": 1.67973028199998, "created_at": "2026-11-22T03:05:25Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-bad3c7269a85", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "2f61d783d87c87b7c9eefb9834ddf30e76d9ddde884a953056f64ba3218efd2f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-bad3c7269a85", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:05:25Z", "seed": 498162082, "treatment": "frontier"}, "priority": 0.21311478792316363, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "629961dc43c1be64429f17db3def91a7fa9f6bfff8de6b430fa336eac99babc1", "run_seconds": 1.7582449913024902, "seed": 498162082, "started_at": "2026-11-22T03:08:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.041816, 0.061982], "effect": 0.051899, "evaluated_at": "2026-11-22T03:08:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-bad3c7269a85", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.90625, 0.934615385, 0.976027397, 0.915384615, 0.995689655, 0.965753425, 0.946969697, 0.942857143, 0.959558824, 0.9296875, 0.940298507, 0.946969697], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.9296875, 0.911538462, 0.965753425, 0.95, 0.956896552, 0.955479452, 0.939393939, 0.957142857, 0.966911765, 0.953125, 0.925373134, 0.965909091]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "2f61d783d87c87b7c9eefb9834ddf30e76d9ddde884a953056f64ba3218efd2f", "primary_control": "random_walk", "result_hash": "629961dc43c1be64429f17db3def91a7fa9f6bfff8de6b430fa336eac99babc1", "signature": "b326b037d76634ea1e607a952279bf3dd7f8662de128d3c9e2f97905ad3ac4a551347e1cbf7bfcdffb147bed8c7b901584d4de525667727475805c37a3cd1603", "state": "VERIFIED", "summary": {"curiosity": 0.946672, "frontier": 1.0, "random_walk": 0.948101}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:08:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:47Z", "cpu_seconds": 1.6351935320000024, "created_at": "2026-11-22T03:05:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-825418b30716", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "40b114c6f12d480c9e2183768dd10e333fc61b57b8f15bde109c62d3f40f0434", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-825418b30716", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:42Z", "seed": 491187557, "treatment": "iw_temperature"}, "priority": 0.3593477211532199, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a51441ed1766c25f4d112a41de1210fbf21abd8a0e1e047fbc6bd88e64db9d8b", "run_seconds": 2.0441131591796875, "seed": 491187557, "started_at": "2026-11-22T03:05:45Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.024302, 0.005426], "effect": -0.009438, "evaluated_at": "2026-11-22T03:05:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-825418b30716", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.307153103, 0.324288303, 0.298930178, 0.260013035, 0.262255017, 0.296829768, 0.267077332, 0.274719672, 0.29120027, 0.25964624, 0.251695555, 0.185299617, 0.225064522, 0.286709535, 0.316585385, 0.22656388], "raw": [0.276846335, 0.294003122, 0.240296712, 0.22687592, 0.27188031, 0.293713845, 0.223034013, 0.224455595, 0.221700295, 0.185350133, 0.247278948, 0.17085588, 0.212740132, 0.26958249, 0.236462968, 0.242202865], "temperature": [0.276846335, 0.30418465, 0.283120805, 0.248255073, 0.31363596, 0.296346025, 0.267077332, 0.224455595, 0.22961775, 0.220836717, 0.26204508, 0.222125802, 0.212740132, 0.283978453, 0.287295608, 0.25046592]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0077, "losses": 11, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.982", "prereg_hash": "40b114c6f12d480c9e2183768dd10e333fc61b57b8f15bde109c62d3f40f0434", "primary_control": "temperature", "result_hash": "a51441ed1766c25f4d112a41de1210fbf21abd8a0e1e047fbc6bd88e64db9d8b", "signature": "80afa844b2d6c2f094f283bea56ec545277a8d1eab78dff9dfff6355226d60385e42f32d780499c2a1109211a303415e48d6d5bb7b84b98db3f1770b929fe10e", "state": "VERIFIED", "summary": {"iw_temperature": 0.270877, "raw": 0.23983, "temperature": 0.261439}, "ties": 1, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:05:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:24Z", "cpu_seconds": 1.4515796659999864, "created_at": "2026-11-22T03:05:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-bcd6ca10ddd5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "7c98f6485145df578430ee9f37fe9027186ddaca2b6069f62d1d911287112710", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-bcd6ca10ddd5", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:42Z", "seed": 34785750, "treatment": "iw_temperature"}, "priority": 0.2054453545024255, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "7914f3c2ba4b40222eba8cb193a53404681ec23893c47e6956cdade87c6fbdbb", "run_seconds": 1.5435478687286377, "seed": 34785750, "started_at": "2026-11-22T03:09:23Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.025383, 0.007673], "effect": -0.008855, "evaluated_at": "2026-11-22T03:09:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-bcd6ca10ddd5", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.244943205, 0.245238462, 0.181589257, 0.190188433, 0.284327153, 0.215674498, 0.249187137, 0.293985273, 0.316275328, 0.146428242, 0.22932263, 0.201830413, 0.263322732, 0.247996497, 0.151007618, 0.213666252], "raw": [0.225823135, 0.236022678, 0.193304085, 0.190188433, 0.286597907, 0.239466512, 0.233334323, 0.26107484, 0.259115245, 0.169168665, 0.209071743, 0.142944198, 0.208616047, 0.239234647, 0.1637209, 0.232411683], "temperature": [0.225823135, 0.252994905, 0.23142359, 0.15457127, 0.286597907, 0.225995105, 0.233334323, 0.28126735, 0.303701352, 0.169168665, 0.22932263, 0.092560228, 0.23410779, 0.239234647, 0.151007618, 0.22219192]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "7c98f6485145df578430ee9f37fe9027186ddaca2b6069f62d1d911287112710", "primary_control": "temperature", "result_hash": "7914f3c2ba4b40222eba8cb193a53404681ec23893c47e6956cdade87c6fbdbb", "signature": "5d00563e576176d0233268c40d01ff288456c7b3b7ad1f021d3dc7a0de6d10f78ff6ad5b1781d09ef709ed8824943c34cd7cb0f6bad372473f7f5d011098b20b", "state": "VERIFIED", "summary": {"iw_temperature": 0.229686, "raw": 0.218131, "temperature": 0.220831}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:09:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:23Z", "cpu_seconds": 1.5568302109999763, "created_at": "2026-11-22T03:05:42Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-ebbd80053a51", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "bcf4c18061c652b48a0c3f49802044b41f6f9da95e3a17f0faf5c7f5d10848c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-ebbd80053a51", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:42Z", "seed": 935537851, "treatment": "iw_temperature"}, "priority": 0.20565721128129677, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a6b2c1649ae3c4d66d6e9434bc18a4677bdf51f57b9f36561d27a26a2f7c723e", "run_seconds": 1.573291540145874, "seed": 935537851, "started_at": "2026-11-22T03:09:21Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014724, 0.001905], "effect": -0.00641, "evaluated_at": "2026-11-22T03:09:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-ebbd80053a51", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.04174844, 0.041686965, 0.034929825, 0.097360725, 0.081666828, 0.094407792, 0.075587685, 0.037791325, 0.045214195, 0.045184733, 0.043556785, 0.033901428, 0.04790537, 0.050373995, 0.05541276, 0.037259462], "raw": [0.037206497, 0.056885772, 0.038117405, 0.074080743, 0.033118087, 0.061430285, 0.075587685, 0.056492718, 0.051169327, 0.046846363, 0.070810138, 0.049546812, 0.075959065, 0.060466505, 0.070464178, 0.053212785], "temperature": [0.059647682, 0.041686965, 0.028346952, 0.059823603, 0.035211675, 0.076577158, 0.057978063, 0.038576072, 0.045214195, 0.050389332, 0.043556785, 0.033901428, 0.035724653, 0.04656325, 0.070464178, 0.037769857]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0936, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.806", "prereg_hash": "bcf4c18061c652b48a0c3f49802044b41f6f9da95e3a17f0faf5c7f5d10848c3", "primary_control": "temperature", "result_hash": "a6b2c1649ae3c4d66d6e9434bc18a4677bdf51f57b9f36561d27a26a2f7c723e", "signature": "61c789c4f9424e125fe18dcf306b85af8caae5abe74b373be373c776bfe18f8af9f983d2b12416d858b99a837f9875e6186739cb1403029f669f66a94d368c02", "state": "VERIFIED", "summary": {"iw_temperature": 0.053999, "raw": 0.056962, "temperature": 0.047589}, "ties": 4, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:09:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:51Z", "cpu_seconds": 7.469896589000001, "created_at": "2026-11-22T03:05:42Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-027138868d69", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "5bbb19f353f1d04c727a2ab5c77b879ef3edc9af8e2c172261e34695856f18b3", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-027138868d69", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:05:42Z", "seed": 1401389335, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "ec251e4e90fbaee8458832657d636db2cdd6a2f3a54dd21c972b3d3a43350e36", "run_seconds": 9.096280813217163, "seed": 1401389335, "started_at": "2026-11-22T03:05:42Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 4}, "better": "higher", "ci95": [-0.321485, -0.078515], "effect": -0.2, "evaluated_at": "2026-11-22T03:05:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-027138868d69", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.3436838824753175, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0001, "losses": 14, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5bbb19f353f1d04c727a2ab5c77b879ef3edc9af8e2c172261e34695856f18b3", "primary_control": "reset", "result_hash": "ec251e4e90fbaee8458832657d636db2cdd6a2f3a54dd21c972b3d3a43350e36", "signature": "9a6cbe960e293fbd699d95186147902e976fa09dc4914fe7fe26fb661ff0af1f008ab6e6978c3ec649d54ddf0a3fda7c43774af681166a47f4faf3035452a90d", "state": "VERIFIED", "summary": {"inherit": 0.5666666666666667, "reset": 0.7666666666666667, "root": 0.5}, "ties": 44, "treatment": "inherit", "wins": 2}, "verified_at": "2026-11-22T03:06:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:05:50Z", "cpu_seconds": 2.9551044689999912, "created_at": "2026-11-22T03:05:42Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-52bd89da1701", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "f70a0bcd19410c508550c8a96f69065fc577fd0c5d8344c3b594653949284cfe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-52bd89da1701", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:05:42Z", "seed": 287408461, "treatment": "frontier"}, "priority": 0.2938929760266817, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "2394572cff35b7df6e5ec7b4c94da44c5531a84b2db4f12ebff992889759a572", "run_seconds": 3.405146837234497, "seed": 287408461, "started_at": "2026-11-22T03:05:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.030563, 0.098679], "effect": 0.064621, "evaluated_at": "2026-11-22T03:05:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-52bd89da1701", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.756329114, 0.868421053, 0.851648352, 0.845959596, 0.75617284, 1.0, 0.873853211, 0.846774194, 0.854545455, 1.0, 0.906818182, 0.815789474], "frontier": [0.89556962, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.829545455, 1.0], "random_walk": [0.870253165, 0.918859649, 0.912087912, 0.881313131, 0.848765432, 1.0, 0.878440367, 0.908602151, 0.972727273, 1.0, 0.870454545, 0.888157895]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.9689, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0107", "prereg_hash": "f70a0bcd19410c508550c8a96f69065fc577fd0c5d8344c3b594653949284cfe", "primary_control": "random_walk", "result_hash": "2394572cff35b7df6e5ec7b4c94da44c5531a84b2db4f12ebff992889759a572", "signature": "6a6d61ad4a985d30addc969e093f0506454095248d24ae9e98e2ada30f907dff845287c03a13fb759c4d8adfaf29d886753f7be51478e58b090e182e53f71701", "state": "VERIFIED", "summary": {"curiosity": 0.864693, "frontier": 0.977093, "random_walk": 0.912472}, "ties": 2, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:05:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:16Z", "cpu_seconds": 1.6355170749999957, "created_at": "2026-11-22T03:05:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-420b3332ae1f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "c3d8a74d1f2b67099fd4301dc83931e59601b2be7364fdb70f785ea784ce6030", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-420b3332ae1f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:58Z", "seed": 545675702, "treatment": "iw_temperature"}, "priority": 0.21027474311131278, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "f47739a71c5d5734e65e1dd85249d327582a14ef935794e28a23bd9542334ed1", "run_seconds": 1.7027742862701416, "seed": 545675702, "started_at": "2026-11-22T03:09:14Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014258, 0.000107], "effect": -0.007076, "evaluated_at": "2026-11-22T03:09:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-420b3332ae1f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.054678275, 0.05696276, 0.040073327, 0.099442177, 0.002300067, 0.21300383, 0.023911132, 0.062127175, 0.045623198, 0.041276725, 0.06684782, 0.049163025, 0.032191127, 0.052879665, 0.058831135, 0.108840513], "raw": [0.044943605, 0.04125226, 0.046470268, 0.093771852, 0.01106474, 0.191234848, 0.041349185, 0.046224603, 0.047946148, 0.03871127, 0.04706018, 0.065933333, 0.04876764, 0.044766893, 0.030392738, 0.049064768], "temperature": [0.047742415, 0.037859103, 0.02336845, 0.069508685, 0.006956428, 0.202358965, 0.034873983, 0.075150107, 0.038651323, 0.061269813, 0.040596282, 0.049163025, 0.032191127, 0.045945065, 0.046336747, 0.082968255]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "c3d8a74d1f2b67099fd4301dc83931e59601b2be7364fdb70f785ea784ce6030", "primary_control": "temperature", "result_hash": "f47739a71c5d5734e65e1dd85249d327582a14ef935794e28a23bd9542334ed1", "signature": "72a9ae0317b1483bdae811148567c6f9e4ac3e3bd5cd24d482559d4d60f5305ac7e68e2b85db90d5e13d7b88d7bc3c5915338c56ed76c307493b372683ec8c0a", "state": "VERIFIED", "summary": {"iw_temperature": 0.063009, "raw": 0.05556, "temperature": 0.055934}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:09:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:16Z", "cpu_seconds": 1.454425452999999, "created_at": "2026-11-22T03:05:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-d87883d4aa27", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "22bf4fa6794f3e55902cc71eeb1f2b6e9696cbd1289d579ac3bcb3df1b8325ee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-d87883d4aa27", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:58Z", "seed": 1155982612, "treatment": "iw_temperature"}, "priority": 0.21614198107337812, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "d4376dd76f21e0dbf900b614dbb2751ef7f00f065244e6c3c1630e42f3cd71ee", "run_seconds": 1.5352702140808105, "seed": 1155982612, "started_at": "2026-11-22T03:08:14Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.02142, 0.00998], "effect": -0.00572, "evaluated_at": "2026-11-22T03:08:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-d87883d4aa27", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.317742505, 0.213743943, 0.302469797, 0.273162738, 0.295701875, 0.285061257, 0.22667363, 0.261282267, 0.176651822, 0.217673312, 0.232878108, 0.266529425, 0.29833361, 0.292838485, 0.310433965, 0.234552385], "raw": [0.28653559, 0.222895265, 0.235649255, 0.232104983, 0.27646327, 0.268654402, 0.175061825, 0.261282267, 0.176651822, 0.25544926, 0.204387987, 0.22684773, 0.317623627, 0.269756948, 0.29720743, 0.250538343], "temperature": [0.312749865, 0.222895265, 0.208383268, 0.273162738, 0.285620012, 0.277162493, 0.175061825, 0.253407295, 0.169484005, 0.244336465, 0.21932707, 0.279646405, 0.340253472, 0.280021287, 0.306286955, 0.266414735]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "22bf4fa6794f3e55902cc71eeb1f2b6e9696cbd1289d579ac3bcb3df1b8325ee", "primary_control": "temperature", "result_hash": "d4376dd76f21e0dbf900b614dbb2751ef7f00f065244e6c3c1630e42f3cd71ee", "signature": "c43f3f4b3f14275f5fec604afed3ca3c52632d267c2b70296ad81fa6354a30cf089338dacd8663459c3ab7d0721f158226b8d83493c9c21ef59c15049d094603", "state": "VERIFIED", "summary": {"iw_temperature": 0.262858, "raw": 0.247319, "temperature": 0.257138}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:08:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:51Z", "cpu_seconds": 1.8626472189999959, "created_at": "2026-11-22T03:05:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f9faaa32803c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "7d09dc36be50b013537b99aa8431867ee2f386c1e7327b0b3907957bcdd6ecc3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f9faaa32803c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:58Z", "seed": 1919744264, "treatment": "iw_temperature"}, "priority": 0.21847584773964868, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "734a3c8d05349139148664551f82a9ea7ae64205c717235e2ffed4853c530533", "run_seconds": 3.504033088684082, "seed": 1919744264, "started_at": "2026-11-22T03:07:48Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.005927, 0.002496], "effect": -0.001716, "evaluated_at": "2026-11-22T03:07:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f9faaa32803c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.051398785, 0.068884047, 0.04691938, 0.03674708, 0.03184292, 0.036696417, 0.075110082, 0.129490282, 0.038995282, 0.03286775, 0.07345411, 0.054184185, 0.078215508, 0.049919432, 0.077336425, 0.064208185], "raw": [0.040106112, 0.08583419, 0.042588478, 0.039746997, 0.045516655, 0.061278775, 0.04579936, 0.15617258, 0.038995282, 0.06129306, 0.080531085, 0.056199697, 0.111608117, 0.041804217, 0.064114358, 0.048409015], "temperature": [0.040106112, 0.07125099, 0.04691938, 0.039746997, 0.03547006, 0.036696417, 0.075110082, 0.139647823, 0.035283295, 0.026121282, 0.075712713, 0.056199697, 0.07786787, 0.049919432, 0.077336425, 0.035428717]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7d09dc36be50b013537b99aa8431867ee2f386c1e7327b0b3907957bcdd6ecc3", "primary_control": "temperature", "result_hash": "734a3c8d05349139148664551f82a9ea7ae64205c717235e2ffed4853c530533", "signature": "a04a25cae31b4acdc861b3229de53225fb47e7666a38cb62090250d2b86a247a67d06db9309181a1f08ad0ce0a90a73b69edbe960e08571575912551d307ab09", "state": "VERIFIED", "summary": {"iw_temperature": 0.059142, "raw": 0.06375, "temperature": 0.057426}, "ties": 5, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:07:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:04Z", "cpu_seconds": 1.7112916969999787, "created_at": "2026-11-22T03:05:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-ff982913da3f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "9c92ab58a6712fe150c336791268624adfa82f8aa6a567626052926178877a16", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-ff982913da3f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:05:58Z", "seed": 1070144896, "treatment": "iw_temperature"}, "priority": 0.33700002019362324, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "fd58af4d7cb556d84d4fd3cc0280f659918005343ce2fd78f704dfed793d8a66", "run_seconds": 1.960211992263794, "seed": 1070144896, "started_at": "2026-11-22T03:06:01Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014988, 0.023801], "effect": 0.004406, "evaluated_at": "2026-11-22T03:06:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-ff982913da3f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.29638596, 0.213537147, 0.315236088, 0.30478825, 0.245035177, 0.228773135, 0.258384405, 0.337758955, 0.244517415, 0.292168082, 0.319165517, 0.278809853, 0.183696715, 0.2643572, 0.306279495, 0.300800505], "raw": [0.2927727, 0.199802967, 0.248090345, 0.27358844, 0.250960882, 0.228773135, 0.25697803, 0.336015893, 0.244517415, 0.248175855, 0.282763225, 0.22749792, 0.231154662, 0.18367853, 0.261342273, 0.21360455], "temperature": [0.30562752, 0.276084787, 0.277169528, 0.297270365, 0.250960882, 0.289215213, 0.234484885, 0.336015893, 0.264540405, 0.272215013, 0.311621845, 0.25848808, 0.28137357, 0.208443627, 0.29587977, 0.300800505]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "9c92ab58a6712fe150c336791268624adfa82f8aa6a567626052926178877a16", "primary_control": "temperature", "result_hash": "fd58af4d7cb556d84d4fd3cc0280f659918005343ce2fd78f704dfed793d8a66", "signature": "0b53f7a9688aebd57eb5836a33977dd87a96a5ec9473e1d5bed806f086ac64f20db6a9dfe610c0e309fb25cb17a069eae37e00341a02900c6ca3c8463831f908", "state": "VERIFIED", "summary": {"iw_temperature": 0.274356, "raw": 0.248732, "temperature": 0.278762}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:06:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:06Z", "cpu_seconds": 6.289293803999982, "created_at": "2026-11-22T03:05:58Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-6d9469a4ea58", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "cba59c1c39c5e7b3ac79b6acbd3d99495c56756adc748c8d512337c09f6b1ddf", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-6d9469a4ea58", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:05:58Z", "seed": 343364738, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "5370669f8a45bcc9304ea265ea9dc9ed6a444b678166d5dfe63a3b6e11b8e98c", "run_seconds": 7.3389365673065186, "seed": 343364738, "started_at": "2026-11-22T03:05:59Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:06:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-6d9469a4ea58", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4547111797575811, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "cba59c1c39c5e7b3ac79b6acbd3d99495c56756adc748c8d512337c09f6b1ddf", "primary_control": "reset", "result_hash": "5370669f8a45bcc9304ea265ea9dc9ed6a444b678166d5dfe63a3b6e11b8e98c", "signature": "28f15675a413733b36d4305330163a788bbba5a1633bd9f59e65e12ac2f45766b7b8691afddf65e18e7725297b105329d30d85d5c5efd777c36c8ad1905aae0a", "state": "PARTIAL", "summary": {"inherit": 0.6833333333333333, "reset": 0.6833333333333333, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:06:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:07Z", "cpu_seconds": 3.3998638019999987, "created_at": "2026-11-22T03:05:58Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-003f10672bd6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "71ccb453f0c849d3eaf8dd0f2fde95335a8811f0afdd28b874fc24e63678393b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-003f10672bd6", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:05:58Z", "seed": 32766830, "treatment": "frontier"}, "priority": 0.28755746733004883, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "17d86b741c4bff81c7587f58df6240f1e3ee4ce77e5fc8184bf6e1404801e0d0", "run_seconds": 4.505573034286499, "seed": 32766830, "started_at": "2026-11-22T03:06:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.039621, 0.103014], "effect": 0.071317, "evaluated_at": "2026-11-22T03:06:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-003f10672bd6", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.829268293, 0.822164948, 0.797794118, 1.0, 0.888888889, 1.0, 0.798295455, 1.0, 0.889784946, 0.836805556, 0.796052632, 0.861386139], "frontier": [0.975609756, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.887195122, 0.912371134, 0.801470588, 1.0, 0.95, 1.0, 0.931818182, 1.0, 0.889784946, 0.920138889, 0.911184211, 0.915841584]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "71ccb453f0c849d3eaf8dd0f2fde95335a8811f0afdd28b874fc24e63678393b", "primary_control": "random_walk", "result_hash": "17d86b741c4bff81c7587f58df6240f1e3ee4ce77e5fc8184bf6e1404801e0d0", "signature": "689232291b7316f4f19c2dbaeb356ce098ff00fbc60a2a53d0a9b0550e09a86d13ec28662db817009857c058c49424ad363bbec63d1e01cc826b0e9655cdd808", "state": "VERIFIED", "summary": {"curiosity": 0.876703, "frontier": 0.997967, "random_walk": 0.92665}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:06:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:23Z", "cpu_seconds": 1.5567638489999922, "created_at": "2026-11-22T03:06:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-3b458ba14128", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "8728b2d5702567880d518eaaf97e5cc911519b5452b777b8e91ffb88efcb19e5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-3b458ba14128", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:14Z", "seed": 1804511186, "treatment": "iw_temperature"}, "priority": 0.21370241356137046, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "1991995611b750188905157a9043b38fdaf67d72666a73c393341974d161d1c6", "run_seconds": 1.882655382156372, "seed": 1804511186, "started_at": "2026-11-22T03:08:21Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.022493, 0.008238], "effect": -0.007128, "evaluated_at": "2026-11-22T03:08:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-3b458ba14128", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.212154232, 0.159259107, 0.175706532, 0.151135282, 0.153976333, 0.126946078, 0.162696905, 0.153655435, 0.13088763, 0.188544413, 0.132009635, 0.156412793, 0.110564268, 0.14316518, 0.219474815, 0.076671278], "raw": [0.137950775, 0.110224745, 0.144172458, 0.131310143, 0.20511895, 0.235294002, 0.160962278, 0.153655435, 0.10923488, 0.188544413, 0.053772535, 0.137287405, 0.110564268, 0.089677978, 0.171089067, 0.106550718], "temperature": [0.174390818, 0.110224745, 0.175706532, 0.13835891, 0.21384249, 0.177285572, 0.162696905, 0.169354285, 0.122617185, 0.188544413, 0.090717458, 0.137287405, 0.10995753, 0.089677978, 0.205710115, 0.07284673]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "8728b2d5702567880d518eaaf97e5cc911519b5452b777b8e91ffb88efcb19e5", "primary_control": "temperature", "result_hash": "1991995611b750188905157a9043b38fdaf67d72666a73c393341974d161d1c6", "signature": "41b0ab13ea3e730914f606f7814de508c5928e9609afccedb1878d0889195b3818e3be9e6292e023a3cb30504c7b09f69af4012ac09af197ef4a388df18fe009", "state": "VERIFIED", "summary": {"iw_temperature": 0.153329, "raw": 0.140338, "temperature": 0.146201}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:08:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:31Z", "cpu_seconds": 1.6059780450000005, "created_at": "2026-11-22T03:06:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-66ac47c0e85f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "92a15aac63c1de103404583370723bc43ab4bf8c14e7d4a9acb65f30abe2cd12", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-66ac47c0e85f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:14Z", "seed": 1376726371, "treatment": "iw_temperature"}, "priority": 0.20272742087588147, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "2dc514bdb475b827763cc9d1fcd23d662aebc3ac45eb9b22ac1a9aed983e5ef4", "run_seconds": 1.6186230182647705, "seed": 1376726371, "started_at": "2026-11-22T03:09:29Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018389, 0.004889], "effect": -0.00675, "evaluated_at": "2026-11-22T03:09:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-66ac47c0e85f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.239571038, 0.20674764, 0.22413575, 0.233951835, 0.220069048, 0.232002165, 0.225015992, 0.23605407, 0.2063765, 0.247734257, 0.260241288, 0.199965767, 0.198342347, 0.195645138, 0.237643478, 0.187141223], "raw": [0.209952528, 0.15866215, 0.17727339, 0.159287623, 0.140021485, 0.186956915, 0.204901673, 0.209532058, 0.187267115, 0.204851837, 0.214468227, 0.198479883, 0.225435568, 0.182345168, 0.185444703, 0.172699672], "temperature": [0.222868137, 0.173631245, 0.22413575, 0.216633282, 0.165132257, 0.207253015, 0.225015992, 0.23605407, 0.21759911, 0.27449395, 0.23513669, 0.198479883, 0.198342347, 0.213980705, 0.21029737, 0.223580098]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "92a15aac63c1de103404583370723bc43ab4bf8c14e7d4a9acb65f30abe2cd12", "primary_control": "temperature", "result_hash": "2dc514bdb475b827763cc9d1fcd23d662aebc3ac45eb9b22ac1a9aed983e5ef4", "signature": "bbde7eee5e1beef2ea61613e162f6870954d6b2a7ea42c32604d9b4420ab72fa6681584f02d271398ef90fa55c235aea12f3ac2b4a68119b001bff307603d009", "state": "VERIFIED", "summary": {"iw_temperature": 0.221915, "raw": 0.188599, "temperature": 0.215165}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:09:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:38Z", "cpu_seconds": 1.6317275359999996, "created_at": "2026-11-22T03:06:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-73dde946c0e7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "dab6023e43d11ba391997ef94b91759b30a9b2ce2f3957ef45b9bf2e4ea9d5fa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-73dde946c0e7", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:14Z", "seed": 1202709238, "treatment": "iw_temperature"}, "priority": 0.18556526311775282, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "29e01636bec01eb74ef1520693d1af6a684a78d77d8d5fb2af5dc9486e3b0c48", "run_seconds": 1.6403677463531494, "seed": 1202709238, "started_at": "2026-11-22T03:10:36Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.006147, 0.005173], "effect": -0.000487, "evaluated_at": "2026-11-22T03:10:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-73dde946c0e7", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.168138865, 0.123510542, 0.117710115, 0.09369463, 0.118116307, 0.13320782, 0.094771302, 0.12482536, 0.07663153, 0.183504088, 0.193673952, 0.117667943, 0.13395093, 0.09773225, 0.15653428, 0.130843082], "raw": [0.139403362, 0.102767583, 0.070235017, 0.08617697, 0.088220498, 0.118940675, 0.062207427, 0.08982647, 0.09351649, 0.129363025, 0.123743602, 0.10030051, 0.118404767, 0.080213225, 0.114953132, 0.09926977], "temperature": [0.168138865, 0.123510542, 0.117710115, 0.09369463, 0.102170355, 0.13320782, 0.094771302, 0.12482536, 0.09351649, 0.172018835, 0.166532565, 0.12873255, 0.155560467, 0.09773225, 0.15653428, 0.128067638]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1116, "losses": 4, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.773", "prereg_hash": "dab6023e43d11ba391997ef94b91759b30a9b2ce2f3957ef45b9bf2e4ea9d5fa", "primary_control": "temperature", "result_hash": "29e01636bec01eb74ef1520693d1af6a684a78d77d8d5fb2af5dc9486e3b0c48", "signature": "0f05cb17e417bb44d0879ccf78a4017915aafaaebeb16f20015dad30d22d17551178426dc6c01db66f5e9305a11c1fb6ffd51aa7f2d48669471e6e5ac512e908", "state": "VERIFIED", "summary": {"iw_temperature": 0.129032, "raw": 0.101096, "temperature": 0.128545}, "ties": 9, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:10:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:29Z", "cpu_seconds": 1.5449928900000032, "created_at": "2026-11-22T03:06:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-e6368945f9c9", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "2afe45e242fa3d9554ef6bbc24497b763ae813654980b87da6e408832a0628e4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-e6368945f9c9", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:14Z", "seed": 358966704, "treatment": "iw_temperature"}, "priority": 0.20470333843174585, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-6c51266cc711", "result_hash": "dd5caf1f656cfb26764f7e26b453f0f2ed9a3a758499bb9e8147e49f18a36c04", "run_seconds": 1.546450138092041, "seed": 358966704, "started_at": "2026-11-22T03:09:28Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.003516, 0.009675], "effect": 0.003079, "evaluated_at": "2026-11-22T03:09:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-e6368945f9c9", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.038420545, 0.046510438, 0.101649593, 0.040379553, 0.097362325, 0.078623927, 0.146209118, 0.127426945, 0.093983787, 0.040996382, 0.04775462, 0.047490113, 0.05398287, 0.034225567, 0.075702033, 0.02429329], "raw": [0.038420545, 0.057720408, 0.055371202, 0.064067058, 0.104257357, 0.077776525, 0.14584208, 0.127426945, 0.109444478, 0.053373558, 0.058359812, 0.048713707, 0.05092812, 0.03481532, 0.0695809, 0.044771532], "temperature": [0.047140272, 0.063575042, 0.06354026, 0.04481896, 0.104257357, 0.077776525, 0.14584208, 0.135416105, 0.113549535, 0.060095035, 0.045304462, 0.051890947, 0.058987967, 0.042324543, 0.0695809, 0.020178798]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.6435, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.227", "prereg_hash": "2afe45e242fa3d9554ef6bbc24497b763ae813654980b87da6e408832a0628e4", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-6c51266cc711", "result_hash": "dd5caf1f656cfb26764f7e26b453f0f2ed9a3a758499bb9e8147e49f18a36c04", "signature": "bcd8ecf3dc84768318d5aa16515e2f1144cf373cb2ac767bf955a380b8254d733a21c57b841e77f4e78a6bd1ba57a3c439d180efb27aa19650f224dee416c70c", "state": "VERIFIED", "summary": {"iw_temperature": 0.068438, "raw": 0.071304, "temperature": 0.071517}, "ties": 0, "treatment": "iw_temperature", "wins": 10}, "verified_at": "2026-11-22T03:09:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:21Z", "cpu_seconds": 1.7334441040000002, "created_at": "2026-11-22T03:06:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f4cb2d904e99", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "a49baf6a54987126fac32ab8853a915472aac286dbbe098e82627c2105d52f45", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f4cb2d904e99", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:14Z", "seed": 54495133, "treatment": "iw_temperature"}, "priority": 0.3441280026650957, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "79a1f644bd795e90e982431c4ae9267002fe8bd60d0966097bc2ff31655dcdf1", "run_seconds": 2.0036861896514893, "seed": 54495133, "started_at": "2026-11-22T03:06:19Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.023129, 0.002509], "effect": -0.01031, "evaluated_at": "2026-11-22T03:06:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f4cb2d904e99", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.304680708, 0.294249043, 0.269737598, 0.201055508, 0.256823837, 0.273172632, 0.281772213, 0.247461537, 0.21735229, 0.268532265, 0.309652827, 0.237691255, 0.260930118, 0.333662243, 0.222130295, 0.24719479], "raw": [0.333964595, 0.276587438, 0.227862768, 0.17381933, 0.178508593, 0.200534143, 0.201490925, 0.209545372, 0.204297055, 0.256713335, 0.287728308, 0.219014255, 0.276966237, 0.315210242, 0.190151198, 0.197413733], "temperature": [0.342356215, 0.291361085, 0.258968233, 0.17381933, 0.222873287, 0.200534143, 0.243987213, 0.235451247, 0.22502434, 0.268532265, 0.301785283, 0.237691255, 0.28821245, 0.329454635, 0.222130295, 0.21895733]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "a49baf6a54987126fac32ab8853a915472aac286dbbe098e82627c2105d52f45", "primary_control": "temperature", "result_hash": "79a1f644bd795e90e982431c4ae9267002fe8bd60d0966097bc2ff31655dcdf1", "signature": "e02fd5a5a6eaa4716bc86dae18b6edea6fcb7acbe1c31cebf38a32217facd6a41f429bcd262d70bf45d3a1b875374e05d3723cece60617ea7f634fcee5115802", "state": "VERIFIED", "summary": {"iw_temperature": 0.264131, "raw": 0.234363, "temperature": 0.253821}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:06:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:23Z", "cpu_seconds": 8.066541685999994, "created_at": "2026-11-22T03:06:14Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3045f155305c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "2ae013fdd05ce8248a1ffb961c79432b083c71a36b9b7001da2885303c60fe97", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3045f155305c", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:06:14Z", "seed": 444767436, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "0138ee323725cb64d8b28f993bb35fe8c952ec737f259ebd400f0890fe96c617", "run_seconds": 8.727478265762329, "seed": 444767436, "started_at": "2026-11-22T03:06:14Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:06:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3045f155305c", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "2ae013fdd05ce8248a1ffb961c79432b083c71a36b9b7001da2885303c60fe97", "primary_control": "reset", "result_hash": "0138ee323725cb64d8b28f993bb35fe8c952ec737f259ebd400f0890fe96c617", "signature": "aa21c9637e1a63590d062b4a13defe1aab7e40b223af780b62d850ee0e051ae8c453335d6f41d027318eeb673485a7eec39d93bd9038ae52119cacc003172805", "state": "PARTIAL", "summary": {"inherit": 0.6833333333333333, "reset": 0.6833333333333333, "root": 0.43333333333333335}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:06:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:58Z", "cpu_seconds": 1.5794983279999997, "created_at": "2026-11-22T03:06:31Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4ec6a6f374a6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "bb0804a96b776dd53d1a5691019f33d86510c26ee513e26e73c6f846c59ededd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4ec6a6f374a6", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:31Z", "seed": 841164255, "treatment": "iw_temperature"}, "priority": 0.21225890539203038, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ad7bfcf3ffca2cc903fec4975975352aba7522648e7b099aff890f59a6477e28", "run_seconds": 1.5940518379211426, "seed": 841164255, "started_at": "2026-11-22T03:08:56Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020349, 0.000559], "effect": -0.009895, "evaluated_at": "2026-11-22T03:08:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4ec6a6f374a6", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.054223868, 0.04704337, 0.036345543, 0.06725729, 0.04991326, 0.082152898, 0.067162358, 0.051433072, 0.074940065, 0.028139185, 0.06785862, 0.00970047, 0.03096453, 0.080823233, 0.086303893, 0.02997774], "raw": [0.047800145, 0.054885018, 0.056710083, 0.058902388, 0.04991326, 0.061199375, 0.07161758, 0.056060122, 0.056511018, 0.039739327, 0.041764703, 0.019269823, 0.043059495, 0.044717752, 0.032767445, 0.049289605], "temperature": [0.056328172, 0.047824, 0.043501533, 0.06725729, 0.040503837, 0.044597458, 0.042134285, 0.049645295, 0.056511018, 0.028139185, 0.09390665, 0.00970047, 0.017842742, 0.03195621, 0.036014612, 0.040058175]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "bb0804a96b776dd53d1a5691019f33d86510c26ee513e26e73c6f846c59ededd", "primary_control": "temperature", "result_hash": "ad7bfcf3ffca2cc903fec4975975352aba7522648e7b099aff890f59a6477e28", "signature": "21eb8c87e6bb8667f30eb728b55f3cd8b355d8a1191c89bc953795e236597391cb5bc672410f6e482451999638d7501b5645a2a2f24bf495bdb1f54577ab2a0a", "state": "VERIFIED", "summary": {"iw_temperature": 0.054015, "raw": 0.049013, "temperature": 0.04412}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:08:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:34Z", "cpu_seconds": 1.738961885000009, "created_at": "2026-11-22T03:06:31Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f37ccc2c2c3d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "125ecf3511870fe5895fa2491a15181edd1f78aaa0e2c73bedc8a34c65d6c267", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f37ccc2c2c3d", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:31Z", "seed": 1125571558, "treatment": "iw_temperature"}, "priority": 0.34275585750389487, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "0e23bdab33e431b3a5ba38bf1d18f238b2b2175e80a36aec11b0d061c19accea", "run_seconds": 2.347885847091675, "seed": 1125571558, "started_at": "2026-11-22T03:06:31Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015827, 0.021562], "effect": 0.002867, "evaluated_at": "2026-11-22T03:06:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f37ccc2c2c3d", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.192835808, 0.32561498, 0.277132865, 0.181757893, 0.22578302, 0.190267578, 0.253363285, 0.20779932, 0.308992828, 0.195735385, 0.34074503, 0.203967885, 0.190279302, 0.300606043, 0.32002325, 0.279289468], "raw": [0.192835808, 0.231211825, 0.216377062, 0.21439942, 0.185581308, 0.159623667, 0.245500753, 0.20323658, 0.23112888, 0.179050185, 0.27128143, 0.203967885, 0.231276697, 0.284257085, 0.297490135, 0.276045272], "temperature": [0.203492815, 0.247834948, 0.239795083, 0.252900103, 0.22578302, 0.209222783, 0.288014123, 0.20323658, 0.28967573, 0.195735385, 0.299290383, 0.228108965, 0.26064402, 0.306983312, 0.310066318, 0.279289468]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "125ecf3511870fe5895fa2491a15181edd1f78aaa0e2c73bedc8a34c65d6c267", "primary_control": "temperature", "result_hash": "0e23bdab33e431b3a5ba38bf1d18f238b2b2175e80a36aec11b0d061c19accea", "signature": "5610ee8eab37b38dc5edd4ceabeb73bfaedce7a300c04ca84cf483ae1e9fb2a4b3f770126a79f5e4b2238d5158756a382d9a8105ac5d60efe7448fc134de1403", "state": "VERIFIED", "summary": {"iw_temperature": 0.249637, "raw": 0.226454, "temperature": 0.252505}, "ties": 3, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:06:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:37Z", "cpu_seconds": 5.234267073000012, "created_at": "2026-11-22T03:06:31Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1dee6946d3be", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "c7a93f2562b190b6e6c78f18e1683f8f77232af78486727bee48b307c570a2ba", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1dee6946d3be", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:06:31Z", "seed": 631521412, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f61db8cd1cc2cbd6432cc088960ffa220a66523fda637d19a374f9c02beb4a90", "run_seconds": 5.9239020347595215, "seed": 631521412, "started_at": "2026-11-22T03:06:31Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:06:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1dee6946d3be", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c7a93f2562b190b6e6c78f18e1683f8f77232af78486727bee48b307c570a2ba", "primary_control": "reset", "result_hash": "f61db8cd1cc2cbd6432cc088960ffa220a66523fda637d19a374f9c02beb4a90", "signature": "718879cd5ef3b3d9b2b3f38810368c06fa735bd394a4f6b2ca3010bdc47ad7edeff6de5df8153ce1c59ed29d30af792b79aad20b5c0a8747c56f5322ec4b9401", "state": "PARTIAL", "summary": {"inherit": 0.5166666666666667, "reset": 0.5166666666666667, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:06:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:35Z", "cpu_seconds": 2.4896457750000422, "created_at": "2026-11-22T03:06:31Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-3fca0e9f26bf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "7b483b2345e668cb50686e22fe1233e63520abfcf3165cfb19de78d4d6e25f64", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-3fca0e9f26bf", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:06:31Z", "seed": 477310756, "treatment": "frontier"}, "priority": 0.2652510963211541, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "436557847e57e70d4e7bfe6e30e1f3da8fff48bedd0b4c50c0a19307e9a33bd7", "run_seconds": 3.359942674636841, "seed": 477310756, "started_at": "2026-11-22T03:06:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.010268, 0.083781], "effect": 0.047024, "evaluated_at": "2026-11-22T03:06:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-3fca0e9f26bf", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 1.0, 0.843457944, 0.856382979, 0.865196078, 0.840909091, 0.983333333, 1.0, 1.0, 0.858433735, 0.8325, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 0.784313725, 1.0, 0.994444444, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 1.0, 0.904205607, 0.816489362, 0.784313725, 0.878787879, 0.977777778, 1.0, 1.0, 0.972891566, 0.88, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.8062, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0156", "prereg_hash": "7b483b2345e668cb50686e22fe1233e63520abfcf3165cfb19de78d4d6e25f64", "primary_control": "random_walk", "result_hash": "436557847e57e70d4e7bfe6e30e1f3da8fff48bedd0b4c50c0a19307e9a33bd7", "signature": "a527f12c87b17520dedbf4e7c448a8d5ec855d1888240fe4955c537181135e8a66435c8a8128db4e8fcf6fa6213e0a1471e7cfde77a5232d7c6a5de0e253870e", "state": "VERIFIED", "summary": {"curiosity": 0.923351, "frontier": 0.981563, "random_walk": 0.934539}, "ties": 6, "treatment": "frontier", "wins": 6}, "verified_at": "2026-11-22T03:06:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:18Z", "cpu_seconds": 0.37238643099999535, "created_at": "2026-11-22T03:06:31Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-45f7b5b66b3c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "786de36f8e8e130fcc6da2c30c1de41fc7cee8dba0f39eb694989c281af4dc4d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-45f7b5b66b3c", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:06:31Z", "seed": 1288301394, "treatment": "frontier"}, "priority": 0.18789560480897513, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "3163d3176d9ba50817840ffe6841e72b6dca923bf8d77d503c2b687844264023", "run_seconds": 0.3769509792327881, "seed": 1288301394, "started_at": "2026-11-22T03:10:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.042271, 0.078431], "effect": 0.060351, "evaluated_at": "2026-11-22T03:10:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-45f7b5b66b3c", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.929104478, 1.0, 0.923913043, 0.925, 0.900735294, 0.887096774, 0.893939394, 0.919230769, 0.924603175, 0.926470588, 0.897727273, 1.0], "frontier": [0.996268657, 1.0, 0.989130435, 0.992857143, 0.985294118, 0.995967742, 0.992424242, 1.0, 1.0, 0.985294118, 0.981060606, 1.0], "random_walk": [0.902985075, 1.0, 0.916666667, 0.932142857, 0.933823529, 0.899193548, 0.90530303, 0.942307692, 0.94047619, 0.900735294, 0.920454545, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "786de36f8e8e130fcc6da2c30c1de41fc7cee8dba0f39eb694989c281af4dc4d", "primary_control": "random_walk", "result_hash": "3163d3176d9ba50817840ffe6841e72b6dca923bf8d77d503c2b687844264023", "signature": "c1f27198c80379729442c6b50099d570f9f5e7a6f7b25d085877fe041f885a0e8659973dc781b66ed5590d21c2618b3ee5d4917e5fce626529005f9430cf4a05", "state": "VERIFIED", "summary": {"curiosity": 0.927318, "frontier": 0.993191, "random_walk": 0.932841}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:10:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:12Z", "cpu_seconds": 1.4644480690000137, "created_at": "2026-11-22T03:06:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1544cc24d5d8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "5531811bf32ec1c6d2f6669fe65344a1b0f05b2a30abaa33d0708a4afac6568a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1544cc24d5d8", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:47Z", "seed": 397874176, "treatment": "iw_temperature"}, "priority": 0.19323505354034134, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "812b5290dfb696c5e5de4a34e9d12a2680e47eb0c95943f5a10d84152c8f5b4d", "run_seconds": 1.6116199493408203, "seed": 397874176, "started_at": "2026-11-22T03:10:10Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015666, -6.1e-05], "effect": -0.007863, "evaluated_at": "2026-11-22T03:10:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1544cc24d5d8", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.15236725, 0.175052592, 0.171077655, 0.161525073, 0.143751412, 0.190198222, 0.142390135, 0.1691242, 0.166823033, 0.152453577, 0.118628195, 0.240736962, 0.064354452, 0.20820557, 0.117194245, 0.117248138], "raw": [0.13821293, 0.166718422, 0.147131642, 0.172575293, 0.14167471, 0.163891437, 0.085352417, 0.1691242, 0.13371751, 0.1194552, 0.102007072, 0.197244065, 0.14554725, 0.134118318, 0.092991425, 0.087749608], "temperature": [0.13821293, 0.175052592, 0.147984327, 0.165676307, 0.14167471, 0.17867405, 0.129213252, 0.174807852, 0.159270378, 0.152453577, 0.140927872, 0.214205877, 0.057565445, 0.159462515, 0.11288518, 0.117248138]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "5531811bf32ec1c6d2f6669fe65344a1b0f05b2a30abaa33d0708a4afac6568a", "primary_control": "temperature", "result_hash": "812b5290dfb696c5e5de4a34e9d12a2680e47eb0c95943f5a10d84152c8f5b4d", "signature": "5d99d5f910f32a67703a7a99d44649b92f686931f9e0c114f472a117e7d5ad896c83a8ff1ac98a9bc8eda8a53ec30b497c6b0914d30eca62871867c53053cd00", "state": "VERIFIED", "summary": {"iw_temperature": 0.155696, "raw": 0.137344, "temperature": 0.147832}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:10:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:38Z", "cpu_seconds": 1.4495126979999782, "created_at": "2026-11-22T03:06:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1d356d10f7ad", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "e44127d08a64cb1cbbf36743a8091f74a5806be73e310a755693829340780565", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1d356d10f7ad", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:47Z", "seed": 376404568, "treatment": "iw_temperature"}, "priority": 0.19621343519176426, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "5239b33117b226dda290ed500fb4729741ea40534af0848499d8c399c409ecc9", "run_seconds": 1.5100739002227783, "seed": 376404568, "started_at": "2026-11-22T03:09:36Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021243, 0.001826], "effect": -0.009709, "evaluated_at": "2026-11-22T03:09:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1d356d10f7ad", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.065144602, 0.058236977, 0.151302825, 0.167868725, 0.059673025, 0.048401873, 0.028888098, 0.039437458, 0.041155008, 0.079768557, 0.024577998, 0.031665583, 0.129348545, 0.076645512, 0.11786267, 0.057724362], "raw": [0.03357693, 0.035845793, 0.196000493, 0.097263622, 0.086123462, 0.089472445, 0.072976243, 0.039437458, 0.03833641, 0.068567925, 0.021939578, 0.022206805, 0.086041888, 0.079093523, 0.08157044, 0.04045967], "temperature": [0.03357693, 0.037806208, 0.169936085, 0.100489797, 0.052407093, 0.048858237, 0.058240095, 0.04296237, 0.03833641, 0.079768557, 0.026668338, 0.03410398, 0.089973998, 0.076645512, 0.08934665, 0.043245435]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "e44127d08a64cb1cbbf36743a8091f74a5806be73e310a755693829340780565", "primary_control": "temperature", "result_hash": "5239b33117b226dda290ed500fb4729741ea40534af0848499d8c399c409ecc9", "signature": "5e64cfd5f01c2da45950a7b18fa582e68bb0d0787ec5f34dab5714ba6080fd97d463d8a9c7cbd597a863140ecd37524ac4f2daa8426efd49ffda427cbd00d802", "state": "VERIFIED", "summary": {"iw_temperature": 0.073606, "raw": 0.068057, "temperature": 0.063898}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:09:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:53Z", "cpu_seconds": 1.6024304719999805, "created_at": "2026-11-22T03:06:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c32d0c429025", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "c976d747baaa5b50847d36b2d93ee9005e38e540945ce70eb200f3bc7c6fa951", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c32d0c429025", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:06:47Z", "seed": 286608935, "treatment": "iw_temperature"}, "priority": 0.3256638333205784, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "8f281aed32d12937613825a3fbb2c3ea2945181e9f3da242a5442cdca01b7560", "run_seconds": 1.6492359638214111, "seed": 286608935, "started_at": "2026-11-22T03:06:51Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.008974, 0.035856], "effect": 0.013441, "evaluated_at": "2026-11-22T03:06:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c32d0c429025", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.33336195, 0.2061981, 0.267914463, 0.28698294, 0.275522483, 0.293106642, 0.29270763, 0.24457916, 0.28531393, 0.045898878, 0.100798547, 0.291167853, 0.257830357, 0.221975612, 0.302619283, 0.270718052], "raw": [0.327747387, 0.230828112, 0.217642158, 0.260062375, 0.263515335, 0.283767953, 0.28591003, 0.21948813, 0.229371623, 0.15536154, 0.199357375, 0.283708043, 0.17125408, 0.206397822, 0.257848065, 0.218289728], "temperature": [0.33147517, 0.27154578, 0.250048433, 0.269420868, 0.275522483, 0.294981442, 0.29270763, 0.24457916, 0.255063057, 0.14939673, 0.229546195, 0.288887037, 0.243713935, 0.23577495, 0.313881792, 0.245206273]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "c976d747baaa5b50847d36b2d93ee9005e38e540945ce70eb200f3bc7c6fa951", "primary_control": "temperature", "result_hash": "8f281aed32d12937613825a3fbb2c3ea2945181e9f3da242a5442cdca01b7560", "signature": "37dc023f5b34eec1502acaf10145121a35607668e7407231f8be68d3dc89941d79dea66308e39062f83caf6cc09cdfdf55e821f3466d302b5fa92bfa85d0e30b", "state": "VERIFIED", "summary": {"iw_temperature": 0.248543, "raw": 0.238159, "temperature": 0.261984}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:06:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:55Z", "cpu_seconds": 7.711872653, "created_at": "2026-11-22T03:06:47Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-033f78f28439", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6ca664fbc6135161f633d5c7699e4beae1cb284fa7b174ff71cd74fc55e097e3", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-033f78f28439", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:06:47Z", "seed": 1211331409, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "c0b9109fd0b2b698b95860765281aaeadab1959ea59571ccf7e5caa492682a1f", "run_seconds": 7.965123653411865, "seed": 1211331409, "started_at": "2026-11-22T03:06:47Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:06:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-033f78f28439", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6ca664fbc6135161f633d5c7699e4beae1cb284fa7b174ff71cd74fc55e097e3", "primary_control": "reset", "result_hash": "c0b9109fd0b2b698b95860765281aaeadab1959ea59571ccf7e5caa492682a1f", "signature": "8693d2029a043758001806c6f56bbdb495a65b17df204b1e60bdf459c98e458b4bb4b79a68da554b67b2154f6177bb821420e8a2ee5e8950c87fd7b5032b2e0c", "state": "PARTIAL", "summary": {"inherit": 0.85, "reset": 0.85, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:07:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:06:55Z", "cpu_seconds": 7.887616312000006, "created_at": "2026-11-22T03:06:47Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-23e2cea5b88a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e5ade7e127dbde1c4505a499df599fe35e0a1bdfc31bb1a9ed52bf382e230cd9", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-23e2cea5b88a", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:06:47Z", "seed": 15972057, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "20ecdaf62a5a10218f72e15cd4492e55557e44e92f9b91f87a446b0784574da7", "run_seconds": 8.109084129333496, "seed": 15972057, "started_at": "2026-11-22T03:06:47Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 4, "reset": 4}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:07:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-23e2cea5b88a", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4839981351660032, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0], "root": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e5ade7e127dbde1c4505a499df599fe35e0a1bdfc31bb1a9ed52bf382e230cd9", "primary_control": "reset", "result_hash": "20ecdaf62a5a10218f72e15cd4492e55557e44e92f9b91f87a446b0784574da7", "signature": "48c92858f23076fac94bde19c99d31d5621ca82eeb787cf1d8637aa92389b2a5f3ec21d277226c74152a0a9c75a24c486f1717c33c45b3f8cc99bd0e38ba740c", "state": "PARTIAL", "summary": {"inherit": 0.7166666666666667, "reset": 0.7166666666666667, "root": 0.45}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:31Z", "cpu_seconds": 0.18870329800000718, "created_at": "2026-11-22T03:06:47Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-e4345bdd7777", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "9fb398c48c59985ba02fbeb01155f0836880b5ceb92761f774f883e8b5458e32", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-e4345bdd7777", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:06:47Z", "seed": 121157571, "treatment": "frontier"}, "priority": 0.18600809729565954, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "b0017014884678210f1f56d19451757b2f969a9cf66d64c3ae056d38cd9b65c2", "run_seconds": 0.2860074043273926, "seed": 121157571, "started_at": "2026-11-22T03:10:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.029544, 0.118954], "effect": 0.074249, "evaluated_at": "2026-11-22T03:10:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-e4345bdd7777", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 1.0, 0.7734375, 0.877118644, 0.811403509, 1.0, 0.834745763, 1.0, 0.847222222, 0.840517241, 0.838235294, 0.826271186], "frontier": [1.0, 1.0, 1.0, 0.93220339, 0.877192982, 1.0, 0.894067797, 1.0, 1.0, 0.900862069, 0.921568627, 0.894067797], "random_walk": [1.0, 1.0, 0.75, 0.868644068, 0.75877193, 1.0, 0.86440678, 1.0, 0.819444444, 0.810344828, 0.848039216, 0.809322034]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "9fb398c48c59985ba02fbeb01155f0836880b5ceb92761f774f883e8b5458e32", "primary_control": "random_walk", "result_hash": "b0017014884678210f1f56d19451757b2f969a9cf66d64c3ae056d38cd9b65c2", "signature": "6d91857c021837f2b577d32f2d86fdaea36d21e1aef603d3b1fe990b9b71b639f21835d5d929935febcf9373b5a7752904f570bf2a588b31d2f1c6619ae24809", "state": "VERIFIED", "summary": {"curiosity": 0.887413, "frontier": 0.951664, "random_walk": 0.877414}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:10:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:25Z", "cpu_seconds": 3.7272812749999957, "created_at": "2026-11-22T03:06:47Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-e90fe3942a7e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "b6b5a4c13651b11453d87ce7b794d8a9e518b94e5a8653b305622f011f9c9364", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-e90fe3942a7e", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:06:47Z", "seed": 1140005688, "treatment": "frontier"}, "priority": 0.1754456491268803, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "a9b7a734b23b33233ab83e71e65c7d5c3e8ac3821a6f21457668fac9bea65732", "run_seconds": 3.838229179382324, "seed": 1140005688, "started_at": "2026-11-22T03:11:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.04778, 0.10264], "effect": 0.07521, "evaluated_at": "2026-11-22T03:11:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-e90fe3942a7e", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.830097087, 0.818421053, 0.834210526, 0.828431373, 0.846330275, 0.856796117, 0.857142857, 0.85106383, 0.800970874, 1.0, 0.919491525], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.90776699, 0.913157895, 0.857894737, 0.931372549, 0.926605505, 0.922330097, 0.909340659, 0.925531915, 0.837378641, 1.0, 0.966101695]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "b6b5a4c13651b11453d87ce7b794d8a9e518b94e5a8653b305622f011f9c9364", "primary_control": "random_walk", "result_hash": "a9b7a734b23b33233ab83e71e65c7d5c3e8ac3821a6f21457668fac9bea65732", "signature": "15ed11f56690ed11e49572a0e653d340cbecd304f1f0be3cd1e430020014b7d28ca0a0239ce94810aa3229f390762f341b1065f94e5b3441e14f499622a6dc08", "state": "VERIFIED", "summary": {"curiosity": 0.870246, "frontier": 1.0, "random_walk": 0.92479}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:11:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:05Z", "cpu_seconds": 0.3846497000000113, "created_at": "2026-11-22T03:07:03Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-15d693db58c0", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "e8129deb57a6337ed1156d4aceb57189dd344fb1fd4ed77be39b344e97fe7224", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-15d693db58c0", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:03Z", "seed": 1008386881, "treatment": "frontier"}, "priority": 0.3480826344998655, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-e513ad9c71bc", "result_hash": "7c2a69b7b97f81f9d688f05466e8a65e289eb55e69f528c60e0af3ac474199af", "run_seconds": 0.4132084846496582, "seed": 1008386881, "started_at": "2026-11-22T03:07:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.021363, 0.04982], "effect": 0.035592, "evaluated_at": "2026-11-22T03:07:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-15d693db58c0", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.824766355, 0.87394958, 0.929133858, 0.882231405, 0.854347826, 0.867924528, 0.872881356, 0.861111111, 0.898373984, 0.932692308, 0.911417323, 0.895833333], "frontier": [0.911214953, 0.907563025, 0.923228346, 0.931818182, 0.919565217, 0.922169811, 0.906779661, 0.908119658, 0.910569106, 0.944230769, 0.927165354, 0.914583333], "random_walk": [0.843457944, 0.852941176, 0.917322835, 0.876033058, 0.863043478, 0.853773585, 0.879237288, 0.886752137, 0.908536585, 0.930769231, 0.919291339, 0.86875]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "e8129deb57a6337ed1156d4aceb57189dd344fb1fd4ed77be39b344e97fe7224", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-e513ad9c71bc", "result_hash": "7c2a69b7b97f81f9d688f05466e8a65e289eb55e69f528c60e0af3ac474199af", "signature": "9bab42869dfb36ac6ab23f096369ad22bb3bf2c8abbfe003322cf44bd89f3992d9ea487da3e3ac3b58e3a2ed411a37580e5cf8db9c0167c5e17a89c79ab5b502", "state": "VERIFIED", "summary": {"curiosity": 0.883722, "frontier": 0.918917, "random_walk": 0.883326}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:10Z", "cpu_seconds": 2.5724821889999703, "created_at": "2026-11-22T03:07:03Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-f61d8051b33e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "795b3dd19794ac1c6b58de86aeda546d897b3324a41fce6e2d9ff269fe2f0233", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-f61d8051b33e", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:03Z", "seed": 1808813076, "treatment": "frontier"}, "priority": 0.2961596963551816, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "ef4b16a101b15eeb0bc5904cbf6efcbcbc641373b835d237476f4bd4a593ca22", "run_seconds": 3.5337867736816406, "seed": 1808813076, "started_at": "2026-11-22T03:07:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.035035, 0.084172], "effect": 0.059603, "evaluated_at": "2026-11-22T03:07:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-f61d8051b33e", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.852380952, 1.0, 0.800675676, 0.841836735, 0.792763158, 0.83, 0.821022727, 0.867788462, 0.853960396, 1.0, 1.0, 0.835858586], "frontier": [1.0, 1.0, 1.0, 1.0, 0.766447368, 1.0, 0.823863636, 1.0, 1.0, 1.0, 1.0, 0.929292929], "random_walk": [0.926190476, 1.0, 0.905405405, 0.903061224, 0.746710526, 0.9075, 0.744318182, 0.90625, 0.886138614, 1.0, 1.0, 0.878787879]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "795b3dd19794ac1c6b58de86aeda546d897b3324a41fce6e2d9ff269fe2f0233", "primary_control": "random_walk", "result_hash": "ef4b16a101b15eeb0bc5904cbf6efcbcbc641373b835d237476f4bd4a593ca22", "signature": "00d6910c10ea642c30f06d0ffa9eff584b2c3f39025d3ea6c8007f3f92ff880382191ad7e84da988b4c386c6788cdab67ad36a2a76927ee32e701e3885e68305", "state": "VERIFIED", "summary": {"curiosity": 0.874691, "frontier": 0.959967, "random_walk": 0.900364}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:37Z", "cpu_seconds": 0.5248920639999994, "created_at": "2026-11-22T03:07:03Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-fcd14904fe61", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "386600dac923ceedf357a94e5fc490c65fededbb2688d1153880d98f2114c679", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-fcd14904fe61", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:03Z", "seed": 1020355058, "treatment": "frontier"}, "priority": 0.17469622334655588, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "21937f1c0bfb27955683b6584416fed052ac6b4d4a8aedc586339185d8274dda", "run_seconds": 0.5263299942016602, "seed": 1020355058, "started_at": "2026-11-22T03:11:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.008441, 0.088579], "effect": 0.04851, "evaluated_at": "2026-11-22T03:11:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-fcd14904fe61", "family": "world_model_frontier", "hypothesis_supported": false, "instance_scores": {"curiosity": [1.0, 1.0, 0.818627451, 0.818627451, 1.0, 0.826923077, 1.0, 0.834821429, 1.0, 0.901639344, 0.90625, 1.0], "frontier": [1.0, 1.0, 1.0, 0.970588235, 1.0, 1.0, 1.0, 0.799107143, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.975, 0.862745098, 0.838235294, 1.0, 0.879807692, 1.0, 0.861607143, 1.0, 0.913934426, 0.85625, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 1, "n": 12, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "386600dac923ceedf357a94e5fc490c65fededbb2688d1153880d98f2114c679", "primary_control": "random_walk", "result_hash": "21937f1c0bfb27955683b6584416fed052ac6b4d4a8aedc586339185d8274dda", "signature": "a89f3e56092a22aa51b958473f7c3af0f65b47d051b3098c5d27b3883c92b580e23e818a3e8030f3c594db003c535d721e2f94104c0e3ea9d62745d442baac02", "state": "VERIFIED", "summary": {"curiosity": 0.925574, "frontier": 0.980808, "random_walk": 0.932298}, "ties": 5, "treatment": "frontier", "wins": 6}, "verified_at": "2026-11-22T03:11:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:32Z", "cpu_seconds": 7.444149461999984, "created_at": "2026-11-22T03:07:19Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-ddf378342cb4", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "89643a54656a9074ab22c3e05904ddbfd2ea08004565b49809f1dc86e55c6388", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-ddf378342cb4", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:07:19Z", "seed": 53388648, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "4bfbc14130fcd0be0700336d09e70c3f3fec11522e1f3e7e55a0b3cead262449", "run_seconds": 9.416303873062134, "seed": 53388648, "started_at": "2026-11-22T03:07:22Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:07:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-ddf378342cb4", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "89643a54656a9074ab22c3e05904ddbfd2ea08004565b49809f1dc86e55c6388", "primary_control": "reset", "result_hash": "4bfbc14130fcd0be0700336d09e70c3f3fec11522e1f3e7e55a0b3cead262449", "signature": "45bce7dabe23cb14b8e0e69cacc44163d0dfc6c77d2559b52ce0cd0bd82e47702129eeba1414a306a85c020997d6be139adfecd8aa6b16355c839d89ec5b2003", "state": "PARTIAL", "summary": {"inherit": 0.6333333333333333, "reset": 0.6333333333333333, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:07:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:32Z", "cpu_seconds": 1.6909494900000084, "created_at": "2026-11-22T03:07:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-641f35dc3648", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "2aed923e38a2f68bc7361b7b15be72075061af3f52274df0f12f83091cdcfdad", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-641f35dc3648", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:07:20Z", "seed": 1790669629, "treatment": "iw_temperature"}, "priority": 0.3364147068053005, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "aad3994246c9de4a95bc49c3e2bb3a1864b37e06c4ddc71b9124194203c1e1ec", "run_seconds": 2.3433165550231934, "seed": 1790669629, "started_at": "2026-11-22T03:07:29Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [0.00479, 0.047867], "effect": 0.026329, "evaluated_at": "2026-11-22T03:07:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-641f35dc3648", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.128432942, 0.283227042, 0.176199042, 0.135474205, 0.195650192, 0.232043583, 0.332452055, 0.275003683, 0.282589712, 0.190651408, 0.274634435, 0.052869325, 0.205126597, 0.248186538, 0.19514324, 0.191747922], "raw": [0.212708895, 0.263733005, 0.160313917, 0.188990325, 0.1816014, 0.193220075, 0.320213385, 0.285367453, 0.234254117, 0.179190377, 0.229564057, 0.147979003, 0.233841825, 0.17592426, 0.205832485, 0.21481701], "temperature": [0.223268252, 0.273681735, 0.160313917, 0.221023265, 0.195650192, 0.232043583, 0.323640748, 0.30591266, 0.2598112, 0.248045488, 0.25988674, 0.164462185, 0.257585407, 0.23462478, 0.22115467, 0.239586593]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.4031, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.395", "prereg_hash": "2aed923e38a2f68bc7361b7b15be72075061af3f52274df0f12f83091cdcfdad", "primary_control": "temperature", "result_hash": "aad3994246c9de4a95bc49c3e2bb3a1864b37e06c4ddc71b9124194203c1e1ec", "signature": "79fcec02884f22ae0e8a10bb6d32f1b2d5b97eaff015759faf480989d8380d9adb9ab0b97f7e83a4ad379e15eab3fdf4d96171dd65941fbb6be9e3ef59896b04", "state": "VERIFIED", "summary": {"iw_temperature": 0.212464, "raw": 0.214222, "temperature": 0.238793}, "ties": 2, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:07:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:33Z", "cpu_seconds": 8.15730800199998, "created_at": "2026-11-22T03:07:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1346167a5cf4", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "cd6c99d3c5e886844539e5c7ed7134decea6230a0e5f338c332a39e06d436311", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1346167a5cf4", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:07:20Z", "seed": 1077064525, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "cb732ce5ba9af868e013ebe8b0658137344e5a5b3fdef4561aba7b4df7c759a4", "run_seconds": 10.119022607803345, "seed": 1077064525, "started_at": "2026-11-22T03:07:23Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.185224, 0.085224], "effect": -0.05, "evaluated_at": "2026-11-22T03:07:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1346167a5cf4", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5751649176138385, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0789, "losses": 10, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.834", "prereg_hash": "cd6c99d3c5e886844539e5c7ed7134decea6230a0e5f338c332a39e06d436311", "primary_control": "reset", "result_hash": "cb732ce5ba9af868e013ebe8b0658137344e5a5b3fdef4561aba7b4df7c759a4", "signature": "eafbf06707a4d9899e32039949844decf5f04cfa817767aab837b88f39f5e15527e28c83cc108f322c7e847a3a86444d947ad39911d3b81158bffb5ce5c9610f", "state": "VERIFIED", "summary": {"inherit": 0.5666666666666667, "reset": 0.6166666666666667, "root": 0.4}, "ties": 43, "treatment": "inherit", "wins": 7}, "verified_at": "2026-11-22T03:07:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:19Z", "cpu_seconds": 0.32383205600001475, "created_at": "2026-11-22T03:07:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-4aa0b273b269", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "3ecdf9a6eda62e0e12d1609d7758f13fb9df5771711bd9fd6e6df9098eb7b9d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-4aa0b273b269", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:20Z", "seed": 1372689766, "treatment": "frontier"}, "priority": 0.18694246923794142, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "e512136fd27d9f1beb9523752711e016cff11cf68c3bcfffe13050f2f91385c5", "run_seconds": 0.3243081569671631, "seed": 1372689766, "started_at": "2026-11-22T03:10:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.04548, 0.090101], "effect": 0.067791, "evaluated_at": "2026-11-22T03:10:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-4aa0b273b269", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.850917431, 0.794270833, 0.792613636, 0.804597701, 0.86965812, 0.841121495, 1.0, 0.784653465, 0.807432432, 0.7575, 0.7875], "frontier": [1.0, 0.903669725, 0.859375, 0.897727273, 0.902298851, 0.908119658, 0.904205607, 1.0, 0.878712871, 0.935810811, 0.8725, 0.885], "random_walk": [1.0, 0.834862385, 0.817708333, 0.795454545, 0.781609195, 0.858974359, 0.838785047, 1.0, 0.792079208, 0.834459459, 0.8, 0.78]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "3ecdf9a6eda62e0e12d1609d7758f13fb9df5771711bd9fd6e6df9098eb7b9d2", "primary_control": "random_walk", "result_hash": "e512136fd27d9f1beb9523752711e016cff11cf68c3bcfffe13050f2f91385c5", "signature": "284ae817d09e0f901f0f0234c611fa3b4d96ef08416aaccdbe7d22981f8e3eb304cc1f05db39178256aed1601ccc28439b87f5061bccb122acf2ade842aeb109", "state": "VERIFIED", "summary": {"curiosity": 0.840855, "frontier": 0.912285, "random_walk": 0.844494}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:10:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:51Z", "cpu_seconds": 2.7479655000000207, "created_at": "2026-11-22T03:07:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-8e5fcc763724", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "ded119fd10998ba57e744afa1dd3e102cb7f19d671bfdd4a41f2412f077c70e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-8e5fcc763724", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:20Z", "seed": 1626603917, "treatment": "frontier"}, "priority": 0.24739473403002613, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "63060da3a83798722cf13d80ab0f6f83025b4125ae6eaf6e47a88a1134ac91d0", "run_seconds": 4.316015958786011, "seed": 1626603917, "started_at": "2026-11-22T03:07:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.030031, 0.102505], "effect": 0.066268, "evaluated_at": "2026-11-22T03:07:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-8e5fcc763724", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 1.0, 0.920454545, 1.0, 0.873684211, 0.893229167, 1.0, 0.829081633, 1.0, 0.804761905, 0.789634146, 0.792553191], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 0.994791667, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 1.0, 0.943181818, 1.0, 0.923684211, 0.84375, 1.0, 0.890306122, 1.0, 0.864285714, 0.856707317, 0.877659574]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.1072, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00781", "prereg_hash": "ded119fd10998ba57e744afa1dd3e102cb7f19d671bfdd4a41f2412f077c70e2", "primary_control": "random_walk", "result_hash": "63060da3a83798722cf13d80ab0f6f83025b4125ae6eaf6e47a88a1134ac91d0", "signature": "99ae2aa57160f8fc3ac0c5406a233727a516a01c5a7ec83aa2d14b572968e8d14ba06bca47b818329532d1eb4bbdc45708e3be2771e54f2e8d3040d562cb6404", "state": "VERIFIED", "summary": {"curiosity": 0.908617, "frontier": 0.999566, "random_walk": 0.933298}, "ties": 5, "treatment": "frontier", "wins": 7}, "verified_at": "2026-11-22T03:07:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:51Z", "cpu_seconds": 0.2792394709999826, "created_at": "2026-11-22T03:07:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-aa885d70d9af", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "83d2ff7f7140aeed3af57241ed1df2630c3c1f4948669d1bbd11084acf31305f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-aa885d70d9af", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:20Z", "seed": 1051385697, "treatment": "frontier"}, "priority": 0.1793256393028515, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "e1d00e255575c36ef37246dfa2e569a10844ab8f719619f9e868b77e2ec2cea0", "run_seconds": 0.27994441986083984, "seed": 1051385697, "started_at": "2026-11-22T03:10:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.048911, 0.077343], "effect": 0.063127, "evaluated_at": "2026-11-22T03:11:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-aa885d70d9af", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.965277778, 0.954861111, 0.894366197, 0.891304348, 0.935714286, 0.957142857, 0.9375, 0.963333333, 0.95, 0.904411765, 0.956666667, 0.914179104], "frontier": [0.989583333, 0.996527778, 0.975352113, 0.992753623, 0.989285714, 0.992857143, 0.996527778, 0.98, 0.992857143, 0.996323529, 0.986666667, 1.0], "random_walk": [0.916666667, 0.923611111, 0.940140845, 0.916666667, 0.935714286, 0.910714286, 0.916666667, 0.95, 0.939285714, 0.889705882, 0.966666667, 0.925373134]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "83d2ff7f7140aeed3af57241ed1df2630c3c1f4948669d1bbd11084acf31305f", "primary_control": "random_walk", "result_hash": "e1d00e255575c36ef37246dfa2e569a10844ab8f719619f9e868b77e2ec2cea0", "signature": "865462ab51bdca839bb98bdf4f54b51121a2058c67ae6c71acd2725bd5bcf755c740d5586e643d0966419b38dca244908d10ac15975da9b235d3fd91aa204c08", "state": "VERIFIED", "summary": {"curiosity": 0.935396, "frontier": 0.990728, "random_walk": 0.927601}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:11:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:24Z", "cpu_seconds": 1.0824262800000213, "created_at": "2026-11-22T03:07:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-ac190d2fb702", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "60380593aff472e56235d55eedb59ef8c93166f28527ab0e082d71af7ddab9f3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-ac190d2fb702", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:20Z", "seed": 378000490, "treatment": "frontier"}, "priority": 0.36712272737727036, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-3502ceb5736b", "result_hash": "621b2604f1b6f1d6785d06b009af90052298f0d7ab4eec274216276c9685cb60", "run_seconds": 1.1562449932098389, "seed": 378000490, "started_at": "2026-11-22T03:07:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.010122, 0.059314], "effect": 0.034718, "evaluated_at": "2026-11-22T03:07:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-ac190d2fb702", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.875, 0.859090909, 0.935483871, 0.9296875, 1.0, 1.0, 1.0, 1.0, 1.0, 0.849056604, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.915178571, 0.909090909, 0.935483871, 0.91796875, 1.0, 1.0, 1.0, 1.0, 1.0, 0.905660377, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.5051, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0313", "prereg_hash": "60380593aff472e56235d55eedb59ef8c93166f28527ab0e082d71af7ddab9f3", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-3502ceb5736b", "result_hash": "621b2604f1b6f1d6785d06b009af90052298f0d7ab4eec274216276c9685cb60", "signature": "530aa4164bb64ee8ca2a345f2b5ded73c94f9599fabdee1e93393c4eea2e2f1fb049e64a28297d9d0fc414d87b2b98700ef79307e5641cdc9a8f99946d956208", "state": "VERIFIED", "summary": {"curiosity": 0.954027, "frontier": 1.0, "random_walk": 0.965282}, "ties": 7, "treatment": "frontier", "wins": 5}, "verified_at": "2026-11-22T03:07:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:43Z", "cpu_seconds": 1.7116190700000118, "created_at": "2026-11-22T03:07:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-2b25f4da1eec", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "867cf33313e1c6ab4b072d46be868b40c61f6f105e25a306d51106c7b8434d4a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-2b25f4da1eec", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:07:36Z", "seed": 1932908493, "treatment": "iw_temperature"}, "priority": 0.3757080895118226, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ef5e724b695a384bc38d979f1a2952a0caaa2e0e4c9cdbd27fec4bb5fb464193", "run_seconds": 5.8389036655426025, "seed": 1932908493, "started_at": "2026-11-22T03:07:37Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:07:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-2b25f4da1eec", "family": "calibration_iw", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"iw_temperature": [0.0, 0.238095], "raw": [0.0, 0.210191], "temperature": [0.0, 0.227692]}, "prereg_hash": "867cf33313e1c6ab4b072d46be868b40c61f6f105e25a306d51106c7b8434d4a", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "ef5e724b695a384bc38d979f1a2952a0caaa2e0e4c9cdbd27fec4bb5fb464193", "signature": "ecb8ef8f99746c473f9a026494b64d7d9432aefd53d37830176df162fb36c8476e162cb740dfe03cd40f0f4e5fc346f40f05644b6da4d7c58b3e335c6f425d0e", "state": "REJECTED", "summary": {"iw_temperature": 0.238095, "raw": 0.210191, "temperature": 0.227692}}, "verified_at": "2026-11-22T03:07:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:42Z", "cpu_seconds": 1.9181048159999818, "created_at": "2026-11-22T03:07:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-8834464d31df", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "9bb96918ca8b06de2509fd63fc0f51ce33a1325f2751bd550be62132b0a47a42", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-8834464d31df", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:07:36Z", "seed": 1731020765, "treatment": "iw_temperature"}, "priority": 0.3303831198257121, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "3c5e2979bb705fe752b2c84b9e18c1a685b3feda7bd0ba6c884b1d3e85d56a37", "run_seconds": 4.947826147079468, "seed": 1731020765, "started_at": "2026-11-22T03:07:37Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020259, 0.020873], "effect": 0.000307, "evaluated_at": "2026-11-22T03:07:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-8834464d31df", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.297923812, 0.217578477, 0.112517332, 0.256048683, 0.309049688, 0.302173045, 0.24208193, 0.272692252, 0.274304825, 0.193015578, 0.14688669, 0.237243312, 0.28475227, 0.290361345, 0.323647998, 0.318965385], "raw": [0.274992195, 0.26555612, 0.2124716, 0.195304315, 0.210331858, 0.282388405, 0.209033173, 0.22765983, 0.197754365, 0.21455401, 0.14688669, 0.224119212, 0.206131583, 0.2421793, 0.325089827, 0.243498952], "temperature": [0.280916792, 0.278287515, 0.203875143, 0.195304315, 0.29133031, 0.31097084, 0.228403725, 0.263176435, 0.20655903, 0.255253812, 0.14688669, 0.221116827, 0.28475227, 0.29955401, 0.32896163, 0.288800212]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "9bb96918ca8b06de2509fd63fc0f51ce33a1325f2751bd550be62132b0a47a42", "primary_control": "temperature", "result_hash": "3c5e2979bb705fe752b2c84b9e18c1a685b3feda7bd0ba6c884b1d3e85d56a37", "signature": "81c8cfe0cc8ddc8db769c98caee7856dd61677f03168b5370750094c3e8816cd492d91053ee1a08fef6a4415998436b278550be2a1479c7396e1cd96592e470e", "state": "VERIFIED", "summary": {"iw_temperature": 0.254953, "raw": 0.229872, "temperature": 0.255259}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:07:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:19Z", "cpu_seconds": 1.7386848000000157, "created_at": "2026-11-22T03:07:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1f64331acd99", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "7c875253d545213462b6d621e162bf397db0e034ad16b98e80a32d77b28fa881", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1f64331acd99", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:07:52Z", "seed": 778302929, "treatment": "iw_temperature"}, "priority": 0.21395239336448857, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "fe441d9ea78606b1016429d21ee896b188fcb57b41d2df0a0f6e3f99b7492320", "run_seconds": 1.9684045314788818, "seed": 778302929, "started_at": "2026-11-22T03:08:17Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.024935, 0.000476], "effect": -0.01223, "evaluated_at": "2026-11-22T03:08:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1f64331acd99", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.099092865, 0.196294417, 0.08652383, 0.109448673, 0.17910219, 0.13740387, 0.153495205, 0.184221458, 0.131108245, 0.108684878, 0.144885608, 0.158170843, 0.107670373, 0.134717275, 0.156070745, 0.081949573], "raw": [0.091632583, 0.09260222, 0.092328615, 0.07832559, 0.10840656, 0.108132312, 0.12505034, 0.157425728, 0.09734092, 0.073404323, 0.100253412, 0.111107262, 0.090639507, 0.080072697, 0.119545367, 0.093040115], "temperature": [0.104836747, 0.11438801, 0.087805215, 0.109448673, 0.1449942, 0.13740387, 0.13574273, 0.184221458, 0.110634125, 0.093745917, 0.100253412, 0.14643888, 0.124705303, 0.111694263, 0.159332292, 0.10751754]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "7c875253d545213462b6d621e162bf397db0e034ad16b98e80a32d77b28fa881", "primary_control": "temperature", "result_hash": "fe441d9ea78606b1016429d21ee896b188fcb57b41d2df0a0f6e3f99b7492320", "signature": "f8255ced8cb39a8e1bb6357cec05bbed22f57985b9589ee916903a87c4a37d140df4ef16f8e9467fc10fae163f83435dca38cecbd1d5d99a4cc4c7f88d71730b", "state": "VERIFIED", "summary": {"iw_temperature": 0.135553, "raw": 0.101207, "temperature": 0.123323}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:08:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:05Z", "cpu_seconds": 1.5651615289999938, "created_at": "2026-11-22T03:07:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4585ac92ef5a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "9a18f458b810d99ab0d3a90edc0954be7b9898de14e0e7af8330f62497998ce8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4585ac92ef5a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:07:52Z", "seed": 977575969, "treatment": "iw_temperature"}, "priority": 0.3063836196332358, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "28ffb2b1e4dc919a8f139f95252eb53dd7a2eab5bc735eba3fa478f85861876a", "run_seconds": 1.706960916519165, "seed": 977575969, "started_at": "2026-11-22T03:08:03Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.003265, 0.019635], "effect": 0.008185, "evaluated_at": "2026-11-22T03:08:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4585ac92ef5a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.217940155, 0.264450732, 0.324957235, 0.263423185, 0.222470152, 0.151515505, 0.185953763, 0.21037425, 0.25238203, 0.127995113, 0.144960102, 0.339131103, 0.291325897, 0.309382722, 0.26924346, 0.282038827], "raw": [0.217940155, 0.232876955, 0.239348922, 0.24742614, 0.211908255, 0.202742472, 0.16039939, 0.19713055, 0.22862288, 0.150691848, 0.170422305, 0.317301727, 0.2879017, 0.29701841, 0.22841011, 0.271432305], "temperature": [0.251379245, 0.256327313, 0.290887022, 0.263423185, 0.211908255, 0.222242155, 0.204411985, 0.222690453, 0.26239698, 0.150691848, 0.170422305, 0.335718255, 0.291325897, 0.303391845, 0.26924346, 0.282038827]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.4121, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.387", "prereg_hash": "9a18f458b810d99ab0d3a90edc0954be7b9898de14e0e7af8330f62497998ce8", "primary_control": "temperature", "result_hash": "28ffb2b1e4dc919a8f139f95252eb53dd7a2eab5bc735eba3fa478f85861876a", "signature": "afb1b9d2a18d509a218def9f8ed29621e14d037b9f318495065c40bf51bfe9fef9ae74fe0c87e72e26b563643269048b6887e4bb6eb9057a4069d2aaa15c8b0b", "state": "VERIFIED", "summary": {"iw_temperature": 0.241097, "raw": 0.228848, "temperature": 0.249281}, "ties": 4, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:08:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:07:58Z", "cpu_seconds": 1.4895589230000041, "created_at": "2026-11-22T03:07:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-8fcf09e138f4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "26c4f912f9276c8be91966b86801dfed4166829361824be7ed3894ecd7352eea", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-8fcf09e138f4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:07:52Z", "seed": 47570475, "treatment": "iw_temperature"}, "priority": 0.3694018753431115, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "b17fc84c2f6d02daf00f5e0f78674b1a9a94e21f837ae71636b04bd50c9e7ed8", "run_seconds": 1.9072155952453613, "seed": 47570475, "started_at": "2026-11-22T03:07:56Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.008894, 0.018707], "effect": 0.004907, "evaluated_at": "2026-11-22T03:08:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-8fcf09e138f4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.244471415, 0.211341512, 0.265221957, 0.215933255, 0.173358717, 0.240795063, 0.226147288, 0.19404876, 0.171578147, 0.155897725, 0.187736443, 0.296696038, 0.226806662, 0.272486405, 0.230062307, 0.20184315], "raw": [0.170889532, 0.23793196, 0.223055493, 0.214973728, 0.208318495, 0.240795063, 0.21583135, 0.15416982, 0.162579652, 0.181565578, 0.19805997, 0.244076072, 0.204228, 0.183856048, 0.185643225, 0.217915982], "temperature": [0.213608475, 0.281143598, 0.265221957, 0.20254788, 0.208318495, 0.254617465, 0.226147288, 0.181228635, 0.162579652, 0.181565578, 0.21479873, 0.272044422, 0.226806662, 0.239368855, 0.22273362, 0.240200295]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "26c4f912f9276c8be91966b86801dfed4166829361824be7ed3894ecd7352eea", "primary_control": "temperature", "result_hash": "b17fc84c2f6d02daf00f5e0f78674b1a9a94e21f837ae71636b04bd50c9e7ed8", "signature": "382e9cb388c80c566ce3805230347c042e6ec91535cdcc90c24e920f89a52f50429c3bc858fd29f3cd11d8b7265e037b5bf16b3e4e6c5adcffc78810d6699b06", "state": "VERIFIED", "summary": {"iw_temperature": 0.219652, "raw": 0.202743, "temperature": 0.224558}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:08:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:02Z", "cpu_seconds": 4.8438697910000315, "created_at": "2026-11-22T03:07:52Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-f204ccffe048", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "822fd8b946c41af3bc001f4507659845e2fc95ea22b98d8100119a71dc046255", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-f204ccffe048", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:07:52Z", "seed": 287577024, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "fa127685a67c013102fb485b777dc53b3f81e564e27b12451377313c339fbaf0", "run_seconds": 5.3383588790893555, "seed": 287577024, "started_at": "2026-11-22T03:07:56Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 1, "reset": 1}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:08:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-f204ccffe048", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0], "root": [1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "822fd8b946c41af3bc001f4507659845e2fc95ea22b98d8100119a71dc046255", "primary_control": "reset", "result_hash": "fa127685a67c013102fb485b777dc53b3f81e564e27b12451377313c339fbaf0", "signature": "79da60b4a480062ba6954a73d9e68fef8bd746845c454cb7d508279cf71fa121436b3783d1550738dde499ecd40b363039979ad9fe773f16e356ce2e7353ef08", "state": "PARTIAL", "summary": {"inherit": 0.5166666666666667, "reset": 0.5166666666666667, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:08:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:02Z", "cpu_seconds": 5.498088921000004, "created_at": "2026-11-22T03:07:52Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-f2c310459dda", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "49a7eb6e8d866f731109825182c188467b04914c65bafd240bb05cae6c768a34", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-f2c310459dda", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:07:52Z", "seed": 1600113771, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "ac568343dfb7a209a299ed87af9d0e98e5213fd0f39c3d7549e78e2962bfac0b", "run_seconds": 5.981699466705322, "seed": 1600113771, "started_at": "2026-11-22T03:07:56Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 4, "reset": 4}, "better": "higher", "ci95": [-0.046588, 0.046588], "effect": 0.0, "evaluated_at": "2026-11-22T03:08:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-f2c310459dda", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4311496462608058, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "49a7eb6e8d866f731109825182c188467b04914c65bafd240bb05cae6c768a34", "primary_control": "reset", "result_hash": "ac568343dfb7a209a299ed87af9d0e98e5213fd0f39c3d7549e78e2962bfac0b", "signature": "22ec7831f2d82bab29e47547e29c67e02e57742af2a3a61e0ab9c630b18452729f4238c2d389722a405d56af439f5c0696bd882e8def7c40aa8f5846277dca0c", "state": "VERIFIED", "summary": {"inherit": 0.9, "reset": 0.9, "root": 0.45}, "ties": 58, "treatment": "inherit", "wins": 1}, "verified_at": "2026-11-22T03:08:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:09Z", "cpu_seconds": 2.310586817000001, "created_at": "2026-11-22T03:07:52Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-42bcf7d44458", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "ea17d4524af1dc6642696814ad4b349c27bf492d8c06f7faf258748f0876ef9f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-42bcf7d44458", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:07:52Z", "seed": 1619085260, "treatment": "frontier"}, "priority": 0.24986542814952242, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "06d86606daba8147c91d62b9420507be387a02fef8a079b1a39622d163c12975", "run_seconds": 2.3538005352020264, "seed": 1619085260, "started_at": "2026-11-22T03:08:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.021079, 0.063492], "effect": 0.042286, "evaluated_at": "2026-11-22T03:08:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-42bcf7d44458", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.889830508, 1.0, 0.882352941, 1.0, 0.81, 0.799479167, 1.0, 1.0, 0.863333333, 0.848684211, 0.838235294, 0.803571429], "frontier": [0.995762712, 1.0, 1.0, 1.0, 1.0, 0.979166667, 1.0, 1.0, 1.0, 1.0, 0.876470588, 0.862244898], "random_walk": [0.961864407, 1.0, 0.941176471, 1.0, 0.8925, 0.901041667, 1.0, 1.0, 0.936666667, 0.914473684, 0.829411765, 0.829081633]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "ea17d4524af1dc6642696814ad4b349c27bf492d8c06f7faf258748f0876ef9f", "primary_control": "random_walk", "result_hash": "06d86606daba8147c91d62b9420507be387a02fef8a079b1a39622d163c12975", "signature": "d9d300b61ed0a1c1d2ac26616eb11b81873db88f6f2d6fea7d5e9a0210328da2777eca3fc7cfe55bf98cd9644266d656ea9b490590e7f16d240a34954d51fa08", "state": "VERIFIED", "summary": {"curiosity": 0.894624, "frontier": 0.976137, "random_walk": 0.933851}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:08:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:12Z", "cpu_seconds": 1.5449132520000148, "created_at": "2026-11-22T03:08:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-648a20005670", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "2dc3b998d0d517e7fa300ca6eb501ab2699ffbe0e282712f955b0c979d0fb09e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-648a20005670", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:08:09Z", "seed": 965841655, "treatment": "iw_temperature"}, "priority": 0.32160466652924447, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "70ad9f469d2b3e19e85fa7e38098cabc9daa247473a09037113e8ea3216e3064", "run_seconds": 1.820068359375, "seed": 965841655, "started_at": "2026-11-22T03:08:10Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019374, 0.004453], "effect": -0.00746, "evaluated_at": "2026-11-22T03:08:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-648a20005670", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.257052302, 0.254895905, 0.234685378, 0.15241425, 0.171909457, 0.120809223, 0.139884465, 0.179241755, 0.266490547, 0.224676645, 0.12531365, 0.234053255, 0.225815485, 0.15734419, 0.242923265, 0.17097843], "raw": [0.200409188, 0.186951537, 0.208390177, 0.172856312, 0.171909457, 0.17313466, 0.197727345, 0.179241755, 0.197758642, 0.214199798, 0.175991673, 0.16643544, 0.211648793, 0.173239202, 0.205148155, 0.227181638], "temperature": [0.243393165, 0.237340577, 0.208390177, 0.183927288, 0.163514897, 0.10884363, 0.165026648, 0.179241755, 0.197758642, 0.214199798, 0.11581971, 0.226816713, 0.225815485, 0.185852837, 0.212205153, 0.17097843]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "2dc3b998d0d517e7fa300ca6eb501ab2699ffbe0e282712f955b0c979d0fb09e", "primary_control": "temperature", "result_hash": "70ad9f469d2b3e19e85fa7e38098cabc9daa247473a09037113e8ea3216e3064", "signature": "bb1838c46894d920a859b1f1b9dc7d96d26adc213dd27cc3f4360bf2932dd2e4d3be538d70c205448fb7ff4eabbc77126b9568cb733fda581d9883fb43dc9906", "state": "VERIFIED", "summary": {"iw_temperature": 0.197406, "raw": 0.191389, "temperature": 0.189945}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:08:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:17Z", "cpu_seconds": 6.199117085999973, "created_at": "2026-11-22T03:08:09Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-4f5524160978", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "a76a0bc8b6006746988873ad2ed04ab45a344821621c97ac6abc5b21142e5d06", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-4f5524160978", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:08:09Z", "seed": 1269257475, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "df8d90f8375e2835ff627f8cc053cc404be358caa93476252f39ab4ec9481362", "run_seconds": 6.785233974456787, "seed": 1269257475, "started_at": "2026-11-22T03:08:10Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:08:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-4f5524160978", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5435625004334507, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a76a0bc8b6006746988873ad2ed04ab45a344821621c97ac6abc5b21142e5d06", "primary_control": "reset", "result_hash": "df8d90f8375e2835ff627f8cc053cc404be358caa93476252f39ab4ec9481362", "signature": "a7b2a78a1199a02eab60cd864eaa62bc172b007962809effdc43e2d8e6656991b2147ff097cad486fbae9b6c0cba3f55f46969fcfabbb300c9e3db420346880b", "state": "PARTIAL", "summary": {"inherit": 0.5666666666666667, "reset": 0.5666666666666667, "root": 0.21666666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:08:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:02Z", "cpu_seconds": 0.20852761099999384, "created_at": "2026-11-22T03:08:09Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-871a569dad6b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "b1bfa33cc074e75d98dc686cf346138c9cbf38e23b402dc9fe647e1078f527c0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-871a569dad6b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:09Z", "seed": 914288034, "treatment": "frontier"}, "priority": 0.1664797764776506, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "bfbb16725d1d19ac19945437314b199f6c6872de9e5a7cdf0d33b832792fd95a", "run_seconds": 0.22287750244140625, "seed": 914288034, "started_at": "2026-11-22T03:13:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.039734, 0.094714], "effect": 0.067224, "evaluated_at": "2026-11-22T03:13:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-871a569dad6b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.805084746, 0.831896552, 1.0, 0.805084746, 0.854166667, 0.856557377, 0.777777778, 0.855769231, 1.0, 0.758928571, 0.806603774], "frontier": [1.0, 0.881355932, 0.913793103, 1.0, 0.881355932, 0.932291667, 0.893442623, 0.916666667, 0.894230769, 1.0, 0.861607143, 0.900943396], "random_walk": [1.0, 0.788135593, 0.784482759, 1.0, 0.788135593, 0.859375, 0.836065574, 0.783333333, 0.855769231, 1.0, 0.78125, 0.79245283]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "b1bfa33cc074e75d98dc686cf346138c9cbf38e23b402dc9fe647e1078f527c0", "primary_control": "random_walk", "result_hash": "bfbb16725d1d19ac19945437314b199f6c6872de9e5a7cdf0d33b832792fd95a", "signature": "fb109d614307085745823903fb3a4be9d56db531a4fae9d4b8a5a893732782417493db90a7915a7508859a774ee053f109bc6d5cda7ee845a30c85a4c74d4707", "state": "VERIFIED", "summary": {"curiosity": 0.862656, "frontier": 0.922974, "random_walk": 0.85575}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:13:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:50Z", "cpu_seconds": 0.177329076999996, "created_at": "2026-11-22T03:08:09Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-bed6469ce5bb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "bf85f2fe99613c72b177590c4fa38d1056c4f99e7fca6260d6dd9d83de493a80", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-bed6469ce5bb", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:09Z", "seed": 1303169104, "treatment": "frontier"}, "priority": 0.1817258584996994, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "0835be1043439b56baca3db2a978ed2c2bb778d1e5de6b22dc12a6806d869e37", "run_seconds": 0.18529081344604492, "seed": 1303169104, "started_at": "2026-11-22T03:10:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019133, 0.036976], "effect": 0.028054, "evaluated_at": "2026-11-22T03:11:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-bed6469ce5bb", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.921641791, 0.914179104, 0.893382353, 0.90234375, 0.901639344, 1.0, 0.878787879, 0.873134328, 0.897058824, 0.911538462, 0.865671642], "frontier": [1.0, 0.936567164, 0.932835821, 0.911764706, 0.91015625, 0.926229508, 1.0, 0.912878788, 0.906716418, 0.926470588, 0.915384615, 0.902985075], "random_walk": [1.0, 0.899253731, 0.899253731, 0.897058824, 0.88671875, 0.897540984, 1.0, 0.863636364, 0.873134328, 0.886029412, 0.880769231, 0.861940299]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "bf85f2fe99613c72b177590c4fa38d1056c4f99e7fca6260d6dd9d83de493a80", "primary_control": "random_walk", "result_hash": "0835be1043439b56baca3db2a978ed2c2bb778d1e5de6b22dc12a6806d869e37", "signature": "88bb9568c989bcb356557c6fdd58a51d22f01c2150310bffc1c521ff8b917839d96de912c39ac4c2c803690235851ecba714b6917de23a47a8ba228d78e70e0b", "state": "VERIFIED", "summary": {"curiosity": 0.913281, "frontier": 0.931832, "random_walk": 0.903778}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:11:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:01Z", "cpu_seconds": 1.4698800170000368, "created_at": "2026-11-22T03:08:25Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-90f98064bf24", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "c3513ab053ba8359cc9ce7b666121867acf6d45d5932875c1e8e073652aed582", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-90f98064bf24", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:08:25Z", "seed": 733322110, "treatment": "iw_temperature"}, "priority": 0.19329017141973254, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "dd1daee1e009c393d3af13088cce9bebcb3e1ae99d7ff6f90f9bb983ace75b93", "run_seconds": 1.5532128810882568, "seed": 733322110, "started_at": "2026-11-22T03:10:00Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010061, 0.024761], "effect": 0.00735, "evaluated_at": "2026-11-22T03:10:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-90f98064bf24", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.162485688, 0.18814729, 0.129903058, 0.147937013, 0.178122587, 0.12616057, 0.147648815, 0.152783007, 0.144830242, 0.135076285, 0.166996798, 0.214849275, 0.133118262, 0.132584995, 0.112930703, 0.059441387], "raw": [0.126848435, 0.127031382, 0.1753431, 0.195688843, 0.132620135, 0.086602717, 0.147648815, 0.142158557, 0.133901467, 0.205424502, 0.103066947, 0.173745367, 0.132495632, 0.103313205, 0.166088922, 0.12886278], "temperature": [0.139465958, 0.145302648, 0.169628987, 0.150159663, 0.15940962, 0.12616057, 0.147648815, 0.129243953, 0.149577943, 0.191572958, 0.146985078, 0.231810135, 0.122874762, 0.121932968, 0.17903919, 0.139808208]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2184, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.605", "prereg_hash": "c3513ab053ba8359cc9ce7b666121867acf6d45d5932875c1e8e073652aed582", "primary_control": "temperature", "result_hash": "dd1daee1e009c393d3af13088cce9bebcb3e1ae99d7ff6f90f9bb983ace75b93", "signature": "f51bd1b9ecf1fa625260ad88552407ffb92f0dfe800c9313aa96b163ad743a84b1d2d3109f6b849dc35b241a2b846526aab448d689a9b0589a231e8b78f1e700", "state": "VERIFIED", "summary": {"iw_temperature": 0.145813, "raw": 0.142553, "temperature": 0.153164}, "ties": 2, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:10:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:33Z", "cpu_seconds": 1.5860251849999827, "created_at": "2026-11-22T03:08:25Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-a8918dbdc7b4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "4f0feb7f375a6b9c4ffe789f77300870a6f96dc6706348a45f8dde7d5d07c968", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-a8918dbdc7b4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:08:25Z", "seed": 398328170, "treatment": "iw_temperature"}, "priority": 0.33711171048249355, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "557995260ab4e50599ef8609f47330575adad3c2afbac91457a73c0977386950", "run_seconds": 1.7678625583648682, "seed": 398328170, "started_at": "2026-11-22T03:08:31Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014232, 0.013767], "effect": -0.000232, "evaluated_at": "2026-11-22T03:08:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-a8918dbdc7b4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.222677702, 0.300707507, 0.270763283, 0.323401153, 0.352687715, 0.203463903, 0.302698638, 0.326296345, 0.284379435, 0.236109983, 0.337210172, 0.210790003, 0.31025892, 0.291102507, 0.088271572, 0.287642585], "raw": [0.199165182, 0.299244055, 0.20296025, 0.305308132, 0.31103237, 0.212852532, 0.250252452, 0.305514663, 0.284379435, 0.185464607, 0.301274895, 0.196220458, 0.294067533, 0.21051017, 0.163494838, 0.26129218], "temperature": [0.228675747, 0.300240005, 0.241625603, 0.321575272, 0.34710209, 0.24365769, 0.284788867, 0.331256998, 0.300415062, 0.2100973, 0.32967594, 0.210790003, 0.30635141, 0.239168758, 0.163494838, 0.28582646]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "4f0feb7f375a6b9c4ffe789f77300870a6f96dc6706348a45f8dde7d5d07c968", "primary_control": "temperature", "result_hash": "557995260ab4e50599ef8609f47330575adad3c2afbac91457a73c0977386950", "signature": "30717eab4451e9695ced0f402b49304573551f99fa37354eb47faf3f8b2e1432af43e2ab991f392c4ff822445eb75c9c2f7719324a66c043fed2e98db14cfd01", "state": "VERIFIED", "summary": {"iw_temperature": 0.271779, "raw": 0.24894, "temperature": 0.271546}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:08:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:45Z", "cpu_seconds": 1.7673855820000028, "created_at": "2026-11-22T03:08:25Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-ab6af457a8e3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "02beac34091aea9902949f8e79634db1af40f311fb94656e9e02950af43b237b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-ab6af457a8e3", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:08:25Z", "seed": 440840936, "treatment": "iw_temperature"}, "priority": 0.19457357328444974, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "331b657523945a9eb81b45d3dccccfda144f2f24c4f46ea3da1d554620655df6", "run_seconds": 2.0456881523132324, "seed": 440840936, "started_at": "2026-11-22T03:09:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018872, 0.017606], "effect": -0.000633, "evaluated_at": "2026-11-22T03:09:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-ab6af457a8e3", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.25086296, 0.243716315, 0.286995132, 0.249902568, 0.271776235, 0.280756435, 0.223476032, 0.243713823, 0.295638043, 0.323243883, 0.235096515, 0.317194297, 0.217167895, 0.264244345, 0.254033405, 0.170766555], "raw": [0.178331195, 0.222618222, 0.290009348, 0.29622434, 0.290328237, 0.270054805, 0.218861133, 0.220145557, 0.272701815, 0.320032373, 0.2361497, 0.253994635, 0.160033412, 0.236081382, 0.1897183, 0.20346691], "temperature": [0.159372617, 0.22629359, 0.288159742, 0.29622434, 0.295321513, 0.293736688, 0.2605445, 0.21038636, 0.287146795, 0.32393717, 0.27583479, 0.26846598, 0.217167895, 0.264244345, 0.23356887, 0.218048668]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.4031, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.395", "prereg_hash": "02beac34091aea9902949f8e79634db1af40f311fb94656e9e02950af43b237b", "primary_control": "temperature", "result_hash": "331b657523945a9eb81b45d3dccccfda144f2f24c4f46ea3da1d554620655df6", "signature": "4a1c9800a1c1ee52d5c36429b577e9973ca20f81d32f4cfb9aa1f72f0e584b69b3411dd4a9ff2ddd4bac574d5c2c0503bc3586c84807686511e03a389c686606", "state": "VERIFIED", "summary": {"iw_temperature": 0.258037, "raw": 0.241172, "temperature": 0.257403}, "ties": 2, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:09:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:35Z", "cpu_seconds": 7.120371273000046, "created_at": "2026-11-22T03:08:25Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-5e6dc175ef88", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "275a992350b5e2efc894a3f176ba7dbbff6aa2f27ba3e0b2257ed92ea3b67ea1", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-5e6dc175ef88", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:08:25Z", "seed": 455754849, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "c3b54ead7de3534662fbe94fb31b29f7bfa7cf6eeccfca3664f39d4d9220f046", "run_seconds": 7.449267864227295, "seed": 455754849, "started_at": "2026-11-22T03:08:28Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:08:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-5e6dc175ef88", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "275a992350b5e2efc894a3f176ba7dbbff6aa2f27ba3e0b2257ed92ea3b67ea1", "primary_control": "reset", "result_hash": "c3b54ead7de3534662fbe94fb31b29f7bfa7cf6eeccfca3664f39d4d9220f046", "signature": "42ec75088e05d5b7305353075398e27824948c7d3b0e7feb38746b1a8fd38d5491612b4f6ed60c0975d26706ff130528cd19d940187247c273e667404080a20f", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:08:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:34Z", "cpu_seconds": 6.757996934000005, "created_at": "2026-11-22T03:08:25Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-e13203cb75db", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "b3ab65b422651b835314b3ddb48d5e4b96d81dc63266e8c575f6a10e18f1db0f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-e13203cb75db", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:08:25Z", "seed": 1082105662, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "72248b8181396d1a442386efe3aa1206f95678fdaee4ac5d4e3e1f2914d70584", "run_seconds": 7.272519111633301, "seed": 1082105662, "started_at": "2026-11-22T03:08:27Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:08:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-e13203cb75db", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b3ab65b422651b835314b3ddb48d5e4b96d81dc63266e8c575f6a10e18f1db0f", "primary_control": "reset", "result_hash": "72248b8181396d1a442386efe3aa1206f95678fdaee4ac5d4e3e1f2914d70584", "signature": "9868a36eb88941927205d085e0ed9b2cd00b4f99a66e02e18a8efe8274534ae2973f82e5dc6a17afbbcf8ba203e480dbda1da24a5d9924dd8c6e87a65b106501", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:08:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:50Z", "cpu_seconds": 1.1659556500000008, "created_at": "2026-11-22T03:08:25Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-835702a640ce", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "70fd4a80f08ed1f06cd56ee7727e59a76a28aecb631076900209c269cb196f0f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-835702a640ce", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:25Z", "seed": 847959585, "treatment": "frontier"}, "priority": 0.17290354436320587, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "3cb0c2517da68e3e23e427e25316556cb53421730b2d1b72f1c8bb3e95da1f2e", "run_seconds": 1.243877649307251, "seed": 847959585, "started_at": "2026-11-22T03:11:48Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:11:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-835702a640ce", "family": "world_model_frontier", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"curiosity": [1.0, 0.834514], "random_walk": [1.0, 0.858828]}, "prereg_hash": "70fd4a80f08ed1f06cd56ee7727e59a76a28aecb631076900209c269cb196f0f", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "3cb0c2517da68e3e23e427e25316556cb53421730b2d1b72f1c8bb3e95da1f2e", "signature": "d0ed8d96070716a2b9c0445b08a54763bcb9bb385c1cb7e8395c315a3953b5500edd10e8267a34a23c95f203fd286a1ee14c91641f90c4fc8013d1e55e7ec10e", "state": "REJECTED", "summary": {"curiosity": 0.834514, "frontier": 1.0, "random_walk": 0.858828}}, "verified_at": "2026-11-22T03:11:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:38Z", "cpu_seconds": 0.3084633189999977, "created_at": "2026-11-22T03:08:25Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-8d2af7e1bf22", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "f4752a6a2b879689db1e1aa3561ed9db860f65886e3c81009e632d37e2451c37", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-8d2af7e1bf22", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:25Z", "seed": 516478058, "treatment": "frontier"}, "priority": 0.23448965156755955, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-d9d5d5ce1fc3", "result_hash": "a573e16e6b5d6a0c15e4fba4bdb1713618bac8317c6ea62640867f0befe31c85", "run_seconds": 0.3416481018066406, "seed": 516478058, "started_at": "2026-11-22T03:08:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019515, 0.042871], "effect": 0.031193, "evaluated_at": "2026-11-22T03:08:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-8d2af7e1bf22", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.881355932, 0.897540984, 1.0, 0.908, 0.91796875, 0.889344262, 0.860849057, 0.927734375, 0.890243902, 0.893442623, 0.907258065, 0.906504065], "frontier": [0.913135593, 0.915983607, 1.0, 0.94, 0.94140625, 0.915983607, 0.926886792, 0.9609375, 0.922764228, 0.922131148, 0.933467742, 0.924796748], "random_walk": [0.88559322, 0.883196721, 1.0, 0.91, 0.919921875, 0.868852459, 0.856132075, 0.91796875, 0.884146341, 0.885245902, 0.899193548, 0.932926829]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.2321, "losses": 1, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00586", "prereg_hash": "f4752a6a2b879689db1e1aa3561ed9db860f65886e3c81009e632d37e2451c37", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-d9d5d5ce1fc3", "result_hash": "a573e16e6b5d6a0c15e4fba4bdb1713618bac8317c6ea62640867f0befe31c85", "signature": "f755a5459fce0392849f1ad6e2f18cc6c0da4ec8fc4c0164195eb9d9c1cf732a50a70a0cd9fe0a1b72684389622dbd82ce4062c5ab1456c375462e5362d35504", "state": "VERIFIED", "summary": {"curiosity": 0.906687, "frontier": 0.934791, "random_walk": 0.903598}, "ties": 1, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:08:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:45Z", "cpu_seconds": 1.7640420800000243, "created_at": "2026-11-22T03:08:41Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1e685ae26722", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "9bcd72d879c1db13c2ab5c5f4eacce406d47c5877b6b38bfed8f9eec13ce84f7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1e685ae26722", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:08:41Z", "seed": 523652809, "treatment": "iw_temperature"}, "priority": 0.3062570945906355, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "f524acc1f3a3cb28eff76845288ca24e50501bf0a1f6a0661167846f5bb943ea", "run_seconds": 1.8182411193847656, "seed": 523652809, "started_at": "2026-11-22T03:08:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.005108, 0.024882], "effect": 0.009887, "evaluated_at": "2026-11-22T03:08:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1e685ae26722", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.265553767, 0.288080095, 0.29692219, 0.275185785, 0.218390517, 0.162211838, 0.27979944, 0.26643599, 0.273661615, 0.18726773, 0.195040925, 0.242848395, 0.333552722, 0.32850109, 0.280089345, 0.264258132], "raw": [0.21015744, 0.2443455, 0.250590605, 0.273700537, 0.22925046, 0.242819105, 0.272187042, 0.24173806, 0.248103003, 0.221690705, 0.195040925, 0.256037518, 0.326999965, 0.318597447, 0.20431813, 0.254025455], "temperature": [0.230089857, 0.30686161, 0.250590605, 0.275185785, 0.266458123, 0.242819105, 0.288743285, 0.26159965, 0.272125933, 0.221690705, 0.232785278, 0.258186195, 0.33460314, 0.32235494, 0.280089345, 0.271805237]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6737, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.212", "prereg_hash": "9bcd72d879c1db13c2ab5c5f4eacce406d47c5877b6b38bfed8f9eec13ce84f7", "primary_control": "temperature", "result_hash": "f524acc1f3a3cb28eff76845288ca24e50501bf0a1f6a0661167846f5bb943ea", "signature": "bf50cb23ff6643440e53088d46e2a72ec3a63b6442a768bffbbfc32293fc9c5730d5402128d36e9b33ee33afd11ecf45ef87df13a4c50f0b9e310bcf2f5f1304", "state": "VERIFIED", "summary": {"iw_temperature": 0.259862, "raw": 0.24935, "temperature": 0.269749}, "ties": 2, "treatment": "iw_temperature", "wins": 9}, "verified_at": "2026-11-22T03:08:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:52Z", "cpu_seconds": 1.6396713249999948, "created_at": "2026-11-22T03:08:41Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-970714146c20", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "38ea40a26e7377ebc1458b312b10b1a57cc1587a072ebbb01ef0d2c4bfbf6ccc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-970714146c20", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:08:41Z", "seed": 1849643309, "treatment": "iw_temperature"}, "priority": 0.22552584180203106, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "54323c6c3e12be054938a29305af721fd5ab6d4b3fc646111d7949478059df23", "run_seconds": 1.8887646198272705, "seed": 1849643309, "started_at": "2026-11-22T03:08:50Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01093, 0.003778], "effect": -0.003576, "evaluated_at": "2026-11-22T03:08:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-970714146c20", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.231718545, 0.17215706, 0.293809422, 0.251689067, 0.289696203, 0.176475445, 0.269656613, 0.230975518, 0.298655895, 0.18905096, 0.220508185, 0.180368192, 0.224560125, 0.211124417, 0.241663262, 0.240883745], "raw": [0.202221285, 0.156547035, 0.240851237, 0.228565808, 0.24231132, 0.176475445, 0.25253974, 0.206267395, 0.270709875, 0.202458525, 0.190777863, 0.199959958, 0.26013306, 0.211124417, 0.194121195, 0.215427323], "temperature": [0.216483755, 0.17215706, 0.27706213, 0.251689067, 0.289696203, 0.176475445, 0.269656613, 0.248301488, 0.298655895, 0.202458525, 0.20652544, 0.15929381, 0.22933044, 0.17454233, 0.263400317, 0.230045557]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0819, "losses": 6, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.828", "prereg_hash": "38ea40a26e7377ebc1458b312b10b1a57cc1587a072ebbb01ef0d2c4bfbf6ccc", "primary_control": "temperature", "result_hash": "54323c6c3e12be054938a29305af721fd5ab6d4b3fc646111d7949478059df23", "signature": "b4ec6243b5fa721077d07a6f3e2a8b739df6a2790c7b82629a838634392b3fccc6c5ec03a493aed20d2fe79616c7381f6a0c18b6530d27775be4bc6effbc690d", "state": "VERIFIED", "summary": {"iw_temperature": 0.232687, "raw": 0.215656, "temperature": 0.229111}, "ties": 6, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:08:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:08:51Z", "cpu_seconds": 3.87727675900004, "created_at": "2026-11-22T03:08:41Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-55c53a89425f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "f458bd2d31b62be0c4c801107ecdf0e708d297f1e8e89033ef04d983d34e2c38", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-55c53a89425f", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:41Z", "seed": 2095358422, "treatment": "frontier"}, "priority": 0.258653287859763, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "427ed288a75e642afa678e2af05fa145555adce0c72290abb35c2688c1fd51be", "run_seconds": 4.0289506912231445, "seed": 2095358422, "started_at": "2026-11-22T03:08:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.055719, 0.125354], "effect": 0.090537, "evaluated_at": "2026-11-22T03:08:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-55c53a89425f", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.848901099, 0.978571429, 0.890776699, 0.778409091, 0.9, 0.799418605, 0.838942308, 1.0, 0.76344086, 1.0, 0.8175, 0.872093023], "frontier": [1.0, 0.995238095, 1.0, 1.0, 1.0, 0.985465116, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.906593407, 0.916666667, 0.912621359, 0.761363636, 0.92, 0.845930233, 0.923076923, 1.0, 0.89516129, 1.0, 0.9175, 0.895348837]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "f458bd2d31b62be0c4c801107ecdf0e708d297f1e8e89033ef04d983d34e2c38", "primary_control": "random_walk", "result_hash": "427ed288a75e642afa678e2af05fa145555adce0c72290abb35c2688c1fd51be", "signature": "e67bab6e958afed61c02f1343311b147c5b818aa608f17346fbe150a78a6544848c099e16408d9befa2d1a24e4a69dccdec4617e2c4e591292fa4c4a1ac5bb01", "state": "VERIFIED", "summary": {"curiosity": 0.874004, "frontier": 0.998392, "random_walk": 0.907855}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:08:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:11Z", "cpu_seconds": 1.4951342530000034, "created_at": "2026-11-22T03:08:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-d65853b2b018", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "730a3915ddc9cf76781465da79907934b6a6039255b88ccbccc49819b0843f98", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-d65853b2b018", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:08:58Z", "seed": 1269125286, "treatment": "iw_temperature"}, "priority": 0.22362760905197865, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "cbe59085dec8d76a8a10497af4358b28541277e19db97838916ae1461d6576f0", "run_seconds": 1.5039424896240234, "seed": 1269125286, "started_at": "2026-11-22T03:09:09Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.013171, 0.010609], "effect": -0.001281, "evaluated_at": "2026-11-22T03:09:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-d65853b2b018", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.212792705, 0.184393965, 0.216758233, 0.23617274, 0.121264662, 0.091106165, 0.264245335, 0.17570096, 0.130933315, 0.199065175, 0.222335133, 0.262551155, 0.108536748, 0.236932428, 0.113670812, 0.254249205], "raw": [0.20432319, 0.17089242, 0.200156785, 0.23617274, 0.17073489, 0.204180997, 0.253149283, 0.18948137, 0.151764363, 0.14356696, 0.25113155, 0.226384062, 0.174870157, 0.154676645, 0.113670812, 0.205068238], "temperature": [0.192240418, 0.184393965, 0.204778465, 0.23617274, 0.137371502, 0.153066028, 0.253149283, 0.18948137, 0.129079665, 0.17187353, 0.222335133, 0.262551155, 0.132264605, 0.188422323, 0.098785417, 0.254249205]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0522, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.887", "prereg_hash": "730a3915ddc9cf76781465da79907934b6a6039255b88ccbccc49819b0843f98", "primary_control": "temperature", "result_hash": "cbe59085dec8d76a8a10497af4358b28541277e19db97838916ae1461d6576f0", "signature": "58ca9d233f34443589ae6bfc013c03e0584ddcf5401e66b5892aee739328f925b81a86d1e92863b9c7a222f22f01224757cba456a0c1b7797e80bb5474ba200a", "state": "VERIFIED", "summary": {"iw_temperature": 0.189419, "raw": 0.190639, "temperature": 0.188138}, "ties": 5, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:09:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:09Z", "cpu_seconds": 5.999580432000016, "created_at": "2026-11-22T03:08:58Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3a90ce406115", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7b5bf0042e09880b21eb442c253d14ef901a54299da3b249c28ed90ecab87cae", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3a90ce406115", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:08:58Z", "seed": 1216739911, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "eb6f42da23828f41abee02595bf7288fcd93cd956e437ef8ce278b7c7f73803c", "run_seconds": 6.415963649749756, "seed": 1216739911, "started_at": "2026-11-22T03:09:03Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.126863, 0.160196], "effect": 0.016667, "evaluated_at": "2026-11-22T03:09:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3a90ce406115", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.301, "losses": 9, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7b5bf0042e09880b21eb442c253d14ef901a54299da3b249c28ed90ecab87cae", "primary_control": "reset", "result_hash": "eb6f42da23828f41abee02595bf7288fcd93cd956e437ef8ce278b7c7f73803c", "signature": "1d9bbfc988963ab3e20fc583bb3307a6f184e024ca8387b4c3953eb84d77728a3893dbaefe49c8b6e14e0c8a93dd4939586e1b08a36b8cf6195a82c63f5ee101", "state": "VERIFIED", "summary": {"inherit": 0.5833333333333334, "reset": 0.5666666666666667, "root": 0.3333333333333333}, "ties": 41, "treatment": "inherit", "wins": 10}, "verified_at": "2026-11-22T03:09:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:08Z", "cpu_seconds": 6.088916288000007, "created_at": "2026-11-22T03:08:58Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-76b8d6e40c87", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e03d0946c4774353ebb1b574dd823867c3e3b83a2a1fb0e1c0fc152cb0d1f88d", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-76b8d6e40c87", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:08:58Z", "seed": 1215417957, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "8ec481001ef540e4983a8dd8e50f5d88f116ce3102d4548461fb9e7b3c92a9e2", "run_seconds": 6.50330662727356, "seed": 1215417957, "started_at": "2026-11-22T03:09:01Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:09:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-76b8d6e40c87", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5007974493672064, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e03d0946c4774353ebb1b574dd823867c3e3b83a2a1fb0e1c0fc152cb0d1f88d", "primary_control": "reset", "result_hash": "8ec481001ef540e4983a8dd8e50f5d88f116ce3102d4548461fb9e7b3c92a9e2", "signature": "90bc547d6a90e4db6032f75a163d65503bd31e686780d3592a8fe37ea1e2e0fa16b96c0152c0926a5e786032eeb1c2ba821b02351f69cae2e80e9896928b3e0c", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.43333333333333335}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:09:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:04Z", "cpu_seconds": 0.37185456200001, "created_at": "2026-11-22T03:08:58Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-4f680e299d05", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "50b643ce04ac958b99467a7c9a748ba5e39e12862aff9ffa4ed47f7be93f0b43", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-4f680e299d05", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:58Z", "seed": 1004450726, "treatment": "frontier"}, "priority": 0.3362488982057387, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-f2b0bc37a614", "result_hash": "dbcd5cc41d9cc29cee43967883be6470baef6853ec5db27f73e9ecd300c53217", "run_seconds": 0.47971653938293457, "seed": 1004450726, "started_at": "2026-11-22T03:09:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.054893, 0.104005], "effect": 0.079449, "evaluated_at": "2026-11-22T03:09:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-4f680e299d05", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.966666667, 0.866666667, 0.793367347, 0.755376344, 0.786111111, 0.801630435, 0.788265306, 1.0, 0.797619048, 0.819306931, 1.0, 0.784210526], "frontier": [1.0, 0.916666667, 0.882653061, 0.887096774, 0.911111111, 0.921195652, 0.885204082, 1.0, 0.89047619, 0.905940594, 1.0, 0.873684211], "random_walk": [0.95, 0.845238095, 0.795918367, 0.774193548, 0.794444444, 0.788043478, 0.783163265, 1.0, 0.802380952, 0.826732673, 1.0, 0.760526316]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "50b643ce04ac958b99467a7c9a748ba5e39e12862aff9ffa4ed47f7be93f0b43", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-f2b0bc37a614", "result_hash": "dbcd5cc41d9cc29cee43967883be6470baef6853ec5db27f73e9ecd300c53217", "signature": "1700a76fba9c2e776885186a6de12a235c5143e7c9aa0762e63d841c33c70e9353790a6f79fd2f62fa55c0fed3e1b90b6f5f0cdc10a9ab67d21d10721c4fac0b", "state": "VERIFIED", "summary": {"curiosity": 0.846602, "frontier": 0.922836, "random_walk": 0.843387}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:09:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:09Z", "cpu_seconds": 0.5247524039999973, "created_at": "2026-11-22T03:08:58Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-5958bc5060bf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "1d96da42d331daf4f41e9e1aa38012cf343d43baab43048f63fe467025c893aa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-5958bc5060bf", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:58Z", "seed": 1815209673, "treatment": "frontier"}, "priority": 0.17750764777487646, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "690b7d43d0e35dcbf56ad56fa397b1c226352216fdcbc04a821ff9f18542f421", "run_seconds": 0.5256385803222656, "seed": 1815209673, "started_at": "2026-11-22T03:11:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.045035, 0.083817], "effect": 0.064426, "evaluated_at": "2026-11-22T03:11:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-5958bc5060bf", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.910714286, 0.910714286, 0.867647059, 0.903225806, 0.899590164, 0.909448819, 0.906504065, 0.972426471, 0.908536585, 0.951923077, 0.848684211], "frontier": [1.0, 0.994047619, 0.988095238, 0.995798319, 0.987903226, 0.99795082, 0.998031496, 0.989837398, 0.994485294, 0.989837398, 1.0, 1.0], "random_walk": [1.0, 0.908730159, 0.928571429, 0.884453782, 0.921370968, 0.922131148, 0.948818898, 0.918699187, 0.974264706, 0.924796748, 0.953846154, 0.877192982]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "1d96da42d331daf4f41e9e1aa38012cf343d43baab43048f63fe467025c893aa", "primary_control": "random_walk", "result_hash": "690b7d43d0e35dcbf56ad56fa397b1c226352216fdcbc04a821ff9f18542f421", "signature": "8ee7d5389cdd38b0d4338812bc9191c63fc1139d471a5cc9ee728fd0073479bffb10a0681df57efe83ab763527e6853f08744a94cd25e6c2377deb7abd0fc006", "state": "VERIFIED", "summary": {"curiosity": 0.915785, "frontier": 0.994666, "random_walk": 0.93024}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:11:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:56Z", "cpu_seconds": 0.5912822370000015, "created_at": "2026-11-22T03:08:58Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-7b25827eb85e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "4b5b75eadcf9c18a5aec4faf43970177eec912f08b53e675c24273268ad4cba7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-7b25827eb85e", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:58Z", "seed": 576059447, "treatment": "frontier"}, "priority": 0.16721994931093812, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "580c7a5d0eb2ddb62502abdc54666d5478bb27fca8a14d90119628dca3800012", "run_seconds": 0.655768632888794, "seed": 576059447, "started_at": "2026-11-22T03:12:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.066606, 0.094221], "effect": 0.080413, "evaluated_at": "2026-11-22T03:13:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-7b25827eb85e", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.902777778, 0.871794872, 0.911290323, 0.916666667, 0.915983607, 0.889583333, 0.904, 0.914, 0.881696429, 0.923828125, 0.880630631, 0.945121951], "frontier": [1.0, 0.982905983, 0.991935484, 0.996031746, 0.991803279, 0.9875, 0.992, 0.988, 1.0, 0.998046875, 0.997747748, 0.995934959], "random_walk": [0.895833333, 0.861111111, 0.933467742, 0.930555556, 0.932377049, 0.902083333, 0.91, 0.942, 0.883928571, 0.939453125, 0.905405405, 0.920731707]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "4b5b75eadcf9c18a5aec4faf43970177eec912f08b53e675c24273268ad4cba7", "primary_control": "random_walk", "result_hash": "580c7a5d0eb2ddb62502abdc54666d5478bb27fca8a14d90119628dca3800012", "signature": "144e2032cec5f0c2637c7b0d40dbbca6b65bf03540e898b2d9a35a098325f413a1a3351dab67b3b5f25eb760fd81ea9e7b9e0081bea4e52b3534fd6b2b568900", "state": "VERIFIED", "summary": {"curiosity": 0.904781, "frontier": 0.993492, "random_walk": 0.913079}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:13:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:28Z", "cpu_seconds": 3.2201300710000282, "created_at": "2026-11-22T03:08:58Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-fc1555ccd016", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "99087df0a0b21eca95672567ff3e380ce3f460bf4c47b5a3653cb28c4d550d89", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-fc1555ccd016", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:08:58Z", "seed": 1015444324, "treatment": "frontier"}, "priority": 0.1753955127546995, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-bdaf69650c7d", "result_hash": "0eb17f4525407c7fe9c78214149ee31600ff9b69cb79330e037fcd2d19fe6dd6", "run_seconds": 3.4271652698516846, "seed": 1015444324, "started_at": "2026-11-22T03:11:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.053805, 0.127757], "effect": 0.090781, "evaluated_at": "2026-11-22T03:11:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-fc1555ccd016", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.774456522, 0.808962264, 0.763020833, 1.0, 1.0, 0.780898876, 0.90625, 0.766666667, 0.804123711, 0.852941176, 0.849537037], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9875, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.891304348, 0.900943396, 0.817708333, 1.0, 1.0, 0.865168539, 0.946875, 0.827777778, 0.87628866, 0.897058824, 0.875]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "99087df0a0b21eca95672567ff3e380ce3f460bf4c47b5a3653cb28c4d550d89", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-bdaf69650c7d", "result_hash": "0eb17f4525407c7fe9c78214149ee31600ff9b69cb79330e037fcd2d19fe6dd6", "signature": "acb1e21b1703e1b307c47df3c2cdb9ba6076fa3e02a58862e8dac243d488eca8db51b741a1cd4145ebc55cf82c0425f533b4e2e86be9da59540c8de6f2ffe608", "state": "VERIFIED", "summary": {"curiosity": 0.858905, "frontier": 0.998958, "random_walk": 0.908177}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:11:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:18Z", "cpu_seconds": 1.6587920399999803, "created_at": "2026-11-22T03:09:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5eb1d85be89f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "1e61c3808777e84fee773cd295abb88323c257285002095c558b57ac3faa60cd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5eb1d85be89f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:14Z", "seed": 542037084, "treatment": "iw_temperature"}, "priority": 0.3164673721559223, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "31f574416d1cd1d3a72d09645f810d2b91c0c18d901f01b8172b8c54e9357187", "run_seconds": 1.6788337230682373, "seed": 542037084, "started_at": "2026-11-22T03:09:16Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021157, 0.014232], "effect": -0.003462, "evaluated_at": "2026-11-22T03:09:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5eb1d85be89f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.194565593, 0.237121985, 0.3013162, 0.283375495, 0.300674973, 0.257225678, 0.21942693, 0.29389076, 0.326680665, 0.311557508, 0.157608705, 0.318615148, 0.286409535, 0.21597177, 0.217859915, 0.26777144], "raw": [0.152999862, 0.211748217, 0.290463622, 0.238857005, 0.270109252, 0.225952765, 0.190218477, 0.209190638, 0.31859575, 0.282684892, 0.157608705, 0.311700057, 0.262807085, 0.18321594, 0.292045907, 0.238611308], "temperature": [0.183204302, 0.211748217, 0.309548518, 0.246571718, 0.298460312, 0.248211265, 0.173543288, 0.281081572, 0.325061068, 0.282684892, 0.262126788, 0.326725733, 0.281989878, 0.18321594, 0.264232348, 0.25627041]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0046, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "1e61c3808777e84fee773cd295abb88323c257285002095c558b57ac3faa60cd", "primary_control": "temperature", "result_hash": "31f574416d1cd1d3a72d09645f810d2b91c0c18d901f01b8172b8c54e9357187", "signature": "f466543b31d5ab96791c5a0df324f39a4e2bbdeeadbaff6f35b4339cd9d11140b5d0521e0c1a0aa9e81c936649dd8abc0623ca3079aba0b158658162695a3105", "state": "VERIFIED", "summary": {"iw_temperature": 0.26188, "raw": 0.239801, "temperature": 0.258417}, "ties": 0, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:09:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:10Z", "cpu_seconds": 1.796915184999989, "created_at": "2026-11-22T03:09:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-a4fc35b19f4a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "79efee52923800053cbce54d2428b326d3ee633d8f0cd4b0d0bb6161fdd65618", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-a4fc35b19f4a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:14Z", "seed": 1768040275, "treatment": "iw_temperature"}, "priority": 0.1712139475715177, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "16be93053c3711bd933d55bd632c101efdbf5d2227847777762add6306dd9997", "run_seconds": 2.2543461322784424, "seed": 1768040275, "started_at": "2026-11-22T03:12:07Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.026707, 0.001253], "effect": -0.012727, "evaluated_at": "2026-11-22T03:12:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-a4fc35b19f4a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.113992813, 0.140616073, 0.160772875, 0.129032765, 0.095377422, 0.187994182, 0.11318228, 0.148156475, 0.201287985, 0.104877798, 0.16292583, 0.117341462, 0.132482315, 0.088753945, 0.207187898, 0.158964413], "raw": [0.110349532, 0.11948289, 0.10407845, 0.095830758, 0.1149024, 0.133934812, 0.093857435, 0.117829302, 0.1636319, 0.073759813, 0.114313695, 0.140632003, 0.087735373, 0.091624783, 0.127997, 0.118052005], "temperature": [0.133458695, 0.140616073, 0.118147645, 0.1071097, 0.140988378, 0.168314635, 0.11318228, 0.120825725, 0.153509328, 0.073759813, 0.15130119, 0.140632003, 0.074042212, 0.105547025, 0.175320885, 0.142556688]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "79efee52923800053cbce54d2428b326d3ee633d8f0cd4b0d0bb6161fdd65618", "primary_control": "temperature", "result_hash": "16be93053c3711bd933d55bd632c101efdbf5d2227847777762add6306dd9997", "signature": "c8f3632014af467180c3a450fdc54b50f386fb8e99ae114be9fd8d731d719c51682c62f7945975ad21f5b022f1cfd960049c3c6ac006fbf6aa39d2c62f88950f", "state": "VERIFIED", "summary": {"iw_temperature": 0.141434, "raw": 0.113001, "temperature": 0.128707}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:12:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:43Z", "cpu_seconds": 1.1320614919999912, "created_at": "2026-11-22T03:09:14Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-b8a862eac292", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "761eb5c93683a41b7c780ea3a3a78284f9409f4e5651bcab2090c34fb59e34cc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-b8a862eac292", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:09:14Z", "seed": 225321354, "treatment": "frontier"}, "priority": 0.16774499947254187, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "68415f68203e57f5662df5fdc215cc2ae2ff64b6a69c2352fdd5d432ead8a23c", "run_seconds": 1.1731491088867188, "seed": 225321354, "started_at": "2026-11-22T03:12:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.040203, 0.098908], "effect": 0.069556, "evaluated_at": "2026-11-22T03:12:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-b8a862eac292", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.912698413, 0.94, 0.873015873, 0.88125, 1.0, 0.87295082, 1.0, 0.855263158, 0.858490566, 1.0, 0.851851852, 0.836538462], "frontier": [0.992063492, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.98245614, 1.0, 1.0, 0.944444444, 1.0], "random_walk": [0.884920635, 0.98, 0.888888889, 0.9125, 1.0, 0.905737705, 1.0, 0.855263158, 0.886792453, 0.983333333, 0.916666667, 0.870192308]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "761eb5c93683a41b7c780ea3a3a78284f9409f4e5651bcab2090c34fb59e34cc", "primary_control": "random_walk", "result_hash": "68415f68203e57f5662df5fdc215cc2ae2ff64b6a69c2352fdd5d432ead8a23c", "signature": "5bea59e4098dea52bcc3b17cf40614c497e941ee622f3541e1879d1897a2a69b1bc073d2dd1134c03a52c7193f5ce2ef0ee13a6a04ea71ad4ce4b640a2dac40c", "state": "VERIFIED", "summary": {"curiosity": 0.906838, "frontier": 0.993247, "random_walk": 0.923691}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:12:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:41Z", "cpu_seconds": 1.8867330190000189, "created_at": "2026-11-22T03:09:30Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-53ee9814c04b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "878b11e42ed8ba4d0e8d1273ff428f81315814a37a4438fd7987fbbee9d7ad9f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-53ee9814c04b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:30Z", "seed": 102615799, "treatment": "iw_temperature"}, "priority": 0.1729569452563281, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "2a3a4568e444c891e52c8bfebaec1c3770c2fa87acef24e7989939aee6f5c982", "run_seconds": 1.9202826023101807, "seed": 102615799, "started_at": "2026-11-22T03:11:39Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.006515, 0.001038], "effect": -0.002739, "evaluated_at": "2026-11-22T03:11:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-53ee9814c04b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.043029555, 0.0592945, 0.057192327, 0.024554465, 0.058833715, 0.086794312, 0.056786217, 0.011533757, 0.015716322, 0.054276505, 0.05212223, 0.072690968, 0.055967612, 0.058167645, 0.044538525, 0.032809867], "raw": [0.054125653, 0.044215182, 0.057192327, 0.039528822, 0.06948149, 0.076186338, 0.060861825, 0.04517781, 0.045433695, 0.039011967, 0.044369793, 0.064375538, 0.0693474, 0.078314102, 0.047852693, 0.059149535], "temperature": [0.041666985, 0.049171938, 0.062016313, 0.02937131, 0.060514635, 0.07246783, 0.057406458, 0.021934165, 0.015716322, 0.041175452, 0.04049707, 0.072690968, 0.045718067, 0.063143038, 0.044538525, 0.022461617]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "878b11e42ed8ba4d0e8d1273ff428f81315814a37a4438fd7987fbbee9d7ad9f", "primary_control": "temperature", "result_hash": "2a3a4568e444c891e52c8bfebaec1c3770c2fa87acef24e7989939aee6f5c982", "signature": "d5d0001e19a735602357dd842c651d94463478cb794c9a9df0f6c72cb832a0f46200ae3c8e1da43811132be9197aa86a15623143e91149df0d925417e5e6b00b", "state": "VERIFIED", "summary": {"iw_temperature": 0.049019, "raw": 0.055914, "temperature": 0.046281}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:11:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:03Z", "cpu_seconds": 1.5016975260000152, "created_at": "2026-11-22T03:09:30Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-d4afc30e8b81", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "6ee1c5eaa20f34f4b037613fb5e8e299bbb6d5db7d9ecafd3b8c14eceb46c984", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-d4afc30e8b81", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:30Z", "seed": 1254099119, "treatment": "iw_temperature"}, "priority": 0.17855507801270848, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "f86a7940a3ed74b13250c6976c9734dbdeb62e3ece5645775c2ba3c85d537d14", "run_seconds": 1.5894994735717773, "seed": 1254099119, "started_at": "2026-11-22T03:11:02Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017402, 0.000875], "effect": -0.008263, "evaluated_at": "2026-11-22T03:11:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-d4afc30e8b81", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.26788334, 0.220556442, 0.210724175, 0.17818735, 0.212922323, 0.282639295, 0.237603513, 0.22021868, 0.237314822, 0.149767842, 0.274161692, 0.287097492, 0.227806115, 0.193009358, 0.270190887, 0.225688487], "raw": [0.20788881, 0.182322687, 0.13423645, 0.152416183, 0.20548923, 0.228848797, 0.21201828, 0.20680598, 0.149753975, 0.138630753, 0.2531877, 0.268400862, 0.21558193, 0.2067444, 0.207442775, 0.225688487], "temperature": [0.293391927, 0.182322687, 0.188555102, 0.17818735, 0.224546342, 0.261772372, 0.237603513, 0.22021868, 0.217444868, 0.178994887, 0.2531877, 0.279188932, 0.21558193, 0.163380763, 0.256912737, 0.212266888]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "6ee1c5eaa20f34f4b037613fb5e8e299bbb6d5db7d9ecafd3b8c14eceb46c984", "primary_control": "temperature", "result_hash": "f86a7940a3ed74b13250c6976c9734dbdeb62e3ece5645775c2ba3c85d537d14", "signature": "b13e97f5d0e8f0ac10343f400c3d08527f373b682c772dd356c1fcd06b7c3f9ae63dc9fe239131cbc46d06fa4df1533a81ad62484116ce00de0a4f3a92bd2f0a", "state": "VERIFIED", "summary": {"iw_temperature": 0.230986, "raw": 0.199716, "temperature": 0.222722}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:11:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:57Z", "cpu_seconds": 1.5731778290000022, "created_at": "2026-11-22T03:09:30Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-e8371de1db16", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "567b6a3da850ed8a6abf6350676399c12496026417afb2855b336a64d62e7128", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-e8371de1db16", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:30Z", "seed": 1561765758, "treatment": "iw_temperature"}, "priority": 0.17236214024358415, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "f65d45f28146340db443dd423c42fbf52c7601bb15f15800f121318298f37030", "run_seconds": 1.9408752918243408, "seed": 1561765758, "started_at": "2026-11-22T03:11:55Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010604, 0.027364], "effect": 0.00838, "evaluated_at": "2026-11-22T03:12:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-e8371de1db16", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.236421743, 0.14621469, 0.14871965, 0.253703793, 0.200961627, 0.248673575, 0.286880378, 0.084015242, 0.298282808, 0.26236567, 0.194253367, 0.25567941, 0.168339697, 0.213188427, 0.242232625, 0.251851875], "raw": [0.185170843, 0.234244092, 0.18102618, 0.220066005, 0.240456123, 0.248673575, 0.242178225, 0.133566287, 0.249835427, 0.205517443, 0.238604305, 0.194083615, 0.230111845, 0.22297784, 0.242232625, 0.19376743], "temperature": [0.19552151, 0.2201984, 0.193735625, 0.244495645, 0.240456123, 0.248673575, 0.25084352, 0.09337624, 0.280594873, 0.222945832, 0.260628507, 0.293072495, 0.208876135, 0.23420898, 0.206597882, 0.231638473]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "567b6a3da850ed8a6abf6350676399c12496026417afb2855b336a64d62e7128", "primary_control": "temperature", "result_hash": "f65d45f28146340db443dd423c42fbf52c7601bb15f15800f121318298f37030", "signature": "953fb0ff5ee7930e08fea59ef9a4e308513a35675593fbc9e4f53f4ffcdce5a24657f069f20b5997bd5174d8e82c36a62831eef422099eeeb9bb35c091fb8107", "state": "VERIFIED", "summary": {"iw_temperature": 0.218237, "raw": 0.216407, "temperature": 0.226616}, "ties": 1, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:12:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:03Z", "cpu_seconds": 1.8619331689999998, "created_at": "2026-11-22T03:09:30Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-ef4eb75de293", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "773357cd10610e60cc66a159a9b0b50d9b01df14a846d0bdb57e054924a37ad9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-ef4eb75de293", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:30Z", "seed": 1781862418, "treatment": "iw_temperature"}, "priority": 0.16213189220758947, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "1d9766ec8528026307c65afa800f1be1b730fec271d980656d228894919e99a0", "run_seconds": 3.1528522968292236, "seed": 1781862418, "started_at": "2026-11-22T03:13:59Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.0196, 0.01909], "effect": -0.000255, "evaluated_at": "2026-11-22T03:14:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-ef4eb75de293", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.067664865, 0.108347595, 0.112109682, 0.083813495, 0.174635257, 0.11683426, 0.14578034, 0.161707347, 0.179467052, 0.17751509, 0.134066833, 0.141342257, 0.09642172, 0.14721615, 0.09658028, 0.147998675], "raw": [0.087107407, 0.112124965, 0.127100915, 0.140306032, 0.063372762, 0.105643867, 0.122392565, 0.130502267, 0.160771122, 0.08199863, 0.115573407, 0.124082918, 0.112942163, 0.145532498, 0.103758343, 0.13984206], "temperature": [0.099970507, 0.126990057, 0.160650267, 0.140306032, 0.10635481, 0.08873919, 0.14578034, 0.149105812, 0.175156867, 0.08199863, 0.129776875, 0.131922363, 0.123395673, 0.15433798, 0.12493741, 0.147998675]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2184, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.605", "prereg_hash": "773357cd10610e60cc66a159a9b0b50d9b01df14a846d0bdb57e054924a37ad9", "primary_control": "temperature", "result_hash": "1d9766ec8528026307c65afa800f1be1b730fec271d980656d228894919e99a0", "signature": "97ff68123775a45eec89c3975ec48395f7e857647193971668b951b421da43a0876d9d5351f34e7bb809b391c477efc373b65c814f97fb5f9ba254f759b8a003", "state": "VERIFIED", "summary": {"iw_temperature": 0.130719, "raw": 0.117066, "temperature": 0.130464}, "ties": 2, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:14:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:36Z", "cpu_seconds": 5.058049007000022, "created_at": "2026-11-22T03:09:30Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-8684efe952b1", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "388783811898b1d8594a1518e96c60e4aebfbc07ce9732c2187558ef69a71a6f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-8684efe952b1", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:09:30Z", "seed": 1410986044, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "238cf81006e70fa7fb7e9b639c922694765da3c29995c8277872d095cf7e0573", "run_seconds": 5.094242095947266, "seed": 1410986044, "started_at": "2026-11-22T03:09:31Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:09:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-8684efe952b1", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "388783811898b1d8594a1518e96c60e4aebfbc07ce9732c2187558ef69a71a6f", "primary_control": "reset", "result_hash": "238cf81006e70fa7fb7e9b639c922694765da3c29995c8277872d095cf7e0573", "signature": "06b05d1e2a9c2f85fd541d7317cf97fa9f294eae4d6d48f539c1a3c129e58af37f2725eefcf0589ad7046c0ad1c139db1ee86e5959144eb13cda2bc9211c530b", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.25}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:09:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:43Z", "cpu_seconds": 8.480682211999977, "created_at": "2026-11-22T03:09:30Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a7975cbff1b5", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ff2bc0257e2564364a2541ba1f456254436f69f88669c147caab628f0164832e", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a7975cbff1b5", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:09:30Z", "seed": 1939977394, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "b3a3ed6bdfd527761f6d3cf452679bb1159637dc6ca15675ed9e0fd57e548f0f", "run_seconds": 8.820903301239014, "seed": 1939977394, "started_at": "2026-11-22T03:09:34Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:09:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a7975cbff1b5", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0], "reset": [0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ff2bc0257e2564364a2541ba1f456254436f69f88669c147caab628f0164832e", "primary_control": "reset", "result_hash": "b3a3ed6bdfd527761f6d3cf452679bb1159637dc6ca15675ed9e0fd57e548f0f", "signature": "4ed83b77c58178cf7b2670552dcf824b2817cf2601c369064a04b3340b4fb7323fd2ed5cda21d383d7ff9d282630df90009a6a429746a1a12ee5b514caed0400", "state": "PARTIAL", "summary": {"inherit": 0.5666666666666667, "reset": 0.5666666666666667, "root": 0.2833333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:09:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:30Z", "cpu_seconds": 0.9855639639999936, "created_at": "2026-11-22T03:09:30Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-2ee4f8efe9e1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "fdf22035d3899f7ccdf187f52f61dff95e5b47a5e3b9c64636a440981586e929", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-2ee4f8efe9e1", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:09:30Z", "seed": 20244597, "treatment": "frontier"}, "priority": 0.16951567059166744, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "82f3dcc18ee518ebb52187d40a14953aefb0ff5336778ae4a8953cd4325dcb9a", "run_seconds": 0.9884917736053467, "seed": 20244597, "started_at": "2026-11-22T03:12:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.088241, 0.18172], "effect": 0.13498, "evaluated_at": "2026-11-22T03:12:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-2ee4f8efe9e1", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.71875, 0.803125, 0.789156627, 0.801020408, 0.79245283, 1.0, 0.788265306, 0.825657895, 0.732142857, 1.0, 0.821428571, 0.846590909], "frontier": [1.0, 0.9625, 1.0, 1.0, 0.992924528, 1.0, 1.0, 1.0, 0.842261905, 1.0, 1.0, 0.96875], "random_walk": [0.710227273, 0.790625, 0.888554217, 0.823979592, 0.79009434, 1.0, 0.831632653, 0.865131579, 0.776785714, 1.0, 0.857142857, 0.8125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "fdf22035d3899f7ccdf187f52f61dff95e5b47a5e3b9c64636a440981586e929", "primary_control": "random_walk", "result_hash": "82f3dcc18ee518ebb52187d40a14953aefb0ff5336778ae4a8953cd4325dcb9a", "signature": "3a222a37ee7a2faccaf73dd2d5b03f6aa09ba7d406cb8a2d0a5ed58b8c5e911d3dff67f53c1606cb653d649fca3a1c358f83d162f2ad883341c1dc0f04894403", "state": "VERIFIED", "summary": {"curiosity": 0.826549, "frontier": 0.980536, "random_walk": 0.845556}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:12:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:59Z", "cpu_seconds": 1.5040628710000021, "created_at": "2026-11-22T03:09:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b39cf5d45909", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "137058952b0ee2756e36e0accab95d40e779e0681a24cbda98a9627e7866c2a0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b39cf5d45909", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:47Z", "seed": 1515573197, "treatment": "iw_temperature"}, "priority": 0.19457954736869218, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "86325f1523c0d0c579577e175cc93dd1fa8db3b525822344431765eee1eccddd", "run_seconds": 1.5112154483795166, "seed": 1515573197, "started_at": "2026-11-22T03:09:57Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015602, 0.000838], "effect": -0.007382, "evaluated_at": "2026-11-22T03:10:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b39cf5d45909", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.08212724, 0.044168425, 0.026187992, 0.062405382, 0.026193172, 0.13053873, 0.009544693, 0.01738887, 0.006304827, 0.084900655, 0.001804765, 0.127548833, 0.118688595, 0.033841348, 0.00552549, 0.062143982], "raw": [0.035415123, 0.04801476, 0.040167473, 0.080835617, 0.034993168, 0.13053873, 0.017667697, 0.022150843, 0.005218322, 0.090510328, 0.00670879, 0.069537915, 0.106550458, 0.03359706, 0.008240758, 0.029462092], "temperature": [0.03514698, 0.057801307, 0.026187992, 0.053027517, 0.01642125, 0.10992508, 0.009528068, 0.01738887, 0.004774852, 0.084900655, 0.00336915, 0.0891122, 0.118688595, 0.04769688, 0.00552549, 0.041711983]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "137058952b0ee2756e36e0accab95d40e779e0681a24cbda98a9627e7866c2a0", "primary_control": "temperature", "result_hash": "86325f1523c0d0c579577e175cc93dd1fa8db3b525822344431765eee1eccddd", "signature": "bdb3e35a4a8f5ac2940e79f0d80fb89f8c1b0367c5bb8d6e9af26903f5ed5d7d22015c3b4eb373deb98b3d962d19f09b7843dc5786f72ffdee3cb98e76ba8c0a", "state": "VERIFIED", "summary": {"iw_temperature": 0.052457, "raw": 0.047476, "temperature": 0.045075}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:10:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:52Z", "cpu_seconds": 1.875077043999994, "created_at": "2026-11-22T03:09:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-fcb9f51549e4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "09d7bb74c1479026a8c89a6a04f9b4c103c858f539bddd950202de7eecb858b8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-fcb9f51549e4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:09:47Z", "seed": 1158418408, "treatment": "iw_temperature"}, "priority": 0.31163855433612486, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "5c3a0460c44518aacd16f4c2b42a87a4fa1e765601fa0df8d5df90dc3c89b971", "run_seconds": 2.253997802734375, "seed": 1158418408, "started_at": "2026-11-22T03:09:50Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019425, 0.003366], "effect": -0.00803, "evaluated_at": "2026-11-22T03:09:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-fcb9f51549e4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.26504448, 0.35038484, 0.324698445, 0.197763858, 0.213028768, 0.327030097, 0.312549368, 0.226273923, 0.340843848, 0.314439858, 0.23682323, 0.28572109, 0.287292278, 0.259598953, 0.31289784, 0.309862552], "raw": [0.19295553, 0.33817992, 0.300771023, 0.209185907, 0.158094435, 0.271453015, 0.300224518, 0.169977943, 0.33623539, 0.266101955, 0.229453522, 0.284086607, 0.255525135, 0.22852827, 0.315821345, 0.296493322], "temperature": [0.203400355, 0.346240725, 0.311686322, 0.245064585, 0.188457353, 0.313532367, 0.306424702, 0.226273923, 0.336936682, 0.287651037, 0.20571735, 0.284413483, 0.278863045, 0.27629571, 0.318568325, 0.306250682]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0016, "losses": 12, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.996", "prereg_hash": "09d7bb74c1479026a8c89a6a04f9b4c103c858f539bddd950202de7eecb858b8", "primary_control": "temperature", "result_hash": "5c3a0460c44518aacd16f4c2b42a87a4fa1e765601fa0df8d5df90dc3c89b971", "signature": "a0c27a2023fa3042e2bd876545073edd0cf6bc8a4abf9b90d39ec3f0f8293fe0056b126de0c8b8ad71cfaeb41852606989aeb16019e05cfe95fd7f63afd5b50e", "state": "VERIFIED", "summary": {"iw_temperature": 0.285266, "raw": 0.259568, "temperature": 0.277236}, "ties": 1, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:09:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:09:55Z", "cpu_seconds": 5.9999405499999625, "created_at": "2026-11-22T03:09:47Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d3ad37f9c2ab", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6ead026d2e1c67b66720c9ee09487cc03f11c9d6aa01faf1cd8de3e52a57f70d", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d3ad37f9c2ab", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:09:47Z", "seed": 451078051, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "86e35a3a0ce4faa00f85868f13e0e01f1ba93ed8e870d223a52e51709dda1f2a", "run_seconds": 6.575340270996094, "seed": 451078051, "started_at": "2026-11-22T03:09:48Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 1, "reset": 4}, "better": "higher", "ci95": [-0.576946, -0.323054], "effect": -0.45, "evaluated_at": "2026-11-22T03:09:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d3ad37f9c2ab", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.22919254914221937, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 27, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "6ead026d2e1c67b66720c9ee09487cc03f11c9d6aa01faf1cd8de3e52a57f70d", "primary_control": "reset", "result_hash": "86e35a3a0ce4faa00f85868f13e0e01f1ba93ed8e870d223a52e51709dda1f2a", "signature": "cd418c1e62d5733d00ca58a4d327342810e6add8963bd4e08202131c3e71b401605a1665a4696fac0ea1594865f676cb0450404d792330c18a3e5707f93bf30b", "state": "VERIFIED", "summary": {"inherit": 0.4, "reset": 0.85, "root": 0.3}, "ties": 33, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:10:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:58Z", "cpu_seconds": 1.6299611499999855, "created_at": "2026-11-22T03:10:03Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-78a8c0d59490", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "acb026a8d85d2f68b2162f8045402765f38a783ddd149e9512376ae0f958600a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-78a8c0d59490", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:03Z", "seed": 828681864, "treatment": "iw_temperature"}, "priority": 0.1786997073075065, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ad5a58b7b619f38ca65689eaaff4ff30d99e2cb92976ddcb547b767994f891c4", "run_seconds": 1.716191291809082, "seed": 828681864, "started_at": "2026-11-22T03:10:56Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017945, 0.00504], "effect": -0.006453, "evaluated_at": "2026-11-22T03:11:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-78a8c0d59490", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.042580045, 0.031884232, 0.035230667, 0.087382358, 0.023288213, 0.0205954, 0.052752377, 0.013465878, 0.071238655, 0.029677048, 0.038324155, 0.046322255, 0.07299897, 0.106151755, 0.04662476, 0.070978302], "raw": [0.038048035, 0.05852396, 0.06197252, 0.05634793, 0.044188297, 0.04221723, 0.052752377, 0.052111425, 0.073981395, 0.030227512, 0.01248539, 0.046322255, 0.06621534, 0.04838053, 0.04166546, 0.098512425], "temperature": [0.042580045, 0.072379555, 0.030092955, 0.055582523, 0.044188297, 0.0205954, 0.043607267, 0.01997553, 0.071238655, 0.02823242, 0.01248539, 0.019583457, 0.07299897, 0.040894177, 0.040840512, 0.070978302]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "acb026a8d85d2f68b2162f8045402765f38a783ddd149e9512376ae0f958600a", "primary_control": "temperature", "result_hash": "ad5a58b7b619f38ca65689eaaff4ff30d99e2cb92976ddcb547b767994f891c4", "signature": "68a8e38e2d2b286326075b690b76c57d983b9b7936cf1f4bd6173121fb8e6e760e71007624036c7c23e28b90a5c7006876cacd0a96de95d4c391f5b371c3c90d", "state": "VERIFIED", "summary": {"iw_temperature": 0.049343, "raw": 0.051497, "temperature": 0.042891}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:11:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:05Z", "cpu_seconds": 1.4589366000000155, "created_at": "2026-11-22T03:10:03Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c3912fde5674", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "22b83c94e3ccd97dac4927a35a7602df39c4063b9f8d72b443ffbdf2266aa8e1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c3912fde5674", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:03Z", "seed": 1527265693, "treatment": "iw_temperature"}, "priority": 0.32769344936722683, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-5cf45a9608bf", "result_hash": "2adb7dc8639af67f2a35a3fae588475dd41a1eb89a3e3c4c3739cc87637bd3ec", "run_seconds": 1.4989755153656006, "seed": 1527265693, "started_at": "2026-11-22T03:10:04Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.026194, 0.00219], "effect": -0.012002, "evaluated_at": "2026-11-22T03:10:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c3912fde5674", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.286816455, 0.128605852, 0.207954825, 0.273180713, 0.222753572, 0.244243208, 0.201182288, 0.315956118, 0.28949423, 0.293407357, 0.27184044, 0.238300425, 0.298794662, 0.324871495, 0.286577867, 0.295530903], "raw": [0.28005618, 0.15630595, 0.237813608, 0.189793777, 0.1657142, 0.239400972, 0.198184712, 0.315062745, 0.262102355, 0.251164315, 0.26379199, 0.158333065, 0.217051252, 0.226336507, 0.224959072, 0.25387453], "temperature": [0.284980072, 0.171392115, 0.26479276, 0.234783502, 0.175373025, 0.24290653, 0.16754616, 0.31555987, 0.281616013, 0.28033278, 0.27055548, 0.22211642, 0.26272625, 0.30399415, 0.254931403, 0.25387453]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0001, "losses": 14, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "22b83c94e3ccd97dac4927a35a7602df39c4063b9f8d72b443ffbdf2266aa8e1", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-5cf45a9608bf", "result_hash": "2adb7dc8639af67f2a35a3fae588475dd41a1eb89a3e3c4c3739cc87637bd3ec", "signature": "9d322fe130788c6552df8681e1ced55542970e41c9a1499f236b196e614ca40433d069847951b0c95ffcb226f5b28530c32bbd3d7d34260c270f6a44401ffd08", "state": "VERIFIED", "summary": {"iw_temperature": 0.261219, "raw": 0.227497, "temperature": 0.249218}, "ties": 0, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:10:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:04Z", "cpu_seconds": 1.6561653089999595, "created_at": "2026-11-22T03:10:03Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f32297674932", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "41e22836c9cbb27aaa6c816c7b21b92722d2f1ba49c8ef2615ef4c1aec07c94d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f32297674932", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:03Z", "seed": 804597545, "treatment": "iw_temperature"}, "priority": 0.17840914702418448, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "9228dad54e5666f3a8d6ea2c0a1dd73b3c25103add2d68175b3d31fd90676dc8", "run_seconds": 1.7307512760162354, "seed": 804597545, "started_at": "2026-11-22T03:11:03Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.003789, 0.016904], "effect": 0.006557, "evaluated_at": "2026-11-22T03:11:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f32297674932", "family": "calibration_iw", "hypothesis_supported": true, "instance_scores": {"iw_temperature": [0.114333968, 0.10210771, 0.196590093, 0.146298492, 0.143691422, 0.215395947, 0.186135777, 0.236220138, 0.239305377, 0.23897411, 0.164336065, 0.212929808, 0.272592328, 0.204214512, 0.173243095, 0.144660547], "raw": [0.127751555, 0.117091835, 0.160548027, 0.171870727, 0.121800325, 0.200814013, 0.198413713, 0.17533966, 0.211421103, 0.203295857, 0.177662607, 0.212929808, 0.206577043, 0.216555143, 0.19487973, 0.18947626], "temperature": [0.120077057, 0.10210771, 0.219867023, 0.163423923, 0.148996655, 0.230022013, 0.212404453, 0.22325602, 0.250207982, 0.256146723, 0.193902295, 0.220629338, 0.217623938, 0.204214512, 0.159331115, 0.173735468]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.5423, "losses": 3, "n": 16, "outcome": "SUPPORTED", "p_sign_display": "0.0287", "prereg_hash": "41e22836c9cbb27aaa6c816c7b21b92722d2f1ba49c8ef2615ef4c1aec07c94d", "primary_control": "temperature", "result_hash": "9228dad54e5666f3a8d6ea2c0a1dd73b3c25103add2d68175b3d31fd90676dc8", "signature": "88b3310df7cc815bab5037df90711967a7598b7928d9153ef020dda47124d8ef6b9912f82371b1f7b583f484eac68d656de7728ed0f1d48e564028b72d81e504", "state": "VERIFIED", "summary": {"iw_temperature": 0.186939, "raw": 0.180402, "temperature": 0.193497}, "ties": 2, "treatment": "iw_temperature", "wins": 11}, "verified_at": "2026-11-22T03:11:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:14Z", "cpu_seconds": 7.149272203999999, "created_at": "2026-11-22T03:10:03Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0149d25f0623", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e97ee482a672ddae19d02a893af014eaf0374d48f8683022cc949c0cd4dca96b", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0149d25f0623", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:10:03Z", "seed": 1843532386, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "d9d2e863c8b271b0b958fad1c21921e5a0afde67c1976be5c727a32e3b9c3b86", "run_seconds": 7.3317296504974365, "seed": 1843532386, "started_at": "2026-11-22T03:10:06Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:10:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0149d25f0623", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0], "root": [1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e97ee482a672ddae19d02a893af014eaf0374d48f8683022cc949c0cd4dca96b", "primary_control": "reset", "result_hash": "d9d2e863c8b271b0b958fad1c21921e5a0afde67c1976be5c727a32e3b9c3b86", "signature": "5997a6ec1c52bf7ea982a6b285a8ec41ae7ecd4dfdaed5dfa9ffc0de9fd8647d43e3a007175b1b3ba62169ba30873af9d2be93a6cdfd09bb7bbeca99a6e22c0b", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:10:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:10Z", "cpu_seconds": 6.148178258999991, "created_at": "2026-11-22T03:10:03Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-7366194177a1", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "56dfa33533333754d9fa69c1681d184000ed69f1c0327f1eb2d25e06a502587b", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-7366194177a1", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:10:03Z", "seed": 452297030, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "fc0db56cb134c03f5e161a48be46f418234dd31a8d137f900bf968e776eeb6bd", "run_seconds": 6.293793201446533, "seed": 452297030, "started_at": "2026-11-22T03:10:04Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.079138, 0.012471], "effect": -0.033333, "evaluated_at": "2026-11-22T03:10:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-7366194177a1", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0], "root": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "56dfa33533333754d9fa69c1681d184000ed69f1c0327f1eb2d25e06a502587b", "primary_control": "reset", "result_hash": "fc0db56cb134c03f5e161a48be46f418234dd31a8d137f900bf968e776eeb6bd", "signature": "e57654a94e23445bb7e11a5c4a43ad1e9f31767ca05cac68991edc6536dd1deaf982ed10a879b063b178f00be282ffb5543f5b5a2035397f7afeb2bdfe492b08", "state": "VERIFIED", "summary": {"inherit": 0.6166666666666667, "reset": 0.65, "root": 0.36666666666666664}, "ties": 58, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:10:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:31Z", "cpu_seconds": 1.4761839319999979, "created_at": "2026-11-22T03:10:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-53a62a79d5da", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "3744e0f3f12a1dacdbf2d94aff9670f00a6a38cb629142e1de6df75ef3c9dba3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-53a62a79d5da", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:20Z", "seed": 1807707644, "treatment": "iw_temperature"}, "priority": 0.18641961818652308, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "da316c96e5f95fce527b1c2b56df1e397abac4ae56a2fe7be60fa338bbc80d09", "run_seconds": 1.4972753524780273, "seed": 1807707644, "started_at": "2026-11-22T03:10:29Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.006639, 0.017706], "effect": 0.005533, "evaluated_at": "2026-11-22T03:10:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-53a62a79d5da", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.06656642, 0.057442743, 0.063594163, 0.039507703, 0.064730548, 0.037268617, 0.10436946, 0.027317135, 0.115002542, 0.00876721, 0.038761497, 0.045907917, 0.03125435, 0.022475778, 0.095144455, 0.03200891], "raw": [0.067350602, 0.027414065, 0.06591223, 0.0288853, 0.07773619, 0.061838718, 0.151181965, 0.030455748, 0.102473505, 0.012138777, 0.050075448, 0.061176452, 0.042890555, 0.044587505, 0.123303705, 0.020718185], "temperature": [0.05881646, 0.024742587, 0.091704885, 0.039507703, 0.07773619, 0.061838718, 0.177819053, 0.022008335, 0.098272255, 0.00876721, 0.027244275, 0.040778672, 0.063641628, 0.017981192, 0.10331829, 0.02447774]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "3744e0f3f12a1dacdbf2d94aff9670f00a6a38cb629142e1de6df75ef3c9dba3", "primary_control": "temperature", "result_hash": "da316c96e5f95fce527b1c2b56df1e397abac4ae56a2fe7be60fa338bbc80d09", "signature": "cd5bd38acc599d7ef4cbe1678fc45ca4a0765d21753584569ba9780682024e777362ea0e539e8fe8073e20441edcd5501632f31cbc82a9b41e8a26368f40080f", "state": "VERIFIED", "summary": {"iw_temperature": 0.053132, "raw": 0.060509, "temperature": 0.058666}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:10:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:24Z", "cpu_seconds": 1.5901247559999945, "created_at": "2026-11-22T03:10:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-7d48f3893342", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "05c3f68f08e63339abc5ee13841cd31f3dfa74eb46fa24ebe90a8aa0ec416493", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-7d48f3893342", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:20Z", "seed": 1642653455, "treatment": "iw_temperature"}, "priority": 0.3203877017083834, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "b5ba753cd0e62cf8f9c53a401462e1fcb28f3a76d0a6df46a2c1cbe1627c1cfd", "run_seconds": 1.6951603889465332, "seed": 1642653455, "started_at": "2026-11-22T03:10:23Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021652, 0.009415], "effect": -0.006118, "evaluated_at": "2026-11-22T03:10:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-7d48f3893342", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.21724716, 0.302396178, 0.3099738, 0.266322552, 0.149117588, 0.294279975, 0.21539293, 0.307365095, 0.20912777, 0.259338933, 0.260157602, 0.287713857, 0.324338137, 0.28391916, 0.243230082, 0.314208475], "raw": [0.150664592, 0.296117875, 0.22566383, 0.2029931, 0.226395643, 0.285800563, 0.197280145, 0.22180733, 0.20912777, 0.21358993, 0.248616735, 0.277963235, 0.268921613, 0.224301538, 0.237796563, 0.307971398], "temperature": [0.179435447, 0.296117875, 0.29795629, 0.217360913, 0.219890065, 0.293812162, 0.23184451, 0.233851017, 0.23004631, 0.259338933, 0.25653106, 0.287713857, 0.306324938, 0.272307405, 0.251261305, 0.312441833]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "05c3f68f08e63339abc5ee13841cd31f3dfa74eb46fa24ebe90a8aa0ec416493", "primary_control": "temperature", "result_hash": "b5ba753cd0e62cf8f9c53a401462e1fcb28f3a76d0a6df46a2c1cbe1627c1cfd", "signature": "9a0e6c86da15bec81fb2c32a335dae96d989da497bf4eeb0e48ce4ffcb68c8a854030631cc2152e5272ab6e2b7d2c6b7a11a597892aa08a7eea6e37fe2585905", "state": "VERIFIED", "summary": {"iw_temperature": 0.265258, "raw": 0.237188, "temperature": 0.25914}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:10:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:26Z", "cpu_seconds": 1.5182912509999937, "created_at": "2026-11-22T03:10:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-cb987c4ddb03", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "e6363683a97fc9d0282b54f941d3eca8314bdc9592731092174acad350a5411a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-cb987c4ddb03", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:20Z", "seed": 943852183, "treatment": "iw_temperature"}, "priority": 0.1946311166870832, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "c36d9a15fd59e173979d8bdcda2b1a1493b3c7c96dbce0bef49b3d11d5ef4456", "run_seconds": 1.7154927253723145, "seed": 943852183, "started_at": "2026-11-22T03:10:24Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011449, 0.00483], "effect": -0.00331, "evaluated_at": "2026-11-22T03:10:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-cb987c4ddb03", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.07535411, 0.061870085, 0.04653487, 0.079436027, 0.041345532, 0.05860713, 0.03305914, 0.093828227, 0.04083262, 0.072309248, 0.089350742, 0.04042666, 0.035426928, 0.053090192, 0.069734875, 0.012258008], "raw": [0.099629188, 0.056408498, 0.076748092, 0.04017382, 0.04225262, 0.074241233, 0.023494403, 0.121780935, 0.061755082, 0.074209832, 0.079623212, 0.038214092, 0.023663637, 0.047258242, 0.065327088, 0.022925595], "temperature": [0.092809155, 0.072636455, 0.036003588, 0.037767915, 0.0256568, 0.068120557, 0.021802995, 0.121780935, 0.04083262, 0.064962285, 0.074489885, 0.04042666, 0.035426928, 0.030707443, 0.069734875, 0.017351162]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0936, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.806", "prereg_hash": "e6363683a97fc9d0282b54f941d3eca8314bdc9592731092174acad350a5411a", "primary_control": "temperature", "result_hash": "c36d9a15fd59e173979d8bdcda2b1a1493b3c7c96dbce0bef49b3d11d5ef4456", "signature": "55af96f00400de0b9cf24e8d90507cd7b4193af88a486cde6232869bd4cf5ee123bda2d2e1bb52a504dd9cba1625833776ec480f545b4e3c385cc4492c8e5201", "state": "VERIFIED", "summary": {"iw_temperature": 0.056467, "raw": 0.059232, "temperature": 0.053157}, "ties": 4, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:10:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:44Z", "cpu_seconds": 1.4610199980000118, "created_at": "2026-11-22T03:10:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-eaafc8025fd6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "0d19527fa3ff0f17dd58187b83ead0914620969d0c300aa3bd725cd4da0b6ad6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-eaafc8025fd6", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:20Z", "seed": 1739841865, "treatment": "iw_temperature"}, "priority": 0.16742544232899853, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "bf3dbd2cdc697216a337f9b59a784a12d8d8673f5e426f6386b9c4b6e7109f6b", "run_seconds": 1.4955220222473145, "seed": 1739841865, "started_at": "2026-11-22T03:12:42Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.000998, 0.0199], "effect": 0.009451, "evaluated_at": "2026-11-22T03:12:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-eaafc8025fd6", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.108616377, 0.13110081, 0.051523782, 0.164754823, 0.157938235, 0.08603558, 0.155644397, 0.13323151, 0.159167212, 0.171272878, 0.157807687, 0.160901392, 0.148422932, 0.137640733, 0.175664423, 0.153464155], "raw": [0.09901035, 0.13110081, 0.102457933, 0.164754823, 0.168815912, 0.133165705, 0.132783257, 0.12933019, 0.157765873, 0.175628668, 0.17406575, 0.143174188, 0.146196005, 0.113394948, 0.151517315, 0.177936248], "temperature": [0.126252787, 0.158859737, 0.11695388, 0.169279798, 0.157938235, 0.099005025, 0.1520644, 0.13323151, 0.159167212, 0.176545435, 0.18687712, 0.153753165, 0.148335845, 0.113394948, 0.166314115, 0.186425733]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5368, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.291", "prereg_hash": "0d19527fa3ff0f17dd58187b83ead0914620969d0c300aa3bd725cd4da0b6ad6", "primary_control": "temperature", "result_hash": "bf3dbd2cdc697216a337f9b59a784a12d8d8673f5e426f6386b9c4b6e7109f6b", "signature": "211814a2ccff9087345d934ba9cdd525f93c83c1f29a2f2da48a41dc13091c29cc6e8a6c622ef47871fe09cafbaf78353b6533dbdb8f516e36c004eb56828d00", "state": "VERIFIED", "summary": {"iw_temperature": 0.140824, "raw": 0.143819, "temperature": 0.150275}, "ties": 3, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:12:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:26Z", "cpu_seconds": 5.649538430000007, "created_at": "2026-11-22T03:10:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-58d9d8d51ed8", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "df587b836dbef370d92bfc251e5a21618d849698c972c17f08a3006a158d7567", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-58d9d8d51ed8", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:10:20Z", "seed": 2110482849, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "19089149f43171927ca977e0f52a764fadf2cf9ba9970c435b351c08c475eb4d", "run_seconds": 5.909780740737915, "seed": 2110482849, "started_at": "2026-11-22T03:10:20Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:10:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-58d9d8d51ed8", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "df587b836dbef370d92bfc251e5a21618d849698c972c17f08a3006a158d7567", "primary_control": "reset", "result_hash": "19089149f43171927ca977e0f52a764fadf2cf9ba9970c435b351c08c475eb4d", "signature": "5c829c0e0011b1288c92e4101f61513f60795fdc537e77ee65893cf475f6649ff34a8dd2a0a096fa4a7ec68d6e9d474c3527afaf9cea86c86a70b5ce511e610d", "state": "PARTIAL", "summary": {"inherit": 0.5166666666666667, "reset": 0.5166666666666667, "root": 0.3}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:10:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:30Z", "cpu_seconds": 1.4490599509999997, "created_at": "2026-11-22T03:10:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-14fe36b64dbf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "4a77254211bf0aaaaa09d34a8269796f200d8116a077720c034fd6db56bfd48e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-14fe36b64dbf", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:36Z", "seed": 1066007010, "treatment": "iw_temperature"}, "priority": 0.15729424652769633, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "854e5c66aca6c27725ca2fa119ad4c7e01fe8797f1ebf860453bebb5942c75c7", "run_seconds": 1.5655665397644043, "seed": 1066007010, "started_at": "2026-11-22T03:14:29Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.008214, 0.003228], "effect": -0.002493, "evaluated_at": "2026-11-22T03:14:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-14fe36b64dbf", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.079242975, 0.043527435, 0.068892927, 0.062563663, 0.088348465, 0.05081439, 0.065933962, 0.05064573, 0.028171377, 0.085291855, 0.018741855, 0.04283685, 0.070791815, 0.054808175, 0.139344385, 0.07899534], "raw": [0.08460699, 0.0494316, 0.070125082, 0.047195825, 0.079567505, 0.070452725, 0.046264458, 0.042074753, 0.054264773, 0.089973825, 0.05835327, 0.079047127, 0.077468093, 0.055111703, 0.131227212, 0.048321528], "temperature": [0.08952818, 0.065661833, 0.068892927, 0.054137482, 0.06927545, 0.05081439, 0.04005773, 0.043968153, 0.028171377, 0.089973825, 0.026694433, 0.042669815, 0.064304703, 0.054808175, 0.137499742, 0.062597295]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "4a77254211bf0aaaaa09d34a8269796f200d8116a077720c034fd6db56bfd48e", "primary_control": "temperature", "result_hash": "854e5c66aca6c27725ca2fa119ad4c7e01fe8797f1ebf860453bebb5942c75c7", "signature": "9aa2dd4c4d6a04d9bbbc3bd1241e9a36cc253d929850e96dcc019c529a5e9f3daf772d2b5fb8c74da87595eafa3378e86c0c68b14f98566f761c481edf6e780a", "state": "VERIFIED", "summary": {"iw_temperature": 0.064309, "raw": 0.067718, "temperature": 0.061816}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:14:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:34Z", "cpu_seconds": 1.6719659689999844, "created_at": "2026-11-22T03:10:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-8cf04a365698", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "c007e01b2b1a815aeed354270b2c78dca77fb83fa22c6d4c5966331622845c2d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-8cf04a365698", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:36Z", "seed": 905572808, "treatment": "iw_temperature"}, "priority": 0.17481847886333604, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "2debe35a18575f056c01074ad90fe8e7d88d1e27be329905e521de1ebb238165", "run_seconds": 1.7097232341766357, "seed": 905572808, "started_at": "2026-11-22T03:11:33Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.016059, 0.000396], "effect": -0.007831, "evaluated_at": "2026-11-22T03:11:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-8cf04a365698", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.058358015, 0.03146774, 0.0438286, 0.05212026, 0.085554038, 0.022701363, 0.056822202, 0.093364692, 0.045924695, 0.073390878, 0.005302565, 0.04515651, 0.0879065, 0.034846145, 0.022332715, 0.02265212], "raw": [0.052570305, 0.048703233, 0.040015038, 0.04679172, 0.076970947, 0.014737775, 0.060229303, 0.07981672, 0.03528766, 0.043824983, 0.023328823, 0.022216127, 0.061704405, 0.036263908, 0.038666407, 0.032978597], "temperature": [0.058358015, 0.03146774, 0.047029007, 0.038994955, 0.05329281, 0.01854588, 0.073495168, 0.065462563, 0.01916537, 0.056665405, 0.00714117, 0.020142545, 0.060338683, 0.057677448, 0.02387528, 0.02477673]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "c007e01b2b1a815aeed354270b2c78dca77fb83fa22c6d4c5966331622845c2d", "primary_control": "temperature", "result_hash": "2debe35a18575f056c01074ad90fe8e7d88d1e27be329905e521de1ebb238165", "signature": "282f73d367b8b0e3199d0d39487fa15b272dcd002f67c80157628ba6e213e49c3ebc9c4221cb78efed93b8c0f7b4c7fbf74aeec93a39e9703790f0d25eadae0c", "state": "VERIFIED", "summary": {"iw_temperature": 0.048858, "raw": 0.044632, "temperature": 0.041027}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:11:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:16Z", "cpu_seconds": 1.8704814899999747, "created_at": "2026-11-22T03:10:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-91306136f6f3", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "602cb0452c0290ac7716145cdcc6e56c200a031de293137978980e0425ba3d53", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-91306136f6f3", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:36Z", "seed": 1819162126, "treatment": "iw_temperature"}, "priority": 0.16986925979926992, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-f27ac428ec79", "result_hash": "cac36cdc4d66780e0e078142658f2206f2df711fc291a534c8be10211342974d", "run_seconds": 2.026134967803955, "seed": 1819162126, "started_at": "2026-11-22T03:12:14Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.014552, -0.000536], "effect": -0.007544, "evaluated_at": "2026-11-22T03:12:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-91306136f6f3", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.072391697, 0.034435155, 0.035629517, 0.025616075, 0.074758355, 0.063928197, 0.070177843, 0.030222303, 0.04439731, 0.023110532, 0.035426765, 0.06891523, 0.047424568, 0.08263361, 0.082222692, 0.039471305], "raw": [0.050462333, 0.0427187, 0.027468217, 0.051326045, 0.059979383, 0.050007868, 0.061313267, 0.030222303, 0.04439731, 0.043597403, 0.034416725, 0.054518363, 0.0427742, 0.069293005, 0.057148625, 0.033849542], "temperature": [0.072391697, 0.037177837, 0.02866144, 0.025616075, 0.074758355, 0.053252645, 0.027330135, 0.028566838, 0.053484233, 0.033013343, 0.035426765, 0.03910093, 0.025609847, 0.070931522, 0.082222692, 0.02251548]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "602cb0452c0290ac7716145cdcc6e56c200a031de293137978980e0425ba3d53", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-f27ac428ec79", "result_hash": "cac36cdc4d66780e0e078142658f2206f2df711fc291a534c8be10211342974d", "signature": "673c60d76764538664ee3d7310b439e9a3a44b7173298eef13023e82612b71de512ee35d88dd5ec2973667b4390c7235ab8466e878099d7f2947737ba450f40b", "state": "VERIFIED", "summary": {"iw_temperature": 0.051923, "raw": 0.047093, "temperature": 0.044379}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:12:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:45Z", "cpu_seconds": 1.8154047280000043, "created_at": "2026-11-22T03:10:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-af3cf510662a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "7b1943932c06fcf45238570cbedecbe0f36a5f3874c4ab3d34392741ce934703", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-af3cf510662a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:36Z", "seed": 1080889106, "treatment": "iw_temperature"}, "priority": 0.29808095370883964, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "75214b207a2465bf36f3a378ace7e5969369461f3137e7b22d830d406d7f8aee", "run_seconds": 1.8942101001739502, "seed": 1080889106, "started_at": "2026-11-22T03:10:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.018896, 0.010308], "effect": -0.004294, "evaluated_at": "2026-11-22T03:10:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-af3cf510662a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.244828445, 0.13540139, 0.27987957, 0.174051542, 0.29930407, 0.283923685, 0.324077365, 0.282291707, 0.34250902, 0.177381757, 0.27477032, 0.290433755, 0.261614413, 0.29974661, 0.2654859, 0.266729818], "raw": [0.22558197, 0.192948225, 0.24004221, 0.145426517, 0.29197007, 0.206109113, 0.261973093, 0.268909458, 0.332357068, 0.191481173, 0.243229808, 0.272524235, 0.235111237, 0.271576788, 0.206380457, 0.232019973], "temperature": [0.283309658, 0.20732368, 0.260297495, 0.177848517, 0.312022565, 0.26195239, 0.28156476, 0.282939755, 0.34250902, 0.177381757, 0.267462047, 0.272524235, 0.263327888, 0.292533888, 0.211263657, 0.239466102]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "7b1943932c06fcf45238570cbedecbe0f36a5f3874c4ab3d34392741ce934703", "primary_control": "temperature", "result_hash": "75214b207a2465bf36f3a378ace7e5969369461f3137e7b22d830d406d7f8aee", "signature": "fbec0718578f7cf25131114c7163308575b8eb09b7bc4dede8b46e50d9ffa49dbc3bb28303feb19dcad9adab68224be6ffd6f57c4a69ac2922d882678012cb0f", "state": "VERIFIED", "summary": {"iw_temperature": 0.262652, "raw": 0.238603, "temperature": 0.258358}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:10:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:46Z", "cpu_seconds": 7.210374710999986, "created_at": "2026-11-22T03:10:36Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1f219320d224", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "5cab0e7f691a5258260f1a343aeb8803ae2e4b8ab51ca2ceb4d6f32746bbbe1b", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1f219320d224", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:10:36Z", "seed": 1857138083, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "988f1292d3f7284eaa5aae1b86e6c1ac5f85da9843daa16b59a129ac9995b726", "run_seconds": 7.404606103897095, "seed": 1857138083, "started_at": "2026-11-22T03:10:38Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:10:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1f219320d224", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.35370979028579796, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "5cab0e7f691a5258260f1a343aeb8803ae2e4b8ab51ca2ceb4d6f32746bbbe1b", "primary_control": "reset", "result_hash": "988f1292d3f7284eaa5aae1b86e6c1ac5f85da9843daa16b59a129ac9995b726", "signature": "a17ac6e63070e8a1d83bce44fe65b9cb930ee8f784145ab6ed1dcd479b501a69563a0d1d6e07f64f3efd83d7b689cd3d4172fc16988754555e62eb11fbcc2e0e", "state": "PARTIAL", "summary": {"inherit": 0.6, "reset": 0.6, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:11:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:45Z", "cpu_seconds": 9.105955782000024, "created_at": "2026-11-22T03:10:36Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-4865d20aa8a3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "2d4e8889f81ae9d463c2c2d399f3b963b2e324187b235e21151f7ed67898132d", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-4865d20aa8a3", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:10:36Z", "seed": 1278423471, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9e7053fb25c2336f7d3e797b81fcd5aa15c5ebc53b9cb2f1b50c592f3e9f849a", "run_seconds": 9.281018018722534, "seed": 1278423471, "started_at": "2026-11-22T03:10:36Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:10:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-4865d20aa8a3", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5561585589867449, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0], "reset": [0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0], "root": [1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "2d4e8889f81ae9d463c2c2d399f3b963b2e324187b235e21151f7ed67898132d", "primary_control": "reset", "result_hash": "9e7053fb25c2336f7d3e797b81fcd5aa15c5ebc53b9cb2f1b50c592f3e9f849a", "signature": "f91da0efef7e014bc702ceaa346308282cb856021d3c548f58cfa4f29f3354b38be2d55b2a50408b96e8499d55c37b8ece79ae48a12003742ab7b3c302b5a509", "state": "PARTIAL", "summary": {"inherit": 0.6, "reset": 0.6, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:10:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:19Z", "cpu_seconds": 4.163586191000036, "created_at": "2026-11-22T03:10:36Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-4a7303ddd4ab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "e170c7ef126c5c121de5fb97750512be012ef1939a23857c8f5de4375a14f1e5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-4a7303ddd4ab", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:10:36Z", "seed": 444642325, "treatment": "frontier"}, "priority": 0.16570573715709083, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "c84092beacddace83e6919aa5de768d7e480c6a9b4c0986cbafd1889c4ab9c78", "run_seconds": 4.36568021774292, "seed": 444642325, "started_at": "2026-11-22T03:13:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.04101, 0.056317], "effect": 0.048663, "evaluated_at": "2026-11-22T03:13:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-4a7303ddd4ab", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.949612403, 0.920081967, 0.944230769, 0.915322581, 0.930769231, 0.883928571, 0.922131148, 0.897826087, 0.918699187, 0.911016949, 0.915254237, 0.933884298], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.972868217, 0.948770492, 0.973076923, 0.953629032, 0.948076923, 0.946428571, 0.950819672, 0.943478261, 0.949186992, 0.959745763, 0.921610169, 0.948347107]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "e170c7ef126c5c121de5fb97750512be012ef1939a23857c8f5de4375a14f1e5", "primary_control": "random_walk", "result_hash": "c84092beacddace83e6919aa5de768d7e480c6a9b4c0986cbafd1889c4ab9c78", "signature": "f07e013b83611256da253bf817b33ffbcd225a298517a0d50142ebffbeddd2254aec4c673bca450ad2e05d47e644582d7196999055a4bc446de18a44b0a38008", "state": "VERIFIED", "summary": {"curiosity": 0.92023, "frontier": 1.0, "random_walk": 0.951337}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:13:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:10:57Z", "cpu_seconds": 1.6856266910000102, "created_at": "2026-11-22T03:10:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-14e997f2230b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "2bb07d4db89e8eb7123fceed7c163409aae7fb7ba4c56cba61e7a573b725cd3b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-14e997f2230b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:10:52Z", "seed": 1709168430, "treatment": "iw_temperature"}, "priority": 0.31062819531294994, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ef2c5402946e82ae0665ddb0d9cb32adefa500f3b78f5d8301df44b00d648232", "run_seconds": 1.7519035339355469, "seed": 1709168430, "started_at": "2026-11-22T03:10:55Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.008915, 0.013891], "effect": 0.002488, "evaluated_at": "2026-11-22T03:11:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-14e997f2230b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.30646665, 0.317999025, 0.208281048, 0.303388987, 0.216126843, 0.306564632, 0.209088625, 0.281701, 0.288534398, 0.30976625, 0.308783078, 0.294761222, 0.236799432, 0.211584235, 0.1633806, 0.183007215], "raw": [0.258591548, 0.300299867, 0.208281048, 0.287306065, 0.19214274, 0.289770877, 0.219601387, 0.24821651, 0.280481953, 0.303638188, 0.28266049, 0.290695227, 0.259318865, 0.173950082, 0.22477726, 0.203485875], "temperature": [0.286789418, 0.317999025, 0.243761813, 0.298883125, 0.222498692, 0.306564632, 0.238569795, 0.24821651, 0.29214222, 0.303638188, 0.302303017, 0.29442387, 0.283841705, 0.19271831, 0.197340067, 0.15635254]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "2bb07d4db89e8eb7123fceed7c163409aae7fb7ba4c56cba61e7a573b725cd3b", "primary_control": "temperature", "result_hash": "ef2c5402946e82ae0665ddb0d9cb32adefa500f3b78f5d8301df44b00d648232", "signature": "03da163c3ac7efd88938f7a2e044f7a108065f9bdb7268a413642806e81a97e231d78e841b4c3dd01f8a3d447784b7e3b4615976ca58605f8d7cf654e77d370f", "state": "VERIFIED", "summary": {"iw_temperature": 0.25914, "raw": 0.251451, "temperature": 0.261628}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:11:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:14Z", "cpu_seconds": 0.8613221490000171, "created_at": "2026-11-22T03:10:52Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-24650e34f313", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "abd8f9edb25067e516967b18939270d693fa6785f716c7d7e4a53d5b847f11e1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-24650e34f313", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:10:52Z", "seed": 1459651179, "treatment": "frontier"}, "priority": 0.1659994915376669, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "aa04d806dafb71f57310a89fe18455812bc067425d726f6c49f85acd2806632e", "run_seconds": 0.8716859817504883, "seed": 1459651179, "started_at": "2026-11-22T03:13:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.08588, 0.163506], "effect": 0.124693, "evaluated_at": "2026-11-22T03:13:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-24650e34f313", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.806122449, 0.805232558, 0.826704545, 0.814903846, 0.844594595, 1.0, 0.848623853, 0.7775, 0.844594595, 0.841954023, 0.79076087, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.836734694, 0.813953488, 0.863636364, 0.850961538, 0.858108108, 1.0, 0.864678899, 0.7975, 0.905405405, 0.913793103, 0.798913043, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "abd8f9edb25067e516967b18939270d693fa6785f716c7d7e4a53d5b847f11e1", "primary_control": "random_walk", "result_hash": "aa04d806dafb71f57310a89fe18455812bc067425d726f6c49f85acd2806632e", "signature": "1f7142170bf6fc3a41ac159cd9b400feadb8ba9f74da36e066ee0b01b204c22694a50138836f04c5c0147167841e8eed7cea4bbf41e5c1b9b7449b7074710f01", "state": "VERIFIED", "summary": {"curiosity": 0.850083, "frontier": 1.0, "random_walk": 0.875307}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:13:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:03Z", "cpu_seconds": 2.914976803000002, "created_at": "2026-11-22T03:10:52Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-d30070092f3e", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "09590cb264675b50b45ec36e2281adb98409506db33353bf26d2222b1985d627", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-d30070092f3e", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:10:52Z", "seed": 464007651, "treatment": "frontier"}, "priority": 0.1667101350493996, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-003f10672bd6", "result_hash": "afc6b9e1dec5655d6b3cfda257f1a055a1b5a13b1467eac2e637b0980caf42b2", "run_seconds": 2.970460891723633, "seed": 464007651, "started_at": "2026-11-22T03:13:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.038289, 0.105454], "effect": 0.071871, "evaluated_at": "2026-11-22T03:13:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-d30070092f3e", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.827981651, 0.840909091, 0.903571429, 0.849462366, 0.917525773, 0.880841121, 1.0, 0.857142857, 0.841145833, 0.816666667, 0.765822785], "frontier": [1.0, 1.0, 0.963068182, 1.0, 1.0, 1.0, 0.813084112, 1.0, 1.0, 0.919270833, 1.0, 1.0], "random_walk": [1.0, 0.880733945, 0.911931818, 0.939285714, 0.879032258, 0.884020619, 0.799065421, 1.0, 0.928571429, 0.903645833, 0.811111111, 0.89556962]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "09590cb264675b50b45ec36e2281adb98409506db33353bf26d2222b1985d627", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-003f10672bd6", "result_hash": "afc6b9e1dec5655d6b3cfda257f1a055a1b5a13b1467eac2e637b0980caf42b2", "signature": "9de0453508e2ec14fa5101a26ba8d9f8085ebf7b174956982db5d1e1e9f3045d7207d655e318f7217f38b19cdd4b2aacada01dbddf2d809c229b42e7cae4540b", "state": "VERIFIED", "summary": {"curiosity": 0.875089, "frontier": 0.974619, "random_walk": 0.902747}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:13:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:35:52Z", "cpu_seconds": 1.4907395829999928, "created_at": "2026-11-22T03:11:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-278ec3a6467f", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "45a8b5dce0b44df9f9320329a74f14128580d3b54465c249e8ea070d89488a48", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-278ec3a6467f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:09Z", "seed": 2113802872, "treatment": "iw_temperature"}, "priority": 0.14972660012401656, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-89f2c03bce14", "result_hash": "fdb1bf95a7a987b86a2a3a8f6d0703ba3cc709945f04c1767adecc1bcbe0ae5f", "run_seconds": 1.5642681121826172, "seed": 2113802872, "started_at": "2026-11-22T04:35:50Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:11:16Z", "cpu_seconds": 1.6675288159999866, "created_at": "2026-11-22T03:11:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-422c0f93fc09", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "706e43b59498ae75c1c55c2fadfd8d0a73b440059972973bea3979f0fd616d52", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-422c0f93fc09", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:09Z", "seed": 1847599706, "treatment": "iw_temperature"}, "priority": 0.18452942280550685, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "4c84f68374bb02c93db18685334119e2d7680a201dfd2426318e2e5b95ef7896", "run_seconds": 1.811208724975586, "seed": 1847599706, "started_at": "2026-11-22T03:11:14Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012979, 0.015235], "effect": 0.001128, "evaluated_at": "2026-11-22T03:11:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-422c0f93fc09", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.159281788, 0.224942202, 0.23086306, 0.184190933, 0.176086083, 0.245481577, 0.24935643, 0.052424137, 0.068690018, 0.26150498, 0.252439397, 0.286694415, 0.22644832, 0.195349877, 0.203690702, 0.19275339], "raw": [0.18206277, 0.191383233, 0.197290213, 0.16589357, 0.1566395, 0.219315965, 0.21166684, 0.178943935, 0.0715426, 0.185848625, 0.163197078, 0.224902528, 0.248443623, 0.180186978, 0.17518374, 0.19275339], "temperature": [0.18206277, 0.205019648, 0.23086306, 0.205095962, 0.176086083, 0.245481577, 0.24935643, 0.133850528, 0.01737495, 0.26150498, 0.21571734, 0.286694415, 0.22644832, 0.180186978, 0.203690702, 0.208811382]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1961, "losses": 4, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.637", "prereg_hash": "706e43b59498ae75c1c55c2fadfd8d0a73b440059972973bea3979f0fd616d52", "primary_control": "temperature", "result_hash": "4c84f68374bb02c93db18685334119e2d7680a201dfd2426318e2e5b95ef7896", "signature": "4bf2d13126f91962ff319a3341b1e236bead5670d6840ca8df703835f596ae647d3c828c1101fbabbf673e0eff03ecdd56072d3bd717e453910c4950558a970f", "state": "VERIFIED", "summary": {"iw_temperature": 0.200637, "raw": 0.184078, "temperature": 0.201765}, "ties": 8, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:11:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:10Z", "cpu_seconds": 1.6744689079999997, "created_at": "2026-11-22T03:11:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-70a6793687a0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "08491419259dae0eb9d5536e921f34f28f0f0541e62626a0bf76f05ed7797d32", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-70a6793687a0", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:09Z", "seed": 1510249003, "treatment": "iw_temperature"}, "priority": 0.15840963969639155, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a41691cf19644c1421b5634a4bbfb90c44ee470a8248157cc3ef7e2eb77f87de", "run_seconds": 1.9763364791870117, "seed": 1510249003, "started_at": "2026-11-22T03:14:08Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011727, 0.019901], "effect": 0.004087, "evaluated_at": "2026-11-22T03:14:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-70a6793687a0", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.103991255, 0.197279523, 0.0547765, 0.05646943, 0.121615292, 0.142431885, 0.136999982, 0.16059582, 0.031653037, 0.104481293, 0.13318198, 0.155360415, 0.199362345, 0.080780305, 0.085991838, 0.111327803], "raw": [0.087146968, 0.07473629, 0.157890445, 0.083322457, 0.137142103, 0.12411534, 0.12580352, 0.120596772, 0.076995013, 0.104481293, 0.112196678, 0.117173805, 0.128551708, 0.104725735, 0.12312023, 0.091482485], "temperature": [0.11677549, 0.149592772, 0.13618062, 0.083322457, 0.137142103, 0.142431885, 0.14929628, 0.132752032, 0.033689555, 0.104481293, 0.110736635, 0.142696967, 0.180854542, 0.078188673, 0.146871643, 0.096681112]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2184, "losses": 7, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.605", "prereg_hash": "08491419259dae0eb9d5536e921f34f28f0f0541e62626a0bf76f05ed7797d32", "primary_control": "temperature", "result_hash": "a41691cf19644c1421b5634a4bbfb90c44ee470a8248157cc3ef7e2eb77f87de", "signature": "53eb478f18a7183b6cb8d16c6a9c0c5dffd433148d67bb730a200272cc24fce61a30cbfec23efbbf6a3c1859d59fc4c265b8305f00ab5fb5524bac94b08eac0c", "state": "VERIFIED", "summary": {"iw_temperature": 0.117269, "raw": 0.110593, "temperature": 0.121356}, "ties": 2, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:14:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:09Z", "cpu_seconds": 1.7410547479999998, "created_at": "2026-11-22T03:11:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-ae5509fe1711", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "dc91fe2944a9ef0b15ac40a2712b8f8c5c87225ca2f701163af803c1c5d861b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-ae5509fe1711", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:09Z", "seed": 1253110819, "treatment": "iw_temperature"}, "priority": 0.15882464398036267, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "181727b1d22a9112c6e4a239945d5269315578dea61301fe911a0b0c982084cc", "run_seconds": 1.9639782905578613, "seed": 1253110819, "started_at": "2026-11-22T03:14:07Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011317, 0.004876], "effect": -0.00322, "evaluated_at": "2026-11-22T03:14:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-ae5509fe1711", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.26090518, 0.20421007, 0.260857793, 0.283201808, 0.273045425, 0.266473372, 0.21534772, 0.217911395, 0.278605922, 0.21754696, 0.29575698, 0.198566767, 0.243601703, 0.284924685, 0.192964655, 0.255080915], "raw": [0.212966677, 0.20421007, 0.249537487, 0.234118477, 0.231707947, 0.250608318, 0.21534772, 0.236085367, 0.278605922, 0.149859935, 0.255554915, 0.23803515, 0.193873295, 0.24638836, 0.16256378, 0.232917138], "temperature": [0.247252205, 0.20421007, 0.260857793, 0.257170493, 0.237381785, 0.250608318, 0.20513723, 0.236085367, 0.30254481, 0.200245855, 0.306127583, 0.210210135, 0.243601703, 0.284924685, 0.207130585, 0.243989525]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0936, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.806", "prereg_hash": "dc91fe2944a9ef0b15ac40a2712b8f8c5c87225ca2f701163af803c1c5d861b6", "primary_control": "temperature", "result_hash": "181727b1d22a9112c6e4a239945d5269315578dea61301fe911a0b0c982084cc", "signature": "a04ba796763e4485fac5dd89877fc1535304e21d184dacbe21ea54c3387bc5887adf86e02999d3d02507027275757e49ecf13a7ff8fe33d801ce1dc8441c890e", "state": "VERIFIED", "summary": {"iw_temperature": 0.246813, "raw": 0.224524, "temperature": 0.243592}, "ties": 4, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:14:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:55Z", "cpu_seconds": 1.5277345690000175, "created_at": "2026-11-22T03:11:09Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-de8ec70b23e4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "573d30923bafcbc45fdd77c8edd869a875ff7f566c1034905065819d81b467a8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-de8ec70b23e4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:09Z", "seed": 1123600843, "treatment": "iw_temperature"}, "priority": 0.17245982174661167, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "5d3998ca7622d55261d7ea3498737f54f41d1f09b4ed3d65dd629989b3dd9fbc", "run_seconds": 1.5676963329315186, "seed": 1123600843, "started_at": "2026-11-22T03:11:53Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015058, 0.014218], "effect": -0.00042, "evaluated_at": "2026-11-22T03:12:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-de8ec70b23e4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.043927477, 0.108875855, 0.125664485, 0.149558333, 0.105595195, 0.168465042, 0.167595208, 0.172996325, 0.20119347, 0.11066418, 0.154984945, 0.13499785, 0.12375983, 0.211498893, 0.109461445, 0.150460365], "raw": [0.113443, 0.126591752, 0.099689957, 0.134894278, 0.099016877, 0.11143333, 0.154485453, 0.144988325, 0.17072648, 0.11066418, 0.136553633, 0.10139696, 0.103643783, 0.14985108, 0.138497545, 0.11418986], "temperature": [0.101921575, 0.15891323, 0.125664485, 0.143249883, 0.105595195, 0.143613397, 0.154485453, 0.168254037, 0.17072648, 0.11066418, 0.13030444, 0.124809325, 0.117330645, 0.185632248, 0.169058415, 0.122761652]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "573d30923bafcbc45fdd77c8edd869a875ff7f566c1034905065819d81b467a8", "primary_control": "temperature", "result_hash": "5d3998ca7622d55261d7ea3498737f54f41d1f09b4ed3d65dd629989b3dd9fbc", "signature": "42f73b1d48d1a6134432cde50a5d99ca95c3bf5c346fec33879776703739397f4d0c4177bad26c220f993542d121840d44cc268865b8a4cdb7783c5e31a7f90f", "state": "VERIFIED", "summary": {"iw_temperature": 0.139981, "raw": 0.125629, "temperature": 0.139562}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:12:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:15Z", "cpu_seconds": 6.294887528999993, "created_at": "2026-11-22T03:11:09Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-9d0b924d879e", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "713365b734445e3d7ccdb59a65d3952cd28076783a94022c0d0d32210a2f1cec", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-9d0b924d879e", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:11:09Z", "seed": 172873004, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "4f2629044b2717097b505c2302bbe410e9c6af481ed1f872d5e9f4e623ec4d5a", "run_seconds": 6.409362077713013, "seed": 172873004, "started_at": "2026-11-22T03:11:09Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [0.189152, 0.444181], "effect": 0.316667, "evaluated_at": "2026-11-22T03:11:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-9d0b924d879e", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.2911050574372532, "rstep": 3}, "hypothesis_supported": true, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -4.9792, "losses": 1, "n": 60, "outcome": "SUPPORTED", "p_sign_display": "1.05e-05", "prereg_hash": "713365b734445e3d7ccdb59a65d3952cd28076783a94022c0d0d32210a2f1cec", "primary_control": "reset", "result_hash": "4f2629044b2717097b505c2302bbe410e9c6af481ed1f872d5e9f4e623ec4d5a", "signature": "21ff7fdd7abf140258993b43f0ab0ff889b00e6f494cffc77f7a97d75020625c650b1c91b2ea96cd046375f3c47147f09aaf38337bc7c8e72990f7da0aefa103", "state": "VERIFIED", "summary": {"inherit": 0.8833333333333333, "reset": 0.5666666666666667, "root": 0.2833333333333333}, "ties": 39, "treatment": "inherit", "wins": 20}, "verified_at": "2026-11-22T03:11:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:19Z", "cpu_seconds": 9.343340266999974, "created_at": "2026-11-22T03:11:09Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d5e3fdc7f948", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "b25eafc04f385a44a34d122726a26b0d2d71c7d00da41544bde0d098c6ddafa8", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d5e3fdc7f948", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:11:09Z", "seed": 768925075, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "6657fa0cdaa4da38452a390dc25d3b9e1d02d1d1eebc04c21c5733a8354d390f", "run_seconds": 9.612840175628662, "seed": 768925075, "started_at": "2026-11-22T03:11:09Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:11:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d5e3fdc7f948", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b25eafc04f385a44a34d122726a26b0d2d71c7d00da41544bde0d098c6ddafa8", "primary_control": "reset", "result_hash": "6657fa0cdaa4da38452a390dc25d3b9e1d02d1d1eebc04c21c5733a8354d390f", "signature": "071de791daebf192e192881f335e67f0f76f441a45c29d8a01127b2e4ac817752d5e36d2d4c316fb30291869667a0d79d898d4cb6525ac6dd42b4d620e976408", "state": "PARTIAL", "summary": {"inherit": 0.45, "reset": 0.45, "root": 0.23333333333333334}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:11:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:35:36Z", "cpu_seconds": 1.4692905730000234, "created_at": "2026-11-22T03:11:25Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5151ca439817", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "3e3d4ab507ce8b0c7125d4603ec40509d98772c0a471008ae92a9e8554aa0c8b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5151ca439817", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:25Z", "seed": 1112964958, "treatment": "iw_temperature"}, "priority": 0.15191179743475436, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "904bfc3e8d82c41d774cc9951e8dd9881bb386bc451b555acf7fc51178603b83", "run_seconds": 2.143242120742798, "seed": 1112964958, "started_at": "2026-11-22T04:35:34Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:12:37Z", "cpu_seconds": 1.7842371200000002, "created_at": "2026-11-22T03:11:25Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-5e49d73d92b6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "16a8b1c680e801ee2fab3ee44d04d2f4e1b2453840f3569cb33cff11a55e6702", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-5e49d73d92b6", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:25Z", "seed": 1399705411, "treatment": "iw_temperature"}, "priority": 0.16909809158037842, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a5912acba27184c5b8444782080dca81526a9b41dd27155d1b50eadae730cea4", "run_seconds": 2.146815776824951, "seed": 1399705411, "started_at": "2026-11-22T03:12:35Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019643, 0.010521], "effect": -0.004561, "evaluated_at": "2026-11-22T03:12:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-5e49d73d92b6", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.222906977, 0.2671933, 0.298468687, 0.198404398, 0.260112503, 0.199507943, 0.224135163, 0.181072317, 0.147325908, 0.263194727, 0.28723011, 0.279878635, 0.18455298, 0.204256162, 0.21131973, 0.271028395], "raw": [0.173867912, 0.19843637, 0.20772795, 0.190184982, 0.19461493, 0.211383017, 0.16991791, 0.14624568, 0.166217423, 0.227478565, 0.230370035, 0.221103375, 0.161291568, 0.163000867, 0.21131973, 0.21643691], "temperature": [0.20546259, 0.245349637, 0.248691535, 0.198404398, 0.260112503, 0.242766195, 0.224135163, 0.14624568, 0.200012777, 0.257193255, 0.27023913, 0.264527287, 0.169401375, 0.184450235, 0.269286618, 0.241332867]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "16a8b1c680e801ee2fab3ee44d04d2f4e1b2453840f3569cb33cff11a55e6702", "primary_control": "temperature", "result_hash": "a5912acba27184c5b8444782080dca81526a9b41dd27155d1b50eadae730cea4", "signature": "8f720fd46f1e9a6aab92c1a3a604493f692451c43352858498c3e7ff7e9297b46711fcfecc08dd2478bf48023f834898ae8205d85ba0eb9993f034cf3beebf08", "state": "VERIFIED", "summary": {"iw_temperature": 0.231287, "raw": 0.1931, "temperature": 0.226726}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:12:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:33Z", "cpu_seconds": 7.556480409000017, "created_at": "2026-11-22T03:11:25Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1631b9f64efe", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "a77ffbd20702d9f65021e3a6314275d0764b0c50d3c96d94b786e3efceb70e44", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1631b9f64efe", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:11:25Z", "seed": 2075539306, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "2882ccb44f32516c3e95d5f3b63a0e96cc67634ee7a168a0c6c9b168ff5032cc", "run_seconds": 7.864777326583862, "seed": 2075539306, "started_at": "2026-11-22T03:11:25Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:11:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1631b9f64efe", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a77ffbd20702d9f65021e3a6314275d0764b0c50d3c96d94b786e3efceb70e44", "primary_control": "reset", "result_hash": "2882ccb44f32516c3e95d5f3b63a0e96cc67634ee7a168a0c6c9b168ff5032cc", "signature": "ef806302d24e6384f9e5586eba5c1e49a92e09b27476d5fb55675a4b908bf0b7c61e87bddd00f2fb07512086424e3e2db72ed403db591e2bbb4c9e4e104c2a03", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.4166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:11:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:31Z", "cpu_seconds": 1.2957858430000044, "created_at": "2026-11-22T03:11:25Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-83e6aa8ab40d", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "2d53d41c8da161a0a20c318a09c6b749c6d19d4fd104396d11e542383ae2afc1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-83e6aa8ab40d", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:11:25Z", "seed": 901684439, "treatment": "frontier"}, "priority": 0.3365148798693171, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-3502ceb5736b", "result_hash": "92d7aeb91150d63f18ef58b03984381ced6ed619698b72862dd29a7f58f092ec", "run_seconds": 1.334249496459961, "seed": 901684439, "started_at": "2026-11-22T03:11:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.060244, 0.123409], "effect": 0.091826, "evaluated_at": "2026-11-22T03:11:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-83e6aa8ab40d", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.842592593, 0.849137931, 0.936170213, 0.882978723, 1.0, 0.87037037, 0.903061224, 0.783333333, 0.892241379, 0.904545455, 0.855], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [1.0, 0.842592593, 0.900862069, 0.920212766, 0.941489362, 1.0, 0.851851852, 0.903061224, 0.827777778, 0.870689655, 0.904545455, 0.935]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.0103, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000977", "prereg_hash": "2d53d41c8da161a0a20c318a09c6b749c6d19d4fd104396d11e542383ae2afc1", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-3502ceb5736b", "result_hash": "92d7aeb91150d63f18ef58b03984381ced6ed619698b72862dd29a7f58f092ec", "signature": "454c7e87879214233006fe58ada56bd92a4b1c92932a453f0142a0ae8a715155c2efd87d360747901a118824a89acd9f8a8949e76d929cf555c986a8ae9a390e", "state": "VERIFIED", "summary": {"curiosity": 0.893286, "frontier": 1.0, "random_walk": 0.908174}, "ties": 2, "treatment": "frontier", "wins": 10}, "verified_at": "2026-11-22T03:11:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:48Z", "cpu_seconds": 1.6297942600000397, "created_at": "2026-11-22T03:11:41Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-707ea75ec5bf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "1737001a3ca1cec00d99eab59f283b310f4782ae2848a0344bc473e9c6075fb6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-707ea75ec5bf", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:41Z", "seed": 1816056350, "treatment": "iw_temperature"}, "priority": 0.19372949740158554, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "82c4ceda5c92890a230fcdbe31d2f131753867b63e9b8dd726ba393dc80c4a3e", "run_seconds": 1.6731135845184326, "seed": 1816056350, "started_at": "2026-11-22T03:11:46Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011707, 0.006094], "effect": -0.002807, "evaluated_at": "2026-11-22T03:11:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-707ea75ec5bf", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.148031465, 0.114322547, 0.09903724, 0.13325194, 0.073102228, 0.154180195, 0.100843325, 0.167005175, 0.149386745, 0.139724145, 0.121594415, 0.120674922, 0.125533805, 0.069609258, 0.146788875, 0.142422885], "raw": [0.1212921, 0.076975377, 0.084973197, 0.081242202, 0.073102228, 0.10663566, 0.100843325, 0.1132632, 0.12311843, 0.091789033, 0.10884612, 0.120674922, 0.082543135, 0.082504025, 0.089872918, 0.119096813], "temperature": [0.1212921, 0.114322547, 0.096532815, 0.13325194, 0.065889048, 0.154180195, 0.100843325, 0.167005175, 0.145414785, 0.155975023, 0.144523062, 0.14665768, 0.082543135, 0.082504025, 0.13057171, 0.119096813]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0522, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.887", "prereg_hash": "1737001a3ca1cec00d99eab59f283b310f4782ae2848a0344bc473e9c6075fb6", "primary_control": "temperature", "result_hash": "82c4ceda5c92890a230fcdbe31d2f131753867b63e9b8dd726ba393dc80c4a3e", "signature": "85c6ac0c939681899500550834c8442f01ab69619be76087a0dad4146dcaf165b04b8d3ab24a7ad890423e00a5de6ada5b4b2afd33411b325c79de6f38ae2c0c", "state": "VERIFIED", "summary": {"iw_temperature": 0.125344, "raw": 0.098548, "temperature": 0.122538}, "ties": 5, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:11:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:43Z", "cpu_seconds": 1.558168543999983, "created_at": "2026-11-22T03:11:41Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-96b14b5a0efa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "bcfcde78c24171ddad2b25e3b8e8d6c7afaafb3022347fd2bbda5ccdf4c0b5fa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-96b14b5a0efa", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:41Z", "seed": 983950045, "treatment": "iw_temperature"}, "priority": 0.34237808428058936, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "3c0f479ff5758dcd759209e72b83b5f5b43d65a600a63060e7daea796fb056fa", "run_seconds": 1.5637857913970947, "seed": 983950045, "started_at": "2026-11-22T03:11:42Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015114, 0.032844], "effect": 0.008865, "evaluated_at": "2026-11-22T03:11:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-96b14b5a0efa", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.27960666, 0.156255928, 0.149091975, 0.284524815, 0.103612945, 0.29834647, 0.29715383, 0.317610675, 0.196475385, 0.268092477, 0.238222348, 0.30647946, 0.204781025, 0.092713712, 0.261619425, 0.244857028], "raw": [0.191580275, 0.172929507, 0.149091975, 0.25956413, 0.235425358, 0.263029513, 0.215643695, 0.323007617, 0.16972562, 0.23378292, 0.306899792, 0.30063361, 0.14126917, 0.19169938, 0.176089817, 0.231792868], "temperature": [0.230978688, 0.156255928, 0.164807957, 0.25956413, 0.235425358, 0.282035087, 0.250341782, 0.328786733, 0.16972562, 0.25990248, 0.306899792, 0.30063361, 0.204781025, 0.184669615, 0.261619425, 0.244857028]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0936, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.806", "prereg_hash": "bcfcde78c24171ddad2b25e3b8e8d6c7afaafb3022347fd2bbda5ccdf4c0b5fa", "primary_control": "temperature", "result_hash": "3c0f479ff5758dcd759209e72b83b5f5b43d65a600a63060e7daea796fb056fa", "signature": "4e7998591173ace00670e85e84404aa9bbd0dd1c255622575990ad483cee0f1257166083d5051ce3e9dd8106c890c0fb33a9100ad81360c2b547e97de3d9c505", "state": "VERIFIED", "summary": {"iw_temperature": 0.231215, "raw": 0.222635, "temperature": 0.24008}, "ties": 4, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:11:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:11:51Z", "cpu_seconds": 7.352840387000015, "created_at": "2026-11-22T03:11:41Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-46af63376785", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "aaedfd504534c95f313719a73ef443e541e4447676b634810a978caeadc67fca", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-46af63376785", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:11:41Z", "seed": 1906987904, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "458147038cca5dccb8f43b5b247333068bc2bef3ad6e7718321bd6cfc437adfd", "run_seconds": 7.388625621795654, "seed": 1906987904, "started_at": "2026-11-22T03:11:43Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:11:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-46af63376785", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "aaedfd504534c95f313719a73ef443e541e4447676b634810a978caeadc67fca", "primary_control": "reset", "result_hash": "458147038cca5dccb8f43b5b247333068bc2bef3ad6e7718321bd6cfc437adfd", "signature": "6a52855eceedfe8da981d995ca674f3dad355f8707112f1457f8c5c67fe2975302a57bf34ffffd489f7708c9ec1ed6812e4a9a2fd5a61d54e259aaed0e685608", "state": "PARTIAL", "summary": {"inherit": 0.5666666666666667, "reset": 0.5666666666666667, "root": 0.25}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:12:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:35Z", "cpu_seconds": 4.511560343000001, "created_at": "2026-11-22T03:11:41Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-49f4c79c7870", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "9ce4d6cfa372bfd808c5b19bfdabe7ebd3114c32bd406433706ddf2642d95153", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-49f4c79c7870", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:11:41Z", "seed": 463071791, "treatment": "frontier"}, "priority": 0.15697169117900026, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "30b86b0713aee71932b8124ad42e602034e25a8a23770f34b0d3b4e632b5404e", "run_seconds": 5.139634132385254, "seed": 463071791, "started_at": "2026-11-22T03:14:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058033, 0.072425], "effect": 0.065229, "evaluated_at": "2026-11-22T03:14:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-49f4c79c7870", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.924796748, 0.934322034, 0.920833333, 0.901639344, 0.90625, 0.932377049, 0.903846154, 0.929487179, 0.876106195, 0.904661017, 0.922268908, 0.891666667], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.945121951, 0.944915254, 0.945833333, 0.926229508, 0.954166667, 0.922131148, 0.938034188, 0.94017094, 0.940265487, 0.911016949, 0.924369748, 0.925]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "9ce4d6cfa372bfd808c5b19bfdabe7ebd3114c32bd406433706ddf2642d95153", "primary_control": "random_walk", "result_hash": "30b86b0713aee71932b8124ad42e602034e25a8a23770f34b0d3b4e632b5404e", "signature": "5b2af56bf45d2cd9ccaaab57b38dc13f00e57d8791bfc4f82e5537e0ab4b2d42c198478965ad9ab595a8785a928e064c5985b72521a764abbb55ed5adc70fc0e", "state": "VERIFIED", "summary": {"curiosity": 0.912355, "frontier": 1.0, "random_walk": 0.934771}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:14:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:10Z", "cpu_seconds": 1.488016149, "created_at": "2026-11-22T03:11:41Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-cf1dcf912c3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "11bd9243ab9ca3b5b0b34d9090d6deb72aa0ffc48b8b7e58d73d595aebed729f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-cf1dcf912c3b", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:11:41Z", "seed": 1659877474, "treatment": "frontier"}, "priority": 0.15607458540941233, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "b5d06465cdd7cc796c799fcab0b588ec06e13501e1e923debcca99ec8399f874", "run_seconds": 1.5073723793029785, "seed": 1659877474, "started_at": "2026-11-22T03:20:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.037776, 0.06826], "effect": 0.053018, "evaluated_at": "2026-11-22T03:22:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-cf1dcf912c3b", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.87890625, 1.0, 0.930147059, 0.957142857, 0.981617647, 0.93442623, 0.9140625, 0.920833333, 0.958333333, 0.974264706, 0.930147059, 0.909090909], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9875, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.9140625, 1.0, 0.948529412, 0.975, 0.959558824, 0.918032787, 0.94921875, 0.954166667, 0.934027778, 0.944852941, 0.948529412, 0.90530303]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "11bd9243ab9ca3b5b0b34d9090d6deb72aa0ffc48b8b7e58d73d595aebed729f", "primary_control": "random_walk", "result_hash": "b5d06465cdd7cc796c799fcab0b588ec06e13501e1e923debcca99ec8399f874", "signature": "870b6e3670072757c17b2aac1be7f29c2699c56a4cc0a6e59758512f51810320bfe5c675a2854f378ed0532305e2e32184748d5a880ab834400c741750c2ee0b", "state": "VERIFIED", "summary": {"curiosity": 0.940748, "frontier": 0.998958, "random_walk": 0.94594}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:22:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:09Z", "cpu_seconds": 1.9418140660000063, "created_at": "2026-11-22T03:11:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-354948230e9e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "dfb676499c95f4ca60f5111d7db1f163a3ff69fd250946e94534b1d258312091", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-354948230e9e", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:58Z", "seed": 2037519133, "treatment": "iw_temperature"}, "priority": 0.17128046764969784, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "3498b19d139d4bc31bdf7cc4948fa50dde28bad5157893b0256a3e4b3390754d", "run_seconds": 2.198857069015503, "seed": 2037519133, "started_at": "2026-11-22T03:12:07Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.002722, 0.005493], "effect": 0.001386, "evaluated_at": "2026-11-22T03:12:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-354948230e9e", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.033043717, 0.105711755, 0.073172865, 0.036869747, 0.086443152, 0.045518148, 0.088406275, 0.026009558, 0.021109585, 0.082844035, 0.189998762, 0.19650171, 0.06620826, 0.04087777, 0.036668285, 0.033061622], "raw": [0.059546838, 0.111406268, 0.096379127, 0.05368656, 0.069553335, 0.03686857, 0.107113145, 0.022221105, 0.033811837, 0.082844035, 0.182700773, 0.191643172, 0.061310348, 0.04452383, 0.048749723, 0.028207735], "temperature": [0.03516191, 0.105711755, 0.090570123, 0.043550968, 0.087490512, 0.031452565, 0.100016215, 0.026009558, 0.021109585, 0.068866645, 0.18631788, 0.192482398, 0.06620826, 0.051154422, 0.037186022, 0.041329722]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.7125, "losses": 4, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.194", "prereg_hash": "dfb676499c95f4ca60f5111d7db1f163a3ff69fd250946e94534b1d258312091", "primary_control": "temperature", "result_hash": "3498b19d139d4bc31bdf7cc4948fa50dde28bad5157893b0256a3e4b3390754d", "signature": "39323646980eebc5ddfd6e352b5de2ea1c57b3c7e336c2d8c7e3fb0857f59c3471e97ca65e68dd81f61ac50c116cefdc74d496715e6d7286b511b0ef9af6a20a", "state": "VERIFIED", "summary": {"iw_temperature": 0.072653, "raw": 0.07691, "temperature": 0.074039}, "ties": 4, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:12:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:21Z", "cpu_seconds": 1.5112848849999807, "created_at": "2026-11-22T03:11:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b51f480d1c25", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "37c753e01895e3981a51a9652fa272033e9bb72e30ee989cff0b9751ca35ddcd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b51f480d1c25", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:58Z", "seed": 349429955, "treatment": "iw_temperature"}, "priority": 0.16435140099472278, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "fecfaa5454bc0cffac533a0ac09807babbaaf94ed9d2e9b833726d73ae9f0e2e", "run_seconds": 1.5287184715270996, "seed": 349429955, "started_at": "2026-11-22T03:13:19Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017283, 0.003354], "effect": -0.006965, "evaluated_at": "2026-11-22T03:13:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b51f480d1c25", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.20834633, 0.254463785, 0.197231418, 0.241438785, 0.254716173, 0.292738237, 0.29490922, 0.265878425, 0.223066013, 0.216205145, 0.25524953, 0.266100745, 0.174432407, 0.209032, 0.171048537, 0.146003958], "raw": [0.20834633, 0.217988683, 0.219297782, 0.204785888, 0.1950641, 0.246112047, 0.233301932, 0.223033365, 0.19272052, 0.20951303, 0.223198515, 0.266100745, 0.19174022, 0.181623143, 0.181511992, 0.178495825], "temperature": [0.223239222, 0.24819734, 0.236332198, 0.192519277, 0.216201022, 0.276548503, 0.262645965, 0.252247923, 0.223066013, 0.216205145, 0.250282535, 0.252869165, 0.174432407, 0.209032, 0.171048537, 0.154559138]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "37c753e01895e3981a51a9652fa272033e9bb72e30ee989cff0b9751ca35ddcd", "primary_control": "temperature", "result_hash": "fecfaa5454bc0cffac533a0ac09807babbaaf94ed9d2e9b833726d73ae9f0e2e", "signature": "ee59f4c6aaa30a0aa32661d5518eb6908a02b85fb08e9160487a7f2db98ac8af34232a4eb6cad2a175b6714abeabaa6eaf7a544e2aa2c3b83a83c99bc3b03a07", "state": "VERIFIED", "summary": {"iw_temperature": 0.229429, "raw": 0.210802, "temperature": 0.222464}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:13:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:02Z", "cpu_seconds": 1.7541673559999822, "created_at": "2026-11-22T03:11:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c23432c31592", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "c6ac11234c193192f96660c9c3f3c56a34d04c0ccc10b9e3f6cf78287c89d849", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c23432c31592", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:58Z", "seed": 1657743068, "treatment": "iw_temperature"}, "priority": 0.3122230518579836, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "29f932efbc320aa2c18784a78c534f52ce878bba94bb73af136c21246e416729", "run_seconds": 1.8096296787261963, "seed": 1657743068, "started_at": "2026-11-22T03:12:00Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019908, 0.00662], "effect": -0.006644, "evaluated_at": "2026-11-22T03:12:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c23432c31592", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.313847985, 0.265286068, 0.208347738, 0.240183447, 0.200368245, 0.244009387, 0.2620283, 0.197710302, 0.26867187, 0.327524627, 0.156245485, 0.290564877, 0.308045223, 0.268677417, 0.296244092, 0.31349003], "raw": [0.268584147, 0.265286068, 0.226673388, 0.230805595, 0.111375992, 0.211999255, 0.219173763, 0.189689045, 0.152357165, 0.316862102, 0.20851493, 0.244839788, 0.299567433, 0.24084386, 0.275924982, 0.282760573], "temperature": [0.29507386, 0.273709592, 0.238594947, 0.252569265, 0.174357658, 0.211999255, 0.246704755, 0.207095752, 0.184560612, 0.325935092, 0.156245485, 0.316647977, 0.302919925, 0.250511288, 0.296244092, 0.321770008]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "c6ac11234c193192f96660c9c3f3c56a34d04c0ccc10b9e3f6cf78287c89d849", "primary_control": "temperature", "result_hash": "29f932efbc320aa2c18784a78c534f52ce878bba94bb73af136c21246e416729", "signature": "56da4829e8500c8e0e48298d2b84e3508dd6dbdf87b5f219304bb8a37c60ccaa02a371ee309a5d2399a4f1341a4bf08d3bb8e62faf5816196f1d16e69dfde203", "state": "VERIFIED", "summary": {"iw_temperature": 0.260078, "raw": 0.234079, "temperature": 0.253434}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:12:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:33Z", "cpu_seconds": 1.5860555410000003, "created_at": "2026-11-22T03:11:58Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-d3fef67e1c9e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "531f60032283c231a5839b85421de4c40cd47be39e68c40f280752808186c6a0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-d3fef67e1c9e", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:11:58Z", "seed": 702754040, "treatment": "iw_temperature"}, "priority": 0.1538070095979751, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "93878e45da70f99fb7f235b5e1a51672468f42316cecc9348f6aeb103aefbb85", "run_seconds": 1.8380012512207031, "seed": 702754040, "started_at": "2026-11-22T03:21:31Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01785, 0.002456], "effect": -0.007697, "evaluated_at": "2026-11-22T03:22:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-d3fef67e1c9e", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.091403478, 0.159350115, 0.181392595, 0.118416643, 0.092678132, 0.055052405, 0.081367555, 0.159798735, 0.172958425, 0.116946192, 0.151031305, 0.17287038, 0.159082875, 0.129422185, 0.08962355, 0.1202292], "raw": [0.083233812, 0.10775783, 0.121711615, 0.108299662, 0.092678132, 0.069827178, 0.063226878, 0.115052047, 0.161810425, 0.07830435, 0.094272572, 0.113117757, 0.109706462, 0.086133595, 0.123712587, 0.089185075], "temperature": [0.10170871, 0.104609275, 0.152419265, 0.118416643, 0.11050122, 0.069827178, 0.081367555, 0.144163187, 0.169902008, 0.12493447, 0.113328685, 0.15245336, 0.14395864, 0.118641998, 0.108693253, 0.11354598]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "531f60032283c231a5839b85421de4c40cd47be39e68c40f280752808186c6a0", "primary_control": "temperature", "result_hash": "93878e45da70f99fb7f235b5e1a51672468f42316cecc9348f6aeb103aefbb85", "signature": "0f478621dec9c3a03a47c446f8bafc3c32283cc992f46b72625abd6542e6762b5aab68200bb2fec2d6296e979f316cbea521175a02b7f8f10600d39a00729f04", "state": "VERIFIED", "summary": {"iw_temperature": 0.128226, "raw": 0.101127, "temperature": 0.120529}, "ties": 2, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:22:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:05Z", "cpu_seconds": 6.183548363, "created_at": "2026-11-22T03:11:58Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0ff8b29c16ac", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6652e04d56762fb88991ef5ed542d1a21e383ed2bd58ed971f9e17f90e60ee92", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0ff8b29c16ac", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:11:58Z", "seed": 556873957, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "865fbde1b82372c09194b4eb950fefa6b4633d749432e1ef4c69e548d8d928ba", "run_seconds": 6.260011196136475, "seed": 556873957, "started_at": "2026-11-22T03:11:59Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:12:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0ff8b29c16ac", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6652e04d56762fb88991ef5ed542d1a21e383ed2bd58ed971f9e17f90e60ee92", "primary_control": "reset", "result_hash": "865fbde1b82372c09194b4eb950fefa6b4633d749432e1ef4c69e548d8d928ba", "signature": "2ba3eddcd15656443108aa3dd8af594667db153eb8113d9ee5c03c2482452b6968dfc18c94a2f6590da400ac2e7ce471ef645a7afc087acbcd9ee9f7ca77e508", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.43333333333333335}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:12:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:02Z", "cpu_seconds": 0.4245626720000075, "created_at": "2026-11-22T03:11:58Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-f498a3a52266", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "b51b4908bc3bb79ab66896223c8cefb4ad2c6a87f0b45a97df9ef1b999001878", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-f498a3a52266", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:11:58Z", "seed": 178178916, "treatment": "frontier"}, "priority": 0.17212198573289533, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "84cde9aab74335697c78b9bda9b530efbb7035fd31ab7b9db7df513a9c728588", "run_seconds": 0.4425930976867676, "seed": 178178916, "started_at": "2026-11-22T03:12:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.078572, 0.148723], "effect": 0.113648, "evaluated_at": "2026-11-22T03:12:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-f498a3a52266", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.869047619, 0.833333333, 0.87755102, 0.8, 0.870689655, 0.868852459, 0.92, 0.808823529, 0.842592593, 0.833333333, 0.924107143], "frontier": [1.0, 0.803571429, 1.0, 1.0, 0.990909091, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.879464286], "random_walk": [1.0, 0.797619048, 0.833333333, 0.87244898, 0.827272727, 0.896551724, 0.881147541, 0.885, 0.833333333, 0.837962963, 0.824074074, 0.821428571]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "b51b4908bc3bb79ab66896223c8cefb4ad2c6a87f0b45a97df9ef1b999001878", "primary_control": "random_walk", "result_hash": "84cde9aab74335697c78b9bda9b530efbb7035fd31ab7b9db7df513a9c728588", "signature": "54c887a589b613582630a5650941accbe6bddeb885bf22d50b3652b08b924a723fef4d95db1c671e51870c519bdac0c2ec77f00c859f931d6f516c7d428dc606", "state": "VERIFIED", "summary": {"curiosity": 0.870694, "frontier": 0.972829, "random_walk": 0.859181}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:12:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:35:49Z", "cpu_seconds": 1.435754836000001, "created_at": "2026-11-22T03:12:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-628e05e52eb6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "95aae3f1b3db9271fd9f9ba92ceaf4c15d05e3ae9362857b0c9c0dc10ba41db4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-628e05e52eb6", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:14Z", "seed": 1383741388, "treatment": "iw_temperature"}, "priority": 0.1516606850802819, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "abdef362f3b32b169e1e2178147a9db405a83e972dd523a004d62f5a7649b3fd", "run_seconds": 1.5067973136901855, "seed": 1383741388, "started_at": "2026-11-22T04:35:48Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:12:24Z", "cpu_seconds": 1.5383262210000055, "created_at": "2026-11-22T03:12:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-9b3a759205e8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "a608a26c21c1a294a7eff669b27d65519a3dfad379e81b285888424804daff34", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-9b3a759205e8", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:14Z", "seed": 1098845095, "treatment": "iw_temperature"}, "priority": 0.16977798092886032, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "f0d41302e95f3b4546c96a581f2c4c352702161727e609299fe161e61de11e65", "run_seconds": 1.7957606315612793, "seed": 1098845095, "started_at": "2026-11-22T03:12:22Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015934, 0.026265], "effect": 0.005166, "evaluated_at": "2026-11-22T03:12:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-9b3a759205e8", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.224228318, 0.118049463, 0.275293643, 0.28327352, 0.258376292, 0.269087853, 0.234350243, 0.270079995, 0.141547005, 0.32412162, 0.279293373, 0.165534255, 0.212021243, 0.298615653, 0.238004823, 0.182705245], "raw": [0.20991036, 0.1636458, 0.237505678, 0.276115737, 0.259893265, 0.18297976, 0.195451435, 0.25988553, 0.226801955, 0.227002225, 0.259154735, 0.191188727, 0.23736901, 0.2451911, 0.222649038, 0.221707873], "temperature": [0.21295127, 0.198686492, 0.23106703, 0.281121813, 0.307319998, 0.231763547, 0.21951833, 0.233982075, 0.189950042, 0.293023158, 0.251247677, 0.238207408, 0.212021243, 0.26507159, 0.2473872, 0.243914602]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "a608a26c21c1a294a7eff669b27d65519a3dfad379e81b285888424804daff34", "primary_control": "temperature", "result_hash": "f0d41302e95f3b4546c96a581f2c4c352702161727e609299fe161e61de11e65", "signature": "67f9e87ce328373d623be805497cc2da3938484cf925fda7b6a0259f2713afc87e06e4b51f73a510145c39f08ddc2a62bb7f15c48b3759fa4d9e3d8deeed9106", "state": "VERIFIED", "summary": {"iw_temperature": 0.235911, "raw": 0.226028, "temperature": 0.241077}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:12:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:17Z", "cpu_seconds": 1.7849103249999985, "created_at": "2026-11-22T03:12:14Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-bf8fa963ed12", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "72fb22a80c593b48f52dd30c8e7ccf0a49936ad63ea60c843f65d7baaa08c448", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-bf8fa963ed12", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:14Z", "seed": 1982733957, "treatment": "iw_temperature"}, "priority": 0.33120212228124696, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "d7593b82388f9377ab737a0d087bbe3c474bb404de7f68f3061cc49f673bd444", "run_seconds": 1.9384379386901855, "seed": 1982733957, "started_at": "2026-11-22T03:12:15Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021319, 0.033579], "effect": 0.00613, "evaluated_at": "2026-11-22T03:12:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-bf8fa963ed12", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.287759015, 0.268453635, 0.08262581, 0.299159562, 0.253376993, 0.142460645, 0.255717393, 0.347882945, 0.16094117, 0.306448335, 0.296128795, 0.206008627, 0.24163188, 0.272782705, 0.345432243, 0.299518652], "raw": [0.149708075, 0.235201625, 0.221897388, 0.294580602, 0.264391065, 0.272089583, 0.263524518, 0.299121075, 0.187380605, 0.237168325, 0.29097884, 0.19093723, 0.253730148, 0.270197732, 0.234974925, 0.29510107], "temperature": [0.172432793, 0.257470505, 0.19231363, 0.298798008, 0.264391065, 0.254884767, 0.278278005, 0.312561185, 0.208699045, 0.282618985, 0.296128795, 0.23605942, 0.253730148, 0.272782705, 0.283743623, 0.299518652]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "72fb22a80c593b48f52dd30c8e7ccf0a49936ad63ea60c843f65d7baaa08c448", "primary_control": "temperature", "result_hash": "d7593b82388f9377ab737a0d087bbe3c474bb404de7f68f3061cc49f673bd444", "signature": "b01e43ef98314a69d71135ded5ff6ae96919bfa7a96692fecc973e2a2c6ec49d90b30c256f9b57322cbc1ad545cd182cf28632dec923bc9b66d91e658995640f", "state": "VERIFIED", "summary": {"iw_temperature": 0.254146, "raw": 0.247561, "temperature": 0.260276}, "ties": 3, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:12:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:18Z", "cpu_seconds": 3.2290437020000127, "created_at": "2026-11-22T03:12:14Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-5fa60c1b8eb0", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ca3b01b74f68fec951d1dc25a44581618c80559e06c8499f59852f234db96cbc", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-5fa60c1b8eb0", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:12:14Z", "seed": 91536944, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "458db76960d54b0358c3581c2f6218553e537742d82211851ad34079fac18ccc", "run_seconds": 3.4974653720855713, "seed": 91536944, "started_at": "2026-11-22T03:12:15Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 1, "reset": 1}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:12:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-5fa60c1b8eb0", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5578856441696775, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ca3b01b74f68fec951d1dc25a44581618c80559e06c8499f59852f234db96cbc", "primary_control": "reset", "result_hash": "458db76960d54b0358c3581c2f6218553e537742d82211851ad34079fac18ccc", "signature": "98c43f7dc44f6c80c7690bf65c8e8bb2da128e5a72a629df37cd30f8241163f7b6f25eea2c5b6ffc5bee9ac1e2a786efd4f7bd08009201bf72eaf4687c206905", "state": "PARTIAL", "summary": {"inherit": 0.3333333333333333, "reset": 0.3333333333333333, "root": 0.3333333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:12:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:29Z", "cpu_seconds": 7.880460019999987, "created_at": "2026-11-22T03:12:14Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-fbc1305e6b88", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7dc64e05ea0f88a1644b104642d3dcdfaa74d52508ca620a4a7f12001d4cd07d", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-fbc1305e6b88", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:12:14Z", "seed": 1053960186, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "c249dc6ec09aeab4b8d3c803baaf6ce3fb8395dbe40b0f1709623f14e5ebea87", "run_seconds": 8.244544267654419, "seed": 1053960186, "started_at": "2026-11-22T03:12:21Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.079138, 0.012471], "effect": -0.033333, "evaluated_at": "2026-11-22T03:12:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-fbc1305e6b88", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6509349684631951, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0], "root": [1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7dc64e05ea0f88a1644b104642d3dcdfaa74d52508ca620a4a7f12001d4cd07d", "primary_control": "reset", "result_hash": "c249dc6ec09aeab4b8d3c803baaf6ce3fb8395dbe40b0f1709623f14e5ebea87", "signature": "10328721a92e46f2b8f9aa5e609c78f25838f77c78b8f6312ce48dbd439b8d6e1e760fe54bdc68b091e145e3f6d94f25d94f4ef64d37c2a43bda01aa0fdb2f0b", "state": "VERIFIED", "summary": {"inherit": 0.5333333333333333, "reset": 0.5666666666666667, "root": 0.3}, "ties": 58, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:12:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:36:58Z", "cpu_seconds": 0.4626418389999998, "created_at": "2026-11-22T03:12:14Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-4d9268ac03f3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "7c1a27eb96ad8cdd2c821c494bacb44c54fadf5885af8f2eed1f17307377806f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-4d9268ac03f3", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:12:14Z", "seed": 1354613978, "treatment": "frontier"}, "priority": 0.1463449800481896, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "23897c9be037208e53fe9187cbe7a81b8d98266e527acbc95185bd78a3d68007", "run_seconds": 0.5026223659515381, "seed": 1354613978, "started_at": "2026-11-22T04:36:57Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:12Z", "cpu_seconds": 0.36309344500000407, "created_at": "2026-11-22T03:12:14Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-8fbfda1be829", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "db8ee2391f91e9db3c98d1532b4520d35bd606e9fd8bbfd13250e5b3b84a43b8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-8fbfda1be829", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:12:14Z", "seed": 43673010, "treatment": "frontier"}, "priority": 0.1478907473111331, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "2d960382135a62f2ab4facc4c27ca951753f7048c44d8099150446dccae8d1fc", "run_seconds": 0.4715263843536377, "seed": 43673010, "started_at": "2026-11-22T04:36:12Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:13:37Z", "cpu_seconds": 1.7529385199999865, "created_at": "2026-11-22T03:12:31Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-4109425bce32", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "db49877980c198bd7d79c29d162ad71c1738a9ae2ec60a060019e23475c60a63", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-4109425bce32", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:31Z", "seed": 566133470, "treatment": "iw_temperature"}, "priority": 0.16428072132429494, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a47d1cdae746ce08429570317453502c09d1eaebcae5b0bf75e027471c7a3635", "run_seconds": 1.80898118019104, "seed": 566133470, "started_at": "2026-11-22T03:13:35Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.004993, 0.00436], "effect": -0.000316, "evaluated_at": "2026-11-22T03:14:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-4109425bce32", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.047775098, 0.034817785, 0.029531697, 0.066571925, 0.0649527, 0.070920662, 0.031707428, 0.051366437, 0.029179905, 0.053227073, 0.049802408, 0.066717327, 0.04737609, 0.0271407, 0.051922753, 0.03794678], "raw": [0.051293765, 0.062969632, 0.068029787, 0.045383545, 0.04966327, 0.068256462, 0.05010834, 0.037702455, 0.06735289, 0.03083715, 0.076056557, 0.08763113, 0.04737609, 0.03206565, 0.036840275, 0.0567627], "temperature": [0.043037135, 0.029212775, 0.039397622, 0.039337482, 0.060498303, 0.070920662, 0.031707428, 0.051366437, 0.033911715, 0.053227073, 0.067294113, 0.073850792, 0.04019815, 0.03206565, 0.051922753, 0.03794678]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.2055, "losses": 5, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.623", "prereg_hash": "db49877980c198bd7d79c29d162ad71c1738a9ae2ec60a060019e23475c60a63", "primary_control": "temperature", "result_hash": "a47d1cdae746ce08429570317453502c09d1eaebcae5b0bf75e027471c7a3635", "signature": "6923e50fd22e07e35add560838be6f8526f278508c1233e8782685497fa82fe63b1f8e0772124260725407f31a47284d1b978daeb004d96794098881bbc54f08", "state": "VERIFIED", "summary": {"iw_temperature": 0.04756, "raw": 0.054271, "temperature": 0.047243}, "ties": 6, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:14:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:36Z", "cpu_seconds": 1.3933652389999907, "created_at": "2026-11-22T03:12:31Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-511247d5106b", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "a0f8a3fff9d60fb521e8d6bd36f7ddd17598ef4a7ef351a510011e8b75022260", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-511247d5106b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:31Z", "seed": 808801842, "treatment": "iw_temperature"}, "priority": 0.3399828081547421, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-9b3a759205e8", "result_hash": "374b57f9ba394546c3ec8aee337131c827ab35d8e9d26bc9c1d85a545a55803d", "run_seconds": 1.8908929824829102, "seed": 808801842, "started_at": "2026-11-22T03:12:34Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012032, 0.022401], "effect": 0.005185, "evaluated_at": "2026-11-22T03:12:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-511247d5106b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.192545773, 0.21156247, 0.278877235, 0.112068125, 0.264315133, 0.236402887, 0.253917945, 0.290949597, 0.21506899, 0.246620983, 0.222974343, 0.085002085, 0.251021617, 0.220590627, 0.226062255, 0.303922755], "raw": [0.2296048, 0.206038283, 0.24759096, 0.112068125, 0.264315133, 0.213300798, 0.17603039, 0.267922197, 0.173310492, 0.246620983, 0.270492843, 0.148304168, 0.2431944, 0.229015058, 0.226062255, 0.257747345], "temperature": [0.240446118, 0.171585537, 0.27067498, 0.166979797, 0.264315133, 0.236402887, 0.190375377, 0.290949597, 0.21506899, 0.246620983, 0.24477422, 0.170064395, 0.251021617, 0.220590627, 0.226062255, 0.288922588]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1961, "losses": 4, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.637", "prereg_hash": "a0f8a3fff9d60fb521e8d6bd36f7ddd17598ef4a7ef351a510011e8b75022260", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-9b3a759205e8", "result_hash": "374b57f9ba394546c3ec8aee337131c827ab35d8e9d26bc9c1d85a545a55803d", "signature": "17842b780a711e78201dd65583274354e94053d5df884927dfd2447351eef2111fc27474e67bf85a472b608da4f2e83350469723a271d0558a167d8a93304904", "state": "VERIFIED", "summary": {"iw_temperature": 0.225744, "raw": 0.219476, "temperature": 0.230928}, "ties": 8, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:12:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:37Z", "cpu_seconds": 4.820201711999971, "created_at": "2026-11-22T03:12:31Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-66f68b59e6b7", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ea9f0c9708fcc11d77b1bd400d21e240c1947a0599b90e41b1c7c223d069adef", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-66f68b59e6b7", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:12:31Z", "seed": 1042761613, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a82e7605ca33cd0eb847a2d4305097cbc33a83a8227b6da95607d67fe9e7fe53", "run_seconds": 5.825882196426392, "seed": 1042761613, "started_at": "2026-11-22T03:12:31Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.079671, 0.213004], "effect": 0.066667, "evaluated_at": "2026-11-22T03:12:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-66f68b59e6b7", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4219656083136083, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.5991, "losses": 8, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.252", "prereg_hash": "ea9f0c9708fcc11d77b1bd400d21e240c1947a0599b90e41b1c7c223d069adef", "primary_control": "reset", "result_hash": "a82e7605ca33cd0eb847a2d4305097cbc33a83a8227b6da95607d67fe9e7fe53", "signature": "e93650197154cc9d6a72ef89a20626e9320c37b486621a446490f22ab44775dc984e1a436844b7289022cc04d375a084df8fcc89b452bd0f3e8390b2dbbdb503", "state": "VERIFIED", "summary": {"inherit": 0.6833333333333333, "reset": 0.6166666666666667, "root": 0.26666666666666666}, "ties": 40, "treatment": "inherit", "wins": 12}, "verified_at": "2026-11-22T03:12:47Z"}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:12:31Z", "error": null, "family": "world_model_frontier", "fence": 0, "id": "E141-world_model_frontier-ce644decd5b2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "e60ed3f6b57778d881f47adad0c668a73553a07860cf14f3da8ac85199df7887", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-ce644decd5b2", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:12:31Z", "seed": 390269440, "treatment": "frontier"}, "priority": 0.14331273060551386, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 390269440, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:25Z", "cpu_seconds": 1.5679745890000163, "created_at": "2026-11-22T03:12:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-21ca45df2103", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "76fa1f2e03d13e7ec0d5656022fe020d26cf98b32d0c11551a344ed91c7ada52", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-21ca45df2103", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:47Z", "seed": 1551142914, "treatment": "iw_temperature"}, "priority": 0.14731580145885412, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "1f06292fa9142c3e093844d32bd3c38488e70027eee0012c54e18003c312f17d", "run_seconds": 1.7388038635253906, "seed": 1551142914, "started_at": "2026-11-22T04:36:23Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:14:02Z", "cpu_seconds": 1.7816361120000002, "created_at": "2026-11-22T03:12:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-3e8559f3339b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "2468f71eabb57e7f87fbca94605d3e2146f2d46ebffe9ded030c0faac3159bce", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-3e8559f3339b", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:47Z", "seed": 1435944629, "treatment": "iw_temperature"}, "priority": 0.1621072447730142, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "22fe4223a6ce937c7e8d6efd355e3e5b35d4e4f418e0018535495d0af3515eeb", "run_seconds": 2.428781747817993, "seed": 1435944629, "started_at": "2026-11-22T03:13:59Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.016908, 0.004386], "effect": -0.006261, "evaluated_at": "2026-11-22T03:14:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-3e8559f3339b", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.228352225, 0.182444135, 0.238835795, 0.246597363, 0.252250647, 0.233046245, 0.199841765, 0.198303632, 0.246908475, 0.193972697, 0.24144491, 0.162167937, 0.225471878, 0.151904465, 0.21340806, 0.216247868], "raw": [0.16373668, 0.205379085, 0.232705183, 0.209929033, 0.198022895, 0.197197605, 0.166165492, 0.140244105, 0.223988843, 0.17664491, 0.2151029, 0.153007388, 0.186029882, 0.208675973, 0.184592233, 0.184922285], "temperature": [0.19276496, 0.22187397, 0.232705183, 0.246597363, 0.225054077, 0.221237623, 0.166165492, 0.18349364, 0.223988843, 0.197554257, 0.24144491, 0.147785525, 0.225471878, 0.19185423, 0.21340806, 0.199623008]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0085, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.981", "prereg_hash": "2468f71eabb57e7f87fbca94605d3e2146f2d46ebffe9ded030c0faac3159bce", "primary_control": "temperature", "result_hash": "22fe4223a6ce937c7e8d6efd355e3e5b35d4e4f418e0018535495d0af3515eeb", "signature": "022976b366226b32471b22034d3c61b974372d00bc3a16b487e954bbabdfe078e396cb05eac8d07cd4e669f06d229277f2f0441dff78388dc082ebe8b836960b", "state": "VERIFIED", "summary": {"iw_temperature": 0.21445, "raw": 0.190397, "temperature": 0.208189}, "ties": 4, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:14:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:12:51Z", "cpu_seconds": 1.651256138000008, "created_at": "2026-11-22T03:12:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-724ad89ff887", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "bba7d170490eb4d13b7865c11fc9ab64fc4dfc8d8f1dc8f9ba75f1c627933bd2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-724ad89ff887", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:47Z", "seed": 462028557, "treatment": "iw_temperature"}, "priority": 0.3083519885349403, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "5a96824aae5447ec6a439f706a00c4548fdc3a2ff1235a0860fb69478a1904e9", "run_seconds": 1.975268840789795, "seed": 462028557, "started_at": "2026-11-22T03:12:49Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.022503, 0.041193], "effect": 0.009345, "evaluated_at": "2026-11-22T03:12:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-724ad89ff887", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.21071032, 0.14213145, 0.2692499, 0.10387615, 0.25201186, 0.226318137, 0.15783019, 0.094572518, 0.335695425, 0.282172242, 0.288631655, 0.30052691, 0.20750282, 0.306853767, 0.149996732, 0.241439065], "raw": [0.18079638, 0.19271724, 0.200445987, 0.130569553, 0.224351195, 0.226318137, 0.185538635, 0.24198728, 0.32466143, 0.246801373, 0.23256574, 0.292843858, 0.1864204, 0.292169715, 0.245500635, 0.182966995], "temperature": [0.148639433, 0.1645775, 0.224563378, 0.118822032, 0.224351195, 0.20696448, 0.234145875, 0.285898648, 0.33359368, 0.270118735, 0.26547055, 0.308270173, 0.171227355, 0.305052147, 0.254995763, 0.202347027]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0482, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.895", "prereg_hash": "bba7d170490eb4d13b7865c11fc9ab64fc4dfc8d8f1dc8f9ba75f1c627933bd2", "primary_control": "temperature", "result_hash": "5a96824aae5447ec6a439f706a00c4548fdc3a2ff1235a0860fb69478a1904e9", "signature": "cb548bd3b81bcbace6dfc59334e8dc15ec7bb75009854d97cd51d8d0f30633e5344ffc7da7fead136fdd124f4b5f540c4f6e1ad10a8501530a84371444ec2004", "state": "VERIFIED", "summary": {"iw_temperature": 0.223095, "raw": 0.224166, "temperature": 0.23244}, "ties": 0, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:12:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:35:29Z", "cpu_seconds": 1.564566192000001, "created_at": "2026-11-22T03:12:47Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-7a351e981ad7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "102f801262db7daf9f7cb80e4be6cbeeea4bf0a5c6279cc02fd780140550d1d0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-7a351e981ad7", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:12:47Z", "seed": 1531252524, "treatment": "iw_temperature"}, "priority": 0.15357796278969077, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "ef48274ef71603eebee800d424a769f5b4661d086183c9fb7b7b22adb8bd4e55", "run_seconds": 1.689026117324829, "seed": 1531252524, "started_at": "2026-11-22T04:35:27Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:12:55Z", "cpu_seconds": 6.915588846999981, "created_at": "2026-11-22T03:12:47Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-ec8a3a628fc6", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "b405c0c98e3fa9b77178dfad1f3687d596f3f5bd769aaaf8cf85b01fc72fd60a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-ec8a3a628fc6", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:12:47Z", "seed": 114988786, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a3299a1c3cef5847bc2eb8784d1d1031d09ce678ce64f6817a30ce09f3610605", "run_seconds": 7.506848573684692, "seed": 114988786, "started_at": "2026-11-22T03:12:48Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:12:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-ec8a3a628fc6", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b405c0c98e3fa9b77178dfad1f3687d596f3f5bd769aaaf8cf85b01fc72fd60a", "primary_control": "reset", "result_hash": "a3299a1c3cef5847bc2eb8784d1d1031d09ce678ce64f6817a30ce09f3610605", "signature": "fcfe9ee8404b6a250e429bc760ed954bffeb3304e7d539a2bf953d39c34a49def62b70bdbd72110b9ce703d4da9cc7857a4e9d400d9d7e024b6c06697d9e9602", "state": "PARTIAL", "summary": {"inherit": 0.6, "reset": 0.6, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:13:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:09Z", "cpu_seconds": 1.6632432199999982, "created_at": "2026-11-22T03:13:03Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-25ab6933d284", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "4d41336ab583c040900279f38ced2c91f4fc76cf1da54a7bf7de656a57e1c19c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-25ab6933d284", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:13:03Z", "seed": 867789471, "treatment": "iw_temperature"}, "priority": 0.29402876831113833, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "95e035c3b6849559f63bcf93cecc61050f62671ad22c423fcdeb375050f2ec2f", "run_seconds": 1.7268459796905518, "seed": 867789471, "started_at": "2026-11-22T03:13:07Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021354, -0.000637], "effect": -0.010996, "evaluated_at": "2026-11-22T03:13:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-25ab6933d284", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.233064278, 0.270593068, 0.240195005, 0.232540255, 0.333118392, 0.221850873, 0.330710565, 0.293851268, 0.282156695, 0.266873352, 0.299525027, 0.21583573, 0.254870793, 0.193797208, 0.290685443, 0.137443482], "raw": [0.197446342, 0.227775552, 0.199989118, 0.232540255, 0.237478573, 0.20887068, 0.285162835, 0.250300138, 0.211992952, 0.247858083, 0.275236425, 0.171268487, 0.246714987, 0.14272207, 0.269461497, 0.150025783], "temperature": [0.222476985, 0.24883927, 0.229788015, 0.267541918, 0.282163602, 0.221850873, 0.31721598, 0.250300138, 0.267165153, 0.266873352, 0.29424336, 0.17521331, 0.263612773, 0.193797208, 0.282654555, 0.137443482]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0014, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.997", "prereg_hash": "4d41336ab583c040900279f38ced2c91f4fc76cf1da54a7bf7de656a57e1c19c", "primary_control": "temperature", "result_hash": "95e035c3b6849559f63bcf93cecc61050f62671ad22c423fcdeb375050f2ec2f", "signature": "5441b93d7d653541b993d14282518ea4b647aa9ee3ce712bb53afba2d444ec949c26b92bfcd509d0bc23ae18cb7ca36034ee93aa5ea38fa9520001fc9ecb4f0d", "state": "VERIFIED", "summary": {"iw_temperature": 0.256069, "raw": 0.222178, "temperature": 0.245074}, "ties": 4, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:13:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:12Z", "cpu_seconds": 1.5480884010000011, "created_at": "2026-11-22T03:13:03Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b959b8c34711", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "836565c82c44ea10498d8ce96a32ea4589d970fc0eb0e45069326ad684237322", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b959b8c34711", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:13:03Z", "seed": 439295068, "treatment": "iw_temperature"}, "priority": 0.1539429105785397, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "6af88b75d2ab8f1d58334aa05d21195b0caa64d1921fd2fc2e38f7e208cd1540", "run_seconds": 1.557598352432251, "seed": 439295068, "started_at": "2026-11-22T03:20:10Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.02748, -0.005461], "effect": -0.01647, "evaluated_at": "2026-11-22T03:22:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b959b8c34711", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.122744888, 0.160221395, 0.137047805, 0.134089045, 0.141305648, 0.15568945, 0.148722513, 0.12626131, 0.179900295, 0.121875817, 0.159637617, 0.174111397, 0.177111402, 0.181701725, 0.15769843, 0.12047181], "raw": [0.093009922, 0.10052347, 0.120902668, 0.113646412, 0.108813952, 0.122481483, 0.12650949, 0.139998918, 0.131471805, 0.093073665, 0.121631105, 0.140897522, 0.10662356, 0.196909775, 0.106222602, 0.099286165], "temperature": [0.12162572, 0.131560663, 0.137047805, 0.12613651, 0.069914937, 0.129623983, 0.148722513, 0.139998918, 0.131471805, 0.093073665, 0.15184142, 0.13559389, 0.177111402, 0.181701725, 0.13916674, 0.12047181]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0002, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "836565c82c44ea10498d8ce96a32ea4589d970fc0eb0e45069326ad684237322", "primary_control": "temperature", "result_hash": "6af88b75d2ab8f1d58334aa05d21195b0caa64d1921fd2fc2e38f7e208cd1540", "signature": "4f5da8e6011ce858f0127c7dda49968b15dadb76b905cf617107a036da200115c361f76229c2ce948ec7274da688910d8b2aa1214a9a1721fd22fb73556aff0a", "state": "VERIFIED", "summary": {"iw_temperature": 0.149912, "raw": 0.120125, "temperature": 0.133441}, "ties": 5, "treatment": "iw_temperature", "wins": 1}, "verified_at": "2026-11-22T03:22:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:10Z", "cpu_seconds": 1.4525916160000065, "created_at": "2026-11-22T03:13:03Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f85a5ee8b0c9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "638263f03b8e64ecfe1a87bea3cfa2fd70877ad1e6ae3f933bf5039bce621d53", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f85a5ee8b0c9", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:13:03Z", "seed": 2139654198, "treatment": "iw_temperature"}, "priority": 0.1693821489836485, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "2dea43c2b48912421de8b9c5caad34747e243ce5dd11db5edb471bde56cb5cf3", "run_seconds": 1.45680570602417, "seed": 2139654198, "started_at": "2026-11-22T03:13:08Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.00441, 0.014999], "effect": 0.005294, "evaluated_at": "2026-11-22T03:13:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f85a5ee8b0c9", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.087515918, 0.143411632, 0.123716767, 0.117061115, 0.10229203, 0.162891565, 0.149111038, 0.071020355, 0.116953835, 0.046527825, 0.118710527, 0.134539265, 0.12771688, 0.121640365, 0.22927125, 0.1385039], "raw": [0.098218185, 0.12056334, 0.123716767, 0.108622713, 0.092885465, 0.162891565, 0.147820185, 0.087321547, 0.09426098, 0.085527443, 0.108399587, 0.108608505, 0.12771688, 0.105997008, 0.132495813, 0.107454742], "temperature": [0.087515918, 0.143411632, 0.133732387, 0.108622713, 0.113288048, 0.181929697, 0.149111038, 0.119918435, 0.116953835, 0.072681005, 0.122316387, 0.134539265, 0.16120023, 0.11432254, 0.19721842, 0.118834703]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5616, "losses": 4, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.274", "prereg_hash": "638263f03b8e64ecfe1a87bea3cfa2fd70877ad1e6ae3f933bf5039bce621d53", "primary_control": "temperature", "result_hash": "2dea43c2b48912421de8b9c5caad34747e243ce5dd11db5edb471bde56cb5cf3", "signature": "dbad12e503d7f31bd2ba3be7a45549a687b5ff353a59c787d1b49d55b27eae6fd7a9b977012b70ec23b03d1a47481f76880fc5efa7e226bf7ece3bb63bd5b002", "state": "VERIFIED", "summary": {"iw_temperature": 0.12443, "raw": 0.113281, "temperature": 0.129725}, "ties": 5, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:13:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:11Z", "cpu_seconds": 6.442867253000031, "created_at": "2026-11-22T03:13:03Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3b25337212e1", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "2fc14e96ce542155e8014dfaa25c8f89cff88aaf530d53fafff6c5d2c414e0d0", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3b25337212e1", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:13:03Z", "seed": 1219649193, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "14dd60c43d75ed750b953823f31d2349022902ce4a13bf7ebb661752274d8ce2", "run_seconds": 6.511362314224243, "seed": 1219649193, "started_at": "2026-11-22T03:13:04Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.046174, 0.179507], "effect": 0.066667, "evaluated_at": "2026-11-22T03:13:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3b25337212e1", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5448576311721528, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.7125, "losses": 4, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.194", "prereg_hash": "2fc14e96ce542155e8014dfaa25c8f89cff88aaf530d53fafff6c5d2c414e0d0", "primary_control": "reset", "result_hash": "14dd60c43d75ed750b953823f31d2349022902ce4a13bf7ebb661752274d8ce2", "signature": "70162116d97b6ed2aa9962dec0546c3bc975636865c275e667046402337f4b587c16040d30076e4de705fc3fecb8dd809ba3cc4f5577dd320c11910e30218d0f", "state": "VERIFIED", "summary": {"inherit": 0.6666666666666666, "reset": 0.6, "root": 0.38333333333333336}, "ties": 48, "treatment": "inherit", "wins": 8}, "verified_at": "2026-11-22T03:13:19Z"}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:13:03Z", "error": null, "family": "world_model_frontier", "fence": 0, "id": "E141-world_model_frontier-f8a6db16527a", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "66d3793347547dc18d71cb642f5b9a87801eb12c641847455453142ca9e6e1ed", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-f8a6db16527a", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:13:03Z", "seed": 723929832, "treatment": "frontier"}, "priority": 0.13971576930700302, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 723929832, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:13:28Z", "cpu_seconds": 1.6685198579999962, "created_at": "2026-11-22T03:13:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-fa4fc30ea3db", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "4784a68041dc4882ea5664ca59c52c2d69eecd7f8ad9bcb6644d39ccd0cdd4c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-fa4fc30ea3db", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:13:20Z", "seed": 1347331554, "treatment": "iw_temperature"}, "priority": 0.293843181019726, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "a0305acdf19f25c1ea6d357ba6f53af542b55c8d3cbab5b13251a44720f9f1e3", "run_seconds": 1.7196640968322754, "seed": 1347331554, "started_at": "2026-11-22T03:13:26Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011471, 0.038076], "effect": 0.013303, "evaluated_at": "2026-11-22T03:13:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-fa4fc30ea3db", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.342307412, 0.034455172, 0.20849484, 0.279963642, 0.235136995, 0.242302438, 0.270620532, 0.24415904, 0.238029183, 0.279452288, 0.317061568, 0.114101117, 0.22204009, 0.280357643, 0.284060527, 0.290873468], "raw": [0.300955553, 0.223266207, 0.193147773, 0.26688491, 0.176372357, 0.242302438, 0.192570148, 0.206306135, 0.195973235, 0.24191344, 0.311082983, 0.17302011, 0.208293402, 0.268505305, 0.25687264, 0.2785688], "temperature": [0.328528655, 0.228015503, 0.24020263, 0.27533836, 0.205795857, 0.247807505, 0.270620532, 0.24415904, 0.260733695, 0.268899692, 0.317061568, 0.114101117, 0.2514274, 0.280357643, 0.273675305, 0.289534605]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1393, "losses": 6, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.726", "prereg_hash": "4784a68041dc4882ea5664ca59c52c2d69eecd7f8ad9bcb6644d39ccd0cdd4c6", "primary_control": "temperature", "result_hash": "a0305acdf19f25c1ea6d357ba6f53af542b55c8d3cbab5b13251a44720f9f1e3", "signature": "7e9a2367977ff09372818c5f0414e785691f5aecaa9546b1fe886ec60d8fdbf654a47722807474438c8bdfdc257a70ddc77430cf223e2ddc03393c5987badb03", "state": "VERIFIED", "summary": {"iw_temperature": 0.242713, "raw": 0.233502, "temperature": 0.256016}, "ties": 5, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:13:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:30Z", "cpu_seconds": 6.098139414000002, "created_at": "2026-11-22T03:13:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-57fb7ea361ca", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e97a067e8087dfb06caae8410e6c33eeaf4564ccdc79442917bae09404110bca", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-57fb7ea361ca", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:13:20Z", "seed": 4461659, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "65203b875de1f4f2d9ba40900fedfdad46dafda8b68461c7d5d85c45925a1f00", "run_seconds": 6.179919242858887, "seed": 4461659, "started_at": "2026-11-22T03:13:24Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.121356, 0.378644], "effect": 0.25, "evaluated_at": "2026-11-22T03:13:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-57fb7ea361ca", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5533667905633937, "rstep": 2}, "hypothesis_supported": true, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0], "root": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -3.4385, "losses": 2, "n": 60, "outcome": "SUPPORTED", "p_sign_display": "0.000364", "prereg_hash": "e97a067e8087dfb06caae8410e6c33eeaf4564ccdc79442917bae09404110bca", "primary_control": "reset", "result_hash": "65203b875de1f4f2d9ba40900fedfdad46dafda8b68461c7d5d85c45925a1f00", "signature": "89941dd90c95323c4ef56551af2ff31346a99096957fadcfafc0a623eb19fb9588aa08565ce3ebcf3173186282e0423bfeebf7a56cfd2c00bcf483ddbddec406", "state": "VERIFIED", "summary": {"inherit": 0.9166666666666666, "reset": 0.6666666666666666, "root": 0.43333333333333335}, "ties": 41, "treatment": "inherit", "wins": 17}, "verified_at": "2026-11-22T03:14:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:13:29Z", "cpu_seconds": 7.891224985999997, "created_at": "2026-11-22T03:13:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d4d7bfade24c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "b6afe51f99e8c11166802505755bc3ddc11eb031e76d2406d586c6f3386bd2be", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d4d7bfade24c", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:13:20Z", "seed": 131938813, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "cbb7a3156171b17bf2f4496f70116283855a5c3b704c6b62e7d188247a826cf6", "run_seconds": 8.022720575332642, "seed": 131938813, "started_at": "2026-11-22T03:13:21Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:13:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d4d7bfade24c", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], "root": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b6afe51f99e8c11166802505755bc3ddc11eb031e76d2406d586c6f3386bd2be", "primary_control": "reset", "result_hash": "cbb7a3156171b17bf2f4496f70116283855a5c3b704c6b62e7d188247a826cf6", "signature": "125e64887fbcf6c92e2c79766810ee807ad19868ce6b2c03adbec7159a351857f876919ba5ab989a8bc9b7cfb226bdfd3195942dd6dbbada8062c28aa3ba260f", "state": "PARTIAL", "summary": {"inherit": 0.6333333333333333, "reset": 0.6333333333333333, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:13:37Z"}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:13:20Z", "error": null, "family": "world_model_frontier", "fence": 0, "id": "E141-world_model_frontier-2ea959dfad22", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "05d0ebb828459985308aa49894c9c116dddb51e1d54b3fc58223654c99a2d937", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-2ea959dfad22", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:13:20Z", "seed": 1464390334, "treatment": "frontier"}, "priority": 0.14308773662170468, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 1464390334, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:13:20Z", "error": null, "family": "world_model_frontier", "fence": 0, "id": "E141-world_model_frontier-d9e52e984b3a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "9638d6997c33921150e3f57b622eef42103875644be855aa43f4868b308b5e99", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-d9e52e984b3a", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:13:20Z", "seed": 853236424, "treatment": "frontier"}, "priority": 0.145615760250983, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 853236424, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:54Z", "cpu_seconds": 0.19195600899999476, "created_at": "2026-11-22T03:13:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-ebbe8fa942ec", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "7263719adb076a190ddc126e368ab59af7be70ff8b3746f53d71f03ba60afc08", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-ebbe8fa942ec", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:13:20Z", "seed": 1601899053, "treatment": "frontier"}, "priority": 0.14959571304476682, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "a0b65d6a68019e6831d1edc8a42bb0d7c77140c6301b80d2e0c4503a52df7085", "run_seconds": 0.23413872718811035, "seed": 1601899053, "started_at": "2026-11-22T04:35:54Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:13:36Z", "cpu_seconds": 2.7030432759999883, "created_at": "2026-11-22T03:13:20Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-fb4b6cdb3bc2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "5098c2149d9baeb627dd5e6f27c25703fa05f6f25a17bd10d3e97fcd72cfd68a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-fb4b6cdb3bc2", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:13:20Z", "seed": 119657376, "treatment": "frontier"}, "priority": 0.261728513079992, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": null, "result_hash": "9a2f6b97b71db82c26b777551313eec7e2f6a4b8b6193407f0f0d8737d77132f", "run_seconds": 2.8227505683898926, "seed": 119657376, "started_at": "2026-11-22T03:13:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.027282, 0.087923], "effect": 0.057602, "evaluated_at": "2026-11-22T03:14:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-fb4b6cdb3bc2", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.89010989, 0.86, 1.0, 0.90060241, 0.86, 0.829207921, 1.0, 1.0, 0.821428571, 0.860091743, 1.0, 1.0], "frontier": [0.986263736, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.901098901, 0.873333333, 1.0, 0.930722892, 0.8775, 0.920792079, 1.0, 1.0, 0.883333333, 0.908256881, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -2.1072, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00781", "prereg_hash": "5098c2149d9baeb627dd5e6f27c25703fa05f6f25a17bd10d3e97fcd72cfd68a", "primary_control": "random_walk", "result_hash": "9a2f6b97b71db82c26b777551313eec7e2f6a4b8b6193407f0f0d8737d77132f", "signature": "0acf826eb8f9e1d2a7c8071161ce4206126664638e3304923c7c30761f85c856274e6ed03c8c6074e162a67e8866523de7d8af35e6bc0a3cedd4fa5c9a86d208", "state": "VERIFIED", "summary": {"curiosity": 0.918453, "frontier": 0.998855, "random_walk": 0.941253}, "ties": 5, "treatment": "frontier", "wins": 7}, "verified_at": "2026-11-22T03:14:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:36:05Z", "cpu_seconds": 1.4996791290000004, "created_at": "2026-11-22T03:13:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-d26cd1a50347", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "69ad78878432c4f970e2979636b943c7c87f98bbc99ad2e6e4eff16508d3528c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-d26cd1a50347", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:13:36Z", "seed": 125516871, "treatment": "iw_temperature"}, "priority": 0.1485198832528437, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": "2c9ff5a3faeb2a3dde5739cb997ec24e76c3204d8a2aebe319594762cace0016", "run_seconds": 2.881742477416992, "seed": 125516871, "started_at": "2026-11-22T04:36:02Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:13:36Z", "error": null, "family": "calibration_iw", "fence": 0, "id": "E141-calibration_iw-dfba88ab19d4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "8bde69ada28cb7d3aa32d8e631880e832658c68c7ec0ccc0310c9bd07b515c70", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-dfba88ab19d4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:13:36Z", "seed": 1194393702, "treatment": "iw_temperature"}, "priority": 0.13600413321801408, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 1194393702, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:13:36Z", "error": null, "family": "calibration_iw", "fence": 0, "id": "E141-calibration_iw-e43da6d6aa42", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "45e296bfac23d525c27d55b142e3c0d37530186aaf252b76ad6892758f139a8e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-e43da6d6aa42", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:13:36Z", "seed": 776099844, "treatment": "iw_temperature"}, "priority": 0.1359646974407672, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 776099844, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:15:18Z", "cpu_seconds": 68.847294066, "created_at": "2026-11-22T03:13:59Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-a2afa01fc203", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "f4325cf0cb1ae962fa871885f60c9ec8f32352457a50900921120ecb7348de43", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-a2afa01fc203", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:13:59Z", "seed": 365171202, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "57aea52643c151083366fd472c64f16d4b897107dc5b83b12ce6e5ef2b38ad15", "run_seconds": 78.86557364463806, "seed": 365171202, "started_at": "2026-11-22T03:13:59Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 1, "inherit_mem": 1, "inherit_nomem": 1, "random_search": 1, "reset_mem": 1, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [-0.211563, 0.033842], "effect": -0.08886, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:15:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-a2afa01fc203", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.69}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"hypdisc": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.879310345, 1.0, 0.98, 0.714869281, 0.612588652, 0.795454545, 0.905797101, 0.221682848, 0.931372549, 0.796385542], "compressed": [0.362068966, 1.0, 0.98, 0.310457516, 0.612588652, 0.806818182, 0.905797101, 0.221682848, 0.931372549, 0.818072289], "inherit_mem": [0.362068966, 1.0, 0.98, 0.310457516, 0.612588652, 0.806818182, 0.905797101, 0.221682848, 0.931372549, 0.818072289], "inherit_nomem": [0.362068966, 1.0, 0.98, 0.310457516, 0.612588652, 0.806818182, 0.905797101, 0.221682848, 0.931372549, 0.818072289], "random_search": [0.642241379, 1.0, 0.986666667, 0.721405229, 0.616489362, 0.909090909, 0.905797101, 0.346682848, 0.931372549, 0.806024096], "reset_mem": [0.362068966, 1.0, 0.98, 0.310457516, 0.612588652, 0.806818182, 0.905797101, 0.221682848, 0.931372549, 0.818072289], "retrieval": [0.879310345, 1.0, 0.98, 0.714869281, 0.612588652, 0.795454545, 0.905797101, 0.221682848, 0.931372549, 0.796385542], "rootcause": [0.362068966, 1.0, 0.98, 0.310457516, 0.612588652, 0.806818182, 0.905797101, 0.221682848, 0.931372549, 0.818072289], "xfer_mem": [0.642241379, 1.0, 0.986666667, 0.81372549, 0.49751773, 0.772727273, 0.920289855, 0.243527508, 0.908496732, 0.819277108]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "per_generation_sealed": [0.783746, 0.783746, 0.694886, 0.694886], "prereg_hash": "f4325cf0cb1ae962fa871885f60c9ec8f32352457a50900921120ecb7348de43", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "compressed", "xfer_mem"], "result_hash": "57aea52643c151083366fd472c64f16d4b897107dc5b83b12ce6e5ef2b38ad15", "signature": "a39632b42a7d28f9734ea3cc064bd9bf6af1c55d3a5cbe9553a17b3ce93dc712bbdc5690bec7ba84cd4a2536a028275dccddc8c9107a81a9d51b31688268570d", "state": "VERIFIED", "summary": {"G0": 0.783746, "compressed": 0.694886, "inherit_mem": 0.694886, "inherit_nomem": 0.694886, "random_search": 0.786577, "reset_mem": 0.694886, "retrieval": 0.783746, "rootcause": 0.694886, "xfer_mem": 0.760447}, "ties": 6, "treatment": "inherit_mem", "wins": 2}, "verified_at": "2026-11-22T03:16:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:09Z", "cpu_seconds": 8.289584508, "created_at": "2026-11-22T03:13:59Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-531d174faa7a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d85a58c26d79b088367494d2417f2d80d39f0a82e1ae8d5d180f98bd0ef3be53", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-531d174faa7a", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:13:59Z", "seed": 612008659, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "1225ecfe6c31f32e2ae415dd0e5db640e869b4ba00401af5cce8a34b09b0d9cb", "run_seconds": 9.906382322311401, "seed": 612008659, "started_at": "2026-11-22T03:13:59Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:14:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-531d174faa7a", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.591034739058577, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0], "reset": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0], "root": [1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d85a58c26d79b088367494d2417f2d80d39f0a82e1ae8d5d180f98bd0ef3be53", "primary_control": "reset", "result_hash": "1225ecfe6c31f32e2ae415dd0e5db640e869b4ba00401af5cce8a34b09b0d9cb", "signature": "8d95a26ed0cd6ba76392a3a409626417d667a109c2da5d6bbd1853fabeece5e240b6ecbfa9ed7e6a029ba1d605c9247ea0f0ddbc22e7e1e50d8afa9f8dcd4d0f", "state": "PARTIAL", "summary": {"inherit": 0.6, "reset": 0.6, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:14:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:17Z", "cpu_seconds": 2.9393323569999996, "created_at": "2026-11-22T03:14:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a781ac4a89eb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "f7d6adcfe56d28f5772a991939ec91d594d897f1b8d8c5fde2bf97088559a343", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a781ac4a89eb", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:14:14Z", "seed": 1405355257, "treatment": "full"}, "priority": 2.2954872459356697, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8b8dee2bcc247ea0c7f93194528c8c4b6c5b1f52f8bac6c026bd63b1fdcb3e32", "run_seconds": 3.1584887504577637, "seed": 1405355257, "started_at": "2026-11-22T03:14:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.094139, 0.844139], "effect": 0.375, "evaluated_at": "2026-11-22T03:14:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a781ac4a89eb", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 0.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "f7d6adcfe56d28f5772a991939ec91d594d897f1b8d8c5fde2bf97088559a343", "primary_control": "e141_best", "result_hash": "8b8dee2bcc247ea0c7f93194528c8c4b6c5b1f52f8bac6c026bd63b1fdcb3e32", "signature": "7e07036a663d18d4f7b78162465f8f495a2e1e83e0eb5a534b7b1db843aa35d7638a9b262d0ad2983bc1bb3816d85a2fb7730beff65cff04ce4444cef4a4100d", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:14:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:24Z", "cpu_seconds": 1.7921919299999995, "created_at": "2026-11-22T03:14:14Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-63ba48569a3f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "01c58b3eda78e232ab7dbb6011bcd2831e8e82ba3a5ab5b73e0d171f646cc351", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-63ba48569a3f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:14:14Z", "seed": 299763484, "treatment": "full"}, "priority": 2.2794974311811944, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "46cc29783adfc9e9172a4c8f11eff6876aa0418dbc2fc5810603f1428bb13881", "run_seconds": 3.02689528465271, "seed": 299763484, "started_at": "2026-11-22T03:14:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.007639, 0.318441], "effect": 0.16304, "evaluated_at": "2026-11-22T03:14:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-63ba48569a3f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.727472527, 0.655394525, 0.293309438, 0.125], "frozen": [0.142857143, 0.293880837, 0.215053763, 0.0], "full": [0.731868132, 0.88647343, 0.645758662, 0.189236111], "retrieval": [0.745054945, 0.934782609, 0.75567503, 0.625], "rules_nodetect": [0.386813187, 0.562801932, 0.261051374, 0.295138889]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "01c58b3eda78e232ab7dbb6011bcd2831e8e82ba3a5ab5b73e0d171f646cc351", "primary_control": "facts", "result_hash": "46cc29783adfc9e9172a4c8f11eff6876aa0418dbc2fc5810603f1428bb13881", "signature": "51a1ce43791ae1be79ab8a58c748ad975ec72c14be243e7bf34a346dd41f3784cf3eb50008499da0c2902f8d45e43faa946658d91dc70d80d4759a2b23437e01", "state": "VERIFIED", "summary": {"facts": 0.450294, "frozen": 0.162948, "full": 0.613334, "retrieval": 0.765128, "rules_nodetect": 0.376451}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:14:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:25Z", "cpu_seconds": 1.766292129, "created_at": "2026-11-22T03:14:14Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5dc2178957ef", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "637c4540d70c486ee425aebdce95eb276ccdcd91eb92c0b097045b8f7ee41540", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5dc2178957ef", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:14:14Z", "seed": 1931247394, "treatment": "gated"}, "priority": 2.2737032515042004, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9f0705906448976ea19ae9d20011cf9b0da7189bc55d79d483f58383a8c0746e", "run_seconds": 2.325169324874878, "seed": 1931247394, "started_at": "2026-11-22T03:14:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01212, 0.086081], "effect": 0.036981, "evaluated_at": "2026-11-22T03:14:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-5dc2178957ef", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.67, 0.625, 0.544642857, 0.511363636], "gated": [0.67, 0.625, 0.633928571, 0.511363636], "naive": [0.77, 0.703125, 0.767857143, 0.568181818], "scratch": [0.6, 0.625, 0.544642857, 0.522727273], "ungated": [0.7, 0.625, 0.633928571, 0.568181818]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "637c4540d70c486ee425aebdce95eb276ccdcd91eb92c0b097045b8f7ee41540", "primary_control": "scratch", "result_hash": "9f0705906448976ea19ae9d20011cf9b0da7189bc55d79d483f58383a8c0746e", "signature": "b118e8b194c5d616f8689a495e013128678497ffd50186d10cf6ddb9d0981503734258b9377b61b6b88078383f1d75e9ec57bca00b625a5acfb558530074d904", "state": "VERIFIED", "summary": {"compressed": 0.587752, "gated": 0.610073, "naive": 0.702291, "scratch": 0.573093, "ungated": 0.631778}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:14:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:23Z", "cpu_seconds": 6.8735235679999995, "created_at": "2026-11-22T03:14:14Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-56b5e99b7600", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ae562f030312dcae46ab7c3fc9473e3e50b6f067d5a817373d39633b58221739", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-56b5e99b7600", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:14:14Z", "seed": 1816174111, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "fe1d1b5a188b28a281b970a5defa39027e053272e078839313794ffcde698bcb", "run_seconds": 9.030011653900146, "seed": 1816174111, "started_at": "2026-11-22T03:14:14Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.005613, 0.105613], "effect": 0.05, "evaluated_at": "2026-11-22T03:14:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-56b5e99b7600", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.631378359172302, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0], "reset": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0], "root": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "ae562f030312dcae46ab7c3fc9473e3e50b6f067d5a817373d39633b58221739", "primary_control": "reset", "result_hash": "fe1d1b5a188b28a281b970a5defa39027e053272e078839313794ffcde698bcb", "signature": "09717e51b99c61d1d3a0719641cccc7c54c2988c2def0f659cdeec0c077dcd6199625818346ee8685dfa31d54bae2064fa9d8de108a46d5b2b90eddb0143fd01", "state": "VERIFIED", "summary": {"inherit": 0.65, "reset": 0.6, "root": 0.25}, "ties": 57, "treatment": "inherit", "wins": 3}, "verified_at": "2026-11-22T03:14:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:16Z", "cpu_seconds": 0.7417669839999999, "created_at": "2026-11-22T03:14:14Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-58496ef4257a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "321be96e0a81bec87729c80745afd974daaab6bc52e10230153f6aeaf50dbb61", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-58496ef4257a", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:14:14Z", "seed": 876076786, "treatment": "frontier"}, "priority": 2.294136171223306, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-6b9c3d1b6c31", "result_hash": "b5684d11f4fcf2118cb05933e96221b7022ecb68dc3dc6db1f431772f9d86fa2", "run_seconds": 0.8792989253997803, "seed": 876076786, "started_at": "2026-11-22T03:14:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.040817, 0.125158], "effect": 0.082987, "evaluated_at": "2026-11-22T03:14:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-58496ef4257a", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.823529412, 1.0, 0.830097087, 1.0, 0.797468354, 0.846698113, 0.853211009, 0.758823529, 0.78021978, 0.857476636, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.843406593, 1.0, 1.0, 1.0], "random_walk": [0.857843137, 1.0, 0.905339806, 1.0, 0.806962025, 0.858490566, 0.885321101, 0.844117647, 0.82967033, 0.859813084, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "prereg_hash": "321be96e0a81bec87729c80745afd974daaab6bc52e10230153f6aeaf50dbb61", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-6b9c3d1b6c31", "result_hash": "b5684d11f4fcf2118cb05933e96221b7022ecb68dc3dc6db1f431772f9d86fa2", "signature": "3cf15ef9da58d82bf61b0633a061aa02a0176168071f50eee93521869cb4137776485d482a0fb9dbcf067f13843bc0f7aac18b3cbf0401d31a98b7895e019e0b", "state": "VERIFIED", "summary": {"curiosity": 0.87896, "frontier": 0.986951, "random_walk": 0.903963}, "ties": 4, "treatment": "frontier", "wins": 8}, "verified_at": "2026-11-22T03:14:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:38Z", "cpu_seconds": 1.5036228880000007, "created_at": "2026-11-22T03:14:30Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-9749b9196707", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "554a9cd5cae0648c66df1adacc39f4dfab78855f52d15aa12ebe1d247370152e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-9749b9196707", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:14:30Z", "seed": 1277762344, "treatment": "iw_temperature"}, "priority": 2.009227555803513, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-c6c284f5e783", "result_hash": "d11be1376ff7c20af11512006d3ef7688be8e93a056f7f2abe0674d4e7ffd66e", "run_seconds": 1.6583313941955566, "seed": 1277762344, "started_at": "2026-11-22T03:15:36Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010231, 0.000137], "effect": -0.005047, "evaluated_at": "2026-11-22T03:16:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-9749b9196707", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.06314923, 0.012787025, 0.025983395, 0.045897435, 0.018559545, 0.037985592, 0.03926083, 0.057807038, 0.010611255, 0.031191357, 0.078092103, 0.131910395, 0.030684378, 0.054747848, 0.06607481, 0.067835552], "raw": [0.082986365, 0.036983297, 0.040465997, 0.043041667, 0.02918909, 0.025448235, 0.03926083, 0.041604828, 0.027105025, 0.039868738, 0.074057255, 0.120748563, 0.046912607, 0.069706565, 0.053870083, 0.053444275], "temperature": [0.06314923, 0.010724825, 0.022362752, 0.02933415, 0.018559545, 0.02810601, 0.043143718, 0.050209937, 0.01034435, 0.018129823, 0.099535873, 0.120748563, 0.023723252, 0.04757764, 0.06607481, 0.040102382]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0007, "losses": 11, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.998", "prereg_hash": "554a9cd5cae0648c66df1adacc39f4dfab78855f52d15aa12ebe1d247370152e", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-c6c284f5e783", "result_hash": "d11be1376ff7c20af11512006d3ef7688be8e93a056f7f2abe0674d4e7ffd66e", "signature": "d6d08e18a2c1050dd07dfc5650410ede555c3435bac2e94c0fd33360458adc0ae8f688b7288418133c09c371003c09b5dd4861c71107fb0daaaaa7e55f52a30e", "state": "VERIFIED", "summary": {"iw_temperature": 0.048286, "raw": 0.051543, "temperature": 0.043239}, "ties": 3, "treatment": "iw_temperature", "wins": 2}, "verified_at": "2026-11-22T03:16:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:37Z", "cpu_seconds": 1.4815586209999996, "created_at": "2026-11-22T03:14:30Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-e5ba53d0422a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "a920ef97ce2396c6372868635b1683f4f80caad386cc10a1e354c555dd7ec9ac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-e5ba53d0422a", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:14:30Z", "seed": 162272214, "treatment": "iw_temperature"}, "priority": 2.281277082543769, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-a17074327d38", "result_hash": "940d6eb74da22e6d458bdbb3f6f008bc1224e98d9f20a2530fb094e3b2f28d36", "run_seconds": 1.588313102722168, "seed": 162272214, "started_at": "2026-11-22T03:14:35Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.013726, 0.011383], "effect": -0.001172, "evaluated_at": "2026-11-22T03:14:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-e5ba53d0422a", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.027961752, 0.055094585, 0.02591547, 0.02698517, 0.091611318, 0.04975554, 0.056281212, 0.08065719, 0.073523225, 0.029034233, 0.048673392, 0.022518035, 0.214131153, 0.046813412, 0.0670847, 0.034599705], "raw": [0.035847095, 0.057547915, 0.035685163, 0.02698517, 0.036488338, 0.066766647, 0.04324924, 0.082322185, 0.050468127, 0.013481295, 0.051631908, 0.059722953, 0.23202849, 0.046813412, 0.045802207, 0.059136625], "temperature": [0.02215297, 0.04962118, 0.022762467, 0.022975825, 0.036488338, 0.0565221, 0.036410822, 0.07580134, 0.046326247, 0.021860568, 0.052737893, 0.08268391, 0.23202849, 0.068583298, 0.045802207, 0.059136625]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0482, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.895", "prereg_hash": "a920ef97ce2396c6372868635b1683f4f80caad386cc10a1e354c555dd7ec9ac", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-a17074327d38", "result_hash": "940d6eb74da22e6d458bdbb3f6f008bc1224e98d9f20a2530fb094e3b2f28d36", "signature": "f05a755a3a8de55de7b95ee6563847d7f5e41158d1df24dde9a779ffe1c09e9cf38fdedaa5ab86dfc1566980a253b98fa684ea8712ec86c45ee3c26b7fc67500", "state": "VERIFIED", "summary": {"iw_temperature": 0.059415, "raw": 0.058999, "temperature": 0.058243}, "ties": 0, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:14:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:44Z", "cpu_seconds": 3.630383689000002, "created_at": "2026-11-22T03:14:30Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ac2394305d3d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "13733f9dc066539d87f8c73ce1c01cf3a1bf8c5e11252f24f1463922b8209796", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ac2394305d3d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:14:30Z", "seed": 71562358, "treatment": "frontier"}, "priority": 2.0098667451366925, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "24ea4716eac4b10a6e9aaf01c7b3925883fc0d969b4977fce744eab777341dd0", "run_seconds": 4.309107542037964, "seed": 71562358, "started_at": "2026-11-22T03:14:40Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:14:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ac2394305d3d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "13733f9dc066539d87f8c73ce1c01cf3a1bf8c5e11252f24f1463922b8209796", "primary_control": "random", "result_hash": "24ea4716eac4b10a6e9aaf01c7b3925883fc0d969b4977fce744eab777341dd0", "signature": "9821c4a31b92658051acf22fc4955020f99d96be4c8ef7f44710d0e31a6db107613b10f463c5c4971dc782748e739e67ab7616a45a7227d0633ca949b2f05b08", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:14:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:52Z", "cpu_seconds": 0.31173873899999194, "created_at": "2026-11-22T03:14:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f3b8506c50ad", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "054fa8e4c1950246366810328c1b9ef4459ffdfde7b7360735413c5d0380d36d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f3b8506c50ad", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:14:30Z", "seed": 113586992, "treatment": "full"}, "priority": 2.0090540066629465, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "875c96a3421986707d069bbd04e65dc2050240d6b67b2b215dbffafd46069b4a", "run_seconds": 0.3298037052154541, "seed": 113586992, "started_at": "2026-11-22T03:15:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:16:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f3b8506c50ad", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 0.5, 1.0], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "054fa8e4c1950246366810328c1b9ef4459ffdfde7b7360735413c5d0380d36d", "primary_control": "e141_best", "result_hash": "875c96a3421986707d069bbd04e65dc2050240d6b67b2b215dbffafd46069b4a", "signature": "84bc80f3ef5266c335232d3303653ce4d12e81e801c5286e7dd4ba838494e45daba30272d034bf7c2ac991201027b08d8969546e6cc7f4c38337f1c697fad004", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.875, "full": 0.9375, "no_world_model": 1.0, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:16:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:42Z", "cpu_seconds": 0.13289934800000047, "created_at": "2026-11-22T03:14:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3a9f78a62746", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "e0899066cde9b60fbaaa7e08dded09134f8b4204f93c69501f448c56bdd1a23b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3a9f78a62746", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:14:30Z", "seed": 488920182, "treatment": "full"}, "priority": 2.0097458670398627, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "98807078af7957df51c6285f37df6b810b6ab01ed5e572500bcd0a110a6d09e3", "run_seconds": 0.14164066314697266, "seed": 488920182, "started_at": "2026-11-22T03:14:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.04446, 0.114631], "effect": 0.079545, "evaluated_at": "2026-11-22T03:14:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3a9f78a62746", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.666666667, 0.636363636, 0.712121212, 0.621212121], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [0.742424242, 0.742424242, 0.818181818, 0.651515152], "retrieval": [0.742424242, 0.727272727, 0.681818182, 0.666666667], "rules_nodetect": [0.257575758, 0.560606061, 0.287878788, 0.287878788]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "e0899066cde9b60fbaaa7e08dded09134f8b4204f93c69501f448c56bdd1a23b", "primary_control": "facts", "result_hash": "98807078af7957df51c6285f37df6b810b6ab01ed5e572500bcd0a110a6d09e3", "signature": "e25093a76b8381fd85c70e8f67c1a71a993c100120d2c23ac377c3caedb883651ce0e9ee29cfa8f7561766e72315ec0daf707bdf29f55a77c0afba36000b4b0f", "state": "VERIFIED", "summary": {"facts": 0.659091, "frozen": 0.030303, "full": 0.738636, "retrieval": 0.704545, "rules_nodetect": 0.348485}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:14:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:03Z", "cpu_seconds": 0.8335077359999978, "created_at": "2026-11-22T03:14:30Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-244f246bdc39", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "f7fb0e83aee553d566293a3228a215ff67ac3e80893982933eb48dcf44997bd5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-244f246bdc39", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:14:30Z", "seed": 857821992, "treatment": "gated"}, "priority": 2.0095435972346545, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c3e68eff39f088aab6d6c4dc7048d8fcca6dc7fe36dcd995076e17d4dbe4568f", "run_seconds": 2.0895256996154785, "seed": 857821992, "started_at": "2026-11-22T03:15:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01451, 0.004706], "effect": -0.004902, "evaluated_at": "2026-11-22T03:15:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-244f246bdc39", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.307189542, 0.411764706, 0.222222222, 0.31372549], "gated": [0.307189542, 0.411764706, 0.222222222, 0.31372549], "naive": [0.307189542, 0.254901961, 0.189542484, 0.31372549], "scratch": [0.307189542, 0.411764706, 0.222222222, 0.333333333], "ungated": [0.307189542, 0.411764706, 0.202614379, 0.254901961]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "f7fb0e83aee553d566293a3228a215ff67ac3e80893982933eb48dcf44997bd5", "primary_control": "scratch", "result_hash": "c3e68eff39f088aab6d6c4dc7048d8fcca6dc7fe36dcd995076e17d4dbe4568f", "signature": "04b7d1ebf2084f0a4c86f4d079f0621459ab86b7cd0a2ce9bb572c440bca4f2c1702e11e4475f9d94fc06e61466da9b26a9377c6a3c4e8f49fb66bb5291dcb06", "state": "VERIFIED", "summary": {"compressed": 0.313725, "gated": 0.313725, "naive": 0.26634, "scratch": 0.318627, "ungated": 0.294118}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:15:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:37Z", "cpu_seconds": 4.765107284999999, "created_at": "2026-11-22T03:14:30Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-c45dff1e5d34", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6ba714e21c2db2365a6e5b570220e556807fd5d39e864ceacd9223786f5abab6", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-c45dff1e5d34", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:14:30Z", "seed": 297683392, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9e225fd67282ea34bc69ac714a0636a95096506a789df4e536f35a0c0e8073d2", "run_seconds": 5.4262471199035645, "seed": 297683392, "started_at": "2026-11-22T03:14:31Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.130317, -0.003016], "effect": -0.066667, "evaluated_at": "2026-11-22T03:14:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-c45dff1e5d34", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.2165396925657679, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "6ba714e21c2db2365a6e5b570220e556807fd5d39e864ceacd9223786f5abab6", "primary_control": "reset", "result_hash": "9e225fd67282ea34bc69ac714a0636a95096506a789df4e536f35a0c0e8073d2", "signature": "04acd3095ab4ce17074a4e048f808c5b40a2ac429fff171aff991d4cce2296a3b3d506a646fe0c418bf0423df3ff3d3e7c5deb61b0c57d0954bd3a5aebebe800", "state": "VERIFIED", "summary": {"inherit": 0.65, "reset": 0.7166666666666667, "root": 0.45}, "ties": 56, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:14:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:56Z", "cpu_seconds": 1.2871291420000013, "created_at": "2026-11-22T03:14:47Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b1634fe46620", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "25622feaaa26269eded071692d091e0c005f0a7faad13322c56db3ab35401ac8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b1634fe46620", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:14:47Z", "seed": 1875245807, "treatment": "frontier"}, "priority": 2.0099249597511895, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e9f3db84761e3e5fc1480d3bc10b2547c703388f5c6a9ab3d11c79ac706c0837", "run_seconds": 1.39231276512146, "seed": 1875245807, "started_at": "2026-11-22T03:14:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.010551, 0.127256], "effect": 0.058352, "evaluated_at": "2026-11-22T03:15:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b1634fe46620", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.975, 0.75, 1.0, 0.772727273], "frontier": [0.975, 0.842105263, 1.0, 0.647727273], "goal": [0.975, 0.842105263, 1.0, 0.647727273], "hypdisc": [0.975, 0.828947368, 0.804347826, 0.647727273], "infogain": [0.975, 0.842105263, 1.0, 0.579545455], "prederr": [0.975, 0.842105263, 1.0, 0.590909091], "random": [0.975, 0.75, 0.858695652, 0.647727273], "risk": [0.975, 0.842105263, 1.0, 0.647727273]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "25622feaaa26269eded071692d091e0c005f0a7faad13322c56db3ab35401ac8", "primary_control": "random", "result_hash": "e9f3db84761e3e5fc1480d3bc10b2547c703388f5c6a9ab3d11c79ac706c0837", "signature": "cfb3bb1a14b6dee75e756cec409f77bfd41370e323dcb449933d94885a34ff947d8633eb08c598feb6d66bd1305a8dfa1e6d620cd95aef241966b7eb166f2302", "state": "VERIFIED", "summary": {"count": 0.874432, "frontier": 0.866208, "goal": 0.866208, "hypdisc": 0.814006, "infogain": 0.849163, "prederr": 0.852004, "random": 0.807856, "risk": 0.866208}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:15:07Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:16:46Z", "cpu_seconds": 33.769055382000005, "created_at": "2026-11-22T03:14:47Z", "error": "lease expired; ", "family": "ow_explore", "fence": 2, "id": "E141-ow_explore-c0089620a040", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "c419f61295a0c6bd25250e2a6ffaccde2b8f5ddd024c2e1fb1d5dea17e5e77c0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c0089620a040", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:14:47Z", "seed": 1805622767, "treatment": "frontier"}, "priority": 2.3057927203272066, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fa2b81ad43eb6f8edae2d2ff43804913f546f6e091f8494322a5b6ca9059df24", "run_seconds": 41.71868276596069, "seed": 1805622767, "started_at": "2026-11-22T03:14:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.05597, 0.312722], "effect": 0.184346, "evaluated_at": "2026-11-22T03:16:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c0089620a040", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.766666667, 0.430952381, 0.722222222, 0.568627451], "frontier": [0.762962963, 0.752380952, 0.933333333, 0.61372549], "goal": [0.659259259, 0.711904762, 0.833333333, 0.498039216], "hypdisc": [0.750617284, 0.597619048, 0.822222222, 0.507843137], "infogain": [0.755555556, 0.761904762, 0.927777778, 0.619607843], "prederr": [0.743209877, 0.176190476, 0.761111111, 0.466666667], "random": [0.758024691, 0.583333333, 0.644444444, 0.339215686], "risk": [0.762962963, 0.752380952, 0.933333333, 0.61372549]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "c419f61295a0c6bd25250e2a6ffaccde2b8f5ddd024c2e1fb1d5dea17e5e77c0", "primary_control": "random", "result_hash": "fa2b81ad43eb6f8edae2d2ff43804913f546f6e091f8494322a5b6ca9059df24", "signature": "5cda5aaf2a07dd3593760c5a8d4b9215cfcead2a661f370eb7ad5426ef7491d23e63823c54a7be4999cd1590f3e86e1805586cdfa24da4bb521439474346e009", "state": "VERIFIED", "summary": {"count": 0.622117, "frontier": 0.765601, "goal": 0.675634, "hypdisc": 0.669575, "infogain": 0.766211, "prederr": 0.536795, "random": 0.581255, "risk": 0.765601}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:16:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:50Z", "cpu_seconds": 0.5098301790000015, "created_at": "2026-11-22T03:14:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-70fd4a1d8ae9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "9087297b76968e795ac3634d315aeee05a152e931770ae3905560b01ab01197c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-70fd4a1d8ae9", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:14:47Z", "seed": 1687415411, "treatment": "full"}, "priority": 2.0099309840546398, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "90890e6ff841ae76b588072333274b310e3d71c7c1ed815af8c28b3fdf4784c4", "run_seconds": 0.5561273097991943, "seed": 1687415411, "started_at": "2026-11-22T03:14:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.42207, 0.04707], "effect": -0.1875, "evaluated_at": "2026-11-22T03:14:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-70fd4a1d8ae9", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.5, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 0.5], "no_world_model": [1.0, 1.0, 0.75, 0.5], "random": [0.75, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "9087297b76968e795ac3634d315aeee05a152e931770ae3905560b01ab01197c", "primary_control": "e141_best", "result_hash": "90890e6ff841ae76b588072333274b310e3d71c7c1ed815af8c28b3fdf4784c4", "signature": "daea935e1687c3a6fe5468454b69a363946144d9b2aadf2db02f5b215caadfced9e2beab854627d1e48621bf04c8918fe5459b3dfd94a5594111d55962303809", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.125, "full": 0.8125, "no_world_model": 0.8125, "random": 0.875}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:14:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:14:55Z", "cpu_seconds": 8.157738855999998, "created_at": "2026-11-22T03:14:47Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-126f41c2bb0d", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "371f8100e8413d0885655655539172570911dd588dc2d04aa1de90c4519600c6", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-126f41c2bb0d", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:14:47Z", "seed": 1646093118, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "05fa3df298856bead0f551ef39d627d32dff983d67247c2bb1251af33de9bf60", "run_seconds": 8.525029420852661, "seed": 1646093118, "started_at": "2026-11-22T03:14:47Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:14:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-126f41c2bb0d", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.49953119136021873, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "371f8100e8413d0885655655539172570911dd588dc2d04aa1de90c4519600c6", "primary_control": "reset", "result_hash": "05fa3df298856bead0f551ef39d627d32dff983d67247c2bb1251af33de9bf60", "signature": "44c3ad02e05001f31b4ccd045d094cca2a4a208c840e9d019d5161f573d6afb59aae6f3cef8b90dcc67d8858d72e4fffaa85ca50c9036e5a53a543cec884ac05", "state": "PARTIAL", "summary": {"inherit": 0.7, "reset": 0.7, "root": 0.26666666666666666}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:15:07Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:16:12Z", "cpu_seconds": 5.98379645, "created_at": "2026-11-22T03:14:47Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-b0c97ff4bcad", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "307c35df4ab625ff6bdde06e7c34d2d061dedcccdaa02e41b2ed43ca90e1cbc4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-b0c97ff4bcad", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:14:47Z", "seed": 881207184, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "2952c6731962480ea029b722a8a6cb4fa4fed43570792d30ae8e8be86ed8855c", "run_seconds": 7.765745162963867, "seed": 881207184, "started_at": "2026-11-22T03:15:00Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-b0c97ff4bcad", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.501464139615556, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "307c35df4ab625ff6bdde06e7c34d2d061dedcccdaa02e41b2ed43ca90e1cbc4", "primary_control": "reset", "result_hash": "2952c6731962480ea029b722a8a6cb4fa4fed43570792d30ae8e8be86ed8855c", "signature": "092c148b143a42c5834714712ec2f9ad7f626be859ba4e89baca1f9f530bc69708a4ab3e9ba37fe4ccd780c9306135d10a863cb43a7cb2976683da650d05c600", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:16:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:28Z", "cpu_seconds": 2.730948232000003, "created_at": "2026-11-22T03:15:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-952c83ae338b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "19b159877e2a15182b39e1104a789465abd4800ade8b4cf25868306f8332a2a1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-952c83ae338b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:15:04Z", "seed": 1818736281, "treatment": "frontier"}, "priority": 2.0098308113700294, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "8296d2527941bec00b3009176224c6bf5a659f7764d6262a12c72eda416a7754", "run_seconds": 4.992897272109985, "seed": 1818736281, "started_at": "2026-11-22T03:15:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.015378, 0.222496], "effect": 0.103559, "evaluated_at": "2026-11-22T03:16:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-952c83ae338b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.641843972, 0.710382514, 0.502109705, 0.581560284], "frontier": [0.780141844, 0.721311475, 0.451476793, 0.617021277], "goal": [0.670212766, 0.579234973, 0.476793249, 0.617021277], "hypdisc": [0.691489362, 0.601092896, 0.518987342, 0.581560284], "infogain": [0.780141844, 0.721311475, 0.451476793, 0.617021277], "prederr": [0.780141844, 0.721311475, 0.451476793, 0.617021277], "random": [0.581560284, 0.540983607, 0.518987342, 0.514184397], "risk": [0.780141844, 0.721311475, 0.493670886, 0.617021277]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "19b159877e2a15182b39e1104a789465abd4800ade8b4cf25868306f8332a2a1", "primary_control": "random", "result_hash": "8296d2527941bec00b3009176224c6bf5a659f7764d6262a12c72eda416a7754", "signature": "352a4e61a22470dea4e32515ea5f0651dc46401415ec9ff55e7016e18634cf28ec8853b99b07587b6b5cfc385a429bfef69e452afafdbd224985babd35274608", "state": "VERIFIED", "summary": {"count": 0.608974, "frontier": 0.642488, "goal": 0.585816, "hypdisc": 0.598282, "infogain": 0.642488, "prederr": 0.642488, "random": 0.538929, "risk": 0.653036}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:16:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:20Z", "cpu_seconds": 0.346594662, "created_at": "2026-11-22T03:15:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9f0135454830", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "a790a188a0b623b0dedfb74cef7a07ff70cb3f8590fed3f994d8d4c0b2c1ba95", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9f0135454830", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:15:04Z", "seed": 1521845083, "treatment": "full"}, "priority": 2.0098793305995675, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "dc9de9fb10a277443bca7e1fb92e765be94f3a89270fb779e4c39f41601bb3ef", "run_seconds": 0.4972553253173828, "seed": 1521845083, "started_at": "2026-11-22T03:15:20Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9f0135454830", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a790a188a0b623b0dedfb74cef7a07ff70cb3f8590fed3f994d8d4c0b2c1ba95", "primary_control": "e141_best", "result_hash": "dc9de9fb10a277443bca7e1fb92e765be94f3a89270fb779e4c39f41601bb3ef", "signature": "e70dace74fce89498c097be3c7106fdd36b6c430ee96bcea0a45868c6ef43a509c3be7c4d4202452424d483e3aa3ae0172fd0b1d93fa1f174c982d367403610c", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:16:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:15Z", "cpu_seconds": 0.220381257, "created_at": "2026-11-22T03:15:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ba85446b0004", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "797cb97d3a3af02f254722258b94a75d21ac945550aeac19b12a6c1ce250002d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ba85446b0004", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:15:04Z", "seed": 379791908, "treatment": "full"}, "priority": 2.009911001792491, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "40fda18c4049a271144de694215a3c20fb0507bfd17f3953be60f02993090ebf", "run_seconds": 0.33302950859069824, "seed": 379791908, "started_at": "2026-11-22T03:15:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:15:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ba85446b0004", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 1.0], "fixed_policy": [0.5, 0.0, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "797cb97d3a3af02f254722258b94a75d21ac945550aeac19b12a6c1ce250002d", "primary_control": "e141_best", "result_hash": "40fda18c4049a271144de694215a3c20fb0507bfd17f3953be60f02993090ebf", "signature": "f460abeb48c0cf481b017ce91d26f223a7caa6c4e4ce185aad8191c585f6e1b3d47f266cf560d5ccd44791417ec4fc3f78fe4fec8226ef75f65682aa5d675808", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:15:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:16Z", "cpu_seconds": 1.034845855, "created_at": "2026-11-22T03:15:04Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3bca8ac59b55", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "12cac1c1d23730deba9e7a2f8ae7f3a34166263d781b3e3e43e46847e73afa3e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3bca8ac59b55", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:15:04Z", "seed": 1876661122, "treatment": "full"}, "priority": 2.0098895489831157, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e8ecf885cf177da8f0cc6142a484206a3fb3329a18e5bd00aab8cd5b17f8e374", "run_seconds": 1.7650809288024902, "seed": 1876661122, "started_at": "2026-11-22T03:15:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005712, 0.05886], "effect": 0.026574, "evaluated_at": "2026-11-22T03:15:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3bca8ac59b55", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.777777778, 1.0, 0.69, 0.911111111], "frozen": [0.325925926, 0.245454545, 0.38, 0.407407407], "full": [0.851851852, 1.0, 0.7, 0.933333333], "retrieval": [0.837037037, 1.0, 0.77, 0.933333333], "rules_nodetect": [0.851851852, 1.0, 0.7, 0.933333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "12cac1c1d23730deba9e7a2f8ae7f3a34166263d781b3e3e43e46847e73afa3e", "primary_control": "facts", "result_hash": "e8ecf885cf177da8f0cc6142a484206a3fb3329a18e5bd00aab8cd5b17f8e374", "signature": "92c09a2951afdb3ac05027c7871da7caaec05b84d0b8c5e3d8d654bf65ac08c038e54037c41aaf0b67d91c29186bab1a6e6c9290516cc1e4e14832b841f0df09", "state": "VERIFIED", "summary": {"facts": 0.844722, "frozen": 0.339697, "full": 0.871296, "retrieval": 0.885093, "rules_nodetect": 0.871296}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:15:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:36Z", "cpu_seconds": 1.7551652359999963, "created_at": "2026-11-22T03:15:20Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-26b9d595e1dd", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "6109447981576dd5a0aadc42152394854b1d3b3111d06a3a17e2c8e87608e727", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-26b9d595e1dd", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:15:20Z", "seed": 1191416030, "treatment": "iw_temperature"}, "priority": 2.0095801194529894, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-5197b72a5372", "result_hash": "5d581c0f9e2e6bbc220e524c681f2e264f1f7f942ce5bf036516fad19b551279", "run_seconds": 2.104332208633423, "seed": 1191416030, "started_at": "2026-11-22T03:15:33Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019328, 0.014424], "effect": -0.002452, "evaluated_at": "2026-11-22T03:16:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-26b9d595e1dd", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.260746662, 0.288611478, 0.286487905, 0.258972562, 0.25705177, 0.336875895, 0.235788075, 0.224199962, 0.193337915, 0.240581878, 0.324566338, 0.311773265, 0.31389011, 0.350873088, 0.182207533, 0.222901347], "raw": [0.266581737, 0.263146863, 0.225035268, 0.215255545, 0.202945017, 0.30735315, 0.2008509, 0.19937851, 0.206822345, 0.208898058, 0.280382857, 0.285375588, 0.30251515, 0.341761683, 0.218574607, 0.24593853], "temperature": [0.276871627, 0.281374555, 0.239223745, 0.29083647, 0.219928983, 0.318873297, 0.235788075, 0.18617787, 0.22079125, 0.191758375, 0.30275089, 0.307323348, 0.311281888, 0.343185052, 0.256815713, 0.266650525]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0265, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.941", "prereg_hash": "6109447981576dd5a0aadc42152394854b1d3b3111d06a3a17e2c8e87608e727", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-5197b72a5372", "result_hash": "5d581c0f9e2e6bbc220e524c681f2e264f1f7f942ce5bf036516fad19b551279", "signature": "16c13a31c30dc9d0ef803b005135fb6e7fc298b194b6cf942b37440bdc07b690e94dc8ed5baabfd80b9dc26405304e491ef835de15cc27cff11ff83332fe9e07", "state": "VERIFIED", "summary": {"iw_temperature": 0.268054, "raw": 0.248176, "temperature": 0.265602}, "ties": 1, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:16:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:31Z", "cpu_seconds": 4.207050335, "created_at": "2026-11-22T03:15:20Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-29c8dfbc6bfe", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 3, "n": 4}, "prereg_hash": "cf6ba577e5a3e6329897bd7486072b2ab0ed6a4bbce95fd89aaf79a707bd0cd0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-29c8dfbc6bfe", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:15:20Z", "seed": 1671422867, "treatment": "frontier"}, "priority": 2.0095849175868445, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b70a60d0e1dd0789f8b67d5ec56f9ade6d0ba8fb9c5e63ce13ab26b11264ec2c", "run_seconds": 6.0871076583862305, "seed": 1671422867, "started_at": "2026-11-22T03:15:25Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-29c8dfbc6bfe", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "adversarial", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "cf6ba577e5a3e6329897bd7486072b2ab0ed6a4bbce95fd89aaf79a707bd0cd0", "primary_control": "random", "result_hash": "b70a60d0e1dd0789f8b67d5ec56f9ade6d0ba8fb9c5e63ce13ab26b11264ec2c", "signature": "4ee5c832b212e4fae84c415b49d1f622b00b2164bb9ba9493f9ecfa4ef705ff1fd1e3fa0949a7d257a10aebfd3191685d5466335e13540747dd4b29b1caf8707", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:16:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:45Z", "cpu_seconds": 19.283660117, "created_at": "2026-11-22T03:15:20Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4a92a286a01b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 6, "n": 4}, "prereg_hash": "d52befd1c8683fb06d248842f061027a31a1a4acf05cb0547cdbcb5fd5927204", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4a92a286a01b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:15:20Z", "seed": 1368536505, "treatment": "full"}, "priority": 2.009833354614845, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "66fd93845507ef7fc2174b22f18f349cdc29a6c559f6dd04728553273d4fadf4", "run_seconds": 23.47487473487854, "seed": 1368536505, "started_at": "2026-11-22T03:15:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.151278, 0.776278], "effect": 0.3125, "evaluated_at": "2026-11-22T03:16:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4a92a286a01b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.0, 0.25], "fixed_policy": [0.25, 0.0, 0.25, 0.0], "full": [0.25, 0.0, 1.0, 0.25], "no_world_model": [0.0, 0.0, 0.0, 0.25], "random": [0.25, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "d52befd1c8683fb06d248842f061027a31a1a4acf05cb0547cdbcb5fd5927204", "primary_control": "e141_best", "result_hash": "66fd93845507ef7fc2174b22f18f349cdc29a6c559f6dd04728553273d4fadf4", "signature": "6d9fd3dab764ea471bab569176b7e756a690a79ee6662a1d5fe7c2dfa7e71589522046a0bbb9617502591e00243ca63ae19b6d945b33ac95abc57ae355ba4007", "state": "VERIFIED", "summary": {"e141_best": 0.0625, "fixed_policy": 0.125, "full": 0.375, "no_world_model": 0.0625, "random": 0.0625}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:16:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:32Z", "cpu_seconds": 7.485296616, "created_at": "2026-11-22T03:15:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-fddbe3802f04", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "58e0c38d695e47cec7002a501019bce905abd509ac9b16fe06a4e1e8386fa5a8", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-fddbe3802f04", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:15:20Z", "seed": 2093948527, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "3b66658d19f7fb549b0ae254fb827779fc060ed42c6c3aaa90e3490787d93112", "run_seconds": 10.297271013259888, "seed": 2093948527, "started_at": "2026-11-22T03:15:22Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-fddbe3802f04", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.39153936022023583, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "58e0c38d695e47cec7002a501019bce905abd509ac9b16fe06a4e1e8386fa5a8", "primary_control": "reset", "result_hash": "3b66658d19f7fb549b0ae254fb827779fc060ed42c6c3aaa90e3490787d93112", "signature": "877f10b21f7cb5e2c4dcce7f508d54772c0f318378caf01ddcf2257d18c79d92349b31268801da6c375eb0185fc23faf1b8ba28c1c64e81779f55fc46317aa03", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.23333333333333334}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:16:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:00Z", "cpu_seconds": 9.154873366000002, "created_at": "2026-11-22T03:15:37Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f553a0672dd1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 7, "n": 4}, "prereg_hash": "6c2aee92170683c133d037da1449f65b8c6c94949f5ee90000b042360ed6dfa8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f553a0672dd1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:15:37Z", "seed": 521473124, "treatment": "frontier"}, "priority": 2.0094554416844854, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d3fa285fcc3f470ddfec708ab215336b9dca398e95d5649d8f44923382819da8", "run_seconds": 9.879272699356079, "seed": 521473124, "started_at": "2026-11-22T03:15:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.045609, 0.112276], "effect": 0.033333, "evaluated_at": "2026-11-22T03:16:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f553a0672dd1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 0.853333333], "frontier": [0.986666667, 1.0, 0.98, 0.891111111], "goal": [0.973333333, 1.0, 0.98, 0.762222222], "hypdisc": [0.986666667, 1.0, 0.9, 0.768888889], "infogain": [0.986666667, 1.0, 0.98, 0.891111111], "prederr": [0.986666667, 1.0, 0.98, 0.891111111], "random": [0.986666667, 1.0, 1.0, 0.737777778], "risk": [0.986666667, 1.0, 0.98, 0.891111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "6c2aee92170683c133d037da1449f65b8c6c94949f5ee90000b042360ed6dfa8", "primary_control": "random", "result_hash": "d3fa285fcc3f470ddfec708ab215336b9dca398e95d5649d8f44923382819da8", "signature": "ec6e8b559f9a26afdb7390f31d55069550bf2d4606a3c44796ecb54702f7c4bdbc4531ca53ebb0ea28413cdc8f1e38564ffc0dd4d7a2666c68367c201bc77700", "state": "VERIFIED", "summary": {"count": 0.963333, "frontier": 0.964444, "goal": 0.928889, "hypdisc": 0.913889, "infogain": 0.964444, "prederr": 0.964444, "random": 0.931111, "risk": 0.964444}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:16:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:41Z", "cpu_seconds": 0.3276203749999951, "created_at": "2026-11-22T03:15:37Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-092168d642e6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "b53678d89d1b05dab41e148cd8de6ed3c45ed02245f4efe8d774f6e48c8247e0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-092168d642e6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:15:37Z", "seed": 1277120426, "treatment": "full"}, "priority": 2.009850996333887, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1a2b50eebe1c3c3bfcabf0a16e01c83a8d6b94458c49c681a4cc64b22eee5dc1", "run_seconds": 0.33631420135498047, "seed": 1277120426, "started_at": "2026-11-22T03:15:41Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-092168d642e6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.5, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b53678d89d1b05dab41e148cd8de6ed3c45ed02245f4efe8d774f6e48c8247e0", "primary_control": "e141_best", "result_hash": "1a2b50eebe1c3c3bfcabf0a16e01c83a8d6b94458c49c681a4cc64b22eee5dc1", "signature": "395141afdd576557982c0f325d635838dd639b6d9e5f0fcd51a105f2ef2b354f92122d4e17d6f5a0e03db9565ffd29a67addee9762ceebbc4c83857c333f3a0b", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:16:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:45Z", "cpu_seconds": 1.4244019179999992, "created_at": "2026-11-22T03:15:37Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1c5756e82692", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "3f43cfe8448e939017342701d552e10b14a3839aa113a13b22f1a827c0914d91", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1c5756e82692", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:15:37Z", "seed": 1775744430, "treatment": "full"}, "priority": 2.009624108303386, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9800be1a62c4803393750b884bf4c28c95d5f8c78d6cd7e1027f8c8300c1b07b", "run_seconds": 1.5441749095916748, "seed": 1775744430, "started_at": "2026-11-22T03:15:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:16:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1c5756e82692", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [1.0, 0.25, 1.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "3f43cfe8448e939017342701d552e10b14a3839aa113a13b22f1a827c0914d91", "primary_control": "e141_best", "result_hash": "9800be1a62c4803393750b884bf4c28c95d5f8c78d6cd7e1027f8c8300c1b07b", "signature": "d7a33876478cd3426ef75e4c5599e57c83e48a602d1ad9a958bd37d7ab9b30208107ebd3a9b82b536292033696aadbc264f7177d55d76bf088d65ccd9f79350f", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.625, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:16:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:52Z", "cpu_seconds": 1.8338413620000011, "created_at": "2026-11-22T03:15:37Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ebe660e948d6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 7, "n": 4}, "prereg_hash": "b5e086cce65cce0252edef3bae50b8636e3e60e7466a5fd7cec537c89a15d78d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ebe660e948d6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:15:37Z", "seed": 1533782564, "treatment": "full"}, "priority": 2.009497044811626, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2ccd11eb6ba8d306838c2fe7427561a8072ee30055f92671c38bad7f8ac4e089", "run_seconds": 1.8779964447021484, "seed": 1533782564, "started_at": "2026-11-22T03:15:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.086122, 0.272311], "effect": 0.179217, "evaluated_at": "2026-11-22T03:16:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ebe660e948d6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.96, 0.433, 0.459666667, 0.3464], "frozen": [0.64, 0.320333333, 0.361666667, 0.32], "full": [1.0, 0.673666667, 0.656666667, 0.5856], "retrieval": [1.0, 0.796, 0.816666667, 0.7784], "rules_nodetect": [1.0, 0.673666667, 0.656666667, 0.5856]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "b5e086cce65cce0252edef3bae50b8636e3e60e7466a5fd7cec537c89a15d78d", "primary_control": "facts", "result_hash": "2ccd11eb6ba8d306838c2fe7427561a8072ee30055f92671c38bad7f8ac4e089", "signature": "5d9c6b72ca9be56289656cc42de988f36d6819ec219b1e5d350731e4fd5b85e0e12f90f881b1c55e38f5b143b32f50b7a417d609533ce22a90fbded41a5dff05", "state": "VERIFIED", "summary": {"facts": 0.549767, "frozen": 0.4105, "full": 0.728983, "retrieval": 0.847767, "rules_nodetect": 0.728983}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:16:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:46Z", "cpu_seconds": 0.23969816500000718, "created_at": "2026-11-22T03:15:37Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-50dbb543d519", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "2cc5d00b82b357e9836b5e3bef1bb98a586e0a52894a816b3450412a4cd464d6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-50dbb543d519", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:15:37Z", "seed": 695691221, "treatment": "frontier"}, "priority": 2.009602629796451, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-c333f1bf81d4", "result_hash": "5e142168e980f58a50575387ac3af28d249d830d3db6831e4d35978b5d538079", "run_seconds": 0.24390459060668945, "seed": 695691221, "started_at": "2026-11-22T03:15:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.056819, 0.100509], "effect": 0.078664, "evaluated_at": "2026-11-22T03:16:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-50dbb543d519", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.810344828, 0.831730769, 0.840909091, 0.87195122, 0.77, 0.844827586, 0.892045455, 1.0, 0.8, 0.794117647, 0.84375, 0.785714286], "frontier": [0.88362069, 0.903846154, 0.806818182, 0.926829268, 0.875, 0.905172414, 0.926136364, 1.0, 0.89, 0.897058824, 1.0, 0.84375], "random_walk": [0.831896552, 0.817307692, 0.772727273, 0.774390244, 0.76, 0.823275862, 0.852272727, 1.0, 0.795, 0.808823529, 0.90625, 0.772321429]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "2cc5d00b82b357e9836b5e3bef1bb98a586e0a52894a816b3450412a4cd464d6", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-c333f1bf81d4", "result_hash": "5e142168e980f58a50575387ac3af28d249d830d3db6831e4d35978b5d538079", "signature": "820da270bff544b008841f89306eb6a3c2120976d1aed9ad8f7b9cf199f6322ecd47065ccab5a182955318e39ad7269e18e498e6955ad6bff0bd73b8009d590f", "state": "VERIFIED", "summary": {"curiosity": 0.840449, "frontier": 0.904853, "random_walk": 0.826189}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:16:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:56Z", "cpu_seconds": 1.5504191429999992, "created_at": "2026-11-22T03:15:53Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-d3a1b6cd41c4", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "1c0a415cbe6bab46e821305eeed7f873b092e23bfd018a18f3ec6d1ec31faf9c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-d3a1b6cd41c4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:15:53Z", "seed": 1130101711, "treatment": "iw_temperature"}, "priority": 2.0092633965077784, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-d87883d4aa27", "result_hash": "2d90b72ab71471ac6fa4bf7b168cbbc6aa6be7972f9a391bc1307e742fe63c5c", "run_seconds": 1.6467342376708984, "seed": 1130101711, "started_at": "2026-11-22T03:16:54Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01708, 0.016131], "effect": -0.000474, "evaluated_at": "2026-11-22T03:16:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-d3a1b6cd41c4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.293047527, 0.174161118, 0.254720542, 0.227153835, 0.263269335, 0.252837853, 0.164724532, 0.186682165, 0.248698065, 0.290535798, 0.295163992, 0.230071678, 0.304774115, 0.29202355, 0.261134177, 0.207142915], "raw": [0.245107938, 0.245862115, 0.238739007, 0.256473662, 0.222034342, 0.206927585, 0.199509535, 0.173192692, 0.26952344, 0.22783791, 0.278833817, 0.195362457, 0.29576194, 0.24403947, 0.261134177, 0.25861986], "temperature": [0.267895435, 0.237523133, 0.248179435, 0.27506401, 0.240529655, 0.206927585, 0.196137123, 0.186682165, 0.239779658, 0.242174488, 0.284114407, 0.241983368, 0.304249028, 0.24403947, 0.280402138, 0.242873268]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.071, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.849", "prereg_hash": "1c0a415cbe6bab46e821305eeed7f873b092e23bfd018a18f3ec6d1ec31faf9c", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-d87883d4aa27", "result_hash": "2d90b72ab71471ac6fa4bf7b168cbbc6aa6be7972f9a391bc1307e742fe63c5c", "signature": "0a53d4df838d9929cfb2612fc10af229ae056906eddbce7814ae652f51c9b3be5ff73d7375758d2a423554fdb303ee33cfbb8195ef7d0523e8ebb5f39c6b6903", "state": "VERIFIED", "summary": {"iw_temperature": 0.246634, "raw": 0.238685, "temperature": 0.24616}, "ties": 1, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:16:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:07Z", "cpu_seconds": 1.9126517229999997, "created_at": "2026-11-22T03:15:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-079379fce9c6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "9ac0a5d6089be3d634771369f0d65aa15f18784b0f71accbda4cf283ec52c6ca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-079379fce9c6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:15:53Z", "seed": 487655185, "treatment": "frontier"}, "priority": 2.009285899322146, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6ee9492ba2d6820507e43df4877dc1493b60fb707c91b1af238410fdb95dfc0c", "run_seconds": 2.4864718914031982, "seed": 487655185, "started_at": "2026-11-22T03:16:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.011999, 0.203356], "effect": 0.107678, "evaluated_at": "2026-11-22T03:16:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-079379fce9c6", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.861111111, 0.794117647, 0.81, 0.931818182], "frontier": [0.791666667, 0.882352941, 0.89, 1.0], "goal": [0.833333333, 0.882352941, 0.92, 1.0], "hypdisc": [0.902777778, 0.867647059, 0.9, 1.0], "infogain": [0.791666667, 0.882352941, 0.89, 1.0], "prederr": [0.791666667, 0.882352941, 0.89, 1.0], "random": [0.805555556, 0.794117647, 0.67, 0.863636364], "risk": [0.791666667, 0.882352941, 0.89, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "9ac0a5d6089be3d634771369f0d65aa15f18784b0f71accbda4cf283ec52c6ca", "primary_control": "random", "result_hash": "6ee9492ba2d6820507e43df4877dc1493b60fb707c91b1af238410fdb95dfc0c", "signature": "13346fc72895262886195810a58436b508adc7b73a617a5c1c3c7d17c6b662536783da6b4c96ca1fa05c6b0876aaeb0974738d7df229811097525b3064dff602", "state": "VERIFIED", "summary": {"count": 0.849262, "frontier": 0.891005, "goal": 0.908922, "hypdisc": 0.917606, "infogain": 0.891005, "prederr": 0.891005, "random": 0.783327, "risk": 0.891005}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:16:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:00Z", "cpu_seconds": 2.622304435999993, "created_at": "2026-11-22T03:15:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3240df3e64dc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "cde7a9cd877b20346c1a8eb5a8a9c695a20d1987c65b5676db75e9b02d2d721e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3240df3e64dc", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:15:53Z", "seed": 1930697381, "treatment": "frontier"}, "priority": 2.0096640752641277, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3cfb281e12607ad0ee74a3e9e1763015fa72e3bda6c7ebdd12fe1b8a1347ea5b", "run_seconds": 2.8791308403015137, "seed": 1930697381, "started_at": "2026-11-22T03:15:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.025568, 0.335543], "effect": 0.180556, "evaluated_at": "2026-11-22T03:16:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-3240df3e64dc", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.916666667, 0.81875, 0.9, 0.944444444], "frontier": [0.791666667, 0.9375, 0.964285714, 1.0], "goal": [0.75, 0.88125, 0.921428571, 0.881944444], "hypdisc": [0.666666667, 0.84375, 0.914285714, 0.923611111], "infogain": [0.791666667, 0.9375, 0.964285714, 1.0], "prederr": [0.791666667, 0.9375, 0.964285714, 1.0], "random": [0.708333333, 0.525, 0.814285714, 0.923611111], "risk": [0.791666667, 0.9375, 0.964285714, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "cde7a9cd877b20346c1a8eb5a8a9c695a20d1987c65b5676db75e9b02d2d721e", "primary_control": "random", "result_hash": "3cfb281e12607ad0ee74a3e9e1763015fa72e3bda6c7ebdd12fe1b8a1347ea5b", "signature": "88b139e41631079a3aa66650e86638c5544bad5c67319ad6800e1c9ae7de7036eb956e70fa49cf20d97706e867cf28d07fe07d03e4127fd466179a0769bd2b07", "state": "VERIFIED", "summary": {"count": 0.894965, "frontier": 0.923363, "goal": 0.858656, "hypdisc": 0.837078, "infogain": 0.923363, "prederr": 0.923363, "random": 0.742808, "risk": 0.923363}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:16:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:16Z", "cpu_seconds": 2.110570451000001, "created_at": "2026-11-22T03:15:53Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5f70eca0e7f4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 3, "n": 4}, "prereg_hash": "6805c5e8282f15d735fd800e8590fc1336a07fe12803d0c64478c2835667750b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5f70eca0e7f4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:15:53Z", "seed": 2077447003, "treatment": "full"}, "priority": 2.0091726922843485, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1afbf2be230beb2f84edd9e38336613bed2e3196eb3c1db0975c61d40df0e0d4", "run_seconds": 2.4091713428497314, "seed": 2077447003, "started_at": "2026-11-22T03:17:13Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:17:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5f70eca0e7f4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.5, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6805c5e8282f15d735fd800e8590fc1336a07fe12803d0c64478c2835667750b", "primary_control": "e141_best", "result_hash": "1afbf2be230beb2f84edd9e38336613bed2e3196eb3c1db0975c61d40df0e0d4", "signature": "43b04212a4b8194ffc757903f514581536a49e240be071b59b037c45bb04cc0f231e489fc0fe305cbdc8a5d10a3cf1d67dc1d61922911ae071bc340b46951a0e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:17:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:19Z", "cpu_seconds": 12.764219783999998, "created_at": "2026-11-22T03:15:53Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ea2d5ee417e4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 9, "n": 4}, "prereg_hash": "b1d9cc5680fd508cfda5049ef9d2aa513cc736273b9f976acfda74d16ebcd3a7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ea2d5ee417e4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:15:53Z", "seed": 632171597, "treatment": "full"}, "priority": 2.0096525278087176, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f0c6df3612948b0309d038f9a27770801d7e4de72e2cec64a370d749f1456f38", "run_seconds": 15.656277894973755, "seed": 632171597, "started_at": "2026-11-22T03:16:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ea2d5ee417e4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.75, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [0.25, 1.0, 1.0, 0.75], "no_world_model": [0.5, 1.0, 1.0, 0.5], "random": [0.75, 1.0, 0.5, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "b1d9cc5680fd508cfda5049ef9d2aa513cc736273b9f976acfda74d16ebcd3a7", "primary_control": "e141_best", "result_hash": "f0c6df3612948b0309d038f9a27770801d7e4de72e2cec64a370d749f1456f38", "signature": "11fa422d3824b4d570156a8563b4f32181b29ae87cf30f2e1932303ab4ab06db2c03a86da444fd01de0c8441de81ff4d444212ff1843b0fb1983ea9dc43ff00d", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.0625, "full": 0.75, "no_world_model": 0.75, "random": 0.5625}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:16:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:15:58Z", "cpu_seconds": 1.203215643, "created_at": "2026-11-22T03:15:53Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ca45d6c96908", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "bfea905993ba172228b90f4381bcfff89b5d093ee1c94fbc1c7924731eeced63", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ca45d6c96908", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:15:53Z", "seed": 2087641044, "treatment": "full"}, "priority": 2.009806789032418, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9d9ffcd01e271cbcbcef851d28aaa678d4d9206be3faddf82734ed05d3c22c0c", "run_seconds": 1.2374870777130127, "seed": 2087641044, "started_at": "2026-11-22T03:15:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.034144, -0.002833], "effect": -0.018489, "evaluated_at": "2026-11-22T03:16:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ca45d6c96908", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.425675676, 0.609977324, 0.360902256, 0.457142857], "frozen": [0.146396396, 0.20861678, 0.261904762, 0.187969925], "full": [0.411036036, 0.578231293, 0.363408521, 0.427067669], "retrieval": [0.431306306, 0.603174603, 0.286967419, 0.464661654], "rules_nodetect": [0.298423423, 0.578231293, 0.363408521, 0.427067669]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "bfea905993ba172228b90f4381bcfff89b5d093ee1c94fbc1c7924731eeced63", "primary_control": "facts", "result_hash": "9d9ffcd01e271cbcbcef851d28aaa678d4d9206be3faddf82734ed05d3c22c0c", "signature": "593aa2aa5fe8f307c4cd07fc7e3241959de4e12c624a8bc3ab554d76d87d0bec4ed2f22a990ce9a99e9e87ec9d88c869e61668d5aad70c83e0d1feb2713b7407", "state": "VERIFIED", "summary": {"facts": 0.463425, "frozen": 0.201222, "full": 0.444936, "retrieval": 0.446527, "rules_nodetect": 0.416783}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:16:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:21Z", "cpu_seconds": 1.1306440560000013, "created_at": "2026-11-22T03:16:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-736cc990d30e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "de1a96dc1eed53872d5185c6adddeac9819f259320f2d80c0b0753dd642e54e8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-736cc990d30e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:16:09Z", "seed": 1405142674, "treatment": "frontier"}, "priority": 2.0098592701299993, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "02dd8102ec2ed70cdd29cfa0af9f6c368b9772ef4a459b956936db32abb246fc", "run_seconds": 1.5030183792114258, "seed": 1405142674, "started_at": "2026-11-22T03:16:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.209579, 0.262023], "effect": 0.235801, "evaluated_at": "2026-11-22T03:16:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-736cc990d30e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.784313725, 0.807692308, 0.68115942, 0.803921569], "frontier": [0.869281046, 0.929487179, 0.804347826, 0.908496732], "goal": [0.718954248, 0.929487179, 0.804347826, 0.843137255], "hypdisc": [0.653594771, 0.66025641, 0.731884058, 0.77124183], "infogain": [0.869281046, 0.929487179, 0.804347826, 0.908496732], "prederr": [0.869281046, 0.929487179, 0.804347826, 0.908496732], "random": [0.666666667, 0.692307692, 0.536231884, 0.673202614], "risk": [0.882352941, 0.929487179, 0.797101449, 0.908496732]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "de1a96dc1eed53872d5185c6adddeac9819f259320f2d80c0b0753dd642e54e8", "primary_control": "random", "result_hash": "02dd8102ec2ed70cdd29cfa0af9f6c368b9772ef4a459b956936db32abb246fc", "signature": "bd862df2e51413ddf7299877d5707a79fc2e9580f8f04c8266b6dcb8d197299e8e2141eaf53fe74879a13312f8b1d98376bae9574ac238c1133fe6dcc3b17509", "state": "VERIFIED", "summary": {"count": 0.769272, "frontier": 0.877903, "goal": 0.823982, "hypdisc": 0.704244, "infogain": 0.877903, "prederr": 0.877903, "random": 0.642102, "risk": 0.87936}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:16:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:29Z", "cpu_seconds": 1.6161676789999966, "created_at": "2026-11-22T03:16:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ab3fd0156077", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "e569a0f494c9790344aa0b2351d45bfe84b8dd4b72416028fb1c592df88d90ca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ab3fd0156077", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:16:09Z", "seed": 673591851, "treatment": "frontier"}, "priority": 2.0097849757737545, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "7679373c75a63a2b0603c4491a250dc2ac3f38fb170c103aff6630c6836261fe", "run_seconds": 3.0720951557159424, "seed": 673591851, "started_at": "2026-11-22T03:16:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.048557, 0.069757], "effect": 0.0106, "evaluated_at": "2026-11-22T03:16:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ab3fd0156077", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.570196078, 0.552941176, 0.552610442, 0.580392157], "frontier": [0.534117647, 0.490196078, 0.513253012, 0.632941176], "goal": [0.534117647, 0.490196078, 0.513253012, 0.632941176], "hypdisc": [0.55372549, 0.530980392, 0.477108434, 0.603137255], "infogain": [0.534117647, 0.490196078, 0.513253012, 0.632941176], "prederr": [0.534117647, 0.490196078, 0.513253012, 0.632941176], "random": [0.468235294, 0.541176471, 0.544578313, 0.574117647], "risk": [0.534117647, 0.490196078, 0.513253012, 0.632941176]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "e569a0f494c9790344aa0b2351d45bfe84b8dd4b72416028fb1c592df88d90ca", "primary_control": "random", "result_hash": "7679373c75a63a2b0603c4491a250dc2ac3f38fb170c103aff6630c6836261fe", "signature": "a047870dbd8f8cb5568c5c25998ab3952bf83eeb6ec395be499373badd46ea89bb5d0167e80862ff410278622ec59295c90c5fa7db17db0933f73d8a7ff92f00", "state": "VERIFIED", "summary": {"count": 0.564035, "frontier": 0.542627, "goal": 0.542627, "hypdisc": 0.541238, "infogain": 0.542627, "prederr": 0.542627, "random": 0.532027, "risk": 0.542627}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:16:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:14Z", "cpu_seconds": 1.2461039830000011, "created_at": "2026-11-22T03:16:09Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1342a86882c9", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "7d52633db95be957f9912625b502afa63dfa3a58b1284a426d3ad12632a505aa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1342a86882c9", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:16:09Z", "seed": 599390268, "treatment": "full"}, "priority": 2.00985994117975, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "eaa4cfa5f6c8b9504847ca86585ef9d63851beecd6be2856cad1ef76a7c0f329", "run_seconds": 1.902907371520996, "seed": 599390268, "started_at": "2026-11-22T03:16:12Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:16:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1342a86882c9", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"no_world_model": [1.0, 0.9375]}, "prereg_hash": "7d52633db95be957f9912625b502afa63dfa3a58b1284a426d3ad12632a505aa", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "eaa4cfa5f6c8b9504847ca86585ef9d63851beecd6be2856cad1ef76a7c0f329", "signature": "b92dd42af55c3dfd22f28d976b6bc1ba192e5fe991a56996a722097b349a8edf1dfb64800af624707c51093ed597cd9ca56945f96d925d5dbb1de1cd9a9d330a", "state": "REJECTED", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}}, "verified_at": "2026-11-22T03:16:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:31Z", "cpu_seconds": 4.244277786000005, "created_at": "2026-11-22T03:16:09Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d2d2272da65e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 7, "n": 4}, "prereg_hash": "01db17c24a523e94122b04a364652518910b1180b97437944fec156af5327294", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d2d2272da65e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:16:09Z", "seed": 1836930811, "treatment": "full"}, "priority": 2.0097919423478894, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "07c4a3ef80b041bcaa888207e3bb252544736fe5d066f960943c422c2bae87c1", "run_seconds": 7.071322679519653, "seed": 1836930811, "started_at": "2026-11-22T03:16:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:16:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d2d2272da65e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "01db17c24a523e94122b04a364652518910b1180b97437944fec156af5327294", "primary_control": "e141_best", "result_hash": "07c4a3ef80b041bcaa888207e3bb252544736fe5d066f960943c422c2bae87c1", "signature": "a98ed561a0c3a3f5bbae4751991ee533586e327796d5d9f8a4afdb1c4ab1572380bb16c31adfe732862933a0dd7f493e5dbdc34d3ab39d52a890002c999c170f", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.75, "full": 1.0, "no_world_model": 0.875, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:16:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:06Z", "cpu_seconds": 41.096445378, "created_at": "2026-11-22T03:16:09Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-ceb925d12105", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "d52414292b94d4b19da8314d090c3da4a347e2eab4600389aa205cf17d61d1ea", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-ceb925d12105", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:16:09Z", "seed": 1161021592, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "fedc47abaf04871ee1b57b1429307db91fd09c465ca800d84a9d97aadc9af151", "run_seconds": 48.71097683906555, "seed": 1161021592, "started_at": "2026-11-22T03:16:17Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"compressed": 0, "inherit_mem": 0, "inherit_nomem": 1, "random_search": 0, "reset_mem": 0, "retrieval": 0, "rootcause": 1, "xfer_mem": 0}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:17:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-ceb925d12105", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.9609375, 1.0, 0.9375, 0.658496732, 0.769791667, 0.75, 0.625, 0.14, 0.964285714, 0.295], "compressed": [0.9609375, 1.0, 0.9375, 0.658496732, 0.769791667, 0.75, 0.625, 0.14, 0.964285714, 0.295], "inherit_mem": [0.9609375, 1.0, 0.9375, 0.658496732, 0.769791667, 0.75, 0.625, 0.14, 0.964285714, 0.295], "inherit_nomem": [0.9609375, 1.0, 0.9609375, 0.658496732, 0.8125, 0.7421875, 0.625, 0.14, 0.964285714, 0.17], "random_search": [0.9609375, 1.0, 0.9375, 0.658496732, 0.769791667, 0.75, 0.625, 0.14, 0.964285714, 0.295], "reset_mem": [0.9609375, 1.0, 0.9375, 0.658496732, 0.769791667, 0.75, 0.625, 0.14, 0.964285714, 0.295], "retrieval": [0.9609375, 1.0, 0.9375, 0.658496732, 0.769791667, 0.75, 0.625, 0.14, 0.964285714, 0.295], "rootcause": [0.9609375, 1.0, 0.9609375, 0.658496732, 0.8125, 0.7421875, 0.625, 0.14, 0.964285714, 0.17], "xfer_mem": [0.9609375, 1.0, 0.9375, 0.658496732, 0.769791667, 0.75, 0.625, 0.14, 0.964285714, 0.295]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "per_generation_sealed": [0.710101, 0.710101, 0.710101, 0.710101], "prereg_hash": "d52414292b94d4b19da8314d090c3da4a347e2eab4600389aa205cf17d61d1ea", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "xfer_mem", "inherit_nomem"], "result_hash": "fedc47abaf04871ee1b57b1429307db91fd09c465ca800d84a9d97aadc9af151", "signature": "9cc3fbd21fe180345473d55b1793b89ed8e8ea22f639f89a8d064d037a86ba4199d195b338acce4bd134a92d63ef2d62713325823746824fd4284357a0977506", "state": "PARTIAL", "summary": {"G0": 0.710101, "compressed": 0.710101, "inherit_mem": 0.710101, "inherit_nomem": 0.703434, "random_search": 0.710101, "reset_mem": 0.710101, "retrieval": 0.710101, "rootcause": 0.703434, "xfer_mem": 0.710101}, "ties": 10, "treatment": "inherit_mem", "wins": 0}, "verified_at": "2026-11-22T03:17:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:40Z", "cpu_seconds": 0.340612857, "created_at": "2026-11-22T03:16:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-425cbfa866d3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "3f99caa14e44cd0fada125e5daffdc7eb820d13ec5ab327418ba8b6781e46dd9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-425cbfa866d3", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:16:26Z", "seed": 2038289309, "treatment": "frontier"}, "priority": 2.0097378596228954, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a656b4e86badf8a682fce38f78ec275abce0c6c9085441e0df4ced45b61f5ca6", "run_seconds": 0.36540794372558594, "seed": 2038289309, "started_at": "2026-11-22T03:16:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.165494, 0.260643], "effect": 0.213068, "evaluated_at": "2026-11-22T03:16:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-425cbfa866d3", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.920454545, 0.954545455, 1.0, 0.977272727], "frontier": [0.943181818, 0.954545455, 1.0, 1.0], "goal": [0.670454545, 0.602272727, 0.875, 0.75], "hypdisc": [0.795454545, 0.647727273, 0.852272727, 0.75], "infogain": [0.943181818, 0.954545455, 1.0, 1.0], "prederr": [0.943181818, 0.954545455, 1.0, 1.0], "random": [0.795454545, 0.704545455, 0.795454545, 0.75], "risk": [0.943181818, 0.954545455, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "3f99caa14e44cd0fada125e5daffdc7eb820d13ec5ab327418ba8b6781e46dd9", "primary_control": "random", "result_hash": "a656b4e86badf8a682fce38f78ec275abce0c6c9085441e0df4ced45b61f5ca6", "signature": "f5e82941b5a7916491af77c624205ec5c8925af605cc3ec040690382dcdd17dbeba8cef978e1da383a8405b97c76645fd33005d0cf4cd6579c7d6898b5c4090e", "state": "VERIFIED", "summary": {"count": 0.963068, "frontier": 0.974432, "goal": 0.724432, "hypdisc": 0.761364, "infogain": 0.974432, "prederr": 0.974432, "random": 0.761364, "risk": 0.974432}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:16:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:34Z", "cpu_seconds": 0.5847455040000007, "created_at": "2026-11-22T03:16:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0a8c727b866e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "c49c7303ac6e525f16500e43f1873d1592a66929999159c83e78d0810a3384e3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0a8c727b866e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:16:26Z", "seed": 1104426844, "treatment": "full"}, "priority": 2.009943090093571, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e63134c9e77fcfa23018465e0ddf4510a115fb7cbc2857956b56d704e4ea70e5", "run_seconds": 0.612708568572998, "seed": 1104426844, "started_at": "2026-11-22T03:16:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.266451, 0.016451], "effect": -0.125, "evaluated_at": "2026-11-22T03:16:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0a8c727b866e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.75, 1.0, 0.75, 1.0], "no_world_model": [0.25, 0.75, 0.75, 1.0], "random": [0.75, 0.0, 0.75, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c49c7303ac6e525f16500e43f1873d1592a66929999159c83e78d0810a3384e3", "primary_control": "e141_best", "result_hash": "e63134c9e77fcfa23018465e0ddf4510a115fb7cbc2857956b56d704e4ea70e5", "signature": "7d11ade3a8057a0ff01c05a8a41f3765baff596ebcacd8f38f0cd098a81f3cac26c3485e9e95bf0c7779d0656f381c1a10dc07a97dc2df3ef0663ad63b7d7e01", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.0, "full": 0.875, "no_world_model": 0.6875, "random": 0.5625}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:16:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:37Z", "cpu_seconds": 0.9808397490000047, "created_at": "2026-11-22T03:16:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e200901fd006", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "4c4683646a0275964105cc5212c1e3e4221f5f9116d98973f8dcdbeaa7e53639", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e200901fd006", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:16:26Z", "seed": 1991604168, "treatment": "full"}, "priority": 2.0098860295727743, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c13b8239b94d96fcb556fbfdb617c4fe85afa6b2a77cce19a68c8063e5ad6278", "run_seconds": 1.052034854888916, "seed": 1991604168, "started_at": "2026-11-22T03:16:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001835, 0.069285], "effect": 0.033725, "evaluated_at": "2026-11-22T03:16:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e200901fd006", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.732954545, 0.925, 0.724137931, 0.976190476], "frozen": [0.357954545, 0.275, 0.362068966, 0.321428571], "full": [0.818181818, 0.925, 0.75, 1.0], "retrieval": [0.926136364, 0.9375, 0.844827586, 1.0], "rules_nodetect": [0.818181818, 0.925, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "4c4683646a0275964105cc5212c1e3e4221f5f9116d98973f8dcdbeaa7e53639", "primary_control": "facts", "result_hash": "c13b8239b94d96fcb556fbfdb617c4fe85afa6b2a77cce19a68c8063e5ad6278", "signature": "8d2867b7028e342ff906af979ac77c8227db7b21e29e4148bf23cc15796feaa48c2233aa654d978ff3cfd171a0979ba5ee2b2ffbdb72f70115031eb7c34ba805", "state": "VERIFIED", "summary": {"facts": 0.839571, "frozen": 0.329113, "full": 0.873295, "retrieval": 0.927116, "rules_nodetect": 0.873295}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:16:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:44Z", "cpu_seconds": 0.7890092620000004, "created_at": "2026-11-22T03:16:26Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ad610d66e87e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "af240e3be399cd3467831fe1072b4074ab2048cfbe84b2d9d8730543548ef65f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ad610d66e87e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:16:26Z", "seed": 487749805, "treatment": "gated"}, "priority": 2.0093674061370943, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "ca92b8a2a7279d3140826a7d255da297f80486274c4e35d96ff9f83e5ed0a085", "run_seconds": 1.5118610858917236, "seed": 487749805, "started_at": "2026-11-22T03:16:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.112629, 0.03093], "effect": -0.04085, "evaluated_at": "2026-11-22T03:16:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-ad610d66e87e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.267973856, 0.25170068, 0.261437908, 0.209150327], "gated": [0.267973856, 0.25170068, 0.261437908, 0.209150327], "naive": [0.248366013, 0.231292517, 0.392156863, 0.261437908], "scratch": [0.267973856, 0.25170068, 0.411764706, 0.222222222], "ungated": [0.281045752, 0.204081633, 0.202614379, 0.274509804]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "af240e3be399cd3467831fe1072b4074ab2048cfbe84b2d9d8730543548ef65f", "primary_control": "scratch", "result_hash": "ca92b8a2a7279d3140826a7d255da297f80486274c4e35d96ff9f83e5ed0a085", "signature": "e82942fdeaa470aabb18aa650f4875d3733844ea93e4fc3e0dfc6533f78c7b69e71368d9bbcac76cb94b84f7e606b5142995484b476bfc617a88c2e2c2b03701", "state": "VERIFIED", "summary": {"compressed": 0.247566, "gated": 0.247566, "naive": 0.283313, "scratch": 0.288415, "ungated": 0.240563}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:16:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:59Z", "cpu_seconds": 7.661696302999999, "created_at": "2026-11-22T03:16:26Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-96744c13dc13", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e78ef8c310cc69cd0ba733a3cd09a7da7d5f4d1a352335f1f0e40ef16c75a8f5", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-96744c13dc13", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:16:26Z", "seed": 677709340, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "1cd9185823ec080ab9394feaf9b3755bd5f4aa1ba838d73b826e8c1d10a85f1b", "run_seconds": 7.825508117675781, "seed": 677709340, "started_at": "2026-11-22T03:16:51Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-96744c13dc13", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.31010443696852297, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e78ef8c310cc69cd0ba733a3cd09a7da7d5f4d1a352335f1f0e40ef16c75a8f5", "primary_control": "reset", "result_hash": "1cd9185823ec080ab9394feaf9b3755bd5f4aa1ba838d73b826e8c1d10a85f1b", "signature": "c9635a02f3be4188e14f557ca3b1808f8762644de33de14f53bd76f69ecced6b3a455b1fe4192d0717f053216637e3542cf9b0af82db2ab69032db8a57f6bf0f", "state": "PARTIAL", "summary": {"inherit": 0.6833333333333333, "reset": 0.6833333333333333, "root": 0.4666666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:17:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:14Z", "cpu_seconds": 8.582865036000001, "created_at": "2026-11-22T03:16:42Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-5e699cbfe853", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "8d1409e55cee79ba59f3056ca35b6e5177107ca759a85537e6047afa29479c6a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-5e699cbfe853", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:16:42Z", "seed": 1066300474, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9cc24f2918276c0ac0d1cb6e6a6aed573255a494770018025b983846d7001fc7", "run_seconds": 10.324089050292969, "seed": 1066300474, "started_at": "2026-11-22T03:17:04Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:17:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-5e699cbfe853", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "root": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8d1409e55cee79ba59f3056ca35b6e5177107ca759a85537e6047afa29479c6a", "primary_control": "reset", "result_hash": "9cc24f2918276c0ac0d1cb6e6a6aed573255a494770018025b983846d7001fc7", "signature": "10f45aefa954f6cb834c72593537d317c225f4544784388c8a838df79cea23aa51b56bfe8cf3730567e0a22f9cdd2f73fd8b6dc19bf34e3d472e4cf02de7b901", "state": "PARTIAL", "summary": {"inherit": 0.7333333333333333, "reset": 0.7333333333333333, "root": 0.48333333333333334}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:17:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:56Z", "cpu_seconds": 1.5909641220000026, "created_at": "2026-11-22T03:16:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e3763ea9e75a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "4cc9daf3eb6b1b1fb1ff15290b649bb3e2c98140c5e58f172b8d82e054e24718", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e3763ea9e75a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:16:43Z", "seed": 576964447, "treatment": "frontier"}, "priority": 2.009239634695544, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4514fb9937122fd9882a7f5ac446559c20e0176b5231e06d8b08a2a0d376952a", "run_seconds": 1.652453899383545, "seed": 576964447, "started_at": "2026-11-22T03:16:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.201376, 0.150287], "effect": -0.025544, "evaluated_at": "2026-11-22T03:16:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e3763ea9e75a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.535294118, 0.334246575, 0.413157895, 0.457142857], "frontier": [0.670588235, 0.317808219, 0.192105263, 0.252380952], "goal": [0.470588235, 0.298630137, 0.268421053, 0.238095238], "hypdisc": [0.511764706, 0.326027397, 0.371052632, 0.395238095], "infogain": [0.647058824, 0.317808219, 0.192105263, 0.233333333], "prederr": [0.141176471, 0.224657534, 0.192105263, 0.1], "random": [0.441176471, 0.356164384, 0.371052632, 0.366666667], "risk": [0.670588235, 0.317808219, 0.192105263, 0.252380952]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "4cc9daf3eb6b1b1fb1ff15290b649bb3e2c98140c5e58f172b8d82e054e24718", "primary_control": "random", "result_hash": "4514fb9937122fd9882a7f5ac446559c20e0176b5231e06d8b08a2a0d376952a", "signature": "38175808ad35359a97157a6c5aead7e2246f2c7aaabbe7ba9b2a27a6bb4428c72bde4723ec61a867c5fa32ecbd7de7511108b79e28e060873ff5d18d766e250b", "state": "VERIFIED", "summary": {"count": 0.43496, "frontier": 0.358221, "goal": 0.318934, "hypdisc": 0.401021, "infogain": 0.347576, "prederr": 0.164485, "random": 0.383765, "risk": 0.358221}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:16:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:49Z", "cpu_seconds": 0.5399691459999971, "created_at": "2026-11-22T03:16:43Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-dca10fdd2f83", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "fd01c729a60c2f0753f4454e401d1f4b5417fd73fd4933c1faa3214e39c402ae", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-dca10fdd2f83", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:16:43Z", "seed": 2007116208, "treatment": "full"}, "priority": 2.0097730127612867, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4d92ebc361a5d5ad49f353e97867ff5b3d709e8680dd7954d816720eddc121b7", "run_seconds": 0.6871397495269775, "seed": 2007116208, "started_at": "2026-11-22T03:16:49Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-dca10fdd2f83", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.0, 1.0], "fixed_policy": [1.0, 0.0, 0.5, 0.0], "full": [1.0, 1.0, 0.0, 1.0], "no_world_model": [1.0, 1.0, 0.0, 1.0], "random": [1.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fd01c729a60c2f0753f4454e401d1f4b5417fd73fd4933c1faa3214e39c402ae", "primary_control": "e141_best", "result_hash": "4d92ebc361a5d5ad49f353e97867ff5b3d709e8680dd7954d816720eddc121b7", "signature": "af96f6ae1ec8b5b656141fe858f2397f93b664002360494388a8714d3b4fdf791233a4b3d447330d23a467b43b414b847f5c38be86290641494ef7868cab1908", "state": "PARTIAL", "summary": {"e141_best": 0.75, "fixed_policy": 0.375, "full": 0.75, "no_world_model": 0.75, "random": 0.5}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:16:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:16:49Z", "cpu_seconds": 3.465167978, "created_at": "2026-11-22T03:16:43Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d6da46520b12", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "prereg_hash": "935f1fa55da514f47886330af124f2d8ec72c0eefa0ce7be81df8a92379a8cfd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d6da46520b12", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:16:43Z", "seed": 1029202220, "treatment": "gated"}, "priority": 2.0098014196111924, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "be1b8f0dc7ee2760652e7fe97885cd2d32c2949dd19c66e9dc6ecd3d90c00fea", "run_seconds": 4.124882459640503, "seed": 1029202220, "started_at": "2026-11-22T03:16:45Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:16:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d6da46520b12", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.126666667, 0.866666667, 0.446511628, 0.84], "gated": [0.126666667, 0.866666667, 0.446511628, 0.84], "naive": [0.088888889, 0.658333333, 0.23255814, 0.41], "scratch": [0.126666667, 0.866666667, 0.446511628, 0.84], "ungated": [0.088888889, 0.658333333, 0.179069767, 0.42]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "935f1fa55da514f47886330af124f2d8ec72c0eefa0ce7be81df8a92379a8cfd", "primary_control": "scratch", "result_hash": "be1b8f0dc7ee2760652e7fe97885cd2d32c2949dd19c66e9dc6ecd3d90c00fea", "signature": "2273ba80a783d912e7d9b52c1847bae8f9aa85f39153803b7f9d1c786af65427cfae1f13aa2205d7bd21dcd9042463dc33205d7ce59b9aaef0c0124952712a03", "state": "PARTIAL", "summary": {"compressed": 0.569961, "gated": 0.569961, "naive": 0.347445, "scratch": 0.569961, "ungated": 0.336573}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:16:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:08Z", "cpu_seconds": 5.636230693999998, "created_at": "2026-11-22T03:16:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5d5f95c42887", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "cb30a7baf85827fa4ef5afbe887565827f268f8071eb3a03b1fe6d48d7bb86c7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5d5f95c42887", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:16:59Z", "seed": 2025523919, "treatment": "frontier"}, "priority": 2.0098770477353436, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1f8e45827a935c5657ed7e9caebd75758485d6873ea2f670cd9543bb762c2fc6", "run_seconds": 7.379504442214966, "seed": 2025523919, "started_at": "2026-11-22T03:17:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001542, 0.0005], "effect": -0.000521, "evaluated_at": "2026-11-22T03:17:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5d5f95c42887", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.994047619, 1.0, 1.0], "frontier": [0.997916667, 1.0, 0.988095238, 1.0], "goal": [0.997916667, 1.0, 0.988095238, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [0.997916667, 1.0, 0.988095238, 1.0], "prederr": [0.997916667, 1.0, 0.988095238, 1.0], "random": [1.0, 1.0, 0.988095238, 1.0], "risk": [0.997916667, 1.0, 0.988095238, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "cb30a7baf85827fa4ef5afbe887565827f268f8071eb3a03b1fe6d48d7bb86c7", "primary_control": "random", "result_hash": "1f8e45827a935c5657ed7e9caebd75758485d6873ea2f670cd9543bb762c2fc6", "signature": "e811a5e4e7c5e95a91baf9bf3a7a823e14821dbf8bcfb837fc70c663f1a411c838e61ec2653953bdb213a800703da4647964fa5a7766419c7fe8f6b65b6e160e", "state": "VERIFIED", "summary": {"count": 0.998512, "frontier": 0.996503, "goal": 0.996503, "hypdisc": 1.0, "infogain": 0.996503, "prederr": 0.996503, "random": 0.997024, "risk": 0.996503}, "ties": 3, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:17:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:03Z", "cpu_seconds": 1.4140220130000003, "created_at": "2026-11-22T03:16:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c63f1c8336a4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "74278502c4384033fc46f547a587e108a51321f2cd13451e14d7eefba0aef687", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c63f1c8336a4", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:16:59Z", "seed": 2060335534, "treatment": "frontier"}, "priority": 2.0099930055873028, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e94b078ad489189ce2c26a140128c5b24d4842ab8ebe2d55989082614d54be94", "run_seconds": 2.1264305114746094, "seed": 2060335534, "started_at": "2026-11-22T03:17:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.113971, 0.461557], "effect": 0.173793, "evaluated_at": "2026-11-22T03:17:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c63f1c8336a4", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.43, 0.75, 0.494949495, 0.68627451], "frontier": [0.796666667, 0.763888889, 0.525252525, 0.882352941], "goal": [0.503333333, 0.791666667, 0.545454545, 0.882352941], "hypdisc": [0.506666667, 0.819444444, 0.757575758, 0.882352941], "infogain": [0.366666667, 0.791666667, 0.393939394, 0.882352941], "prederr": [0.46, 0.791666667, 0.737373737, 0.882352941], "random": [0.43, 0.694444444, 0.717171717, 0.431372549], "risk": [0.796666667, 0.763888889, 0.525252525, 0.882352941]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "74278502c4384033fc46f547a587e108a51321f2cd13451e14d7eefba0aef687", "primary_control": "random", "result_hash": "e94b078ad489189ce2c26a140128c5b24d4842ab8ebe2d55989082614d54be94", "signature": "ed3e0f9aa5bed3500d4dba5c85071377d3b4a15f60a03060be1390635363b3608535d21e134ac8d2113aae9e5977dde856f426c2c1aac1d1f6a1067e63db9606", "state": "VERIFIED", "summary": {"count": 0.590306, "frontier": 0.74204, "goal": 0.680702, "hypdisc": 0.74151, "infogain": 0.608656, "prederr": 0.717848, "random": 0.568247, "risk": 0.74204}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:17:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:08Z", "cpu_seconds": 0.1455940360000021, "created_at": "2026-11-22T03:16:59Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2712a4505cb7", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "a2c0ab2fd8fa6bb7975553965dc9981cbb704fcd4d8293ab41f5c4ec5744f844", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2712a4505cb7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:16:59Z", "seed": 1657935374, "treatment": "full"}, "priority": 2.009864365873848, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-3a9f78a62746", "result_hash": "4b2cf08b2704a281a30118ce43bb632720c8f7e3fc2dc03eccbbdc79519daad4", "run_seconds": 0.24075913429260254, "seed": 1657935374, "started_at": "2026-11-22T03:17:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.101965, 0.163187], "effect": 0.132576, "evaluated_at": "2026-11-22T03:17:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2712a4505cb7", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.712121212, 0.53030303, 0.696969697, 0.575757576], "frozen": [0.015151515, 0.03030303, 0.03030303, 0.03030303], "full": [0.878787879, 0.666666667, 0.787878788, 0.712121212], "retrieval": [0.863636364, 0.651515152, 0.772727273, 0.681818182], "rules_nodetect": [0.575757576, 0.46969697, 0.378787879, 0.590909091]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a2c0ab2fd8fa6bb7975553965dc9981cbb704fcd4d8293ab41f5c4ec5744f844", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-3a9f78a62746", "result_hash": "4b2cf08b2704a281a30118ce43bb632720c8f7e3fc2dc03eccbbdc79519daad4", "signature": "f0b63d3042b82a5053884aa08d1ff9c2b5c3565c1874d3000df90f7f8b0d809807212fd5ae69b8fa0ace884be5f61e41b0f373d41bdae10323358c7c5235560e", "state": "VERIFIED", "summary": {"facts": 0.628788, "frozen": 0.026515, "full": 0.761364, "retrieval": 0.742424, "rules_nodetect": 0.503788}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:17:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:14Z", "cpu_seconds": 0.4276479300000027, "created_at": "2026-11-22T03:16:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-fa43be557454", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "prereg_hash": "0497cf89f325aba13d0eed680ff075c4fc1eb952e14cf064c8ccc8fc487c531d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-fa43be557454", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:16:59Z", "seed": 1449187966, "treatment": "gated"}, "priority": 2.0098350417278503, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "1dfffe605622c4cf150710e102f7d06bf3c50ac2700e70adfd7976062f9b7b11", "run_seconds": 0.6455991268157959, "seed": 1449187966, "started_at": "2026-11-22T03:17:13Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:17:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-fa43be557454", "family": "ow_transfer", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"compressed": [1.0, 0.602083], "gated": [1.0, 0.602083], "naive": [1.0, 0.145833], "scratch": [1.0, 0.577083], "ungated": [1.0, 0.338542]}, "prereg_hash": "0497cf89f325aba13d0eed680ff075c4fc1eb952e14cf064c8ccc8fc487c531d", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "1dfffe605622c4cf150710e102f7d06bf3c50ac2700e70adfd7976062f9b7b11", "signature": "37a2a7217ada742f002bfdeaf8e758899f164f28968e43e57d47934edb15d04a1e00ea939e61596c47c268eab361f522cfed916ed4a8900d9b558d3da0e8300c", "state": "REJECTED", "summary": {"compressed": 0.602083, "gated": 0.602083, "naive": 0.145833, "scratch": 0.577083, "ungated": 0.338542}}, "verified_at": "2026-11-22T03:17:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:21Z", "cpu_seconds": 4.105646038000003, "created_at": "2026-11-22T03:17:14Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3ab12ba7a4dd", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "5f86b5163bbc49f28354824f60b24cfc761a2255d1b0bf746cb4a414abdd4b9a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3ab12ba7a4dd", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:17:14Z", "seed": 1808145215, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "410988a13b11c9490863f781e40e304f4bbaf10e5139227d49cc4ce71ee03093", "run_seconds": 4.489995718002319, "seed": 1808145215, "started_at": "2026-11-22T03:17:16Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:17:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3ab12ba7a4dd", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0], "root": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "5f86b5163bbc49f28354824f60b24cfc761a2255d1b0bf746cb4a414abdd4b9a", "primary_control": "reset", "result_hash": "410988a13b11c9490863f781e40e304f4bbaf10e5139227d49cc4ce71ee03093", "signature": "d6c6e3339f08ac46ca202bfb829f3f6cb56b16213b99f153f34a31320e82c1781f39a13ec9754d859017383438ef25b2ae7fbfccbab6c9e1f19ea58102065906", "state": "PARTIAL", "summary": {"inherit": 0.5666666666666667, "reset": 0.5666666666666667, "root": 0.25}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:17:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:57Z", "cpu_seconds": 1.6954833020000004, "created_at": "2026-11-22T03:17:15Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-694075ef4ada", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "7d370cc5e67253e7fa4b43bc3fb4a39cabf0d39f92769dcb46e8a4b85047004b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-694075ef4ada", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:17:15Z", "seed": 81433701, "treatment": "iw_temperature"}, "priority": 2.0095209519854396, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-7f328ec3a5ba", "result_hash": "91b2ee8f5cc52082467982b8dc262fbcd67f4d45e1e79216ede2cea18bf13866", "run_seconds": 1.8389196395874023, "seed": 81433701, "started_at": "2026-11-22T03:17:55Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.01544, -0.003857], "effect": -0.009649, "evaluated_at": "2026-11-22T03:18:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-694075ef4ada", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.044880417, 0.040520003, 0.058342338, 0.054709792, 0.070211883, 0.054046128, 0.049998023, 0.020958663, 0.035334385, 0.07774125, 0.045439443, 0.03745689, 0.037550712, 0.058127187, 0.046569453, 0.065845858], "raw": [0.044880417, 0.05179, 0.051682318, 0.077971585, 0.051383325, 0.048541052, 0.051181252, 0.053817495, 0.040333822, 0.06292239, 0.039245355, 0.046439925, 0.02763521, 0.03395461, 0.057464603, 0.031538085], "temperature": [0.031322455, 0.030805265, 0.058342338, 0.049587787, 0.066733948, 0.03436544, 0.036859052, 0.012795567, 0.035334385, 0.044861092, 0.02917651, 0.035393912, 0.035720275, 0.06792346, 0.042596508, 0.031538085]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 13, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7d370cc5e67253e7fa4b43bc3fb4a39cabf0d39f92769dcb46e8a4b85047004b", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-7f328ec3a5ba", "result_hash": "91b2ee8f5cc52082467982b8dc262fbcd67f4d45e1e79216ede2cea18bf13866", "signature": "3178418a83833c408091234f8b717e683954c8e0ecadbe59dd3772c83616da9b1aa15d09a89ba76f08a9a67c01c7f24d9a46e6218087a6ea4d49265cf36b2b0e", "state": "VERIFIED", "summary": {"iw_temperature": 0.049858, "raw": 0.048174, "temperature": 0.04021}, "ties": 2, "treatment": "iw_temperature", "wins": 1}, "verified_at": "2026-11-22T03:18:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:24Z", "cpu_seconds": 2.1463744279999943, "created_at": "2026-11-22T03:17:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-568ad45cffe2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "4cb8224da83f5202f1909ebaf0b5954f315dd1b2a1a5308ee823b7b93e78184f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-568ad45cffe2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:17:15Z", "seed": 2043692349, "treatment": "frontier"}, "priority": 2.009711461484686, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "11081ae4f5667849778f64ef539da201e969f45fa005ba4f6ed695823d051560", "run_seconds": 3.1281492710113525, "seed": 2043692349, "started_at": "2026-11-22T03:17:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.167628, 0.273059], "effect": 0.052716, "evaluated_at": "2026-11-22T03:17:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-568ad45cffe2", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.571794872, 0.573770492, 0.539805825, 0.53592233], "frontier": [0.430769231, 0.590163934, 0.386407767, 0.741747573], "goal": [0.320512821, 0.554098361, 0.386407767, 0.741747573], "hypdisc": [0.284615385, 0.491803279, 0.510679612, 0.415533981], "infogain": [0.633333333, 0.580327869, 0.269902913, 0.186407767], "prederr": [0.566666667, 0.580327869, 0.172815534, 0.186407767], "random": [0.569230769, 0.504918033, 0.47961165, 0.384466019], "risk": [0.430769231, 0.590163934, 0.386407767, 0.153398058]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "4cb8224da83f5202f1909ebaf0b5954f315dd1b2a1a5308ee823b7b93e78184f", "primary_control": "random", "result_hash": "11081ae4f5667849778f64ef539da201e969f45fa005ba4f6ed695823d051560", "signature": "d8ffe973950c6531320e18dcedb6d4c5ba05ad8f28824aa1744f24c88f2be6aad1676e0fc79f3b6ca19433531924cdbb5029f7ace9044788168fb790e3e12707", "state": "VERIFIED", "summary": {"count": 0.555323, "frontier": 0.537272, "goal": 0.500692, "hypdisc": 0.425658, "infogain": 0.417493, "prederr": 0.376554, "random": 0.484557, "risk": 0.390185}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:17:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:25Z", "cpu_seconds": 3.382869229999983, "created_at": "2026-11-22T03:17:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7e685ff2df25", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "92e735e745a7e89326d1fce4634fad3e07e27afff18f5d66b1647cf489a2dd1a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7e685ff2df25", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:17:15Z", "seed": 1657504967, "treatment": "frontier"}, "priority": 2.0097413940552302, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4b85cd6abe2f3c544d7205bc1d71aadabdb423f9f61c7658e84e1066a18e9778", "run_seconds": 4.945580244064331, "seed": 1657504967, "started_at": "2026-11-22T03:17:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.20464, 0.050517], "effect": -0.077062, "evaluated_at": "2026-11-22T03:17:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7e685ff2df25", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.377308707, 0.62755102, 1.0, 0.888888889], "frontier": [0.341688654, 0.520408163, 0.916666667, 0.435185185], "goal": [0.337730871, 0.423469388, 0.854166667, 0.462962963], "hypdisc": [0.349604222, 0.56122449, 0.989583333, 0.62962963], "infogain": [0.341688654, 0.520408163, 0.916666667, 0.435185185], "prederr": [0.341688654, 0.520408163, 0.916666667, 0.435185185], "random": [0.360158311, 0.5, 0.958333333, 0.703703704], "risk": [0.341688654, 0.520408163, 0.916666667, 0.435185185]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "92e735e745a7e89326d1fce4634fad3e07e27afff18f5d66b1647cf489a2dd1a", "primary_control": "random", "result_hash": "4b85cd6abe2f3c544d7205bc1d71aadabdb423f9f61c7658e84e1066a18e9778", "signature": "b8edd264a52c622dd236153c4cf2095d178645518938c68ddaee7857f93798d2506b8f8d9aad29ed11498f3a73b18d51fed8f87056b6cf0fd58ef14527b9ac0c", "state": "VERIFIED", "summary": {"count": 0.723437, "frontier": 0.553487, "goal": 0.519582, "hypdisc": 0.63251, "infogain": 0.553487, "prederr": 0.553487, "random": 0.630549, "risk": 0.553487}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:17:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:21Z", "cpu_seconds": 1.9245154390000003, "created_at": "2026-11-22T03:17:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-eca2c46621b4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "5ea5c74421f9edea3171166beb0f88f755b0738f7980f8a7d34fe91887acb5be", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-eca2c46621b4", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:17:15Z", "seed": 1167965685, "treatment": "frontier"}, "priority": 2.0097500685313654, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "0f02ac4c006d66cf696c243c6cc13d52fee2d235928f8e070c3b04a9248f266d", "run_seconds": 2.2951157093048096, "seed": 1167965685, "started_at": "2026-11-22T03:17:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.007793, 0.054707], "effect": 0.03125, "evaluated_at": "2026-11-22T03:17:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-eca2c46621b4", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.966666667, 0.983333333, 1.0, 0.966666667], "goal": [0.825, 0.9, 1.0, 0.825], "hypdisc": [0.958333333, 0.983333333, 1.0, 0.966666667], "infogain": [0.966666667, 0.983333333, 1.0, 0.966666667], "prederr": [0.966666667, 0.983333333, 1.0, 0.966666667], "random": [0.916666667, 0.933333333, 1.0, 0.941666667], "risk": [0.966666667, 0.983333333, 1.0, 0.966666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "5ea5c74421f9edea3171166beb0f88f755b0738f7980f8a7d34fe91887acb5be", "primary_control": "random", "result_hash": "0f02ac4c006d66cf696c243c6cc13d52fee2d235928f8e070c3b04a9248f266d", "signature": "d5c7a39f719c9bd0c5a24621d2ed4a5fabfaf7cbc14d2aa789765c3b69847160d36e6fed8a7b7ad51fb47f8bb675011dd09e2e9812ffc6c7afdb6a91a613f100", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.979167, "goal": 0.8875, "hypdisc": 0.977083, "infogain": 0.979167, "prederr": 0.979167, "random": 0.947917, "risk": 0.979167}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:17:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:05Z", "cpu_seconds": 1.0985703119999997, "created_at": "2026-11-22T03:17:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-57e76a9bc8af", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "f537c53ebf784a2b6d785d128ed65c65a0f9486a60bc6b7c8b8c7e99e981c0d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-57e76a9bc8af", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:17:15Z", "seed": 1827245238, "treatment": "full"}, "priority": 2.0089307984002485, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ec57f7c7439332bb4bfd1089637976bed8aa506229109e9f67d58d914774585c", "run_seconds": 1.2592198848724365, "seed": 1827245238, "started_at": "2026-11-22T03:19:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.066868, 0.047188], "effect": -0.00984, "evaluated_at": "2026-11-22T03:20:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-57e76a9bc8af", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.552941176, 0.684210526, 0.610666667, 0.414117647], "frozen": [0.035294118, 0.136842105, 0.128, 0.12], "full": [0.629411765, 0.642105263, 0.584, 0.367058824], "retrieval": [0.582352941, 0.673684211, 0.605333333, 0.392941176], "rules_nodetect": [0.629411765, 0.642105263, 0.584, 0.367058824]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "f537c53ebf784a2b6d785d128ed65c65a0f9486a60bc6b7c8b8c7e99e981c0d2", "primary_control": "facts", "result_hash": "ec57f7c7439332bb4bfd1089637976bed8aa506229109e9f67d58d914774585c", "signature": "5cf91f5d5a95483b2439356cd4ec633c29501ed08fc9b23c254e5fe8e5f30b035f92a676ee2bad556e841462f4f6af3e15a6d3a9123bd307fa2dfcbff831f20b", "state": "VERIFIED", "summary": {"facts": 0.565484, "frozen": 0.105034, "full": 0.555644, "retrieval": 0.563578, "rules_nodetect": 0.555644}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:20:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:27Z", "cpu_seconds": 1.0097739110000035, "created_at": "2026-11-22T03:17:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c9cdda06711d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "48e16f18690f204be3f19a43396a1e46108bb1b6e5ae9ca382836f8105b0c187", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c9cdda06711d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:17:15Z", "seed": 1248619469, "treatment": "full"}, "priority": 2.009681454896022, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "58775ef7ce6cef9a1133547b44026027d9010e05adf64a7b286876e0e757d855", "run_seconds": 1.0221927165985107, "seed": 1248619469, "started_at": "2026-11-22T03:17:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.009617, 0.08205], "effect": 0.045833, "evaluated_at": "2026-11-22T03:17:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c9cdda06711d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.916666667, 1.0, 0.916666667, 0.916666667], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.983333333, 1.0, 0.95, 1.0], "retrieval": [1.0, 1.0, 1.0, 0.95], "rules_nodetect": [0.983333333, 1.0, 0.95, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "48e16f18690f204be3f19a43396a1e46108bb1b6e5ae9ca382836f8105b0c187", "primary_control": "facts", "result_hash": "58775ef7ce6cef9a1133547b44026027d9010e05adf64a7b286876e0e757d855", "signature": "c6037ae6c0a567ff087a4df3b76a54584fd65ffe825587e1f020c7c6f68270eaf36277fbcf2029e0e88eda17ce9a130eaab73d4d5d959466879d366172f3f80b", "state": "VERIFIED", "summary": {"facts": 0.9375, "frozen": 0.0, "full": 0.983333, "retrieval": 0.9875, "rules_nodetect": 0.983333}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:17:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:42Z", "cpu_seconds": 1.877631603, "created_at": "2026-11-22T03:17:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0bb7b35ed4d4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "d0d3213865238a39f920ebdbc1d4383c143796e01e4e92f303b1fd822709aa96", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0bb7b35ed4d4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:17:30Z", "seed": 2020268027, "treatment": "full"}, "priority": 2.009987474499197, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f2bd168bc537bbc56a7984d0d4fc79c79819efe3a97f420ae3884fb39d379fac", "run_seconds": 2.670820713043213, "seed": 2020268027, "started_at": "2026-11-22T03:17:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T03:17:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0bb7b35ed4d4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.5, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 0.0, 0.0, 0.0], "no_world_model": [0.0, 0.0, 0.75, 0.0], "random": [0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d0d3213865238a39f920ebdbc1d4383c143796e01e4e92f303b1fd822709aa96", "primary_control": "e141_best", "result_hash": "f2bd168bc537bbc56a7984d0d4fc79c79819efe3a97f420ae3884fb39d379fac", "signature": "55ade119e92462b8dddaec31754866ac3038ebaf052aeb93fd4451615a5c5e260bc4280446d27eeee739c481b98c925f47dfd12c9be6f79b80e959476bdcfa02", "state": "VERIFIED", "summary": {"e141_best": 0.125, "fixed_policy": 0.0, "full": 0.0, "no_world_model": 0.1875, "random": 0.0}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:18:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:40Z", "cpu_seconds": 0.5564851369999999, "created_at": "2026-11-22T03:17:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-6a2ebe0ce5c0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "9ab8b3693ac3db70a523dc03e1e2f58632c393326dce7124098d16d66860f664", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-6a2ebe0ce5c0", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:17:30Z", "seed": 2044427005, "treatment": "full"}, "priority": 2.0099551779402494, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8b7860f5f3bff68539af89eb287683e1ac84a9644fc4e07837e371ba463bbc19", "run_seconds": 0.6399943828582764, "seed": 2044427005, "started_at": "2026-11-22T03:17:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T03:17:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-6a2ebe0ce5c0", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.5, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.25, 1.0, 0.5, 1.0], "no_world_model": [0.75, 1.0, 0.0, 1.0], "random": [0.75, 0.5, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "9ab8b3693ac3db70a523dc03e1e2f58632c393326dce7124098d16d66860f664", "primary_control": "e141_best", "result_hash": "8b7860f5f3bff68539af89eb287683e1ac84a9644fc4e07837e371ba463bbc19", "signature": "28d9907a12a6cb636f95682f5a97ff4307ffc305200addfe37223c1af993c0987b8114ab0049a6c8777323d2c30b926200a1a4d213d0e915c0b65c4c2a3c9504", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0, "full": 0.6875, "no_world_model": 0.6875, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:17:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:41Z", "cpu_seconds": 0.69467556, "created_at": "2026-11-22T03:17:31Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c367e76c250e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "81f42ce014fa0aba94c95d159c77fb5b1dd0bf5f4bc254821be03b798e26144d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c367e76c250e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:17:31Z", "seed": 1875935525, "treatment": "full"}, "priority": 2.009889426501275, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3c903af1d9417c3e4128dbcc5ecabf0e1f26a4e51d6b80c32a928538d5a51918", "run_seconds": 1.2734949588775635, "seed": 1875935525, "started_at": "2026-11-22T03:17:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014248, 0.009265], "effect": -0.002492, "evaluated_at": "2026-11-22T03:17:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c367e76c250e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.464052288, 0.441176471, 0.736434109, 0.466666667], "frozen": [0.241830065, 0.333333333, 0.085271318, 0.3875], "full": [0.477124183, 0.426470588, 0.736434109, 0.458333333], "retrieval": [0.575163399, 0.441176471, 0.833333333, 0.658333333], "rules_nodetect": [0.464052288, 0.31372549, 0.736434109, 0.4625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "81f42ce014fa0aba94c95d159c77fb5b1dd0bf5f4bc254821be03b798e26144d", "primary_control": "facts", "result_hash": "3c903af1d9417c3e4128dbcc5ecabf0e1f26a4e51d6b80c32a928538d5a51918", "signature": "e4f8b80aabed68c6d3c34936fd9087ada8a294419e1707be1e7b37810a256e5acb078073af8162f7c92b056cc5f1cb2dbe7d9874d6e5eadbd2d5998d0a9b9506", "state": "VERIFIED", "summary": {"facts": 0.527082, "frozen": 0.261984, "full": 0.524591, "retrieval": 0.627002, "rules_nodetect": 0.494178}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:17:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:53Z", "cpu_seconds": 2.125110747, "created_at": "2026-11-22T03:17:31Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-656b2730cc97", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "e8105fb95abed9c24763410fa3c91819ada9d9bffe0554c9dd6b7f2a69267b73", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-656b2730cc97", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:17:31Z", "seed": 1232935632, "treatment": "frontier"}, "priority": 2.009671722720524, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-fb4b6cdb3bc2", "result_hash": "d9f82fd1f92f3879955cb458966f937c776a0602b32d7a3834f3597495d2bf4c", "run_seconds": 2.4055004119873047, "seed": 1232935632, "started_at": "2026-11-22T03:17:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.016878, 0.085258], "effect": 0.051068, "evaluated_at": "2026-11-22T03:18:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-656b2730cc97", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [1.0, 0.926470588, 0.853092784, 0.826388889, 0.838383838, 0.783980583, 1.0, 1.0, 1.0, 1.0, 1.0, 0.765789474], "frontier": [1.0, 1.0, 1.0, 0.918981481, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.981578947], "random_walk": [1.0, 0.891176471, 0.87371134, 0.856481481, 0.954545455, 0.890776699, 1.0, 1.0, 1.0, 1.0, 1.0, 0.821052632]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.8062, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0156", "prereg_hash": "e8105fb95abed9c24763410fa3c91819ada9d9bffe0554c9dd6b7f2a69267b73", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-fb4b6cdb3bc2", "result_hash": "d9f82fd1f92f3879955cb458966f937c776a0602b32d7a3834f3597495d2bf4c", "signature": "73b81ddb2db515ba8b6d37ae35bf304917ef4a5a00680c0f82ae5a376f3105f889814bc5a610c1f85a99f153d0e394d7176f2722a010a80d3d8897c18199b606", "state": "VERIFIED", "summary": {"curiosity": 0.916176, "frontier": 0.991713, "random_walk": 0.940645}, "ties": 6, "treatment": "frontier", "wins": 6}, "verified_at": "2026-11-22T03:18:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:50Z", "cpu_seconds": 3.367664899, "created_at": "2026-11-22T03:17:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b4cb817af41a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "bb9efccfedc2e7e2129e2424f7f9f0b1cdb8bbd8f90dbe0aaf651aed2a2d74f1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b4cb817af41a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:17:45Z", "seed": 544903833, "treatment": "full"}, "priority": 2.009752206994043, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a26ac81eecbacd0ba601250cdad5ed57d24e374b6acefe7374ca59cf75bff62b", "run_seconds": 4.07446813583374, "seed": 544903833, "started_at": "2026-11-22T03:17:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.299352, 0.549352], "effect": 0.125, "evaluated_at": "2026-11-22T03:18:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b4cb817af41a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.0, 0.5, 0.75], "fixed_policy": [1.0, 0.5, 0.25, 1.0], "full": [1.0, 0.25, 0.0, 1.0], "no_world_model": [0.5, 0.25, 0.75, 0.5], "random": [1.0, 0.75, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "bb9efccfedc2e7e2129e2424f7f9f0b1cdb8bbd8f90dbe0aaf651aed2a2d74f1", "primary_control": "e141_best", "result_hash": "a26ac81eecbacd0ba601250cdad5ed57d24e374b6acefe7374ca59cf75bff62b", "signature": "f0e50a479fa55c59480cc0cb2a0fbd51b2e27671440d4fa5372db4c5050ecbef55c1dbe9f3c6043a2a88e770f96fa837154e9449e2b0936f1babfd64b844ed08", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.6875, "full": 0.5625, "no_world_model": 0.5, "random": 0.8125}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:18:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:59Z", "cpu_seconds": 0.8419051689999999, "created_at": "2026-11-22T03:17:45Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6f524a036255", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "639764d3e956cf8407e70c4f64d815ca33b5189a265fa92cf4adec899f7b049c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6f524a036255", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:17:45Z", "seed": 114851680, "treatment": "full"}, "priority": 2.009345410662331, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "696bce467a826e2eb9fd473afa7dd730b6ad332e4b734ddfcd9bfe098252716c", "run_seconds": 0.8978934288024902, "seed": 114851680, "started_at": "2026-11-22T03:17:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026926, 0.105487], "effect": 0.039281, "evaluated_at": "2026-11-22T03:18:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6f524a036255", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.793103448, 0.5, 0.782945736, 0.775862069], "frozen": [0.333333333, 0.288690476, 0.348837209, 0.281609195], "full": [0.793103448, 0.639880952, 0.782945736, 0.793103448], "retrieval": [0.793103448, 0.556547619, 0.782945736, 0.798850575], "rules_nodetect": [0.816091954, 0.75, 0.782945736, 0.793103448]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "639764d3e956cf8407e70c4f64d815ca33b5189a265fa92cf4adec899f7b049c", "primary_control": "facts", "result_hash": "696bce467a826e2eb9fd473afa7dd730b6ad332e4b734ddfcd9bfe098252716c", "signature": "8275287f1be7728b0433bbf40a571cb53f251618f7860161c2758e453db241d19f09e8515ae30fdbeaa76d057f365137a2e171eaf84c42614f819d12144d1d03", "state": "VERIFIED", "summary": {"facts": 0.712978, "frozen": 0.313118, "full": 0.752258, "retrieval": 0.732862, "rules_nodetect": 0.785535}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:18:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:45Z", "cpu_seconds": 106.968484753, "created_at": "2026-11-22T03:17:45Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-d7fbaa30f6ff", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "b5853deb9653ea0720fb5cf28a16259f7a96b3b56252c12770278f31aef3e3dc", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-d7fbaa30f6ff", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:17:45Z", "seed": 1904023618, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "449b93a46e6e04873b5cdbb24d9b7a26310747697756e55819d984e83bf7acf9", "run_seconds": 117.95114088058472, "seed": 1904023618, "started_at": "2026-11-22T03:17:47Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 1, "inherit_mem": 1, "inherit_nomem": 0, "random_search": 0, "reset_mem": 1, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [-0.038439, 0.11852], "effect": 0.040041, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:20:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-d7fbaa30f6ff", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.243902439, 0.237053571], "compressed": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.644308943, 0.237053571], "inherit_mem": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.644308943, 0.237053571], "inherit_nomem": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.243902439, 0.237053571], "random_search": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.243902439, 0.237053571], "reset_mem": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.644308943, 0.237053571], "retrieval": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.243902439, 0.237053571], "rootcause": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.644308943, 0.237053571], "xfer_mem": [0.353658537, 1.0, 0.979166667, 0.697674419, 0.886153846, 1.0, 0.810606061, 0.758333333, 0.644308943, 0.237053571]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "per_generation_sealed": [0.696655, 0.736696, 0.736696, 0.736696], "prereg_hash": "b5853deb9653ea0720fb5cf28a16259f7a96b3b56252c12770278f31aef3e3dc", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "reset_mem", "random_search"], "result_hash": "449b93a46e6e04873b5cdbb24d9b7a26310747697756e55819d984e83bf7acf9", "signature": "997ddf006f44c6fbcc473433182a48f862a37bddffc9520cf5c8aa6b8a3410af1e94d3d82e677c42cb92b247c0b6d62be440aa407dd91212a46dc1c50d554105", "state": "VERIFIED", "summary": {"G0": 0.696655, "compressed": 0.736696, "inherit_mem": 0.736696, "inherit_nomem": 0.696655, "random_search": 0.696655, "reset_mem": 0.736696, "retrieval": 0.696655, "rootcause": 0.736696, "xfer_mem": 0.736696}, "ties": 9, "treatment": "inherit_mem", "wins": 1}, "verified_at": "2026-11-22T03:22:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:45Z", "cpu_seconds": 0.4557532169999998, "created_at": "2026-11-22T03:17:45Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-bc258f0fdce4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "prereg_hash": "06dc396c31b8f73273194ea8848827190d4b69d2a965397e5da0206c8352b8c1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-bc258f0fdce4", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:17:45Z", "seed": 1382578522, "treatment": "gated"}, "priority": 2.0098513553856123, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "90b3187bb49f12e02a6c66e86eb86bb448e0cf95f173d9bad1575d259f8ffe04", "run_seconds": 0.4797639846801758, "seed": 1382578522, "started_at": "2026-11-22T03:17:45Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:18:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-bc258f0fdce4", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.75, 0.75, 0.766666667, 1.0], "gated": [0.75, 0.75, 0.766666667, 1.0], "naive": [0.516666667, 0.683333333, 0.4, 0.4], "scratch": [0.75, 0.75, 0.766666667, 1.0], "ungated": [0.516666667, 0.683333333, 0.366666667, 0.4]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "06dc396c31b8f73273194ea8848827190d4b69d2a965397e5da0206c8352b8c1", "primary_control": "scratch", "result_hash": "90b3187bb49f12e02a6c66e86eb86bb448e0cf95f173d9bad1575d259f8ffe04", "signature": "662ad2e3f22888f7431ff6622d83c7f03b42248524cf58e8cb0645df5ebd3b596360e09b915f6718440f6dc0f2bb00e7c9e0c4118d3e3051c6f6e44aec137a00", "state": "PARTIAL", "summary": {"compressed": 0.816667, "gated": 0.816667, "naive": 0.5, "scratch": 0.816667, "ungated": 0.491667}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:18:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:17:55Z", "cpu_seconds": 5.999854674000005, "created_at": "2026-11-22T03:17:45Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-eda35ba5fd6f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "cf9c7096a3280b9b665d547bbd75e99504038bd54b2417cacc28e90225159760", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-eda35ba5fd6f", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:17:45Z", "seed": 1926999266, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a18c6178b78f0d62a34fec109aae2185b7065e9fe4749ffd7392c18eabdb676d", "run_seconds": 6.725833892822266, "seed": 1926999266, "started_at": "2026-11-22T03:17:48Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:18:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-eda35ba5fd6f", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.3593499880679335, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0], "root": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "cf9c7096a3280b9b665d547bbd75e99504038bd54b2417cacc28e90225159760", "primary_control": "reset", "result_hash": "a18c6178b78f0d62a34fec109aae2185b7065e9fe4749ffd7392c18eabdb676d", "signature": "133e3ed68853e2c16767ebf19e743d8546a0964cf166e780e212f4cc6474069d306d7df4d89b603c104e970013654f71ee139b0f48df7754771d5410805b900e", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.4166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:18:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:18Z", "cpu_seconds": 1.8268961839999998, "created_at": "2026-11-22T03:18:01Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b38b14a531f4", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "0924fe1187bbf33108cb9d442eede9c3d1cf781bab0260b8da37b4ab464cf5e8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b38b14a531f4", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:18:01Z", "seed": 1178245451, "treatment": "iw_temperature"}, "priority": 2.009451917576108, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-5eb1d85be89f", "result_hash": "399dcdae9153815748d3056e8fabbdafd33775539fa82e9deb6174617f201776", "run_seconds": 2.4445340633392334, "seed": 1178245451, "started_at": "2026-11-22T03:18:15Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012591, 0.027962], "effect": 0.007686, "evaluated_at": "2026-11-22T03:20:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b38b14a531f4", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.182288687, 0.316049345, 0.06917078, 0.315437452, 0.188979448, 0.232232897, 0.26965268, 0.25909929, 0.304299963, 0.244077325, 0.205144738, 0.3041101, 0.226159735, 0.293707368, 0.26793997, 0.261480988], "raw": [0.231599437, 0.300517365, 0.18710614, 0.256844198, 0.225115958, 0.258312705, 0.241836418, 0.2237761, 0.202619687, 0.169156048, 0.214736217, 0.275776103, 0.226159735, 0.269429735, 0.27374217, 0.242375042], "temperature": [0.231599437, 0.311529197, 0.19823484, 0.28671796, 0.225115958, 0.222940708, 0.26965268, 0.249337593, 0.246814598, 0.244077325, 0.2310578, 0.275776103, 0.24402808, 0.287803958, 0.26793997, 0.27017272]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "0924fe1187bbf33108cb9d442eede9c3d1cf781bab0260b8da37b4ab464cf5e8", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-5eb1d85be89f", "result_hash": "399dcdae9153815748d3056e8fabbdafd33775539fa82e9deb6174617f201776", "signature": "f07ff83f7825d7f4c49fd549a0c3395d2bb0e2efafddf14d46224b05330cc3fef1fa23afa485febad387fa922ccdbf33cea36396e914f602587d99ff43d31b01", "state": "VERIFIED", "summary": {"iw_temperature": 0.246239, "raw": 0.237444, "temperature": 0.253925}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:20:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:05Z", "cpu_seconds": 0.6294756129999994, "created_at": "2026-11-22T03:18:01Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-18546d2074a0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "6e09d2182b97b970e73d205630a041ed2eead8cd319787d3dd645ef9a654e701", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-18546d2074a0", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:18:01Z", "seed": 925100571, "treatment": "full"}, "priority": 2.0097072589343696, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6935e226a320364b4cb131dcf6ca556abe9faf3ddca673ac653e412dca63b598", "run_seconds": 1.0166912078857422, "seed": 925100571, "started_at": "2026-11-22T03:18:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:18:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-18546d2074a0", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.25, 0.5], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [1.0, 1.0, 0.25, 0.25], "no_world_model": [1.0, 1.0, 0.0, 0.25], "random": [1.0, 1.0, 0.5, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "6e09d2182b97b970e73d205630a041ed2eead8cd319787d3dd645ef9a654e701", "primary_control": "e141_best", "result_hash": "6935e226a320364b4cb131dcf6ca556abe9faf3ddca673ac653e412dca63b598", "signature": "8ba72e628d1358fd3850c8cb6a420161cfe2f4685088a44f0189be1c37742641a38ce51e542d6db962c623e5b94e2f88f4735b84a9d8ab3d5e0590de42ec720e", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.0625, "full": 0.625, "no_world_model": 0.5625, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:18:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:05Z", "cpu_seconds": 2.0920503989999997, "created_at": "2026-11-22T03:18:01Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-41d8420dc9e5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 3, "n": 4}, "prereg_hash": "f2a39d33ffff6465242d352e5c0e0e699885dd5cbcd11bea90826bc8c570cae5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-41d8420dc9e5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:18:01Z", "seed": 2009317432, "treatment": "full"}, "priority": 2.009712154617181, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bd30dea2b390d7931e28296440b068608d3e6ccb3b5f915d52352dc3f16224df", "run_seconds": 3.1063144207000732, "seed": 2009317432, "started_at": "2026-11-22T03:18:02Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:18:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-41d8420dc9e5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.0, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 0.0, 0.0, 0.0], "no_world_model": [1.0, 0.0, 0.0, 0.0], "random": [0.5, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f2a39d33ffff6465242d352e5c0e0e699885dd5cbcd11bea90826bc8c570cae5", "primary_control": "e141_best", "result_hash": "bd30dea2b390d7931e28296440b068608d3e6ccb3b5f915d52352dc3f16224df", "signature": "d98ddde982cbaa7a1234bf6400c7ade950fecaf4cc3e024535e50f42b4c2db2213a54e6ba44c555e2265603e52b019ad1514817a3d7651c6de046c1f7ab28b03", "state": "PARTIAL", "summary": {"e141_best": 0.25, "fixed_policy": 0.0, "full": 0.25, "no_world_model": 0.25, "random": 0.125}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:20:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:10Z", "cpu_seconds": 0.45111848800000054, "created_at": "2026-11-22T03:18:01Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-486dd1639a54", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "ab1ead47efb40948f1fe0268ac30ed493de8ef9d43599d29e4c647cb59caadc2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-486dd1639a54", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:18:01Z", "seed": 1905514001, "treatment": "full"}, "priority": 2.009597716041058, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bec82d37a8ec009a518abd517ef3f4459574a49db522d9e5d963d09bd9fdb746", "run_seconds": 0.45485806465148926, "seed": 1905514001, "started_at": "2026-11-22T03:18:10Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:20:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-486dd1639a54", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 1.0, 1.0], "random": [1.0, 0.5, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ab1ead47efb40948f1fe0268ac30ed493de8ef9d43599d29e4c647cb59caadc2", "primary_control": "e141_best", "result_hash": "bec82d37a8ec009a518abd517ef3f4459574a49db522d9e5d963d09bd9fdb746", "signature": "982c2e5d9645ab328dffe4df56627a23470cd3e4df8d31f9707f772c8886ea7c8a1468688e6d0add73e756fa79fd3ceb9c41e5c669e939dfd6e32ad57f57c202", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 0.875, "random": 0.625}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:20:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:14Z", "cpu_seconds": 2.361402224999999, "created_at": "2026-11-22T03:18:01Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a56a76034979", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "74d6bb56e851dac6a44177745c0f1774f8fd7c2ee5be9c55fb302973ec756ee3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a56a76034979", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:18:01Z", "seed": 1335844662, "treatment": "full"}, "priority": 2.00955040040395, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "823308fdf6d0e0649162e607f6218eff0a858d6ad71a35ea0f124f2e817306ea", "run_seconds": 3.1557390689849854, "seed": 1335844662, "started_at": "2026-11-22T03:18:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001212, 0.003737], "effect": 0.001263, "evaluated_at": "2026-11-22T03:20:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a56a76034979", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.398989899, 0.981220657, 0.943661972, 0.937198068], "frozen": [0.090909091, 0.084507042, 0.084507042, 0.077294686], "full": [0.404040404, 0.981220657, 0.943661972, 0.937198068], "retrieval": [0.601010101, 0.981220657, 0.976525822, 0.942028986], "rules_nodetect": [0.404040404, 0.981220657, 0.943661972, 0.937198068]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "74d6bb56e851dac6a44177745c0f1774f8fd7c2ee5be9c55fb302973ec756ee3", "primary_control": "facts", "result_hash": "823308fdf6d0e0649162e607f6218eff0a858d6ad71a35ea0f124f2e817306ea", "signature": "7b1e52875b0a401d1b464f4149d39b8002d4a908d647377ea80a283ee5198c1d6769c29bded53f94a55028a8afaa4318a329a536883a9ee1b6302aa35358630b", "state": "VERIFIED", "summary": {"facts": 0.815268, "frozen": 0.084304, "full": 0.81653, "retrieval": 0.875196, "rules_nodetect": 0.81653}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:20:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:09Z", "cpu_seconds": 6.273972348000001, "created_at": "2026-11-22T03:18:01Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-618bb21da74e", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "b20e728cafbfed5990d1880efa8397c7a889040f2fc1dff6f1f5297d3d878912", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-618bb21da74e", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:18:01Z", "seed": 70437398, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "01facf658f0baf6131868e8c5bdd426c525e1cf1e04008b6ff3ac4ee008c52c2", "run_seconds": 7.89863395690918, "seed": 70437398, "started_at": "2026-11-22T03:18:02Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.092508, 0.125841], "effect": 0.016667, "evaluated_at": "2026-11-22T03:20:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-618bb21da74e", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5901563906994808, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0], "reset": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.301, "losses": 5, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "b20e728cafbfed5990d1880efa8397c7a889040f2fc1dff6f1f5297d3d878912", "primary_control": "reset", "result_hash": "01facf658f0baf6131868e8c5bdd426c525e1cf1e04008b6ff3ac4ee008c52c2", "signature": "13aba133cdf98778b86afd8ce31f851936dea848f5708c2b4ab36be13cda5ff0fe4e9b9c48c4bd6518aca4694e6fbcbbf9f621fc521838f40aab7b5db9771a00", "state": "VERIFIED", "summary": {"inherit": 0.55, "reset": 0.5333333333333333, "root": 0.26666666666666666}, "ties": 49, "treatment": "inherit", "wins": 6}, "verified_at": "2026-11-22T03:20:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:46Z", "cpu_seconds": 1.853235305, "created_at": "2026-11-22T03:18:16Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-21dfc43b9a26", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "15143b147de42b3dbf70e277b201aa8c000bd61eb4099cd2e5d6103d7d84879d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-21dfc43b9a26", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:18:16Z", "seed": 1527138732, "treatment": "iw_temperature"}, "priority": 2.0094504805359246, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-c7cd6d2c2c2c", "result_hash": "ea4a0aee450423b0e76cf505a3c80e2919fc515560e428c31f2f3102b2d5c8fb", "run_seconds": 2.8298654556274414, "seed": 1527138732, "started_at": "2026-11-22T03:18:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.016467, 0.003437], "effect": -0.006515, "evaluated_at": "2026-11-22T03:20:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-21dfc43b9a26", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.03714343, 0.02056218, 0.02851004, 0.046793918, 0.01748715, 0.025674265, 0.103821855, 0.06049405, 0.034580047, 0.053312755, 0.026009992, 0.020882243, 0.025726503, 0.029362697, 0.025904312, 0.02412013], "raw": [0.049862515, 0.063093682, 0.02716648, 0.046793918, 0.026317152, 0.029870743, 0.047986465, 0.05234951, 0.041748972, 0.023784805, 0.025954248, 0.055612805, 0.022898913, 0.073618663, 0.024082738, 0.038599155], "temperature": [0.031330505, 0.028325543, 0.035041142, 0.046793918, 0.01748715, 0.025674265, 0.039204065, 0.023282427, 0.030642627, 0.037787083, 0.036361267, 0.037497473, 0.007121785, 0.03905608, 0.016424263, 0.02412013]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0936, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.806", "prereg_hash": "15143b147de42b3dbf70e277b201aa8c000bd61eb4099cd2e5d6103d7d84879d", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-c7cd6d2c2c2c", "result_hash": "ea4a0aee450423b0e76cf505a3c80e2919fc515560e428c31f2f3102b2d5c8fb", "signature": "82923ede49e9103684af18ef5e3d5b79611b1c4fd6c6b83f0b7ba3656817c32fe29a2eea0c0300cc1b453730ec327d273e6a87b91ae6fc6f2165e6ff409d8a00", "state": "VERIFIED", "summary": {"iw_temperature": 0.036274, "raw": 0.040609, "temperature": 0.029759}, "ties": 4, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:20:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:35Z", "cpu_seconds": 4.946155441999999, "created_at": "2026-11-22T03:18:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a5d107c5e444", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "c05df6b4cb893f9010670373a0297f723ed84d151865dd554c053a9f8437fc49", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a5d107c5e444", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:18:16Z", "seed": 889258492, "treatment": "frontier"}, "priority": 2.0095709289683428, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a917deb7271b89b43bcd919cec090d88a04bb622e76abdadd65329a58de2c5f4", "run_seconds": 5.937642812728882, "seed": 889258492, "started_at": "2026-11-22T03:18:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016979, 0.193339], "effect": 0.08818, "evaluated_at": "2026-11-22T03:20:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a5d107c5e444", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.625287356, 1.0, 0.637777778, 0.664444444], "frontier": [0.829885057, 0.98, 0.515555556, 0.602222222], "goal": [0.724137931, 0.97, 0.435555556, 0.406666667], "hypdisc": [0.682758621, 0.96, 0.511111111, 0.44], "infogain": [0.829885057, 0.98, 0.515555556, 0.602222222], "prederr": [0.829885057, 0.98, 0.515555556, 0.602222222], "random": [0.581609195, 0.96, 0.477777778, 0.555555556], "risk": [0.829885057, 0.98, 0.515555556, 0.602222222]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "c05df6b4cb893f9010670373a0297f723ed84d151865dd554c053a9f8437fc49", "primary_control": "random", "result_hash": "a917deb7271b89b43bcd919cec090d88a04bb622e76abdadd65329a58de2c5f4", "signature": "f14d1dba04735662349a6e2f6b4023ff75314df52c5d14b54156607a04cc4dc5324ee3b15c9e674c7b6dabbe333d6a7ff298ff19973c8545ca684c64e7ebd400", "state": "VERIFIED", "summary": {"count": 0.731877, "frontier": 0.731916, "goal": 0.63409, "hypdisc": 0.648467, "infogain": 0.731916, "prederr": 0.731916, "random": 0.643736, "risk": 0.731916}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:20:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:48Z", "cpu_seconds": 1.1995672049999992, "created_at": "2026-11-22T03:18:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c23f6e4a56a0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "29a26c5a1ad3e76b2990990541c31b3c21fdb10be9114fdbb62703158c739153", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c23f6e4a56a0", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:18:16Z", "seed": 1945289375, "treatment": "full"}, "priority": 2.009421986604187, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "74b3febe619c2ac772ebacfab8e2021bd593eb2443fb8e6eae7f41e1e891a00d", "run_seconds": 1.5513060092926025, "seed": 1945289375, "started_at": "2026-11-22T03:18:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.064966, 0.006871], "effect": -0.029048, "evaluated_at": "2026-11-22T03:20:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c23f6e4a56a0", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.838095238, 1.0, 0.9625, 0.88], "frozen": [0.6, 0.261363636, 0.325, 0.4], "full": [0.761904762, 1.0, 0.9625, 0.84], "retrieval": [0.876190476, 1.0, 0.9625, 0.82], "rules_nodetect": [0.580952381, 1.0, 0.325, 0.48]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "29a26c5a1ad3e76b2990990541c31b3c21fdb10be9114fdbb62703158c739153", "primary_control": "facts", "result_hash": "74b3febe619c2ac772ebacfab8e2021bd593eb2443fb8e6eae7f41e1e891a00d", "signature": "82586e5668b489c4dd84ebd932e3b75590db6546299c4e5a7e811f03ea5da3bdb3c6a70ce1975da541f2a55449716c3ee1ea12de5a9f9fefb12b7f8f25486507", "state": "VERIFIED", "summary": {"facts": 0.920149, "frozen": 0.396591, "full": 0.891101, "retrieval": 0.914673, "rules_nodetect": 0.596488}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:20:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:30Z", "cpu_seconds": 0.9507253569999999, "created_at": "2026-11-22T03:18:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-fa1a15ebe977", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "57d3504c820af88b1d644cbe8e94ac61ac85828d19b630947de21aa8c6aed7a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-fa1a15ebe977", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:18:16Z", "seed": 399435336, "treatment": "full"}, "priority": 2.009947601206199, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-3bca8ac59b55", "result_hash": "ed2866022441f7510da11c1e9f005ad3c6683665635c0a3511cc1e543fcde233", "run_seconds": 1.1293833255767822, "seed": 399435336, "started_at": "2026-11-22T03:18:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004364, 0.013455], "effect": 0.004545, "evaluated_at": "2026-11-22T03:20:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-fa1a15ebe977", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.527272727, 0.986206897, 1.0, 1.0], "frozen": [0.3, 0.227586207, 0.278571429, 0.383333333], "full": [0.545454545, 0.986206897, 1.0, 1.0], "retrieval": [0.609090909, 0.993103448, 1.0, 1.0], "rules_nodetect": [0.545454545, 0.986206897, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "57d3504c820af88b1d644cbe8e94ac61ac85828d19b630947de21aa8c6aed7a2", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-3bca8ac59b55", "result_hash": "ed2866022441f7510da11c1e9f005ad3c6683665635c0a3511cc1e543fcde233", "signature": "91ab52dc75d430d8fe21522cc674140ec9e1fd97b7bb3953b46c53ae3d8897aeac7e2d9fff457d832ff96e234999d109e5c2b8ebb0b15d1df29fd16ec3a6c500", "state": "VERIFIED", "summary": {"facts": 0.87837, "frozen": 0.297373, "full": 0.882915, "retrieval": 0.900549, "rules_nodetect": 0.882915}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:20:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:03Z", "cpu_seconds": 0.48948917199999986, "created_at": "2026-11-22T03:18:16Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-576df9d25218", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "prereg_hash": "bb1e7eb64c82c7b38207a06fa097490c9296911aa9be8adf7425464ac1e26edf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-576df9d25218", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:18:16Z", "seed": 1200872750, "treatment": "gated"}, "priority": 2.00922252567539, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "b2e877867367944befcf2301f39ca4f9538f019d931719cd0106f52b4a0499d3", "run_seconds": 0.5131688117980957, "seed": 1200872750, "started_at": "2026-11-22T03:19:02Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:20:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-576df9d25218", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.708333333, 1.0, 0.8125, 0.833333333], "gated": [0.708333333, 1.0, 0.8125, 0.833333333], "naive": [0.208333333, 0.375, 0.229166667, 0.4375], "scratch": [0.708333333, 1.0, 0.8125, 0.833333333], "ungated": [0.333333333, 0.520833333, 0.229166667, 0.791666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "bb1e7eb64c82c7b38207a06fa097490c9296911aa9be8adf7425464ac1e26edf", "primary_control": "scratch", "result_hash": "b2e877867367944befcf2301f39ca4f9538f019d931719cd0106f52b4a0499d3", "signature": "609c1fee2850935e1fce3fe87196084f6bf76b5b6b2dfc4eafb30b5d9331e0469c2a7cfb16e9ee05c595f4465e5926d8a6451c7076acc413b237b23d6657170d", "state": "PARTIAL", "summary": {"compressed": 0.838542, "gated": 0.838542, "naive": 0.3125, "scratch": 0.838542, "ungated": 0.46875}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:20:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:41Z", "cpu_seconds": 0.6221907580000003, "created_at": "2026-11-22T03:18:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9949c73dbea5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "a48aa4f54e0fe70617f0614d865103d60fe2ecedf68a4a8ece743184b15d051e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9949c73dbea5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:18:33Z", "seed": 1529696445, "treatment": "full"}, "priority": 2.009750378193679, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6511457f9078173d8ee4887204e955611f6285c210d236e16cc848927e0d09a5", "run_seconds": 0.6516807079315186, "seed": 1529696445, "started_at": "2026-11-22T03:18:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T03:20:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9949c73dbea5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 1.0], "fixed_policy": [0.25, 0.25, 0.25, 0.0], "full": [1.0, 0.25, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [0.25, 0.75, 0.25, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "a48aa4f54e0fe70617f0614d865103d60fe2ecedf68a4a8ece743184b15d051e", "primary_control": "e141_best", "result_hash": "6511457f9078173d8ee4887204e955611f6285c210d236e16cc848927e0d09a5", "signature": "5a47c614a63a1726139143b415778c30caaf45566112e484e83f5e019df980be7469a72a33ee3b344e3021fcce081a96dc5dfe2b7e593649208eea59b68faa0f", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.1875, "full": 0.8125, "no_world_model": 1.0, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:20:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:38Z", "cpu_seconds": 2.497732364, "created_at": "2026-11-22T03:18:33Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-514e8e82bc32", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lights|lights|1|1"}, "prereg_hash": "d45e9891cf54595770455762955cd01330c247c1bc517e25a6f6c3d3a68245fb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-514e8e82bc32", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lights|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:18:33Z", "seed": 449712898, "treatment": "gated"}, "priority": 2.0099246195713976, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "acd72674bffc96349e98af04558caa99204eb2a247985e59fae06527defb0ed0", "run_seconds": 2.602540969848633, "seed": 449712898, "started_at": "2026-11-22T03:18:35Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:20:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-514e8e82bc32", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.843915344, 1.0, 0.843915344, 0.85483871], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [1.0, 1.0, 0.843915344, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d45e9891cf54595770455762955cd01330c247c1bc517e25a6f6c3d3a68245fb", "primary_control": "scratch", "result_hash": "acd72674bffc96349e98af04558caa99204eb2a247985e59fae06527defb0ed0", "signature": "221d61be3267d64934ee08ccbeef19b684fe5e2f164761aedf7f3890cd1fb5cce0ab0f6bacea1b8e2d3bd300148d0b0c3f2661575a959c5fd89b538ef2f27309", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.885667, "scratch": 1.0, "ungated": 0.960979}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:20:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:40Z", "cpu_seconds": 4.710790505, "created_at": "2026-11-22T03:18:33Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-36564f37805f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "4bd4f21f86f28711650c45812d2273574de468c08ce7a64b504eefebf780d851", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-36564f37805f", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:18:33Z", "seed": 1628605841, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "386d271bfac32d4190aeba3fa37a20dcd0d4010f6b7a863c50da2d50013157aa", "run_seconds": 5.305842399597168, "seed": 1628605841, "started_at": "2026-11-22T03:18:34Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:20:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-36564f37805f", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5701599607306979, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4bd4f21f86f28711650c45812d2273574de468c08ce7a64b504eefebf780d851", "primary_control": "reset", "result_hash": "386d271bfac32d4190aeba3fa37a20dcd0d4010f6b7a863c50da2d50013157aa", "signature": "667c8384ded6c963a8fa07f75f9c54940625e737b56482bc3167a15ce5dde143db713581e72faac65d011429d5acbf29f0b5e7947c772eabfd158adba653150f", "state": "PARTIAL", "summary": {"inherit": 0.7666666666666667, "reset": 0.7666666666666667, "root": 0.45}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:20:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:53Z", "cpu_seconds": 0.6832807169999988, "created_at": "2026-11-22T03:18:49Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0b479c48246a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "b7ed7f3b684b401ea84f21c0db8d424f5f2cc63ba1ba59b2e8c286d2559cb3e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0b479c48246a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:18:49Z", "seed": 436951028, "treatment": "full"}, "priority": 2.0095357796172757, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "47afd07d61b9b959d7845cd159184a521eabc39f681098ad3063cca1bc1e5c7c", "run_seconds": 0.6892626285552979, "seed": 436951028, "started_at": "2026-11-22T03:18:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.004215, 0.620785], "effect": 0.3125, "evaluated_at": "2026-11-22T03:20:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0b479c48246a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.75, 0.75, 0.5], "fixed_policy": [0.25, 0.0, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 0.5], "no_world_model": [0.5, 1.0, 0.75, 0.5], "random": [0.5, 0.5, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b7ed7f3b684b401ea84f21c0db8d424f5f2cc63ba1ba59b2e8c286d2559cb3e2", "primary_control": "e141_best", "result_hash": "47afd07d61b9b959d7845cd159184a521eabc39f681098ad3063cca1bc1e5c7c", "signature": "651fcaba1ca5d5f55c72af829aaec958ff9ad8598078354d48b21ca752259686d1e1aab004669cdb37b7c0cdda5f6788c34198fb0541fd413c988d9173164c0f", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.125, "full": 0.875, "no_world_model": 0.6875, "random": 0.75}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:20:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:59Z", "cpu_seconds": 0.26365380899999913, "created_at": "2026-11-22T03:18:49Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ba402b5b5634", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "dd71b41abc6eac720a1d02387e43e7b3c2f310066bd7ea905900d5172fe906b5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ba402b5b5634", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:18:49Z", "seed": 1786656740, "treatment": "full"}, "priority": 2.0092667950135974, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "fbce898ff8b0fb7dfc2acd2d6e8062d791732881fd783b205a297541fd9907ca", "run_seconds": 0.27370381355285645, "seed": 1786656740, "started_at": "2026-11-22T03:18:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.32793, 0.79707], "effect": 0.5625, "evaluated_at": "2026-11-22T03:20:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ba402b5b5634", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 0.75, 0.25], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "dd71b41abc6eac720a1d02387e43e7b3c2f310066bd7ea905900d5172fe906b5", "primary_control": "e141_best", "result_hash": "fbce898ff8b0fb7dfc2acd2d6e8062d791732881fd783b205a297541fd9907ca", "signature": "8d8119dee320cac412858d5f42980d32ec4dd886fa3742ea5268841dc130bf55c0ec8b48d51adb598fa3c45d32a86e72eabdc46c9b6243a15278051954725705", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:20:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:57Z", "cpu_seconds": 0.7565763580000002, "created_at": "2026-11-22T03:18:49Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3cccc7ffff06", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|crafting|crafting|1|1"}, "prereg_hash": "9ff416e889bd5276a30cde0709453b487c304be58d148c8c3fb102e31725bb2d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3cccc7ffff06", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|crafting|crafting|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:18:49Z", "seed": 1397226345, "treatment": "gated"}, "priority": 2.0094360186624, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c60d3d656ff7a5fb7d2191e0c335caf76445b7181953930d268812d5f91c08ad", "run_seconds": 0.796353816986084, "seed": 1397226345, "started_at": "2026-11-22T03:18:56Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:20:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-3cccc7ffff06", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.545098039, 0.494901961, 0.453333333, 0.515294118], "gated": [0.545098039, 0.494901961, 0.453333333, 0.515294118], "naive": [0.524705882, 0.586666667, 0.430588235, 0.454117647], "scratch": [0.545098039, 0.494901961, 0.453333333, 0.515294118], "ungated": [0.399215686, 0.586666667, 0.458823529, 0.258823529]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9ff416e889bd5276a30cde0709453b487c304be58d148c8c3fb102e31725bb2d", "primary_control": "scratch", "result_hash": "c60d3d656ff7a5fb7d2191e0c335caf76445b7181953930d268812d5f91c08ad", "signature": "ee077fe519c317bc9ddae45279da3131052586672065c8e0d0c99879a1c28e7fe7a14cb6bbaae7cb2e7ad3eea86610c19366a014b973f6bdecedde4f8e71d10d", "state": "PARTIAL", "summary": {"compressed": 0.502157, "gated": 0.502157, "naive": 0.49902, "scratch": 0.502157, "ungated": 0.425882}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:20:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:18:51Z", "cpu_seconds": 0.2394666869999993, "created_at": "2026-11-22T03:18:49Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-aaafd7dceb0a", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "prereg_hash": "60dbd6b9d9ee734ad7eef7a06928a73d138be33e38c6da15e297eef64b179fb5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-aaafd7dceb0a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:18:49Z", "seed": 642288434, "treatment": "gated"}, "priority": 2.009969554021455, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "df9785f0120022747c64f7881a66eb5c5e808229a08b7ab248ead796d824b8d8", "run_seconds": 0.24081683158874512, "seed": 642288434, "started_at": "2026-11-22T03:18:51Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:20:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-aaafd7dceb0a", "family": "ow_transfer", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"compressed": [1.0, 0.439394], "gated": [1.0, 0.439394], "naive": [1.0, 0.280303], "scratch": [1.0, 0.439394], "ungated": [1.0, 0.280303]}, "prereg_hash": "60dbd6b9d9ee734ad7eef7a06928a73d138be33e38c6da15e297eef64b179fb5", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "df9785f0120022747c64f7881a66eb5c5e808229a08b7ab248ead796d824b8d8", "signature": "67c2e89367966185f0ec2f5603afe78d93d48be33e7bcc17dd6d3d0030b37e646a872a5641021ddfbdbeadd58eee2c36ceebd79230a283a4ab7d0a763457a50f", "state": "REJECTED", "summary": {"compressed": 0.439394, "gated": 0.439394, "naive": 0.280303, "scratch": 0.439394, "ungated": 0.280303}}, "verified_at": "2026-11-22T03:20:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:21Z", "cpu_seconds": 3.852307669, "created_at": "2026-11-22T03:19:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6312af802341", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "74fe3e3f7b28fd1b4a0e99262aee1e970472ba0c7da4451aac00fa86d5e45dfc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6312af802341", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:19:06Z", "seed": 999070233, "treatment": "frontier"}, "priority": 2.009339432431363, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "255e0fb2fe81658b74604a63a1651a6afc16e763fa96e86a2640130592fafc13", "run_seconds": 5.005720615386963, "seed": 999070233, "started_at": "2026-11-22T03:19:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.010894, 0.049825], "effect": 0.019465, "evaluated_at": "2026-11-22T03:20:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6312af802341", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.717948718, 0.909090909, 0.87804878, 0.784722222], "frontier": [0.717948718, 0.909090909, 0.87804878, 0.777777778], "goal": [0.705128205, 0.909090909, 0.87804878, 0.784722222], "hypdisc": [0.346153846, 0.909090909, 0.87804878, 0.729166667], "infogain": [0.717948718, 0.909090909, 0.87804878, 0.777777778], "prederr": [0.705128205, 0.909090909, 0.87804878, 0.777777778], "random": [0.705128205, 0.909090909, 0.81300813, 0.777777778], "risk": [0.717948718, 0.909090909, 0.87804878, 0.777777778]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "74fe3e3f7b28fd1b4a0e99262aee1e970472ba0c7da4451aac00fa86d5e45dfc", "primary_control": "random", "result_hash": "255e0fb2fe81658b74604a63a1651a6afc16e763fa96e86a2640130592fafc13", "signature": "3cb71b17d6c4f8880edd5b2a9f78c987a7560afaf31a5bb125991d64039bf373fd83dfb1335ec2ba6034c865db1b0c08d800dd7af5ded4f4d90009a162647d06", "state": "VERIFIED", "summary": {"count": 0.822453, "frontier": 0.820717, "goal": 0.819248, "hypdisc": 0.715615, "infogain": 0.820717, "prederr": 0.817511, "random": 0.801251, "risk": 0.820717}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:20:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:38Z", "cpu_seconds": 0.33216229299999966, "created_at": "2026-11-22T03:19:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c255b00cb3ad", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "99e7b6ef31fea0ab9ca433007d169de2a2fd832a8f4a49b3e89e5da878d096a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c255b00cb3ad", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:19:06Z", "seed": 169408012, "treatment": "frontier"}, "priority": 2.00908316194369, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ad67a370b8842093739f083beb18ba3337e7a2c0f44ac3db037bf81b9cd6db97", "run_seconds": 0.332777738571167, "seed": 169408012, "started_at": "2026-11-22T03:19:37Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:20:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c255b00cb3ad", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.992424], "frontier": [1.0, 0.924242], "goal": [1.0, 0.734848], "hypdisc": [1.0, 0.901515], "infogain": [1.0, 0.924242], "prederr": [1.0, 0.924242], "random": [1.0, 0.825758], "risk": [1.0, 0.924242]}, "prereg_hash": "99e7b6ef31fea0ab9ca433007d169de2a2fd832a8f4a49b3e89e5da878d096a2", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "ad67a370b8842093739f083beb18ba3337e7a2c0f44ac3db037bf81b9cd6db97", "signature": "6564a0ce1d43c204019e40f1466ab0d6714d464901445c54c3033a6ba2bf9edc326185c2a0d6ecdf97c2b172ed440abf8b615ef3fcd4d80bb593cc284b747000", "state": "REJECTED", "summary": {"count": 0.992424, "frontier": 0.924242, "goal": 0.734848, "hypdisc": 0.901515, "infogain": 0.924242, "prederr": 0.924242, "random": 0.825758, "risk": 0.924242}}, "verified_at": "2026-11-22T03:20:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:49Z", "cpu_seconds": 0.8312724990000007, "created_at": "2026-11-22T03:19:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d9e091d31cda", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "b5b46833ccef1326c69a299a55cbbd5790ad60399344f8119f5bbf5c5d68a124", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d9e091d31cda", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:19:06Z", "seed": 223636440, "treatment": "frontier"}, "priority": 2.009000381743631, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "132673a9ce5b6c8decfee9d44aae8ea5c27cfc36f23d11816f688a7900b9e196", "run_seconds": 1.0673596858978271, "seed": 223636440, "started_at": "2026-11-22T03:19:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017747, 0.076081], "effect": 0.029167, "evaluated_at": "2026-11-22T03:22:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d9e091d31cda", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.983333333, 1.0], "frontier": [1.0, 1.0, 1.0, 0.983333333], "goal": [1.0, 0.9, 0.933333333, 0.983333333], "hypdisc": [1.0, 0.983333333, 0.983333333, 1.0], "infogain": [1.0, 1.0, 1.0, 0.983333333], "prederr": [1.0, 1.0, 1.0, 0.983333333], "random": [1.0, 0.983333333, 0.9, 0.983333333], "risk": [1.0, 1.0, 1.0, 0.983333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "b5b46833ccef1326c69a299a55cbbd5790ad60399344f8119f5bbf5c5d68a124", "primary_control": "random", "result_hash": "132673a9ce5b6c8decfee9d44aae8ea5c27cfc36f23d11816f688a7900b9e196", "signature": "17c11e4bc7811d04d3b2ca83f6161d5513acae81cd2073b82280865d7abdb03d2c3f40535f5d70d79edda6c030d033a105c17e1c7e919e4aeb7bfcfa5392dd08", "state": "VERIFIED", "summary": {"count": 0.995833, "frontier": 0.995833, "goal": 0.954167, "hypdisc": 0.991667, "infogain": 0.995833, "prederr": 0.995833, "random": 0.966667, "risk": 0.995833}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:22:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:11Z", "cpu_seconds": 1.249047082999999, "created_at": "2026-11-22T03:19:06Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4c3c8bd0666f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "82c96520ac16d893fef95ef215bf0eb8afae4e452d917d3ec0088eb2bbebafbf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4c3c8bd0666f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:19:06Z", "seed": 47348166, "treatment": "full"}, "priority": 2.00963151909184, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "45ad40f79992e030495c305b0373275f9bf73a114cd2d6f66eaebf3090ef24ce", "run_seconds": 1.344728708267212, "seed": 47348166, "started_at": "2026-11-22T03:19:10Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:20:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4c3c8bd0666f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.25, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "82c96520ac16d893fef95ef215bf0eb8afae4e452d917d3ec0088eb2bbebafbf", "primary_control": "e141_best", "result_hash": "45ad40f79992e030495c305b0373275f9bf73a114cd2d6f66eaebf3090ef24ce", "signature": "3f17b86c42c88f8cad269a0aa34f77044c1066b22f1820ebf5ff2780389f40b38b1b90636c13576ccee27725af9219a5d9229cb57522d41fe579a91ba6e5ff08", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:20:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:21Z", "cpu_seconds": 1.5343397250000006, "created_at": "2026-11-22T03:19:06Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8e96dfdf639f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "1ad844aa5a0680e91cefd65ee5ccf1d97b75c93aa63ceaaf293839e7ae6b5a6e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8e96dfdf639f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:19:06Z", "seed": 1136431196, "treatment": "full"}, "priority": 2.009309379031947, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "98c7f4577aec6a83fb0dd82968e6ab2b3483b7d6ae588e8d4da7d95bda38ef3d", "run_seconds": 2.353384494781494, "seed": 1136431196, "started_at": "2026-11-22T03:19:19Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:20:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8e96dfdf639f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1ad844aa5a0680e91cefd65ee5ccf1d97b75c93aa63ceaaf293839e7ae6b5a6e", "primary_control": "e141_best", "result_hash": "98c7f4577aec6a83fb0dd82968e6ab2b3483b7d6ae588e8d4da7d95bda38ef3d", "signature": "4f7d482d2987412a9592551cf105b0ef3224df3d7e4940539fdc058819ecb705de5fa7fe93b8aa7005fe1daa0c3b402e2b1735f9a8e39e43ef2fc4913bdb8203", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:20:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:14Z", "cpu_seconds": 5.561973299, "created_at": "2026-11-22T03:19:06Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-872e4a3a5555", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 7, "n": 4}, "prereg_hash": "9dc38e9e8f3d63734343a0c98d7dce521e673f1afadaff651adb6c0f41540fd2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-872e4a3a5555", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:19:06Z", "seed": 379588566, "treatment": "full"}, "priority": 2.0096315573755246, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "739129155a22eea7d99782baacb787e120518abcac47fa0d93bff087ba30e630", "run_seconds": 5.797514200210571, "seed": 379588566, "started_at": "2026-11-22T03:19:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.157733, 0.716291], "effect": 0.437012, "evaluated_at": "2026-11-22T03:20:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-872e4a3a5555", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.6875, 0.139973958, 0.760416667, 0.6640625], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.96875, 0.723958333, 0.9296875, 0.958333333], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "9dc38e9e8f3d63734343a0c98d7dce521e673f1afadaff651adb6c0f41540fd2", "primary_control": "facts", "result_hash": "739129155a22eea7d99782baacb787e120518abcac47fa0d93bff087ba30e630", "signature": "b3f7266b552866bd7dbefc875dd88a05b6e5fa93f9712b645c244ad2dba024eb1adae2424acf19841e659aeb781763815e452b0a1b2473afeeddce4be84f4f07", "state": "VERIFIED", "summary": {"facts": 0.562988, "frozen": 0.0, "full": 1.0, "retrieval": 0.895182, "rules_nodetect": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:20:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:28Z", "cpu_seconds": 1.3621180489999993, "created_at": "2026-11-22T03:19:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5b23263d7a11", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "4a4ac47ca2abf75a4e2fc00d3b20c459e0d10a33852c4cd4d8bbd33f425a7374", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5b23263d7a11", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:19:22Z", "seed": 1671900779, "treatment": "frontier"}, "priority": 2.0098494021009965, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ac7fbe45939adf2a036f96bd3b80e249d0858ec604c2f3e4c334e509d3029067", "run_seconds": 1.574284553527832, "seed": 1671900779, "started_at": "2026-11-22T03:19:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.09872, 0.264403], "effect": 0.181561, "evaluated_at": "2026-11-22T03:20:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5b23263d7a11", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.6875, 0.463917526, 0.604166667, 0.640625], "frontier": [0.8375, 0.559278351, 0.766666667, 0.671875], "goal": [0.8375, 0.518041237, 0.708333333, 0.578125], "hypdisc": [0.525, 0.484536082, 0.604166667, 0.5703125], "infogain": [0.8375, 0.559278351, 0.766666667, 0.65625], "prederr": [0.4375, 0.559278351, 0.766666667, 0.4140625], "random": [0.6, 0.378865979, 0.520833333, 0.609375], "risk": [0.8375, 0.559278351, 0.766666667, 0.671875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "4a4ac47ca2abf75a4e2fc00d3b20c459e0d10a33852c4cd4d8bbd33f425a7374", "primary_control": "random", "result_hash": "ac7fbe45939adf2a036f96bd3b80e249d0858ec604c2f3e4c334e509d3029067", "signature": "67398df46fb209ee1c793413eb0a010eb17e5bc6fb1c548a3226e9252bb3623c3b810072747867a5bd1dafe7605ee0b768f04f8b602e5cbf877226c0333eb306", "state": "VERIFIED", "summary": {"count": 0.599052, "frontier": 0.70883, "goal": 0.6605, "hypdisc": 0.546004, "infogain": 0.704924, "prederr": 0.544377, "random": 0.527269, "risk": 0.70883}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:20:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:32Z", "cpu_seconds": 1.0876081709999994, "created_at": "2026-11-22T03:19:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-932c0b25a613", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "f69673088b603adeb237f66371169aff541dfbf1258f5a9e765c51a3ede5dfb3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-932c0b25a613", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:19:22Z", "seed": 2109343000, "treatment": "full"}, "priority": 2.009628059020449, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "20fd0595cfb9ce7a89e472f1a5fb59b92bb74180e3e54d2342ee7103a093579d", "run_seconds": 1.131807565689087, "seed": 2109343000, "started_at": "2026-11-22T03:19:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.223722, 1.151278], "effect": 0.6875, "evaluated_at": "2026-11-22T03:20:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-932c0b25a613", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.25, 1.0], "fixed_policy": [0.5, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 0.75], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "f69673088b603adeb237f66371169aff541dfbf1258f5a9e765c51a3ede5dfb3", "primary_control": "e141_best", "result_hash": "20fd0595cfb9ce7a89e472f1a5fb59b92bb74180e3e54d2342ee7103a093579d", "signature": "a5ab8da4e7d3fbd4aa9c03b13cd35ba70335b9f312bffdf5ea4242d529d018e94fe12c557d28cdff4ce947f4bbf8be7b93c7201c38ad3d4b3277e602ae37750f", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:20:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:26Z", "cpu_seconds": 0.123058094000001, "created_at": "2026-11-22T03:19:22Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6b86c7059b05", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "862a4f5f0f55682b9e0281893ec28df3ebda52851b0dacccb38027c4d35c3574", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6b86c7059b05", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:19:22Z", "seed": 203106925, "treatment": "full"}, "priority": 2.0097337126613337, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "736cf3691301d944642b2fe6e8760c77c328769cf4dc0389e8bad00fe18d9802", "run_seconds": 0.12961268424987793, "seed": 203106925, "started_at": "2026-11-22T03:19:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.013637, 0.084848], "effect": 0.049242, "evaluated_at": "2026-11-22T03:20:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6b86c7059b05", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.803030303, 0.636363636, 0.795454545, 0.818181818], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [0.803030303, 0.704545455, 0.878787879, 0.863636364], "retrieval": [0.825757576, 0.803030303, 0.901515152, 0.863636364], "rules_nodetect": [0.803030303, 0.704545455, 0.878787879, 0.863636364]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "862a4f5f0f55682b9e0281893ec28df3ebda52851b0dacccb38027c4d35c3574", "primary_control": "facts", "result_hash": "736cf3691301d944642b2fe6e8760c77c328769cf4dc0389e8bad00fe18d9802", "signature": "051e06bd43a57cdf2e659dd300991fcdeed5be6ee97f7a6e76a6d6bbfe3e65a64300b1080d6c011046d94bfedd2bdc1015ad232e3db34ed5a473d44980371300", "state": "VERIFIED", "summary": {"facts": 0.763258, "frozen": 0.030303, "full": 0.8125, "retrieval": 0.848485, "rules_nodetect": 0.8125}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:20:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:34Z", "cpu_seconds": 1.0029340560000009, "created_at": "2026-11-22T03:19:22Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1a3741e5991b", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "prereg_hash": "c613493afbf01e3450d5d2e3e76f62a9bfb6c266c7f4a479dfe593ac8c37375d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1a3741e5991b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:19:22Z", "seed": 1401883388, "treatment": "gated"}, "priority": 2.0095825312677063, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d790068c15bb9456a67b2330311f1076fc8ee1bbb2167e1c148cbd8a36806ad9", "run_seconds": 1.1475622653961182, "seed": 1401883388, "started_at": "2026-11-22T03:19:33Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:20:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1a3741e5991b", "family": "ow_transfer", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"compressed": [1.0, 0.563474], "gated": [1.0, 0.563474], "naive": [1.0, 0.445583], "scratch": [1.0, 0.575974], "ungated": [1.0, 0.54072]}, "prereg_hash": "c613493afbf01e3450d5d2e3e76f62a9bfb6c266c7f4a479dfe593ac8c37375d", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "d790068c15bb9456a67b2330311f1076fc8ee1bbb2167e1c148cbd8a36806ad9", "signature": "6f9642482e049b7e80d2717e609e3e8efd51fe3c8401d078a9396c19151903cffe63e2df12260529e6cce68e8b2a6b313e256d5fd78405efd003c757d807750a", "state": "REJECTED", "summary": {"compressed": 0.563474, "gated": 0.563474, "naive": 0.445583, "scratch": 0.575974, "ungated": 0.54072}}, "verified_at": "2026-11-22T03:20:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:09Z", "cpu_seconds": 1.8884868570000002, "created_at": "2026-11-22T03:19:39Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-91b45abb2faf", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "f5c9ef7783af34f045f50ee53edaf8605030827ee744b32ae8742ed94c561d1d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-91b45abb2faf", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:19:39Z", "seed": 310228620, "treatment": "iw_temperature"}, "priority": 2.008867104314905, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-90f98064bf24", "result_hash": "b0483430b580fb6a345eb3243a5fec7d664ac7ba36c1b0f67737c5b2ab9fcaac", "run_seconds": 1.9113125801086426, "seed": 310228620, "started_at": "2026-11-22T03:20:07Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011962, 0.020579], "effect": 0.004309, "evaluated_at": "2026-11-22T03:22:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-91b45abb2faf", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.088647677, 0.10648896, 0.133595958, 0.122520697, 0.11274329, 0.149508333, 0.041303198, 0.15379316, 0.089565647, 0.139371675, 0.162381135, 0.148856248, 0.191223823, 0.163000023, 0.092558635, 0.123194848], "raw": [0.145017095, 0.099656008, 0.18626752, 0.08861981, 0.12056383, 0.161313575, 0.11687425, 0.154869145, 0.120236203, 0.10975376, 0.105834075, 0.148856248, 0.269371595, 0.14813996, 0.13277256, 0.150170843], "temperature": [0.145017095, 0.129285275, 0.174883445, 0.116996245, 0.080514443, 0.1110486, 0.07839647, 0.154869145, 0.087038432, 0.133357528, 0.135221517, 0.103646065, 0.22554397, 0.128928793, 0.13277256, 0.150170843]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.2232, "losses": 8, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.598", "prereg_hash": "f5c9ef7783af34f045f50ee53edaf8605030827ee744b32ae8742ed94c561d1d", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-90f98064bf24", "result_hash": "b0483430b580fb6a345eb3243a5fec7d664ac7ba36c1b0f67737c5b2ab9fcaac", "signature": "9ae3f1228c34ebc20c3bc3ad47ad4ca27486e7a6e57f35987cd038a7716c59776eaa813c0eb0fb9578a30f479e264cff7c1a0375b92cd2458f0e9a4581471e09", "state": "VERIFIED", "summary": {"iw_temperature": 0.126172, "raw": 0.141145, "temperature": 0.130481}, "ties": 0, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:22:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:41Z", "cpu_seconds": 2.3848725319999993, "created_at": "2026-11-22T03:19:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3d3f31969702", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "418101579ecc3a6ae74236e361a83a0fe65aba29578453d32e6f6f67ffda1c94", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3d3f31969702", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:19:39Z", "seed": 1270646794, "treatment": "frontier"}, "priority": 2.009896714182073, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "8a7e7169a2e9a7cf3905414ea0bfefa66917172999722c2d347880c72629fa58", "run_seconds": 2.4397690296173096, "seed": 1270646794, "started_at": "2026-11-22T03:19:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.388126, 0.344772], "effect": -0.021677, "evaluated_at": "2026-11-22T03:20:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-3d3f31969702", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.840909091, 0.830645161, 0.819444444, 0.852941176], "frontier": [0.295454545, 0.935483871, 0.958333333, 0.955882353], "goal": [0.295454545, 0.75, 0.875, 0.823529412], "hypdisc": [0.295454545, 0.89516129, 0.930555556, 0.852941176], "infogain": [0.295454545, 0.919354839, 0.986111111, 0.985294118], "prederr": [0.295454545, 0.241935484, 0.958333333, 0.647058824], "random": [0.863636364, 0.830645161, 0.861111111, 0.676470588], "risk": [0.295454545, 0.935483871, 0.958333333, 0.955882353]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "418101579ecc3a6ae74236e361a83a0fe65aba29578453d32e6f6f67ffda1c94", "primary_control": "random", "result_hash": "8a7e7169a2e9a7cf3905414ea0bfefa66917172999722c2d347880c72629fa58", "signature": "47ef2611ae8f4ebafe477c329afff0d46117685fdf19310d3683643abd71186b9d2a064e4459967da98bb30a1774f3751ef2efa8e93fa42bbe4251b56220c203", "state": "VERIFIED", "summary": {"count": 0.835985, "frontier": 0.786289, "goal": 0.685996, "hypdisc": 0.743528, "infogain": 0.796554, "prederr": 0.535696, "random": 0.807966, "risk": 0.786289}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:20:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:49Z", "cpu_seconds": 2.59243077, "created_at": "2026-11-22T03:19:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5ecebb80e991", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 5, "n": 4}, "prereg_hash": "0f7bd4c2fe4b169c04b45168c95ff0d772f5d9295d7e0ec8b8cd7888c27d97f0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5ecebb80e991", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:19:39Z", "seed": 372620924, "treatment": "full"}, "priority": 2.0094744824740394, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "363418cc96549c290f73f3a4731c083939f12cb9885c6818dbb9174c7f435256", "run_seconds": 2.921119451522827, "seed": 372620924, "started_at": "2026-11-22T03:19:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.020801, 0.152682], "effect": 0.06594, "evaluated_at": "2026-11-22T03:22:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5ecebb80e991", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.429090909, 0.624137931, 1.0, 1.0], "frozen": [0.069090909, 0.165517241, 0.38, 0.35], "full": [0.581818182, 0.755172414, 0.98, 1.0], "retrieval": [0.636363636, 0.75862069, 1.0, 1.0], "rules_nodetect": [0.581818182, 0.734482759, 0.98, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "0f7bd4c2fe4b169c04b45168c95ff0d772f5d9295d7e0ec8b8cd7888c27d97f0", "primary_control": "facts", "result_hash": "363418cc96549c290f73f3a4731c083939f12cb9885c6818dbb9174c7f435256", "signature": "4e759a27e285b391d984b3607a9175e00109d251a083a6c92d4fd3b990c8209586d9c2c63ee090d45bc5e5e846fa9e91f9c7f1f24cdc376f494e17c61be3bb07", "state": "VERIFIED", "summary": {"facts": 0.763307, "frozen": 0.241152, "full": 0.829248, "retrieval": 0.848746, "rules_nodetect": 0.824075}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:43Z", "cpu_seconds": 0.4762491290000028, "created_at": "2026-11-22T03:19:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c5f26f8d46ac", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "aa3225518767f766cf7c8c6d31a5cac70cbcff44f259000a87ba632628010b94", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c5f26f8d46ac", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:19:39Z", "seed": 785883224, "treatment": "full"}, "priority": 2.0095541411002102, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5b1c32d47ba696b616bbb07b75011ca6fc580f76302febc829bf05bdac02db45", "run_seconds": 0.4780759811401367, "seed": 785883224, "started_at": "2026-11-22T03:19:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004585, 0.122232], "effect": 0.058824, "evaluated_at": "2026-11-22T03:20:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c5f26f8d46ac", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.908496732, 0.712418301, 0.503267974, 0.882352941], "frozen": [0.091503268, 0.091503268, 0.091503268, 0.091503268], "full": [0.921568627, 0.77124183, 0.653594771, 0.895424837], "retrieval": [0.941176471, 0.810457516, 0.784313725, 0.941176471], "rules_nodetect": [0.921568627, 0.77124183, 0.653594771, 0.895424837]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "aa3225518767f766cf7c8c6d31a5cac70cbcff44f259000a87ba632628010b94", "primary_control": "facts", "result_hash": "5b1c32d47ba696b616bbb07b75011ca6fc580f76302febc829bf05bdac02db45", "signature": "fbffc7f744a5ada1c64d693a9082f76bab53f9f29f59f1e8245d73268c864c8750e74ba48adbb4d94dc9ea453f249f6f2666544a9e1ddc603f5c854f1fa3b703", "state": "VERIFIED", "summary": {"facts": 0.751634, "frozen": 0.091503, "full": 0.810458, "retrieval": 0.869281, "rules_nodetect": 0.810458}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:20:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:52Z", "cpu_seconds": 1.2309036970000022, "created_at": "2026-11-22T03:19:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d82d8cd0d4a7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "ac71b0fad9a97a0d34ffb9f9de20c8a1f7cad616199f7911ee5aee5003335bd9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d82d8cd0d4a7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:19:39Z", "seed": 1123609335, "treatment": "full"}, "priority": 2.0089398502692855, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "32cb7be61c01e5fb4f42e8de43243bc7affbc448c831c03f4b4cd49029cecec8", "run_seconds": 1.2687153816223145, "seed": 1123609335, "started_at": "2026-11-22T03:19:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.050253, 0.243534], "effect": 0.146893, "evaluated_at": "2026-11-22T03:22:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d82d8cd0d4a7", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.196183953, 0.985714286, 0.219178082, 0.218688845], "frozen": [0.011741683, 0.0, 0.038649706, 0.039138943], "full": [0.401663405, 0.985714286, 0.421722114, 0.398238748], "retrieval": [0.3962818, 0.985714286, 0.435420744, 0.433463796], "rules_nodetect": [0.401663405, 0.985714286, 0.421722114, 0.398238748]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "ac71b0fad9a97a0d34ffb9f9de20c8a1f7cad616199f7911ee5aee5003335bd9", "primary_control": "facts", "result_hash": "32cb7be61c01e5fb4f42e8de43243bc7affbc448c831c03f4b4cd49029cecec8", "signature": "83e6da7b171152bb742c0dbfa33abb774d31936bbd5b47f9ed1bff91d592f0c19e8a2f633ed5965154bdac9e8adb83ebe6f9375d24cfd15ac6f407efcb85260f", "state": "VERIFIED", "summary": {"facts": 0.404941, "frozen": 0.022383, "full": 0.551835, "retrieval": 0.56272, "rules_nodetect": 0.551835}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:22:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:59Z", "cpu_seconds": 1.6459344189999996, "created_at": "2026-11-22T03:19:54Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-a2899d17e77c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "b64916336567015c0f02ac900a90165cf0f696e4e03f6a01af0cc23f25d2975f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-a2899d17e77c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:19:54Z", "seed": 224206123, "treatment": "iw_temperature"}, "priority": 2.009732977200443, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-fa4fc30ea3db", "result_hash": "264a8b9ca4dbbb295ef9d2323340d83aa809d461bd7a239df4f5c2ee8bca779b", "run_seconds": 1.8722360134124756, "seed": 224206123, "started_at": "2026-11-22T03:19:57Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.025756, 0.015996], "effect": -0.00488, "evaluated_at": "2026-11-22T03:22:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-a2899d17e77c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.098295925, 0.265445267, 0.285847995, 0.283380765, 0.102514945, 0.259414272, 0.311325517, 0.29531111, 0.290734335, 0.302236948, 0.294925575, 0.337482443, 0.171090078, 0.26279916, 0.258996808, 0.258681245], "raw": [0.14869233, 0.297350333, 0.221774992, 0.236496127, 0.17539801, 0.245042847, 0.242112072, 0.252735267, 0.275029623, 0.284030005, 0.276141465, 0.336069397, 0.18201528, 0.24028196, 0.183721395, 0.24358002], "temperature": [0.167044997, 0.311521775, 0.2493605, 0.26636805, 0.153855137, 0.25354346, 0.268556817, 0.260030827, 0.28129479, 0.299069773, 0.295937163, 0.33720034, 0.18201528, 0.267712717, 0.144389975, 0.262494755]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.112, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.773", "prereg_hash": "b64916336567015c0f02ac900a90165cf0f696e4e03f6a01af0cc23f25d2975f", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-fa4fc30ea3db", "result_hash": "264a8b9ca4dbbb295ef9d2323340d83aa809d461bd7a239df4f5c2ee8bca779b", "signature": "de61427b8fa2e72afda3a5141bc86432d8595d61759ae7c8970a9eed09b5133e4f10b3c8a11fb5eaedd09cf49f2bedff373ffcadee7a351fd7bb6ea2d1e0bb03", "state": "VERIFIED", "summary": {"iw_temperature": 0.254905, "raw": 0.240029, "temperature": 0.250025}, "ties": 0, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:22:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:04Z", "cpu_seconds": 7.891825848, "created_at": "2026-11-22T03:19:54Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ee29e3f12b31", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "6cc47b19f9fbf7ab5a8c2bb458647b5bcd905974acef198f39732129f509020f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ee29e3f12b31", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:19:54Z", "seed": 789792236, "treatment": "full"}, "priority": 2.0098270982845965, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "503e411ffc6cfeabe930c005c790cb14146ce3bb75dc28112a67cacfbd6f1a47", "run_seconds": 9.401922702789307, "seed": 789792236, "started_at": "2026-11-22T03:19:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.400083, 0.400083], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ee29e3f12b31", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.5, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 1.0, 0.75, 0.0], "no_world_model": [0.0, 0.5, 0.75, 0.0], "random": [0.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "6cc47b19f9fbf7ab5a8c2bb458647b5bcd905974acef198f39732129f509020f", "primary_control": "e141_best", "result_hash": "503e411ffc6cfeabe930c005c790cb14146ce3bb75dc28112a67cacfbd6f1a47", "signature": "9875a5e9684411f3468eb3f64a59699e81d4071d194d8890e11924b0061f9d56faf22c0089e492e7c39a32677cd78f7480be5d7c4c1bd980caa72abea866220b", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.0, "full": 0.4375, "no_world_model": 0.3125, "random": 0.5}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:19:57Z", "cpu_seconds": 2.011746636999998, "created_at": "2026-11-22T03:19:54Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0e7826805f48", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "052171669beb970a6dba57405771b137343bf23e847e67031d6bcebcdc8c5ae4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0e7826805f48", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:19:54Z", "seed": 1118503473, "treatment": "full"}, "priority": 2.0098713111398747, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "795b2af1d07ce971134dc79d1143a5d6cac23aa4614e5db09304d2954e17c8a2", "run_seconds": 2.4106945991516113, "seed": 1118503473, "started_at": "2026-11-22T03:19:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.16447, 0.300808], "effect": 0.232639, "evaluated_at": "2026-11-22T03:22:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0e7826805f48", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.868055556, 0.743055556, 0.708333333, 0.75], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.951388889, 0.9375, 0.930555556, 0.930555556], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "052171669beb970a6dba57405771b137343bf23e847e67031d6bcebcdc8c5ae4", "primary_control": "facts", "result_hash": "795b2af1d07ce971134dc79d1143a5d6cac23aa4614e5db09304d2954e17c8a2", "signature": "d9b964c61110897b1cbbb3abc3b02c520562231bfb29b6891bde4de3b86234b3ae9b3c9d8b6c6c2de4dd9d0c30f17f23b1555346ef8fa3532b8e3cf00911b903", "state": "VERIFIED", "summary": {"facts": 0.767361, "frozen": 0.0, "full": 1.0, "retrieval": 0.9375, "rules_nodetect": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:22:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:02Z", "cpu_seconds": 0.19646750599999763, "created_at": "2026-11-22T03:19:54Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2625ccd9e88b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "prereg_hash": "cdec450755550a2160b18b688e649c573c26cf659773affd6d0ea8d560ecd901", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2625ccd9e88b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:19:54Z", "seed": 1361592055, "treatment": "gated"}, "priority": 2.0096632216754675, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "436a6ff6f25dc0a62f518edc0ce4f6d7a884f08e75bb0e7fac4e065fa3cca3f0", "run_seconds": 0.2423853874206543, "seed": 1361592055, "started_at": "2026-11-22T03:20:02Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2625ccd9e88b", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.333333333, 0.318181818, 0.106060606, 0.303030303], "gated": [0.333333333, 0.318181818, 0.106060606, 0.303030303], "naive": [0.106060606, 0.090909091, 0.106060606, 0.303030303], "scratch": [0.333333333, 0.318181818, 0.106060606, 0.303030303], "ungated": [0.106060606, 0.090909091, 0.106060606, 0.303030303]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "cdec450755550a2160b18b688e649c573c26cf659773affd6d0ea8d560ecd901", "primary_control": "scratch", "result_hash": "436a6ff6f25dc0a62f518edc0ce4f6d7a884f08e75bb0e7fac4e065fa3cca3f0", "signature": "274fb2980b662cd4cfa2f4b8a9d06ad6cf64e7315058b795982dbbf69ef5698ce3e0501626313aa1b839dbb129d4d7e96ac6b16f10058669f187589620ec1d0e", "state": "PARTIAL", "summary": {"compressed": 0.265152, "gated": 0.265152, "naive": 0.151515, "scratch": 0.265152, "ungated": 0.151515}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:22:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:05Z", "cpu_seconds": 0.7801388139999972, "created_at": "2026-11-22T03:19:54Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-26e4ee543c7a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|crafting|crafting|1|1"}, "prereg_hash": "d9a6c00e135edfe4e214e67b109166496b91f06db7c4ef266165354d40889896", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-26e4ee543c7a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|crafting|crafting|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:19:54Z", "seed": 1738870427, "treatment": "gated"}, "priority": 2.0095658859579486, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "18cb271bb7ecc95f9ed75eceb536303f54c6421a7a45c525ed4d8bac21efc8ad", "run_seconds": 1.0251872539520264, "seed": 1738870427, "started_at": "2026-11-22T03:20:04Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-26e4ee543c7a", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.444705882, 0.52173913, 0.59372549, 0.472156863], "gated": [0.444705882, 0.52173913, 0.59372549, 0.472156863], "naive": [0.494901961, 0.536758893, 0.636862745, 0.375686275], "scratch": [0.444705882, 0.52173913, 0.59372549, 0.472156863], "ungated": [0.396862745, 0.646640316, 0.534117647, 0.460392157]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d9a6c00e135edfe4e214e67b109166496b91f06db7c4ef266165354d40889896", "primary_control": "scratch", "result_hash": "18cb271bb7ecc95f9ed75eceb536303f54c6421a7a45c525ed4d8bac21efc8ad", "signature": "a2a72b55a0d8f849c8e23078f24b784cfd45998925271619fd0a4367422a108856dd2347a03b406b93e137d4083a93cc8cd8ae805cfcdfe98f1d8cbd64584108", "state": "PARTIAL", "summary": {"compressed": 0.508082, "gated": 0.508082, "naive": 0.511052, "scratch": 0.508082, "ungated": 0.509503}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:22:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:21Z", "cpu_seconds": 0.5193351420000027, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9b57c21b9416", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "f4f02cb057324e232b9c1d9d970fccfb947805c930a603a52506ed4c88f1decc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9b57c21b9416", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:20:10Z", "seed": 1667707690, "treatment": "frontier"}, "priority": 2.0093411674763972, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3e7bcae34f04eeeae1a56cc544f8352744fe008365e708f79ea2e558d7b09619", "run_seconds": 0.7656588554382324, "seed": 1667707690, "started_at": "2026-11-22T03:20:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.159113, 0.234826], "effect": 0.19697, "evaluated_at": "2026-11-22T03:22:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9b57c21b9416", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.863636364, 0.75, 0.803030303, 0.871212121], "frontier": [0.825757576, 0.825757576, 0.848484848, 0.909090909], "goal": [0.636363636, 0.454545455, 0.681818182, 0.674242424], "hypdisc": [0.681818182, 0.53030303, 0.71969697, 0.765151515], "infogain": [0.825757576, 0.825757576, 0.848484848, 0.909090909], "prederr": [0.825757576, 0.825757576, 0.848484848, 0.909090909], "random": [0.681818182, 0.590909091, 0.636363636, 0.712121212], "risk": [0.825757576, 0.825757576, 0.848484848, 0.909090909]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f4f02cb057324e232b9c1d9d970fccfb947805c930a603a52506ed4c88f1decc", "primary_control": "random", "result_hash": "3e7bcae34f04eeeae1a56cc544f8352744fe008365e708f79ea2e558d7b09619", "signature": "b7cc47fdb0058058f62fb68ad13b1166bdf239523e42a81026d0704c20964d9c7fcbc7be8e08c9d18f0a4b9f3c26c79b89635fc60346947fb9188de832bb5e06", "state": "VERIFIED", "summary": {"count": 0.82197, "frontier": 0.852273, "goal": 0.611742, "hypdisc": 0.674242, "infogain": 0.852273, "prederr": 0.852273, "random": 0.655303, "risk": 0.852273}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:22:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:36Z", "cpu_seconds": 17.596170090999998, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-04489a7a2f9e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 8, "n": 4}, "prereg_hash": "463a4a186e61dd49b1240d45d9f2a41f895b367553e3ec6b11ec364f6acb510d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-04489a7a2f9e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:20:10Z", "seed": 1717200315, "treatment": "full"}, "priority": 2.0098384227443273, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ef8afcd77461fb3589268993690910301b6696a825f8974ee0444daed05bc8de", "run_seconds": 20.588826656341553, "seed": 1717200315, "started_at": "2026-11-22T03:20:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.447307, 0.447307], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-04489a7a2f9e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.5, 1.0, 0.25], "fixed_policy": [0.75, 0.25, 1.0, 1.0], "full": [1.0, 0.0, 0.75, 0.5], "no_world_model": [0.5, 0.0, 1.0, 0.5], "random": [1.0, 0.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "463a4a186e61dd49b1240d45d9f2a41f895b367553e3ec6b11ec364f6acb510d", "primary_control": "e141_best", "result_hash": "ef8afcd77461fb3589268993690910301b6696a825f8974ee0444daed05bc8de", "signature": "14a6e9ce7fdc519a13ecc401a502191c2830001f6679125433dc31cf2ed3d6c75c29ab9628be3fd8615608cc7b90a725abd3aef52fd48e02712d46ed8707920e", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.75, "full": 0.5625, "no_world_model": 0.5, "random": 0.6875}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:15Z", "cpu_seconds": 0.9298248939999993, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-41e50ac6aace", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "4cf90d86ff175c3d638431b16981ec93bfae836e69af7fb47cd7ebf7333d0f1f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-41e50ac6aace", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:20:10Z", "seed": 1159099072, "treatment": "full"}, "priority": 2.0099485909031203, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b63b15155e66618fad0a4d0cde40af4dfbcb620b847bfca258429f8c24dd6fa9", "run_seconds": 1.2848587036132812, "seed": 1159099072, "started_at": "2026-11-22T03:20:14Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:22:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-41e50ac6aace", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.75], "fixed_policy": [1.0, 0.625], "full": [1.0, 0.875], "no_world_model": [1.0, 0.5], "random": [1.0, 0.8125]}, "prereg_hash": "4cf90d86ff175c3d638431b16981ec93bfae836e69af7fb47cd7ebf7333d0f1f", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "b63b15155e66618fad0a4d0cde40af4dfbcb620b847bfca258429f8c24dd6fa9", "signature": "b6449d21b06016a23b923950c4915bc9025213e20be0d6248cad3e0d05c36d47c2559ebffb8b2c2e3bf89250571af925c1072d45a5d1e1fbe2ad0e0809c98b05", "state": "REJECTED", "summary": {"e141_best": 0.75, "fixed_policy": 0.625, "full": 0.875, "no_world_model": 0.5, "random": 0.8125}}, "verified_at": "2026-11-22T03:22:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:42Z", "cpu_seconds": 14.763623366000004, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8da404128ff5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 4, "n": 4}, "prereg_hash": "d52af38e213fe1345b07e09c594c696ebbad69239ce78e1991b96ac8722a32d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8da404128ff5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:20:10Z", "seed": 1199321815, "treatment": "full"}, "priority": 2.0092207832350293, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "18bc1304a64f148c1e6b72d4547436115787283fbbeaea999818653cc33ba57b", "run_seconds": 16.66304898262024, "seed": 1199321815, "started_at": "2026-11-22T03:20:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:22:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8da404128ff5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d52af38e213fe1345b07e09c594c696ebbad69239ce78e1991b96ac8722a32d2", "primary_control": "e141_best", "result_hash": "18bc1304a64f148c1e6b72d4547436115787283fbbeaea999818653cc33ba57b", "signature": "7e48bc2140300608bbbba00028ee060dc62d89f4d62dbdad20de81e03466bf2fc04e9cf295fda3ac4c9749e2f2e0d2106ef878f563436d6242db926eaa45b809", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:27Z", "cpu_seconds": 0.5340145069999984, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b6f16b521768", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "07045c47fb4ae879aff571fc463e8e5d5b33d971aedd977132cb62d39accf9db", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b6f16b521768", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:20:10Z", "seed": 376914285, "treatment": "full"}, "priority": 2.0091985264957573, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-18546d2074a0", "result_hash": "aba00737f98e71480ecf92fa7ce4d0c210396a6e5c388b476f7a9e77e0f9afb7", "run_seconds": 0.603684663772583, "seed": 376914285, "started_at": "2026-11-22T03:20:26Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b6f16b521768", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 1.0, 1.0], "fixed_policy": [0.75, 0.25, 0.5, 0.25], "full": [1.0, 0.0, 1.0, 1.0], "no_world_model": [1.0, 0.0, 1.0, 1.0], "random": [1.0, 0.5, 1.0, 0.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "07045c47fb4ae879aff571fc463e8e5d5b33d971aedd977132cb62d39accf9db", "primary_control": "e141_best", "replicated": false, "replication_of": "E141-ow_goal-18546d2074a0", "result_hash": "aba00737f98e71480ecf92fa7ce4d0c210396a6e5c388b476f7a9e77e0f9afb7", "signature": "e62a519e9758697a3726211d62657e2328e6be8d71298cfc14c32e97946d3d29a47a15946e897417a6fddfe77f5e0fa1f8f7aadcf8945426f7cac6a0da383500", "state": "PARTIAL", "summary": {"e141_best": 0.75, "fixed_policy": 0.4375, "full": 0.75, "no_world_model": 0.75, "random": 0.625}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:52Z", "cpu_seconds": 3.1614467750000017, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-39682ed40972", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "eabea2dcff9a9bc9ba1f697de35e6393ce74496f09f75bc0b2499457fc9fdcf0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-39682ed40972", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:20:10Z", "seed": 2112674033, "treatment": "full"}, "priority": 2.009163303005423, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3df6a5f42598cf0b6b70e1b8b6725d191b85de25a6c9e57e70d1bfdfeb81079f", "run_seconds": 4.189672231674194, "seed": 2112674033, "started_at": "2026-11-22T03:20:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017481, 0.070935], "effect": 0.026727, "evaluated_at": "2026-11-22T03:22:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-39682ed40972", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.986842105, 1.0, 0.520833333], "frozen": [0.230769231, 0.197368421, 0.340909091, 0.322916667], "full": [1.0, 1.0, 1.0, 0.614583333], "retrieval": [1.0, 1.0, 1.0, 0.59375], "rules_nodetect": [1.0, 1.0, 1.0, 0.614583333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "eabea2dcff9a9bc9ba1f697de35e6393ce74496f09f75bc0b2499457fc9fdcf0", "primary_control": "facts", "result_hash": "3df6a5f42598cf0b6b70e1b8b6725d191b85de25a6c9e57e70d1bfdfeb81079f", "signature": "545e5a340085014505b9a42b463777b134f04d76acf9c055d1f06435c345d84ad0223cf916e26e32645a1cc2137b3974b6d49d9dc2a2f34383611b00f9306607", "state": "VERIFIED", "summary": {"facts": 0.876919, "frozen": 0.272991, "full": 0.903646, "retrieval": 0.898438, "rules_nodetect": 0.903646}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:49Z", "cpu_seconds": 0.18780984099999998, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-bab1f5b896c3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "6ec1d8ec05332ca849edfe1ecc79c02b9eab9b45cb24f8547c825f6d99a48886", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bab1f5b896c3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:20:10Z", "seed": 214553104, "treatment": "full"}, "priority": 2.0091511752133533, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3d93793e3d3e9962958ec05b76eae81e077d4a0d8c3556ee410297ed918bcae1", "run_seconds": 0.4405808448791504, "seed": 214553104, "started_at": "2026-11-22T03:20:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003636, 0.011212], "effect": 0.003788, "evaluated_at": "2026-11-22T03:22:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-bab1f5b896c3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.909090909, 0.348484848, 1.0, 1.0], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [0.924242424, 0.348484848, 1.0, 1.0], "retrieval": [0.893939394, 0.484848485, 1.0, 1.0], "rules_nodetect": [0.924242424, 0.348484848, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6ec1d8ec05332ca849edfe1ecc79c02b9eab9b45cb24f8547c825f6d99a48886", "primary_control": "facts", "result_hash": "3d93793e3d3e9962958ec05b76eae81e077d4a0d8c3556ee410297ed918bcae1", "signature": "8ec651cc9d5de237598eebb25c3c26b967daef7cabb7ad08bc1016d8dd181251ecb11929f5e55aefd30174d547d2b2275ecbee7231f5243389301fcedd251b0e", "state": "VERIFIED", "summary": {"facts": 0.814394, "frozen": 0.030303, "full": 0.818182, "retrieval": 0.844697, "rules_nodetect": 0.818182}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:20Z", "cpu_seconds": 1.5129712450000028, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ca48068e491b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "4c2792e3e7e6632943f1e46b838a87f9cd496dd62fee7cdaa2e0f14a7927aef3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ca48068e491b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:20:10Z", "seed": 1865880957, "treatment": "full"}, "priority": 2.0096352127829116, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a982ca252de4bf72135c2e3a9bd26dd17437c11e358ee89047ffc2a008d800e8", "run_seconds": 3.196815013885498, "seed": 1865880957, "started_at": "2026-11-22T03:20:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01002, 0.155997], "effect": 0.072989, "evaluated_at": "2026-11-22T03:22:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ca48068e491b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.814141414, 0.816161616, 0.609195402, 0.652873563], "frozen": [0.0, 0.0, 0.096551724, 0.098850575], "full": [0.814141414, 0.816161616, 0.765517241, 0.788505747], "retrieval": [0.943434343, 0.961616162, 0.8, 0.825287356], "rules_nodetect": [0.814141414, 0.816161616, 0.765517241, 0.788505747]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "4c2792e3e7e6632943f1e46b838a87f9cd496dd62fee7cdaa2e0f14a7927aef3", "primary_control": "facts", "result_hash": "a982ca252de4bf72135c2e3a9bd26dd17437c11e358ee89047ffc2a008d800e8", "signature": "4ddb15133a54d34790f5486e743627f47e591fbc2c0c35deaaad4c50e3bfc5632670c4b7adccf49ab241cea49d56d2aaba9b2a74b319e15df3455803c8736c02", "state": "VERIFIED", "summary": {"facts": 0.723093, "frozen": 0.048851, "full": 0.796082, "retrieval": 0.882584, "rules_nodetect": 0.796082}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:19Z", "cpu_seconds": 4.792217318, "created_at": "2026-11-22T03:20:10Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d057aa174454", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6d12ab384f9d6490b2c3a27d42bd07af1c9ad6194f04f639c5d4d11d836694c9", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d057aa174454", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:20:10Z", "seed": 637681152, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "724db3c488e2da2ccc2e9b0d2cdbc4b2d298dade391340662d354c6cea552c64", "run_seconds": 7.090100526809692, "seed": 637681152, "started_at": "2026-11-22T03:20:12Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d057aa174454", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6d12ab384f9d6490b2c3a27d42bd07af1c9ad6194f04f639c5d4d11d836694c9", "primary_control": "reset", "result_hash": "724db3c488e2da2ccc2e9b0d2cdbc4b2d298dade391340662d354c6cea552c64", "signature": "3f526810744b45f9c59601ee86d563d9da323498510f2daa639ef60a65d4990c984fa0cae0f4b8584e09177bdb7eb1ab7d95d5cd622a8adb305fb41d3294dc0b", "state": "PARTIAL", "summary": {"inherit": 0.7, "reset": 0.7, "root": 0.2833333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:22:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:55Z", "cpu_seconds": 1.820162248999992, "created_at": "2026-11-22T03:20:27Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c760e7ce9d09", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "57f1329b4691e9ac4c893eca8ad2762fba2d8551ba98524a18d4cf9fa669f4db", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c760e7ce9d09", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:20:27Z", "seed": 773448967, "treatment": "iw_temperature"}, "priority": 2.0091121609398006, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-dd183dae9f61", "result_hash": "2d8581c3f3a75e990fa967e68befe4e2a524e421f9ab17a08e40abb81df3cfee", "run_seconds": 2.655097484588623, "seed": 773448967, "started_at": "2026-11-22T03:20:52Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.02936, 0.023675], "effect": -0.002842, "evaluated_at": "2026-11-22T03:22:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c760e7ce9d09", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.280055768, 0.290681953, 0.269066688, 0.12256765, 0.23705578, 0.104305198, 0.298352235, 0.239702488, 0.299698635, 0.266379848, 0.284615178, 0.260360713, 0.163869542, 0.182343753, 0.21890971, 0.23667609], "raw": [0.266130488, 0.25365484, 0.16666615, 0.197617117, 0.164904773, 0.186119845, 0.214516055, 0.18577802, 0.285661945, 0.21840386, 0.24072256, 0.260360713, 0.245313915, 0.11069291, 0.15929483, 0.23064754], "temperature": [0.271594303, 0.273233262, 0.24247974, 0.208859472, 0.12279616, 0.155105712, 0.24488034, 0.22916886, 0.286557385, 0.266379848, 0.30130196, 0.302645415, 0.262721548, 0.145467235, 0.15929483, 0.23667609]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0409, "losses": 9, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.91", "prereg_hash": "57f1329b4691e9ac4c893eca8ad2762fba2d8551ba98524a18d4cf9fa669f4db", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-dd183dae9f61", "result_hash": "2d8581c3f3a75e990fa967e68befe4e2a524e421f9ab17a08e40abb81df3cfee", "signature": "6dcb24e0f56420e41e904c397f7fbfcd5e0cd05ff783f480eb1b123f6dd3a090702f955140e0a242ae10bb6984e2ceaccf91a4794bd2fd9d6c80c8b219be880a", "state": "VERIFIED", "summary": {"iw_temperature": 0.234665, "raw": 0.211655, "temperature": 0.231823}, "ties": 2, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:22:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:44Z", "cpu_seconds": 0.7138743980000015, "created_at": "2026-11-22T03:20:27Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4d53b3c305ab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "2991ea8f3087ad8e6f6394e99ecfa949df176477e835d445661acf423ad51af0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4d53b3c305ab", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:20:27Z", "seed": 482561234, "treatment": "frontier"}, "priority": 2.009220710939967, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b875758a39f25cd28fd11422e4ac52340aadf2e9f8ce8caf02ec0d5acc432f84", "run_seconds": 0.7404747009277344, "seed": 482561234, "started_at": "2026-11-22T03:20:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.049429, 0.0405], "effect": -0.004464, "evaluated_at": "2026-11-22T03:22:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4d53b3c305ab", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 0.979166667], "frontier": [0.964285714, 0.9375, 0.958333333, 0.9375], "goal": [0.946428571, 0.791666667, 0.895833333, 0.958333333], "hypdisc": [0.982142857, 0.958333333, 0.875, 0.979166667], "infogain": [0.964285714, 0.9375, 0.958333333, 0.9375], "prederr": [0.964285714, 0.9375, 0.958333333, 0.9375], "random": [0.982142857, 0.979166667, 0.895833333, 0.958333333], "risk": [0.964285714, 0.9375, 0.958333333, 0.9375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "2991ea8f3087ad8e6f6394e99ecfa949df176477e835d445661acf423ad51af0", "primary_control": "random", "result_hash": "b875758a39f25cd28fd11422e4ac52340aadf2e9f8ce8caf02ec0d5acc432f84", "signature": "007d90215c8a02267ed791c85237a6f0b87d46c8f50e32d129ae0a676fc7cbfa48ca0e5f3e70f0edd8f409e19af28268dbaf19af45309d646dc70a2e92f2a208", "state": "VERIFIED", "summary": {"count": 0.994792, "frontier": 0.949405, "goal": 0.898065, "hypdisc": 0.948661, "infogain": 0.949405, "prederr": 0.949405, "random": 0.953869, "risk": 0.949405}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:22:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:42Z", "cpu_seconds": 1.4024124629999974, "created_at": "2026-11-22T03:20:27Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a4ce761a822d", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "43a0aa35fa25fd75ccb5a3b75daea9416a0c854fb7e4e601d24c176df14bf96a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a4ce761a822d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:20:27Z", "seed": 2023566682, "treatment": "frontier"}, "priority": 2.0099107640817784, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "170267d4afa3d55c6e5ee56e80652df3a2521b8fda4f0edd3df37256d299fe68", "run_seconds": 1.404369592666626, "seed": 2023566682, "started_at": "2026-11-22T03:20:41Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:22:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a4ce761a822d", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.497945], "frontier": [1.0, 0.488719], "goal": [1.0, 0.487248], "hypdisc": [1.0, 0.50977], "infogain": [1.0, 0.409247], "prederr": [1.0, 0.48328], "random": [1.0, 0.299517], "risk": [1.0, 0.47502]}, "prereg_hash": "43a0aa35fa25fd75ccb5a3b75daea9416a0c854fb7e4e601d24c176df14bf96a", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "170267d4afa3d55c6e5ee56e80652df3a2521b8fda4f0edd3df37256d299fe68", "signature": "01d87e3179863400ba022960e7bda509c93cf42b3fdda593f776e4bb29b0190306eae5355b57648ceb52498fa5600d5d3649c08a45e7021ff502457541a93b0f", "state": "REJECTED", "summary": {"count": 0.497945, "frontier": 0.488719, "goal": 0.487248, "hypdisc": 0.50977, "infogain": 0.409247, "prederr": 0.48328, "random": 0.299517, "risk": 0.47502}}, "verified_at": "2026-11-22T03:22:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:32Z", "cpu_seconds": 0.5429719069999983, "created_at": "2026-11-22T03:20:27Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-66f51299a8c2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "acee4420d2164c0c53460e1912fad6c7e5f6066f3e66aad7910ac62a80a6fa0e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-66f51299a8c2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:20:27Z", "seed": 489301378, "treatment": "full"}, "priority": 2.009997941379837, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2523ef1a2b5c4661a8401c1f4094b894271239cf4a7344537aa0de88bda5e7df", "run_seconds": 0.6026644706726074, "seed": 489301378, "started_at": "2026-11-22T03:20:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.844139, 0.094139], "effect": -0.375, "evaluated_at": "2026-11-22T03:22:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-66f51299a8c2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.5, 0.5], "full": [0.75, 0.25, 0.25, 1.0], "no_world_model": [1.0, 0.25, 0.25, 1.0], "random": [0.75, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "acee4420d2164c0c53460e1912fad6c7e5f6066f3e66aad7910ac62a80a6fa0e", "primary_control": "e141_best", "result_hash": "2523ef1a2b5c4661a8401c1f4094b894271239cf4a7344537aa0de88bda5e7df", "signature": "4061efb643374ddd97284df4987a4ef2d8ec9f940db45415d17a7be25f0a0c90f4437cc4b4e7d18cec89d2a6599d05d6c6c6bdcce20c086c4a534f845e3cb503", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.25, "full": 0.5625, "no_world_model": 0.625, "random": 0.9375}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:38Z", "cpu_seconds": 0.7072543969999998, "created_at": "2026-11-22T03:20:27Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8ad50116d430", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "4b006557966733b98d0c404eda4aab66c92d64fb20f4f3d173c1ad7744e7bf15", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8ad50116d430", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:20:27Z", "seed": 1153915452, "treatment": "full"}, "priority": 2.0099788303504638, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ae81395b12b001ef4353856814abc5da43c394d16da52f8e4acd164c1acb34aa", "run_seconds": 0.9591498374938965, "seed": 1153915452, "started_at": "2026-11-22T03:20:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:22:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8ad50116d430", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 1.0, 0.75], "fixed_policy": [1.0, 0.25, 0.25, 1.0], "full": [0.25, 1.0, 1.0, 0.75], "no_world_model": [0.25, 1.0, 1.0, 0.25], "random": [0.75, 0.5, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "4b006557966733b98d0c404eda4aab66c92d64fb20f4f3d173c1ad7744e7bf15", "primary_control": "e141_best", "result_hash": "ae81395b12b001ef4353856814abc5da43c394d16da52f8e4acd164c1acb34aa", "signature": "0188a493abd98ca19f15523cebfb16cf2e2bd8c3ced31dfaae731fc95f44b8b88f42861b39197b027f5a5ddffeae15aff2a56bffef81610f12d1f9aec3e99703", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.625, "full": 0.75, "no_world_model": 0.625, "random": 0.8125}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:58Z", "cpu_seconds": 1.4143251849999956, "created_at": "2026-11-22T03:20:27Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d7075ca58762", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "17a55f8962f82eb8a8f0fe5109eff89f2b13796589f358282e7833013b3dac8b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d7075ca58762", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:20:27Z", "seed": 1051619192, "treatment": "gated"}, "priority": 2.008955448854446, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "2b4d8c7631bed92f2cdfc5eebc648a04110a0b6ccf4f38c4a66f0ca6d53171e4", "run_seconds": 1.6164062023162842, "seed": 1051619192, "started_at": "2026-11-22T03:20:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013524, 0.119293], "effect": 0.052885, "evaluated_at": "2026-11-22T03:22:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d7075ca58762", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.28125, 0.471153846, 0.475, 0.528846154], "gated": [0.40625, 0.567307692, 0.475, 0.528846154], "naive": [0.635416667, 0.701923077, 0.6375, 0.701923077], "scratch": [0.28125, 0.471153846, 0.475, 0.538461538], "ungated": [0.510416667, 0.596153846, 0.6625, 0.701923077]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "17a55f8962f82eb8a8f0fe5109eff89f2b13796589f358282e7833013b3dac8b", "primary_control": "scratch", "result_hash": "2b4d8c7631bed92f2cdfc5eebc648a04110a0b6ccf4f38c4a66f0ca6d53171e4", "signature": "a92e7e857f14ef246f2fa719393f5d575a91c44f1df5787ed2fb74d1c801135af0c40b651fd6b065866a63eada1e40a7d0e20e28773800064a7564319841ce0a", "state": "VERIFIED", "summary": {"compressed": 0.439063, "gated": 0.494351, "naive": 0.669191, "scratch": 0.441466, "ungated": 0.617748}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:22:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:33Z", "cpu_seconds": 5.2920231399999995, "created_at": "2026-11-22T03:20:27Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0f027a578e84", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d88f7252933341897a7924a269145d6abc164bada143e6af35f8cb8f1a288dbe", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0f027a578e84", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:20:27Z", "seed": 1696146626, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "b7bdbd85e59a7a656137144a5861adf22829e2cb9651d4eaece0d66be3b9dce6", "run_seconds": 5.6105077266693115, "seed": 1696146626, "started_at": "2026-11-22T03:20:27Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0f027a578e84", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.7165018433776399, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0], "reset": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d88f7252933341897a7924a269145d6abc164bada143e6af35f8cb8f1a288dbe", "primary_control": "reset", "result_hash": "b7bdbd85e59a7a656137144a5861adf22829e2cb9651d4eaece0d66be3b9dce6", "signature": "bd5f3d9ee3f1b8a6ab7f20ebd7ff9aacd55443454e3429590423ed296acdf84b5c6032bda4c601bd27ba924f00cb9c09502b49d9326bfa30095ae1fffee6b304", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:22:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:12Z", "cpu_seconds": 0.2547252040000103, "created_at": "2026-11-22T03:20:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8e135da2a484", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "ecee6fe283ec717aa62d470d576a1851dbc969a3c88d26cb4aac2da3694067ab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8e135da2a484", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:20:43Z", "seed": 814805080, "treatment": "frontier"}, "priority": 2.0088529225603655, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f9470388313eeea830acd4f7b7b83147e70248553cf09a7a76c44eaeda7c401b", "run_seconds": 0.2663609981536865, "seed": 814805080, "started_at": "2026-11-22T03:21:12Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-8e135da2a484", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ecee6fe283ec717aa62d470d576a1851dbc969a3c88d26cb4aac2da3694067ab", "primary_control": "random", "result_hash": "f9470388313eeea830acd4f7b7b83147e70248553cf09a7a76c44eaeda7c401b", "signature": "6ce2d2e3335466c963ecb78664a8830ba25e45766a1677b0fdc596e5b4b025436b6ad27a70c5876b3fd98cb0ea2ef2488a7ce64a8ceb5a565a63c77941c94008", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:22:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:20:47Z", "cpu_seconds": 0.48800490599998625, "created_at": "2026-11-22T03:20:43Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-38926040bceb", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "37b62d6095c397f6df1e039087636cc80c0a25cd11f363fcb5fb4e51edad266c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-38926040bceb", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:20:43Z", "seed": 1560571791, "treatment": "full"}, "priority": 2.0097463363969754, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6fdd645319f307a65e423856ea09d90266dc7fc5ff608df4d1b8d3ae0ed3d5b2", "run_seconds": 0.4914829730987549, "seed": 1560571791, "started_at": "2026-11-22T03:20:47Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:22:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-38926040bceb", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.3125], "fixed_policy": [1.0, 0.3125], "full": [1.0, 0.4375], "no_world_model": [1.0, 0.0625], "random": [1.0, 0.3125]}, "prereg_hash": "37b62d6095c397f6df1e039087636cc80c0a25cd11f363fcb5fb4e51edad266c", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "6fdd645319f307a65e423856ea09d90266dc7fc5ff608df4d1b8d3ae0ed3d5b2", "signature": "261e752a06a59e2363e6c452488c37f93e7405c3ae59cec85a104838d7553ec01459fe327b3b4d06ab1ef05ba8cf2de9a98484779084cf343834159344e6dd08", "state": "REJECTED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.3125, "full": 0.4375, "no_world_model": 0.0625, "random": 0.3125}}, "verified_at": "2026-11-22T03:22:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:00Z", "cpu_seconds": 4.398272803000001, "created_at": "2026-11-22T03:20:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1ce1f1416bd6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 2, "n": 4}, "prereg_hash": "48f4e4d671a42a9640650a9ab965bceb6c8615023c092f61b14db761fb2fdb24", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1ce1f1416bd6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:20:43Z", "seed": 328884417, "treatment": "full"}, "priority": 2.0090214624749834, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e12ba6970726e80ab9f485a3e8137558a65286f6cc22eda0310e65defac47939", "run_seconds": 4.997707843780518, "seed": 328884417, "started_at": "2026-11-22T03:20:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037821, 0.367871], "effect": 0.165025, "evaluated_at": "2026-11-22T03:22:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1ce1f1416bd6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.303448276, 0.341964286, 0.804040404], "frozen": [0.366666667, 0.103448276, 0.252678571, 0.0], "full": [0.991666667, 0.554022989, 0.759821429, 0.804040404], "retrieval": [0.991666667, 0.577011494, 0.828571429, 0.947474747], "rules_nodetect": [0.991666667, 0.554022989, 0.759821429, 0.804040404]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "48f4e4d671a42a9640650a9ab965bceb6c8615023c092f61b14db761fb2fdb24", "primary_control": "facts", "result_hash": "e12ba6970726e80ab9f485a3e8137558a65286f6cc22eda0310e65defac47939", "signature": "8a94cf842e395a67988288fc1c875034ed26ce34335d9c0e8369e7c73a5e08997062057ede027a2c8da742fafc47dd73eb942d89e7c488f1dcaa16cf9aa54104", "state": "VERIFIED", "summary": {"facts": 0.612363, "frozen": 0.180698, "full": 0.777388, "retrieval": 0.836181, "rules_nodetect": 0.777388}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:01Z", "cpu_seconds": 0.7383911400000045, "created_at": "2026-11-22T03:20:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-aae8805e1322", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 6, "n": 4}, "prereg_hash": "9476fe6628fb2a3a8bd15a70a2587cc66fbd934abfb092bc6757b4ab8c9a903d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-aae8805e1322", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:20:43Z", "seed": 1584840883, "treatment": "full"}, "priority": 2.0089493253747435, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e1f992317d3106908aed52680d20f4375993ded7856e1fe83f308670367e978e", "run_seconds": 0.7954373359680176, "seed": 1584840883, "started_at": "2026-11-22T03:21:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.011538, 0.146538], "effect": 0.0675, "evaluated_at": "2026-11-22T03:22:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-aae8805e1322", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.441666667, 0.9, 0.48, 0.4825], "frozen": [0.37, 0.55, 0.413333333, 0.420833333], "full": [0.603333333, 0.95, 0.448333333, 0.5725], "retrieval": [0.774166667, 0.975, 0.766666667, 0.7875], "rules_nodetect": [0.4175, 0.7125, 0.440833333, 0.433333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "9476fe6628fb2a3a8bd15a70a2587cc66fbd934abfb092bc6757b4ab8c9a903d", "primary_control": "facts", "result_hash": "e1f992317d3106908aed52680d20f4375993ded7856e1fe83f308670367e978e", "signature": "6dd2db6c4292d4455bc3df424d567b53ddf3219e66e0681fcb4498dd9db1f4e39d79e476e4b7f3d479d2d89248915ae8c61e112cd6ec21feccea7ba5daadde0c", "state": "VERIFIED", "summary": {"facts": 0.576042, "frozen": 0.438542, "full": 0.643542, "retrieval": 0.825833, "rules_nodetect": 0.501042}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:22:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:32Z", "cpu_seconds": 2.556231208, "created_at": "2026-11-22T03:21:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1e739793484a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "73837f5d47be3c0ae6c3a010e9b2c62fd95282ad3b209512447a84068081ed60", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1e739793484a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:21:00Z", "seed": 775514995, "treatment": "frontier"}, "priority": 2.008795030946911, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9bbdd160c8b1551011954d78b6703f4ace84546d672ded36c2a50dd195cc8b75", "run_seconds": 2.8509397506713867, "seed": 775514995, "started_at": "2026-11-22T03:21:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.080848, -0.001348], "effect": -0.041098, "evaluated_at": "2026-11-22T03:22:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1e739793484a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.322690217, 0.760416667, 1.0, 0.555232558], "frontier": [0.289402174, 0.552083333, 1.0, 0.406976744], "goal": [0.279211957, 0.572916667, 1.0, 0.380813953], "hypdisc": [0.300951087, 0.541666667, 1.0, 0.456395349], "infogain": [0.289402174, 0.552083333, 1.0, 0.406976744], "prederr": [0.289402174, 0.552083333, 1.0, 0.406976744], "random": [0.30638587, 0.609375, 1.0, 0.497093023], "risk": [0.289402174, 0.552083333, 1.0, 0.406976744]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "73837f5d47be3c0ae6c3a010e9b2c62fd95282ad3b209512447a84068081ed60", "primary_control": "random", "result_hash": "9bbdd160c8b1551011954d78b6703f4ace84546d672ded36c2a50dd195cc8b75", "signature": "9d26af6b8b6e3074ec066042a86f7cca8c4c16a032824172af4129358c9114ef5eb45200faa793f9c313ef2f3938db1509535d59379ca62ff88697774b0b0b07", "state": "VERIFIED", "summary": {"count": 0.659585, "frontier": 0.562116, "goal": 0.558236, "hypdisc": 0.574753, "infogain": 0.562116, "prederr": 0.562116, "random": 0.603213, "risk": 0.562116}, "ties": 1, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:22:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:07Z", "cpu_seconds": 0.9071298469999931, "created_at": "2026-11-22T03:21:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-efa6513e4102", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "ed6e59772bb0f1d7072f800be1255a9fe634d720d27e6e3d64fa4075b77579cf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-efa6513e4102", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:21:00Z", "seed": 1468676116, "treatment": "frontier"}, "priority": 2.0097639251911894, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "694d3ee69bef96701eb7ef67598b069ffdec7d1462f7b91142bc3f81fe97fa0d", "run_seconds": 1.3314855098724365, "seed": 1468676116, "started_at": "2026-11-22T03:21:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.042937, 0.037728], "effect": -0.002604, "evaluated_at": "2026-11-22T03:22:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-efa6513e4102", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.95, 0.7, 1.0, 1.0], "frontier": [0.95, 0.8, 0.96875, 0.984375], "goal": [0.95, 0.5, 0.958333333, 0.953125], "hypdisc": [0.925, 0.7, 0.979166667, 0.984375], "infogain": [0.95, 0.725, 0.96875, 0.984375], "prederr": [0.95, 0.75, 0.96875, 0.984375], "random": [0.9, 0.85, 0.979166667, 0.984375], "risk": [0.95, 0.8, 0.96875, 0.984375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "ed6e59772bb0f1d7072f800be1255a9fe634d720d27e6e3d64fa4075b77579cf", "primary_control": "random", "result_hash": "694d3ee69bef96701eb7ef67598b069ffdec7d1462f7b91142bc3f81fe97fa0d", "signature": "356de46333e55a79a3b77bf0b17c963d3dfffc768f3a05cc10e9914980b71f4c3366c876ed7d9408a838d88b0e10d47460e5a502f2efa4b327c9734a48a58a05", "state": "VERIFIED", "summary": {"count": 0.9125, "frontier": 0.925781, "goal": 0.840365, "hypdisc": 0.897135, "infogain": 0.907031, "prederr": 0.913281, "random": 0.928385, "risk": 0.925781}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:22:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:12Z", "cpu_seconds": 0.7578116259999987, "created_at": "2026-11-22T03:21:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0cb858ee983a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "bc80b7f5428f48d221d1f46958be419434fc9dc42376f17ef75012deeed19c42", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0cb858ee983a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:00Z", "seed": 1499051726, "treatment": "full"}, "priority": 2.009589336513968, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2ca348a848e20cc5837e72c361c4dd916093fdbb40a961f19d5ce629c270b76b", "run_seconds": 0.809321403503418, "seed": 1499051726, "started_at": "2026-11-22T03:21:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.030267, 0.101107], "effect": 0.03542, "evaluated_at": "2026-11-22T03:22:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0cb858ee983a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.615120275, 0.653846154, 0.614035088, 0.780068729], "frozen": [0.281786942, 0.397435897, 0.30994152, 0.281786942], "full": [0.725085911, 0.615384615, 0.684210526, 0.780068729], "retrieval": [0.697594502, 0.679487179, 0.649122807, 0.780068729], "rules_nodetect": [0.725085911, 0.871794872, 0.684210526, 0.780068729]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "bc80b7f5428f48d221d1f46958be419434fc9dc42376f17ef75012deeed19c42", "primary_control": "facts", "result_hash": "2ca348a848e20cc5837e72c361c4dd916093fdbb40a961f19d5ce629c270b76b", "signature": "4fee6e083ae5a7e6ecc20fa74e08f931320004957d230e8ffcab5f75e12a7a1953c2e8ac89800cd2c1b4315b01c3dff95949b835278d3a2a2c06bba3536a5a09", "state": "VERIFIED", "summary": {"facts": 0.665768, "frozen": 0.317738, "full": 0.701187, "retrieval": 0.701568, "rules_nodetect": 0.76529}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:13Z", "cpu_seconds": 1.2775853220000002, "created_at": "2026-11-22T03:21:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2e89ae98a72d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "11f0425eafecea6765c25096b2ab6b2c188dfa66a89e1d6013b7e5da949e4b1f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2e89ae98a72d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:00Z", "seed": 1003603158, "treatment": "full"}, "priority": 2.009026622048115, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1beeb9c7ec07212cfcdd951352a7f4651413d364d40704efa7b70db368622779", "run_seconds": 1.3790462017059326, "seed": 1003603158, "started_at": "2026-11-22T03:21:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.051613, -0.029697], "effect": -0.040655, "evaluated_at": "2026-11-22T03:22:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2e89ae98a72d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.4375, 0.457928803, 0.43452381, 0.384057971], "frozen": [0.166666667, 0.181229773, 0.180059524, 0.175362319], "full": [0.400297619, 0.415857605, 0.40625, 0.328985507], "retrieval": [0.453869048, 0.482200647, 0.450892857, 0.37826087], "rules_nodetect": [0.400297619, 0.733009709, 0.40625, 0.328985507]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "11f0425eafecea6765c25096b2ab6b2c188dfa66a89e1d6013b7e5da949e4b1f", "primary_control": "facts", "result_hash": "1beeb9c7ec07212cfcdd951352a7f4651413d364d40704efa7b70db368622779", "signature": "aa6340cd7e54383d053fd3f83cfb75acb94e9b05be9ec836ed0fb75ab7925fb92907f3a550ebac482e3faab0172a74c97af786637f65577ef11ed70df3ebef0e", "state": "VERIFIED", "summary": {"facts": 0.428503, "frozen": 0.17583, "full": 0.387848, "retrieval": 0.441306, "rules_nodetect": 0.467136}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:32Z", "cpu_seconds": 1.7425212329999908, "created_at": "2026-11-22T03:21:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-83e4e802ef57", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "6f0c26f5221ca59dbb37c5acea81bc5a384e95b00b093bc4f4144d5ba01eba8a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-83e4e802ef57", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:00Z", "seed": 1442365542, "treatment": "full"}, "priority": 2.0085661486374766, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "af726481b49f88f18c4c315cf148b9d91b54200880648c2ad19d423e81338fa1", "run_seconds": 2.0161521434783936, "seed": 1442365542, "started_at": "2026-11-22T03:21:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.248414, 0.424503], "effect": 0.336458, "evaluated_at": "2026-11-22T03:22:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-83e4e802ef57", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.741666667, 0.729166667, 0.633333333, 0.55], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.883333333, 0.925, 0.908333333, 0.895833333], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "6f0c26f5221ca59dbb37c5acea81bc5a384e95b00b093bc4f4144d5ba01eba8a", "primary_control": "facts", "result_hash": "af726481b49f88f18c4c315cf148b9d91b54200880648c2ad19d423e81338fa1", "signature": "2e19bc91beb06c55c61cc425467bf153a949c3c3561924da73b7f7970b5b6f74165366af8838df8688af7fa9b18279dc5c7eb31241179e764ffa45227a1c1a0c", "state": "VERIFIED", "summary": {"facts": 0.663542, "frozen": 0.0, "full": 1.0, "retrieval": 0.903125, "rules_nodetect": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:22:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:06Z", "cpu_seconds": 2.2489570090000015, "created_at": "2026-11-22T03:21:00Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-763beb20f13e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lights|lights|1|1"}, "prereg_hash": "57f779a652d0bb7909fc7b3e8aa81df86fa338c0050f00250cb9296bee162b39", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-763beb20f13e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lights|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:21:00Z", "seed": 244574784, "treatment": "gated"}, "priority": 2.009865170032556, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6792864607a430277c2b4fccd7039d6ab4eb113503c6e1fca282a365ef6ee7a8", "run_seconds": 2.967453718185425, "seed": 244574784, "started_at": "2026-11-22T03:21:03Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-763beb20f13e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.523809524, 0.258064516, 0.247311828, 0.408602151], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.682539683, 0.247311828, 0.844086022, 0.543010753]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "57f779a652d0bb7909fc7b3e8aa81df86fa338c0050f00250cb9296bee162b39", "primary_control": "scratch", "result_hash": "6792864607a430277c2b4fccd7039d6ab4eb113503c6e1fca282a365ef6ee7a8", "signature": "fc06effc26762cdd5359cd423624530cf2df3e504c2031e09ed31207627bba6a1fb4e39b8bcf6f0094dc6be7b61837137bca51d9fd4a305dac25f4243b3c1308", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.359447, "scratch": 1.0, "ungated": 0.579237}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:22:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:06Z", "cpu_seconds": 1.1316664900000006, "created_at": "2026-11-22T03:21:00Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ee42555633c2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "prereg_hash": "13837e7e6ec02ddc99acf28652bc24619548c3a8276d97a2549b43664b597f8d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ee42555633c2", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:21:00Z", "seed": 1611956850, "treatment": "gated"}, "priority": 2.0098049310221096, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "a64cff32390234d9263ce651c199d0a8402d46fa555188c5762d30790c54c661", "run_seconds": 1.6012396812438965, "seed": 1611956850, "started_at": "2026-11-22T03:21:05Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-ee42555633c2", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.328125, 0.192307692, 0.308743169, 0.208333333], "gated": [0.328125, 0.192307692, 0.308743169, 0.208333333], "naive": [0.30078125, 0.163461538, 0.258196721, 0.177083333], "scratch": [0.328125, 0.192307692, 0.308743169, 0.208333333], "ungated": [0.30078125, 0.163461538, 0.258196721, 0.197916667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "13837e7e6ec02ddc99acf28652bc24619548c3a8276d97a2549b43664b597f8d", "primary_control": "scratch", "result_hash": "a64cff32390234d9263ce651c199d0a8402d46fa555188c5762d30790c54c661", "signature": "f3349300d03c90fd031ab2e0a7b2967b18caed2ea7c7a0f9ec113df48587008073babb670f1001fe010a77b194d28f9234989136862a84c7d00584add5c4c702", "state": "PARTIAL", "summary": {"compressed": 0.259377, "gated": 0.259377, "naive": 0.224881, "scratch": 0.259377, "ungated": 0.230089}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:22:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:19Z", "cpu_seconds": 0.9321583139999987, "created_at": "2026-11-22T03:21:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7acb7bdeb989", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "94a2622fe511b0ae7eb47e38e835560fccee4aae25244d15dc9d85a62f7b5caf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7acb7bdeb989", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:17Z", "seed": 1316268787, "treatment": "full"}, "priority": 2.009365967633273, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1a4384de4e08345d322e2e90af2664e779ac65c76c44e31a88376c7a265c4f58", "run_seconds": 1.1964678764343262, "seed": 1316268787, "started_at": "2026-11-22T03:21:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T03:22:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7acb7bdeb989", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.25], "fixed_policy": [1.0, 0.25, 1.0, 1.0], "full": [1.0, 1.0, 0.5, 0.25], "no_world_model": [1.0, 1.0, 0.75, 0.5], "random": [1.0, 1.0, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "94a2622fe511b0ae7eb47e38e835560fccee4aae25244d15dc9d85a62f7b5caf", "primary_control": "e141_best", "result_hash": "1a4384de4e08345d322e2e90af2664e779ac65c76c44e31a88376c7a265c4f58", "signature": "eea292b85ffd33fdd77773370ceed945bd816839bf5a6dedbd39766a039cbbe1e4ce2b898969a52666467a0c4f711ee9b04135a2d5f7eaa06229c83259d2570d", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.8125, "full": 0.6875, "no_world_model": 0.8125, "random": 0.875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:26Z", "cpu_seconds": 1.6422289389999989, "created_at": "2026-11-22T03:21:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9440f454c347", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 6, "n": 4}, "prereg_hash": "73595e956278e196410ea2144fb413c622419628f2415beebccc419ee63dd1f2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9440f454c347", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:17Z", "seed": 311355863, "treatment": "full"}, "priority": 2.0091247572263558, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8b87f2e75ba57e1b4abe87608fc1228082ec5326053b48dc6f3163a14059b5ed", "run_seconds": 3.136051893234253, "seed": 311355863, "started_at": "2026-11-22T03:21:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.094139, 0.844139], "effect": 0.375, "evaluated_at": "2026-11-22T03:22:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9440f454c347", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 1.0, 0.0], "fixed_policy": [0.0, 0.25, 0.5, 0.25], "full": [0.25, 1.0, 1.0, 1.0], "no_world_model": [0.25, 0.75, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "73595e956278e196410ea2144fb413c622419628f2415beebccc419ee63dd1f2", "primary_control": "e141_best", "result_hash": "8b87f2e75ba57e1b4abe87608fc1228082ec5326053b48dc6f3163a14059b5ed", "signature": "7f24184484aead8493d30b013968c6747e078865ef758a6b5f95d451fa854ca05ba40c726890835d882e57b3c5abae6549872bc2523c925b4b1301ae1270a807", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.25, "full": 0.8125, "no_world_model": 0.6875, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:25Z", "cpu_seconds": 0.4444619520000117, "created_at": "2026-11-22T03:21:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c5336f3c6b1d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "fa45473818c171692d931031c72b861be1d16e29e38c66c1156ed16b984b2b2f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c5336f3c6b1d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:17Z", "seed": 136317359, "treatment": "full"}, "priority": 2.008836253581437, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4e14c9e6e38917a5a7998d0cce21adf80d1b41ba290d4cc615c4b6b66897f608", "run_seconds": 1.3827629089355469, "seed": 136317359, "started_at": "2026-11-22T03:21:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.42207, 0.04707], "effect": -0.1875, "evaluated_at": "2026-11-22T03:22:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c5336f3c6b1d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 0.25, 0.75], "fixed_policy": [0.25, 0.25, 0.0, 0.75], "full": [0.25, 0.25, 0.25, 0.25], "no_world_model": [0.25, 1.0, 0.0, 0.25], "random": [0.25, 1.0, 0.25, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "fa45473818c171692d931031c72b861be1d16e29e38c66c1156ed16b984b2b2f", "primary_control": "e141_best", "result_hash": "4e14c9e6e38917a5a7998d0cce21adf80d1b41ba290d4cc615c4b6b66897f608", "signature": "eff583cb595d4edae8222d3c828b641eaf591354a66ee00b405d156477d7c6c705de96b62ab53fcd9eb9baed947cb0b8da3fbc551dfee89faf09169d6a9c1d02", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.3125, "full": 0.25, "no_world_model": 0.375, "random": 0.375}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:24Z", "cpu_seconds": 0.18310589800000088, "created_at": "2026-11-22T03:21:17Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2c8696f8416d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "a5fbf6523987838b907badff89879fcc6c3b4edd22010c215466e3c5a42e6e93", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2c8696f8416d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:17Z", "seed": 507560759, "treatment": "full"}, "priority": 2.008867360857586, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "99b516f7f78d560e3cf2af941be6ab8733cd9af4a739f3355ab8b425ffcef0e3", "run_seconds": 0.3724987506866455, "seed": 507560759, "started_at": "2026-11-22T03:21:24Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2c8696f8416d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.988636364, 0.954545455, 1.0], "frozen": [0.034090909, 0.034090909, 0.034090909, 0.022727273], "full": [1.0, 0.988636364, 0.954545455, 1.0], "retrieval": [1.0, 0.988636364, 0.931818182, 1.0], "rules_nodetect": [1.0, 0.988636364, 0.954545455, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a5fbf6523987838b907badff89879fcc6c3b4edd22010c215466e3c5a42e6e93", "primary_control": "facts", "result_hash": "99b516f7f78d560e3cf2af941be6ab8733cd9af4a739f3355ab8b425ffcef0e3", "signature": "aafedf7ac08a2d3d14680f64f471d5fcc4263cd7160cdaba871b6baef74bfdf69be9ae3a0ab8d0d91cd70d502e863803632fd29a219d4f8b1064b8492d449a0f", "state": "PARTIAL", "summary": {"facts": 0.985795, "frozen": 0.03125, "full": 0.985795, "retrieval": 0.980114, "rules_nodetect": 0.985795}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:18Z", "cpu_seconds": 0.7443349710000007, "created_at": "2026-11-22T03:21:17Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a28e308a5278", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "10c3c6374544532f31ffaae55b36177a7854407a87c594db580dec64b0c7df5b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a28e308a5278", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:17Z", "seed": 1010739323, "treatment": "full"}, "priority": 2.009729651526826, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "01dffcb4f0f28c90c182c91c483f0ce9a0d96c03838aa52e13e6328851912256", "run_seconds": 0.821925163269043, "seed": 1010739323, "started_at": "2026-11-22T03:21:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.013423, 0.093113], "effect": 0.053268, "evaluated_at": "2026-11-22T03:22:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a28e308a5278", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.433986928, 0.488235294, 0.467320261, 0.490196078], "frozen": [0.305228758, 0.34379085, 0.344444444, 0.350980392], "full": [0.545751634, 0.520915033, 0.487581699, 0.538562092], "retrieval": [0.729411765, 0.78496732, 0.761437908, 0.779738562], "rules_nodetect": [0.545751634, 0.520915033, 0.487581699, 0.538562092]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "10c3c6374544532f31ffaae55b36177a7854407a87c594db580dec64b0c7df5b", "primary_control": "facts", "result_hash": "01dffcb4f0f28c90c182c91c483f0ce9a0d96c03838aa52e13e6328851912256", "signature": "e9d57d6d9f9e2e85b3805394b27bbebcf58771c7206a598990d89a152c8cabc0bbb9da1df91a52d994e87198dd3821da51e680d2cafc12e2a6aae2475cf9c70b", "state": "VERIFIED", "summary": {"facts": 0.469935, "frozen": 0.336111, "full": 0.523203, "retrieval": 0.763889, "rules_nodetect": 0.523203}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:22:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:19Z", "cpu_seconds": 1.285544855000012, "created_at": "2026-11-22T03:21:17Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f9cd6275d89c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "836fdae34b0bc65e252f1ab59ba6272801e8aaea6bbb64f2584d6ce4f270b12e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f9cd6275d89c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:17Z", "seed": 957367699, "treatment": "full"}, "priority": 2.0094317972360023, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "eeba1e52ee7d1f0fb1ae4c7ff52baaf92b0a348c918689e25f48b1c484c83ed4", "run_seconds": 1.5939667224884033, "seed": 957367699, "started_at": "2026-11-22T03:21:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.022285, 0.147829], "effect": 0.085057, "evaluated_at": "2026-11-22T03:22:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f9cd6275d89c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.475833333, 0.679166667, 0.8625, 0.451818182], "frozen": [0.380833333, 0.591666667, 0.420833333, 0.420909091], "full": [0.629166667, 0.75, 0.865833333, 0.564545455], "retrieval": [0.740833333, 0.804166667, 0.874166667, 0.651818182], "rules_nodetect": [0.625, 0.7625, 0.865833333, 0.694545455]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "836fdae34b0bc65e252f1ab59ba6272801e8aaea6bbb64f2584d6ce4f270b12e", "primary_control": "facts", "result_hash": "eeba1e52ee7d1f0fb1ae4c7ff52baaf92b0a348c918689e25f48b1c484c83ed4", "signature": "3177a554c7683a05a93ef92df7d0fb796f69b7734c0dad3ed232697812849ff72358bfef260d50267c284673863a192adcc8dec692a46e4b3ae40ebf931b8909", "state": "VERIFIED", "summary": {"facts": 0.61733, "frozen": 0.453561, "full": 0.702386, "retrieval": 0.767746, "rules_nodetect": 0.73697}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:22:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:40Z", "cpu_seconds": 1.6099665590000143, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-73409848ad3c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "prereg_hash": "77d75019857f88784c47b5b5fa3fedf0ea6209e85b91dad8e2e159ae489fcef5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-73409848ad3c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:21:33Z", "seed": 219703038, "treatment": "iw_temperature"}, "priority": 2.0098803540379633, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-c487ee48da85", "result_hash": "b69058e3d9f1724a619ba8cfd2fb696ce038d6b9f4855f0595fe1f2ac58c5a29", "run_seconds": 2.8626255989074707, "seed": 219703038, "started_at": "2026-11-22T03:21:37Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.010242, 0.01757], "effect": 0.003664, "evaluated_at": "2026-11-22T03:22:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-73409848ad3c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.126871098, 0.143269443, 0.130570015, 0.169570027, 0.136089595, 0.16459675, 0.17109568, 0.232896752, 0.18919254, 0.033531978, 0.10968116, 0.096698572, 0.14921902, 0.15959743, 0.163727853, 0.068538775], "raw": [0.112621605, 0.143792433, 0.12917359, 0.219518342, 0.146716282, 0.142019345, 0.15075703, 0.166498885, 0.17693469, 0.1002598, 0.141620415, 0.139610453, 0.166478733, 0.139860912, 0.183945342, 0.12037665], "temperature": [0.102936877, 0.142284695, 0.130570015, 0.226153872, 0.146716282, 0.142019345, 0.176295092, 0.166498885, 0.18919254, 0.069678405, 0.1338583, 0.109049957, 0.132445472, 0.15676204, 0.183945342, 0.09536487]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.4031, "losses": 6, "n": 16, "outcome": "INCONCLUSIVE", "p_sign_display": "0.395", "prereg_hash": "77d75019857f88784c47b5b5fa3fedf0ea6209e85b91dad8e2e159ae489fcef5", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-c487ee48da85", "result_hash": "b69058e3d9f1724a619ba8cfd2fb696ce038d6b9f4855f0595fe1f2ac58c5a29", "signature": "7026bb33265f5027a844b6886a4136bb18e1e995940eed54d38af66fef5a375b30f0f4db67c32b4e503b81aa724e7e891e5e90b8997a65d48ab3c856353c9002", "state": "VERIFIED", "summary": {"iw_temperature": 0.140322, "raw": 0.148762, "temperature": 0.143986}, "ties": 2, "treatment": "iw_temperature", "wins": 8}, "verified_at": "2026-11-22T03:22:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:54Z", "cpu_seconds": 8.122135741999998, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a74bedf3fc42", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 2, "n": 4}, "prereg_hash": "737f4d8f2d5167e972c137d1a3add20e23f60bbcfd2f597d6da4f396bff747f5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a74bedf3fc42", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:21:33Z", "seed": 1449251639, "treatment": "frontier"}, "priority": 2.009610191824788, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f5fbf27b2f4747f383e52bd0faa3b63ac414019c0edd923dbc7512ac08b748b6", "run_seconds": 8.867116212844849, "seed": 1449251639, "started_at": "2026-11-22T03:21:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005461, 0.032041], "effect": 0.01329, "evaluated_at": "2026-11-22T03:22:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a74bedf3fc42", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.758928571, 0.554022989, 0.554022989, 1.0], "frontier": [0.758928571, 0.554022989, 0.554022989, 0.991666667], "goal": [0.758928571, 0.554022989, 0.554022989, 0.966666667], "hypdisc": [0.678571429, 0.434482759, 0.464367816, 0.966666667], "infogain": [0.758928571, 0.554022989, 0.554022989, 0.991666667], "prederr": [0.758928571, 0.554022989, 0.554022989, 0.991666667], "random": [0.758928571, 0.547126437, 0.549425287, 0.95], "risk": [0.758928571, 0.554022989, 0.554022989, 0.991666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "737f4d8f2d5167e972c137d1a3add20e23f60bbcfd2f597d6da4f396bff747f5", "primary_control": "random", "result_hash": "f5fbf27b2f4747f383e52bd0faa3b63ac414019c0edd923dbc7512ac08b748b6", "signature": "93c95f6a30078149ae11183de38521e361e7fa212238797f41f60fb820a8c02ff1198be2f14c2282d975f2e60bdf072326439f69bc22a247301d89c729b42001", "state": "VERIFIED", "summary": {"count": 0.716744, "frontier": 0.71466, "goal": 0.70841, "hypdisc": 0.636022, "infogain": 0.71466, "prederr": 0.71466, "random": 0.70137, "risk": 0.71466}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:22:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:44Z", "cpu_seconds": 0.2649329059999985, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1028b7891982", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "3e6e7e8bfb512d5a079b89cde691bd253a23bfd6a46ff5e9b783c9df4d47a034", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1028b7891982", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:33Z", "seed": 559136074, "treatment": "full"}, "priority": 2.009727378418989, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9e670ba1d7f467b63493b632d26022ca9c9e54708861e7b4d6bee03f4945e04e", "run_seconds": 0.2707822322845459, "seed": 559136074, "started_at": "2026-11-22T03:21:43Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1028b7891982", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.75, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3e6e7e8bfb512d5a079b89cde691bd253a23bfd6a46ff5e9b783c9df4d47a034", "primary_control": "e141_best", "result_hash": "9e670ba1d7f467b63493b632d26022ca9c9e54708861e7b4d6bee03f4945e04e", "signature": "63f35746ad4e8d3a29853efd47ce93af782fdef6d909fed16a28b3b4ce2cd7f9b50fbd26452bd7acdbc9bf0a18445b046464baca56fb7101246b39ea7c06d904", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.9375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:49Z", "cpu_seconds": 0.3647173349999946, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-344d3cb7635b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "4ad2735dfcabbad4bd760a97c0402e7a7bd56a4ef823174ba2aa2fecff4ee530", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-344d3cb7635b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:33Z", "seed": 2025880114, "treatment": "full"}, "priority": 2.0092759042091854, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c80bc3cc7339783ab9b3de14044ed5fa6a56e5f0adf8d2747fe4fe0d3ae3efef", "run_seconds": 0.4222898483276367, "seed": 2025880114, "started_at": "2026-11-22T03:21:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.094139, 0.844139], "effect": 0.375, "evaluated_at": "2026-11-22T03:22:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-344d3cb7635b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 0.0], "fixed_policy": [1.0, 0.5, 0.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "4ad2735dfcabbad4bd760a97c0402e7a7bd56a4ef823174ba2aa2fecff4ee530", "primary_control": "e141_best", "result_hash": "c80bc3cc7339783ab9b3de14044ed5fa6a56e5f0adf8d2747fe4fe0d3ae3efef", "signature": "cdf21ec82d47beb1f7ae5ace073f3c1764c5920195f9c20193bc565b091a463bf785b396f8e771ae1a147cae364960f18b6e03613cb5fb69ebe5db5b05b39c08", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.625, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:22:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:40Z", "cpu_seconds": 2.1323803060000017, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d845e61a6f06", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 2, "n": 4}, "prereg_hash": "dce13498657c43bd150875fc5e239a711d4c8c55960daa2411801629a8216ea2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d845e61a6f06", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:33Z", "seed": 1630744472, "treatment": "full"}, "priority": 2.009944560045637, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5279ace4785ff8afb3efdc2b363325e899a99193f4fbc1afb31401d593849d92", "run_seconds": 3.0646939277648926, "seed": 1630744472, "started_at": "2026-11-22T03:21:37Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d845e61a6f06", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.5, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "dce13498657c43bd150875fc5e239a711d4c8c55960daa2411801629a8216ea2", "primary_control": "e141_best", "result_hash": "5279ace4785ff8afb3efdc2b363325e899a99193f4fbc1afb31401d593849d92", "signature": "c815d335c360bfe9161bf488be3bfc45b300a059a7ed66a0ec40c6a22477af17c7095bf833b1cc565fbb215055c27362d6ec6e024f2876f5ab83c0ebd490e00e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.125, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:04Z", "cpu_seconds": 0.6631248299999868, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f1bb93e9c833", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "e57bc8cd9c6014db6bfd1842a6ec2dcaee90b759ccffc7914c466e358e0c8c04", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f1bb93e9c833", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:33Z", "seed": 1705318653, "treatment": "full"}, "priority": 2.0091965454274057, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "eb3a656233067fc4533b707db73a9cb065a6883748963c1f72cc8a3466f47b41", "run_seconds": 0.7476632595062256, "seed": 1705318653, "started_at": "2026-11-22T03:22:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.370785, 0.245785], "effect": -0.0625, "evaluated_at": "2026-11-22T03:22:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f1bb93e9c833", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 0.0], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [1.0, 0.0, 1.0, 0.25], "no_world_model": [1.0, 0.0, 1.0, 0.25], "random": [1.0, 0.0, 0.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "e57bc8cd9c6014db6bfd1842a6ec2dcaee90b759ccffc7914c466e358e0c8c04", "primary_control": "e141_best", "result_hash": "eb3a656233067fc4533b707db73a9cb065a6883748963c1f72cc8a3466f47b41", "signature": "f6e804d129d04fe834c2f3ee73eea8da0fe252ddedca3c97067be365b4c956ed32a3545348e0971ec5ea633b686e7df9820644b4110d7ee3e8ff9d4de9d54905", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.0625, "full": 0.5625, "no_world_model": 0.5625, "random": 0.3125}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:38Z", "cpu_seconds": 0.19142135599999932, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4cc9a5cb51e6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "6f49dcf3c2775b369f58c7bccaee610ff3aed256d364ec4644dde5fcc267356c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4cc9a5cb51e6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:33Z", "seed": 1759026654, "treatment": "full"}, "priority": 2.0097281005964915, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "12133a1839d626047b13b50e6d42ba569f44f8bc1257859f660bb787db24ac66", "run_seconds": 0.41877317428588867, "seed": 1759026654, "started_at": "2026-11-22T03:21:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003636, 0.011212], "effect": 0.003788, "evaluated_at": "2026-11-22T03:22:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4cc9a5cb51e6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.954545455, 0.712121212, 0.96969697, 1.0], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [0.954545455, 0.712121212, 0.984848485, 1.0], "retrieval": [0.939393939, 0.696969697, 0.984848485, 1.0], "rules_nodetect": [0.954545455, 0.712121212, 0.984848485, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6f49dcf3c2775b369f58c7bccaee610ff3aed256d364ec4644dde5fcc267356c", "primary_control": "facts", "result_hash": "12133a1839d626047b13b50e6d42ba569f44f8bc1257859f660bb787db24ac66", "signature": "cb34030d19977d47ccc3441f83f42a8933c70a288356b2fbaac1cfd91216df50d607aec62420f055f8b921b6bc72b61a46c03f0130f394f6298918003a06180f", "state": "VERIFIED", "summary": {"facts": 0.909091, "frozen": 0.030303, "full": 0.912879, "retrieval": 0.905303, "rules_nodetect": 0.912879}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:46Z", "cpu_seconds": 0.52197181599999, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f7f631a44ef4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "35bc6b23812afaca78fe586155bf19439907bd12204fff301d8e77e5258766fa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f7f631a44ef4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:33Z", "seed": 1909420049, "treatment": "full"}, "priority": 2.0093462843801855, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "27a07d99c2534b688a082926c00daa1835c0a566596b83047b0cb3f48743b5c9", "run_seconds": 0.5581154823303223, "seed": 1909420049, "started_at": "2026-11-22T03:21:45Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f7f631a44ef4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.916666667, 0.916666667, 0.80952381, 0.838095238], "frozen": [0.5, 0.5, 0.685714286, 0.666666667], "full": [0.916666667, 0.916666667, 0.80952381, 0.838095238], "retrieval": [0.916666667, 0.916666667, 0.80952381, 0.838095238], "rules_nodetect": [0.916666667, 0.916666667, 0.80952381, 0.838095238]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "35bc6b23812afaca78fe586155bf19439907bd12204fff301d8e77e5258766fa", "primary_control": "facts", "result_hash": "27a07d99c2534b688a082926c00daa1835c0a566596b83047b0cb3f48743b5c9", "signature": "75bd06ba462f84839fa5c1216e6c29fe92edfdfe942a909d1af3f9e177c80ec8f5b6347137d181208d064f66c63d413ef7262ea20d83fb26233cc73d5309820d", "state": "PARTIAL", "summary": {"facts": 0.870238, "frozen": 0.588095, "full": 0.870238, "retrieval": 0.870238, "rules_nodetect": 0.870238}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:22Z", "cpu_seconds": 0.9843380530000019, "created_at": "2026-11-22T03:21:33Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-132e2ffc7ec5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "prereg_hash": "bede457c43c49c2b2a610a91bda6da135552ce0f92d5b24becc2427efa5a89f8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-132e2ffc7ec5", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:21:33Z", "seed": 1082616011, "treatment": "gated"}, "priority": 2.0091589257824416, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "e27b0e2d3f25baaab339a0cc07f717bcf0460ea2277c9ec1f671a6600ef4ea1c", "run_seconds": 2.1887261867523193, "seed": 1082616011, "started_at": "2026-11-22T03:22:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005, 0.015417], "effect": 0.005208, "evaluated_at": "2026-11-22T03:22:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-132e2ffc7ec5", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.631578947, 0.671875, 0.476190476, 0.833333333], "gated": [0.631578947, 0.671875, 0.476190476, 0.833333333], "naive": [0.5, 0.609375, 0.476190476, 0.583333333], "scratch": [0.631578947, 0.671875, 0.476190476, 0.8125], "ungated": [0.631578947, 0.609375, 0.476190476, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "bede457c43c49c2b2a610a91bda6da135552ce0f92d5b24becc2427efa5a89f8", "primary_control": "scratch", "result_hash": "e27b0e2d3f25baaab339a0cc07f717bcf0460ea2277c9ec1f671a6600ef4ea1c", "signature": "e56f0547468e68d03d3192e331746b703c1332afe93d4ce96c70f325def6201419439c654ba990f543486d991c6f437e77dcd665eafcedc8d944ad82dc53fc00", "state": "VERIFIED", "summary": {"compressed": 0.653244, "gated": 0.653244, "naive": 0.542225, "scratch": 0.648036, "ungated": 0.616786}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:22:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:00Z", "cpu_seconds": 1.003389091999992, "created_at": "2026-11-22T03:21:50Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ee962a66cbd7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "5d35c9bade4e1a304fb6f4f7dbeb6d3dfae975beeaebfba067c58631aea809ea", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ee962a66cbd7", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:21:50Z", "seed": 857996908, "treatment": "frontier"}, "priority": 2.0093642940893908, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4b23063ff9dc5fed18f53be698491b35a9e9bcd00a8cfaa09268c1c76fe1e532", "run_seconds": 1.0566051006317139, "seed": 857996908, "started_at": "2026-11-22T03:21:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035898, 0.040949], "effect": 0.002525, "evaluated_at": "2026-11-22T03:22:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ee962a66cbd7", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.698412698, 0.828282828, 0.952380952, 0.882352941], "frontier": [0.753968254, 0.545454545, 0.904761905, 0.882352941], "goal": [0.698412698, 0.494949495, 0.952380952, 0.882352941], "hypdisc": [0.69047619, 0.494949495, 0.904761905, 0.882352941], "infogain": [0.666666667, 0.545454545, 0.904761905, 0.882352941], "prederr": [0.69047619, 0.545454545, 0.904761905, 0.882352941], "random": [0.706349206, 0.535353535, 0.952380952, 0.882352941], "risk": [0.753968254, 0.545454545, 0.904761905, 0.882352941]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "5d35c9bade4e1a304fb6f4f7dbeb6d3dfae975beeaebfba067c58631aea809ea", "primary_control": "random", "result_hash": "4b23063ff9dc5fed18f53be698491b35a9e9bcd00a8cfaa09268c1c76fe1e532", "signature": "89c8f3f41bfce23a437450ade8a01cf291aba3fb267aca5039b54647e0639a57a6859df2b27218b9e87bdec626837bc258a38ef77fbc0f91e3907248fc122d03", "state": "VERIFIED", "summary": {"count": 0.840357, "frontier": 0.771634, "goal": 0.757024, "hypdisc": 0.743135, "infogain": 0.749809, "prederr": 0.755761, "random": 0.769109, "risk": 0.771634}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:22:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:07Z", "cpu_seconds": 2.8200008729999997, "created_at": "2026-11-22T03:21:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1c79b3fca1f4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 2, "n": 4}, "prereg_hash": "afdd3db42f6eb1a8373ab2696b08b12d06a55044e592c3518e0d6d9914d32a84", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1c79b3fca1f4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:50Z", "seed": 876546399, "treatment": "full"}, "priority": 2.0075890725476864, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "324341d7b48aef860b4334f4e8f53b600b31915ad3d5cb5380e1bf59d49cc88f", "run_seconds": 3.208653688430786, "seed": 876546399, "started_at": "2026-11-22T03:35:03Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:35:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1c79b3fca1f4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 0.0, 0.0], "fixed_policy": [0.0, 0.25, 0.25, 0.0], "full": [0.0, 1.0, 0.0, 0.0], "no_world_model": [0.0, 1.0, 0.0, 0.5], "random": [0.0, 1.0, 0.5, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "afdd3db42f6eb1a8373ab2696b08b12d06a55044e592c3518e0d6d9914d32a84", "primary_control": "e141_best", "result_hash": "324341d7b48aef860b4334f4e8f53b600b31915ad3d5cb5380e1bf59d49cc88f", "signature": "76c64f9c0a7b52f51b6c7de5eff9e7522e7cda854e715aa6871d7d5e8a1bb4ec7debf1a6e816b5f21303643a59746d5bbe6faea5f3b4278a7eee2563056f2300", "state": "PARTIAL", "summary": {"e141_best": 0.25, "fixed_policy": 0.125, "full": 0.25, "no_world_model": 0.375, "random": 0.375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:37:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:52Z", "cpu_seconds": 0.6074052419999987, "created_at": "2026-11-22T03:21:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-81336ff909ba", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "ad9d0dd60e86acf0b69605679ae1538dde1d518e43843bc3c0c3c091c8a5ae5f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-81336ff909ba", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:50Z", "seed": 482209521, "treatment": "full"}, "priority": 2.0099840121218984, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9fd102132522f2bd9ac5727185a0c305b1ded31f576acc6a8ff0fb846a80af89", "run_seconds": 0.6684038639068604, "seed": 482209521, "started_at": "2026-11-22T03:21:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:22:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-81336ff909ba", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.25, 1.0], "fixed_policy": [0.0, 0.75, 0.25, 0.0], "full": [0.0, 0.0, 0.0, 1.0], "no_world_model": [0.0, 0.0, 0.0, 1.0], "random": [0.0, 0.0, 0.25, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "ad9d0dd60e86acf0b69605679ae1538dde1d518e43843bc3c0c3c091c8a5ae5f", "primary_control": "e141_best", "result_hash": "9fd102132522f2bd9ac5727185a0c305b1ded31f576acc6a8ff0fb846a80af89", "signature": "e7aacea4624ac42b1fa69d23b8000a556659fd61cba625187fe99c0264e5a9bb3ae388a9372f1bb58be91350076d2d8f076d7658d02a606a9c7a371abf500d0b", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.25, "full": 0.25, "no_world_model": 0.25, "random": 0.3125}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:22:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:58Z", "cpu_seconds": 0.44050691199998937, "created_at": "2026-11-22T03:21:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d177f619d161", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "096b08286c964ec811f6153ebc53e8eab76e3537eeca50c79c1cd9899289d670", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d177f619d161", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:50Z", "seed": 2090949392, "treatment": "full"}, "priority": 2.0096431082464403, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e2b7ca984febf09040fe9c37e2ccd24e712211e9a46ef50181b89b0ac9a1e435", "run_seconds": 1.1293742656707764, "seed": 2090949392, "started_at": "2026-11-22T03:21:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.403518, 1.096482], "effect": 0.75, "evaluated_at": "2026-11-22T03:22:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d177f619d161", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.0, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.5, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "096b08286c964ec811f6153ebc53e8eab76e3537eeca50c79c1cd9899289d670", "primary_control": "e141_best", "result_hash": "e2b7ca984febf09040fe9c37e2ccd24e712211e9a46ef50181b89b0ac9a1e435", "signature": "1d065fefd1d226d09bcd8fc333838956927baf1ea61aafdd6fc18e382af45dea81f5b0fab52189b734ea7c265bb495894cf360d25d8919966eb86821ec1fec04", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 0.875, "random": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:22:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:07Z", "cpu_seconds": 1.5182233840000094, "created_at": "2026-11-22T03:21:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e0b5a36652a5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "89aaf829e1c2b3ad32b52fab295d2509b791f36bc347cb19e60b9777155e6a39", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e0b5a36652a5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:21:50Z", "seed": 854017174, "treatment": "full"}, "priority": 2.00917527695404, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "abda2e523846c32c9faaa564a5e16450c47ef4975a9d03a4fb2aad1c1fcbc2d1", "run_seconds": 1.5457346439361572, "seed": 854017174, "started_at": "2026-11-22T03:22:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.052693, 0.947307], "effect": 0.5, "evaluated_at": "2026-11-22T03:22:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e0b5a36652a5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.75, 0.25], "fixed_policy": [0.0, 0.0, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 0.75, 0.5], "random": [1.0, 0.0, 0.5, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "89aaf829e1c2b3ad32b52fab295d2509b791f36bc347cb19e60b9777155e6a39", "primary_control": "e141_best", "result_hash": "abda2e523846c32c9faaa564a5e16450c47ef4975a9d03a4fb2aad1c1fcbc2d1", "signature": "be406c9ab56a2d305e8a78fd90e85f0ffba0a6a4de2bb431e44ea4d2ed2f3d097ff2e0f310a303d2a08f635f894692bd1b3cdcbe2d34263c1d023a75b0b2b700", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 0.5625, "random": 0.5625}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:22:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:02Z", "cpu_seconds": 0.7814185479999978, "created_at": "2026-11-22T03:21:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2fedb5bedd4e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "b83f0efc234fdf06235116177294462dcd7df9f258373020bf0c06fb6b0a2f60", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2fedb5bedd4e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:21:50Z", "seed": 1184205491, "treatment": "full"}, "priority": 2.009230361863804, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "30b305bf13c8d9640a0aaea9dc7853d43682f745dc3aa5f559cb26d759f9dee5", "run_seconds": 0.833733320236206, "seed": 1184205491, "started_at": "2026-11-22T03:22:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023994, 0.11608], "effect": 0.046043, "evaluated_at": "2026-11-22T03:22:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2fedb5bedd4e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.3625, 0.293333333, 0.35942029, 0.229411765], "frozen": [0.23, 0.224, 0.220289855, 0.035294118], "full": [0.3775, 0.306666667, 0.362318841, 0.382352941], "retrieval": [0.3325, 0.282666667, 0.295652174, 0.411764706], "rules_nodetect": [0.4, 0.306666667, 0.362318841, 0.405882353]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "b83f0efc234fdf06235116177294462dcd7df9f258373020bf0c06fb6b0a2f60", "primary_control": "facts", "result_hash": "30b305bf13c8d9640a0aaea9dc7853d43682f745dc3aa5f559cb26d759f9dee5", "signature": "925a189782d2d688f619d0cffd4cf0b4258e5c4c8aeda8939a21279571cee11285d90be221bc6b5276a7ceac1234f1408007f1b899908d1d6e90e23d2a6f3b06", "state": "VERIFIED", "summary": {"facts": 0.311166, "frozen": 0.177396, "full": 0.35721, "retrieval": 0.330646, "rules_nodetect": 0.368717}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:22:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:21:56Z", "cpu_seconds": 1.2613986909999966, "created_at": "2026-11-22T03:21:50Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-40ab75ea05e6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "6d5ebde8af3c7363017c1e5ba544ee6d9e282e094cc931701393f9aa8e4eb670", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-40ab75ea05e6", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:21:50Z", "seed": 1272158833, "treatment": "gated"}, "priority": 2.0096773910767065, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "369cf41408600b4fca229aae5ec1b6dbf0b0cf6f088a3aa809782efb44eb1a16", "run_seconds": 1.4748072624206543, "seed": 1272158833, "started_at": "2026-11-22T03:21:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.012669, 0.01859], "effect": 0.002961, "evaluated_at": "2026-11-22T03:22:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-40ab75ea05e6", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.477272727, 0.543103448, 0.4625, 0.473684211], "gated": [0.477272727, 0.543103448, 0.4625, 0.473684211], "naive": [0.75, 0.784482759, 0.6, 0.631578947], "scratch": [0.477272727, 0.543103448, 0.4375, 0.486842105], "ungated": [0.602272727, 0.784482759, 0.6375, 0.631578947]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "6d5ebde8af3c7363017c1e5ba544ee6d9e282e094cc931701393f9aa8e4eb670", "primary_control": "scratch", "result_hash": "369cf41408600b4fca229aae5ec1b6dbf0b0cf6f088a3aa809782efb44eb1a16", "signature": "0f1e5a0e7befaf839014883bc4f5694911abb121c5243db2daf0a4a4c00b9fc75a92d53b0ea87ce0b5952899c75d3782c253fb022923c0ad638a33eae6cc270c", "state": "VERIFIED", "summary": {"compressed": 0.48914, "gated": 0.48914, "naive": 0.691515, "scratch": 0.48618, "ungated": 0.663959}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:22:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:08Z", "cpu_seconds": 0.9791930900000025, "created_at": "2026-11-22T03:22:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0be5c77c883a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "439fc9906fa241c11d0cc34eb289a9d4d06227bb7a5287deb0ebc405179b5acc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0be5c77c883a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:06Z", "seed": 1986341079, "treatment": "frontier"}, "priority": 2.009873060395358, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "67174e5b9a4fc67c9683426d0dff919de3fd8171f62fe5db469bd3a8421a3348", "run_seconds": 1.0540471076965332, "seed": 1986341079, "started_at": "2026-11-22T03:22:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.143643, 0.028174], "effect": -0.057734, "evaluated_at": "2026-11-22T03:22:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0be5c77c883a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.986928105, 0.944444444, 0.96031746, 0.747126437], "frontier": [0.941176471, 0.759259259, 0.96031746, 0.735632184], "goal": [0.941176471, 0.944444444, 0.96031746, 0.735632184], "hypdisc": [0.986928105, 0.944444444, 0.96031746, 0.735632184], "infogain": [0.934640523, 0.944444444, 0.96031746, 0.735632184], "prederr": [0.934640523, 0.944444444, 0.96031746, 0.735632184], "random": [0.986928105, 0.944444444, 0.96031746, 0.735632184], "risk": [0.941176471, 0.759259259, 0.96031746, 0.735632184]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "439fc9906fa241c11d0cc34eb289a9d4d06227bb7a5287deb0ebc405179b5acc", "primary_control": "random", "result_hash": "67174e5b9a4fc67c9683426d0dff919de3fd8171f62fe5db469bd3a8421a3348", "signature": "25513f493294f273cc399d03bc9bf7493f9ab4600f79fa0d632d62d539fddac44480b3b891074e5f47c1f41e5d0d6c3dcae659e26bc3e52e3955aec87f137e03", "state": "VERIFIED", "summary": {"count": 0.909704, "frontier": 0.849096, "goal": 0.895393, "hypdisc": 0.906831, "infogain": 0.893759, "prederr": 0.893759, "random": 0.906831, "risk": 0.849096}, "ties": 2, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:22:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:15Z", "cpu_seconds": 1.6280698679999972, "created_at": "2026-11-22T03:22:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9e08fa011043", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "aec3b63adb4eedd7e4f34a3d4b71c9945ca9e5b12e864484d6274998cef8ae0e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9e08fa011043", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:06Z", "seed": 766562943, "treatment": "frontier"}, "priority": 2.0094174792660415, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ebcb719261234a93217cdd3a38ba21a7cbd67d874fdc139bdbf122511c61a87d", "run_seconds": 2.0835440158843994, "seed": 766562943, "started_at": "2026-11-22T03:22:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.009605, 0.150394], "effect": 0.070395, "evaluated_at": "2026-11-22T03:22:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9e08fa011043", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.958333333, 0.605263158, 0.727699531, 0.625], "frontier": [0.958333333, 0.710526316, 0.769953052, 0.875], "goal": [0.958333333, 0.736842105, 0.727699531, 0.9], "hypdisc": [0.958333333, 0.697368421, 0.708920188, 0.85], "infogain": [0.958333333, 0.710526316, 0.727699531, 0.875], "prederr": [0.958333333, 0.710526316, 0.516431925, 0.6], "random": [0.958333333, 0.578947368, 0.769953052, 0.725], "risk": [0.958333333, 0.710526316, 0.769953052, 0.875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "aec3b63adb4eedd7e4f34a3d4b71c9945ca9e5b12e864484d6274998cef8ae0e", "primary_control": "random", "result_hash": "ebcb719261234a93217cdd3a38ba21a7cbd67d874fdc139bdbf122511c61a87d", "signature": "2195cce62b5ade0d6627ed9ffd23e398242627e6f5ac577992ecf16dde04866ef6f2cc3c4e68eeb23660ff549a4665a3035a162322a174aff2f6cff810475d0d", "state": "VERIFIED", "summary": {"count": 0.729074, "frontier": 0.828453, "goal": 0.830719, "hypdisc": 0.803655, "infogain": 0.81789, "prederr": 0.696323, "random": 0.758058, "risk": 0.828453}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:22:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:25Z", "cpu_seconds": 3.815513323999994, "created_at": "2026-11-22T03:22:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a94121b0578a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 4, "n": 4}, "prereg_hash": "23d2768fb0e7356c34a848388ce4c79e141e8394aae050f1e7f3183a65c305df", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a94121b0578a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:06Z", "seed": 689349221, "treatment": "frontier"}, "priority": 2.0094046822633462, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f0d54dcea53e3521ede72acc99668fb044cc279eba50f17b189355e23637df95", "run_seconds": 5.854892015457153, "seed": 689349221, "started_at": "2026-11-22T03:22:19Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:22:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a94121b0578a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "23d2768fb0e7356c34a848388ce4c79e141e8394aae050f1e7f3183a65c305df", "primary_control": "random", "result_hash": "f0d54dcea53e3521ede72acc99668fb044cc279eba50f17b189355e23637df95", "signature": "b7fb86b5f6c8edaa652703d4417074ffd331dabcf8e5f300421e01402e645aa79d21d6d87254f66f1b558184b9378547b3e9f064dbecfbba1bd29dc7cacf5c0d", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:23:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:37Z", "cpu_seconds": 1.9183581669999938, "created_at": "2026-11-22T03:22:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-cc1b837bb709", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "e9309e581049868a50a4b50cf54982798cfbfa2312dd4e875c49f284c9a524d6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-cc1b837bb709", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:06Z", "seed": 1283324480, "treatment": "frontier"}, "priority": 2.008722155516956, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "164fb5dbf3e8be0de8a87d5feeae21504a1a038c7c2df18a40a3a33bbe0f755e", "run_seconds": 2.0705690383911133, "seed": 1283324480, "started_at": "2026-11-22T03:22:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.032194, 0.165663], "effect": 0.066735, "evaluated_at": "2026-11-22T03:23:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-cc1b837bb709", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.457364341, 0.595744681, 0.503267974, 0.22875817], "frontier": [0.317829457, 0.617021277, 0.666666667, 0.594771242], "goal": [0.310077519, 0.404255319, 0.673202614, 0.522875817], "hypdisc": [0.403100775, 0.510638298, 0.607843137, 0.529411765], "infogain": [0.333333333, 0.617021277, 0.673202614, 0.62745098], "prederr": [0.310077519, 0.184397163, 0.45751634, 0.431372549], "random": [0.395348837, 0.475177305, 0.535947712, 0.522875817], "risk": [0.317829457, 0.617021277, 0.666666667, 0.594771242]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "e9309e581049868a50a4b50cf54982798cfbfa2312dd4e875c49f284c9a524d6", "primary_control": "random", "result_hash": "164fb5dbf3e8be0de8a87d5feeae21504a1a038c7c2df18a40a3a33bbe0f755e", "signature": "d0e0139f9bea4eb3f0ed83a36f1eb1c2602ce9f116a3dabebec93070dc463b078003d5d28cdb67ace516bb7e956b8ee6edd711a81e577d8210982829f39fb105", "state": "VERIFIED", "summary": {"count": 0.446284, "frontier": 0.549072, "goal": 0.477603, "hypdisc": 0.512748, "infogain": 0.562752, "prederr": 0.345841, "random": 0.482337, "risk": 0.549072}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:23:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:41Z", "cpu_seconds": 7.581754842999999, "created_at": "2026-11-22T03:22:06Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e02b7f210fc8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "8a8c3e7a3b8c7237ab8b735c4718c53cfde8618c62f0c6b91ed29235ba549de2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e02b7f210fc8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:06Z", "seed": 1181165521, "treatment": "frontier"}, "priority": 2.009128110095208, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1641fa625f77359c87dd273cca872231849922177a9018832f96a93894755f51", "run_seconds": 8.80677580833435, "seed": 1181165521, "started_at": "2026-11-22T03:22:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.129248, 0.00085], "effect": -0.064199, "evaluated_at": "2026-11-22T03:23:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e02b7f210fc8", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.746153846, 0.646666667, 0.686666667, 0.726666667], "frontier": [0.678846154, 0.61, 0.526666667, 0.696666667], "goal": [0.571153846, 0.613333333, 0.488333333, 0.696666667], "hypdisc": [0.705769231, 0.63, 0.508333333, 0.726666667], "infogain": [0.584615385, 0.611666667, 0.563333333, 0.696666667], "prederr": [0.620192308, 0.69, 0.608333333, 0.696666667], "random": [0.692307692, 0.666666667, 0.686666667, 0.723333333], "risk": [0.678846154, 0.61, 0.526666667, 0.696666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "8a8c3e7a3b8c7237ab8b735c4718c53cfde8618c62f0c6b91ed29235ba549de2", "primary_control": "random", "result_hash": "1641fa625f77359c87dd273cca872231849922177a9018832f96a93894755f51", "signature": "43b2f43a63249d679961eec2825feca8b14fe9e428cfc37184b0d0aa4452581abd2b6e4c09ac08d4d682c8534cd3030ac1a1fdf44685fbb819fc3d1ddcd88c0c", "state": "VERIFIED", "summary": {"count": 0.701538, "frontier": 0.628045, "goal": 0.592372, "hypdisc": 0.642692, "infogain": 0.614071, "prederr": 0.653798, "random": 0.692244, "risk": 0.628045}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:23:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:14Z", "cpu_seconds": 1.7568521990000079, "created_at": "2026-11-22T03:22:06Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8a28a851f449", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 8, "n": 4}, "prereg_hash": "2eab022db9828dc677b69a94316bb4b1a9cf2577ffd73beb912008070d0c7f1c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8a28a851f449", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:22:06Z", "seed": 1008026129, "treatment": "full"}, "priority": 2.0095071180519493, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "92cd268f8b1675ef5b4ade3361af75ba362c873ccd656b1bd9e1d28b0849aa3c", "run_seconds": 1.9235844612121582, "seed": 1008026129, "started_at": "2026-11-22T03:22:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:22:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8a28a851f449", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.5, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 0.5], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "2eab022db9828dc677b69a94316bb4b1a9cf2577ffd73beb912008070d0c7f1c", "primary_control": "e141_best", "result_hash": "92cd268f8b1675ef5b4ade3361af75ba362c873ccd656b1bd9e1d28b0849aa3c", "signature": "333cf1b80714d5d7bf6902a7da189aa08f3ccf42900e23d7441678277ec48c44572a616dd5d9d945a69bf641697daa61d9a268da3d948b4f9098afbc9d25fe01", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.125, "full": 1.0, "no_world_model": 0.875, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:22:59Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:28:38Z", "cpu_seconds": 1.9332256019999998, "created_at": "2026-11-22T03:22:06Z", "error": "lease expired; ", "family": "ow_model", "fence": 2, "id": "E141-ow_model-3ec8ee6c989a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 6, "n": 4}, "prereg_hash": "9b6efd41370352bfa55a1ca1a57bac98f616c464e092f5b0e1d13f06806042e5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3ec8ee6c989a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:22:06Z", "seed": 1926528421, "treatment": "full"}, "priority": 2.008556239291361, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7cab8dc4851c6b5147d8434aa1d2d03aef5e7aca9cd47718d3b7dc9b7c0b8a61", "run_seconds": 1.9775586128234863, "seed": 1926528421, "started_at": "2026-11-22T03:25:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.047799, 0.173799], "effect": 0.063, "evaluated_at": "2026-11-22T03:28:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3ec8ee6c989a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.328, 0.98, 0.516666667, 0.4], "frozen": [0.12, 0.0, 0.0, 0.0], "full": [0.56, 0.98, 0.536666667, 0.4], "retrieval": [0.56, 0.98, 0.833333333, 0.4], "rules_nodetect": [0.197333333, 0.12, 0.316666667, 0.4]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "9b6efd41370352bfa55a1ca1a57bac98f616c464e092f5b0e1d13f06806042e5", "primary_control": "facts", "result_hash": "7cab8dc4851c6b5147d8434aa1d2d03aef5e7aca9cd47718d3b7dc9b7c0b8a61", "signature": "07e12c64bf31e566f323a9de7e738d043372f704feecabc1f8ccd94240f5b08d20a8b95b518ec052550f7811bd6f7e6d18b98cd8700b7a81954ecd961ed21a05", "state": "VERIFIED", "summary": {"facts": 0.556167, "frozen": 0.03, "full": 0.619167, "retrieval": 0.693333, "rules_nodetect": 0.2585}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:28:40Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:22:06Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_rsi", "fence": 3, "id": "E141-ow_rsi-0a8bcbd37154", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "95449c5d6cde732c61772d5de9278b39aec93855bdbb4d517cedd0d29113392c", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-0a8bcbd37154", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:22:06Z", "seed": 317390713, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 317390713, "started_at": "2026-11-22T03:22:09Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 2, "completed_at": "2026-11-22T03:28:39Z", "cpu_seconds": 0.8373173769999998, "created_at": "2026-11-22T03:22:23Z", "error": "lease expired; ", "family": "ow_goal", "fence": 2, "id": "E141-ow_goal-95dbb50e0e41", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "62a5999464f20aeb02c5b4cb2098f8a1bfd1d7b8c2d1fd81b5500c55cbba2c49", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-95dbb50e0e41", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:22:23Z", "seed": 1881179542, "treatment": "full"}, "priority": 2.0083613943196363, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f05c6ecd2fa50926477f025c32b2c38cc1a431893a58844dfd87c4d93aa00cc2", "run_seconds": 0.855724573135376, "seed": 1881179542, "started_at": "2026-11-22T03:25:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T03:28:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-95dbb50e0e41", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [0.75, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "62a5999464f20aeb02c5b4cb2098f8a1bfd1d7b8c2d1fd81b5500c55cbba2c49", "primary_control": "e141_best", "result_hash": "f05c6ecd2fa50926477f025c32b2c38cc1a431893a58844dfd87c4d93aa00cc2", "signature": "05c98dc53e1ed55342eab70ac4041944eb2192f0c225f4660dcde76e27e0c22aba488a1b37a411da16facb9dea708d70a4af01393076fa9a03b39fcf04cd3f00", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.9375, "random": 0.6875}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:28:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:00Z", "cpu_seconds": 2.305591305, "created_at": "2026-11-22T03:22:23Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b4864334b044", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "db91a8ca4cba7f5b6c8de813e9e70271d7445328edd5881cf188d212ad17b206", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b4864334b044", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:22:23Z", "seed": 557074078, "treatment": "full"}, "priority": 2.0086159797328125, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "cec8c28024f62e5755d1a64b6f7f15d4fa8a9d066fe4c4c385d37832be08befc", "run_seconds": 4.168580055236816, "seed": 557074078, "started_at": "2026-11-22T03:22:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.299958, 0.700042], "effect": 0.5, "evaluated_at": "2026-11-22T03:23:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b4864334b044", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 0.75, 0.25], "fixed_policy": [1.0, 0.75, 1.0, 0.75], "full": [0.75, 0.75, 1.0, 0.75], "no_world_model": [0.25, 0.5, 0.25, 0.25], "random": [0.5, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "db91a8ca4cba7f5b6c8de813e9e70271d7445328edd5881cf188d212ad17b206", "primary_control": "e141_best", "result_hash": "cec8c28024f62e5755d1a64b6f7f15d4fa8a9d066fe4c4c385d37832be08befc", "signature": "b1361733b8ffc57ac2055d790a1e00f84e0b830d28984a09045028be63dae1772ff5d1755ac1992054e7ee752c04b16de510324b9fd60eaae8a2b867b047260e", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.875, "full": 0.8125, "no_world_model": 0.3125, "random": 0.8125}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:23:19Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:22:23Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_goal", "fence": 3, "id": "E141-ow_goal-d3ec04041ee0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 7, "n": 4}, "prereg_hash": "bb41130dd1b8f2eff0371d15bf474acb459f0ba4a3c913b99d54bec4ef05e7bf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d3ec04041ee0", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:22:23Z", "seed": 23368799, "treatment": "full"}, "priority": 2.0082911834761683, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 23368799, "started_at": "2026-11-22T03:26:15Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:22:30Z", "cpu_seconds": 0.699411985999987, "created_at": "2026-11-22T03:22:23Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-dd5f03520c17", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "381f5c7d0a998c02c92b1774426ef5a598180442a7319b42497775c9becd956a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-dd5f03520c17", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:22:23Z", "seed": 672723282, "treatment": "full"}, "priority": 2.0096725679375784, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f85ddcd96c80fea87f2d64256826fc45e2828dc5655148ab5a4dd86dbc513853", "run_seconds": 0.7236335277557373, "seed": 672723282, "started_at": "2026-11-22T03:22:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T03:23:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-dd5f03520c17", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 0.75, 1.0], "fixed_policy": [1.0, 1.0, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 0.0, 0.75], "random": [1.0, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "381f5c7d0a998c02c92b1774426ef5a598180442a7319b42497775c9becd956a", "primary_control": "e141_best", "result_hash": "f85ddcd96c80fea87f2d64256826fc45e2828dc5655148ab5a4dd86dbc513853", "signature": "c037a87f6a0d077f5e01a3d6fa79767f9332b5eb0962289c375b3d413ff3c730e7ed9f45af160c23ef9b548fe043628ccbe0853cd0f0587eb05a68f981d13a0b", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.5625, "full": 1.0, "no_world_model": 0.5625, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:23:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:28Z", "cpu_seconds": 0.4068383220000058, "created_at": "2026-11-22T03:22:23Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-330c50dd95a3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "a9fe284b9a3864c4e9f87483f86485612f499ab6ede1de93c82e756265358173", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-330c50dd95a3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:22:23Z", "seed": 1104576142, "treatment": "full"}, "priority": 2.00976646387393, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5ef858d2d0e73e728ccd825585679b2dbab46b648a761358f72b2003d586e42c", "run_seconds": 0.6408512592315674, "seed": 1104576142, "started_at": "2026-11-22T03:22:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.047825, 0.204258], "effect": 0.126042, "evaluated_at": "2026-11-22T03:23:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-330c50dd95a3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.783333333, 0.770833333, 0.8, 0.883333333], "frozen": [0.2, 0.25, 0.133333333, 0.0], "full": [0.966666667, 0.958333333, 0.816666667, 1.0], "retrieval": [0.9, 0.9375, 0.8, 0.95], "rules_nodetect": [0.966666667, 0.958333333, 0.433333333, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a9fe284b9a3864c4e9f87483f86485612f499ab6ede1de93c82e756265358173", "primary_control": "facts", "result_hash": "5ef858d2d0e73e728ccd825585679b2dbab46b648a761358f72b2003d586e42c", "signature": "9018330f4e3a927df7223d9e3722abaaf398f0da70f48b102c19e67a4bc8c95c8840d2e110fecdd962b4d7145bbd244c4604da1ff6de89420dca58d00d48ec06", "state": "VERIFIED", "summary": {"facts": 0.809375, "frozen": 0.145833, "full": 0.935417, "retrieval": 0.896875, "rules_nodetect": 0.839583}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:23:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:35Z", "cpu_seconds": 7.428701138000001, "created_at": "2026-11-22T03:22:23Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-6a3da5ff9d19", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "451e4b548596e3145d0ede3311a322bf0de17cb83ba795ac9fac26e7212eb574", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-6a3da5ff9d19", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:22:23Z", "seed": 716081119, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "c904e5bc50080f761919be01f04d5413a408586089f9b7d64a75389d0cc6551d", "run_seconds": 9.622618198394775, "seed": 716081119, "started_at": "2026-11-22T03:22:25Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:23:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-6a3da5ff9d19", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "451e4b548596e3145d0ede3311a322bf0de17cb83ba795ac9fac26e7212eb574", "primary_control": "reset", "result_hash": "c904e5bc50080f761919be01f04d5413a408586089f9b7d64a75389d0cc6551d", "signature": "679f78098063e41d90fed7b6080b8ee1bf8bc147dad21ce122705b0152135749339d3a8795c786b3c223159b3b0539a158bb3f019082f3c2381eab7a09e50a0c", "state": "PARTIAL", "summary": {"inherit": 0.7, "reset": 0.7, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:23:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:51Z", "cpu_seconds": 4.018717416000001, "created_at": "2026-11-22T03:22:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-2daade321a03", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "7e71ef4136a32dd1d925b50e8a0fc65b55aecb181b0bb18f16046ddb652f18bc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2daade321a03", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:39Z", "seed": 386867578, "treatment": "frontier"}, "priority": 2.009188896558737, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e1d14a116730c14d0f505e75da7dff9dc8c57f4c7ec57ef63d764c1d5cb6fe30", "run_seconds": 4.625902414321899, "seed": 386867578, "started_at": "2026-11-22T03:22:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.113557, 0.116246], "effect": 0.001345, "evaluated_at": "2026-11-22T03:23:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-2daade321a03", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.794642857, 0.882352941, 0.94], "frontier": [1.0, 0.866071429, 0.602941176, 0.93], "goal": [1.0, 0.875, 0.779411765, 0.93], "hypdisc": [1.0, 0.848214286, 0.764705882, 0.84], "infogain": [1.0, 0.866071429, 0.602941176, 0.93], "prederr": [1.0, 0.866071429, 0.602941176, 0.93], "random": [1.0, 0.758928571, 0.764705882, 0.87], "risk": [1.0, 0.866071429, 0.602941176, 0.93]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7e71ef4136a32dd1d925b50e8a0fc65b55aecb181b0bb18f16046ddb652f18bc", "primary_control": "random", "result_hash": "e1d14a116730c14d0f505e75da7dff9dc8c57f4c7ec57ef63d764c1d5cb6fe30", "signature": "af36780f010be988fc30a80f428c876b60988dcbf664327b01b51ecf5ce58be7d3ef36fad4fc3aa2b2cde04414384d354bb9c82664c0f627128e647e460a3003", "state": "VERIFIED", "summary": {"count": 0.904249, "frontier": 0.849753, "goal": 0.896103, "hypdisc": 0.86323, "infogain": 0.849753, "prederr": 0.849753, "random": 0.848409, "risk": 0.849753}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:23:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:43Z", "cpu_seconds": 0.697176779000003, "created_at": "2026-11-22T03:22:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4671cb3be4ca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "272c4400e7cd4336331a0b70888121bcd836882eaae1e318d18b4e6445f9acb3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4671cb3be4ca", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:39Z", "seed": 31120508, "treatment": "frontier"}, "priority": 2.0098784689275213, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a90e0c7cb8939839b77c40f0a32fa254f738f0267f473a00939dfcca25e94a6b", "run_seconds": 0.942096471786499, "seed": 31120508, "started_at": "2026-11-22T03:22:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.021862, 0.053138], "effect": 0.0375, "evaluated_at": "2026-11-22T03:23:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4671cb3be4ca", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 0.983333333, 0.983333333, 0.983333333], "hypdisc": [0.95, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.95, 0.983333333, 0.966666667, 0.95], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "272c4400e7cd4336331a0b70888121bcd836882eaae1e318d18b4e6445f9acb3", "primary_control": "random", "result_hash": "a90e0c7cb8939839b77c40f0a32fa254f738f0267f473a00939dfcca25e94a6b", "signature": "bc9bb9eb054b04dae2cdf4d3486366123a7cfee33124a25f91a3b3f0a20f73d99e5d2ab433e5804fcda9dc32be34c5e7950dc75833b99c33963fdb9ddc6cb507", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 1.0, "goal": 0.9875, "hypdisc": 0.9875, "infogain": 1.0, "prederr": 1.0, "random": 0.9625, "risk": 1.0}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:23:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:55Z", "cpu_seconds": 0.6471563879999991, "created_at": "2026-11-22T03:22:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a88670569749", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "fd98d889606d6b5f0f4e9d8a98fdfc50ec666f01a06867e6bca54a4254be3725", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a88670569749", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:22:39Z", "seed": 879544260, "treatment": "full"}, "priority": 2.009083381266576, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0a875af62aea514d97794023770a2bdc17db4f00c1079bfb86f9cf0af6eef86a", "run_seconds": 0.7569637298583984, "seed": 879544260, "started_at": "2026-11-22T03:22:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.305, 0.43], "effect": 0.0625, "evaluated_at": "2026-11-22T03:23:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a88670569749", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 1.0, 0.75], "fixed_policy": [1.0, 1.0, 1.0, 0.25], "full": [1.0, 1.0, 0.5, 1.0], "no_world_model": [0.75, 1.0, 0.75, 0.5], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "fd98d889606d6b5f0f4e9d8a98fdfc50ec666f01a06867e6bca54a4254be3725", "primary_control": "e141_best", "result_hash": "0a875af62aea514d97794023770a2bdc17db4f00c1079bfb86f9cf0af6eef86a", "signature": "76c34278a5a0d64d68cd5ac651cc5e33b9652577adc34e18914f6c3c97a0d3d159524965dcefae52bfaf5d6169ea8ea566632dd207907938ab95fcce7ebf7809", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.8125, "full": 0.875, "no_world_model": 0.75, "random": 1.0}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:23:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:49Z", "cpu_seconds": 0.3405663500000031, "created_at": "2026-11-22T03:22:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2d8ab98a6a23", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "f287468b0a3b6847e4760a8af02ede3b5e22052bc8eb42f0fb7417d07bcd8465", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2d8ab98a6a23", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:22:39Z", "seed": 472016987, "treatment": "full"}, "priority": 2.0091882484394232, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "feee8294e55b53e128318d6e454354d1c02b339919848b7eb5c4054ff12b0668", "run_seconds": 0.5817146301269531, "seed": 472016987, "started_at": "2026-11-22T03:22:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.011237, 0.176263], "effect": 0.09375, "evaluated_at": "2026-11-22T03:23:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2d8ab98a6a23", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.791666667, 0.791666667, 0.9375, 0.895833333], "frozen": [0.0, 0.0, 0.083333333, 0.166666667], "full": [0.958333333, 0.958333333, 0.958333333, 0.916666667], "retrieval": [0.916666667, 0.875, 0.9375, 0.916666667], "rules_nodetect": [0.958333333, 0.958333333, 0.958333333, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f287468b0a3b6847e4760a8af02ede3b5e22052bc8eb42f0fb7417d07bcd8465", "primary_control": "facts", "result_hash": "feee8294e55b53e128318d6e454354d1c02b339919848b7eb5c4054ff12b0668", "signature": "494ea31d07c3aa5abdd64a6b459f6d1614e26951eb8364ce57528087d55ba660c15cddafa11e014a5c1cf2434985960885a34a2bf454e94658397e401766940a", "state": "VERIFIED", "summary": {"facts": 0.854167, "frozen": 0.0625, "full": 0.947917, "retrieval": 0.911458, "rules_nodetect": 0.947917}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:23:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:22:49Z", "cpu_seconds": 7.593452542999998, "created_at": "2026-11-22T03:22:39Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-801d73255338", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "56eef29ad89c9c0d2a73fcc9b5740b95060677f0a799552781ac2bb8b18aeb2e", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-801d73255338", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:22:39Z", "seed": 1598275728, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "809cee67fc187b5c612c35e42c2ec1aa7af21e678f153425e8c8adb6b277c119", "run_seconds": 9.306848764419556, "seed": 1598275728, "started_at": "2026-11-22T03:22:40Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.105613, 0.005613], "effect": -0.05, "evaluated_at": "2026-11-22T03:23:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-801d73255338", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.1940834013503538, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "56eef29ad89c9c0d2a73fcc9b5740b95060677f0a799552781ac2bb8b18aeb2e", "primary_control": "reset", "result_hash": "809cee67fc187b5c612c35e42c2ec1aa7af21e678f153425e8c8adb6b277c119", "signature": "3e1fc9df54b4e1aef3c9dd3924e24722d5bdddc87c8380694c22a242e2a9ccaeb7d4bef36bbe704f0f9c4cf668ce283c4e4c9e931e58be4342740cb2de8e560a", "state": "VERIFIED", "summary": {"inherit": 0.65, "reset": 0.7, "root": 0.36666666666666664}, "ties": 57, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:23:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:01Z", "cpu_seconds": 1.0254996960000113, "created_at": "2026-11-22T03:22:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-163ad0b43591", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "277c196e376c521bc711cd0e1d9bb8c277f15fcd6eefa60dfa712c63b40fa608", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-163ad0b43591", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:56Z", "seed": 936916416, "treatment": "frontier"}, "priority": 2.00989253901857, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "99bb6b7b21d0faf5927d7896f8d40c7b144b1389b0e3e28822cfa4d16e1333ab", "run_seconds": 1.564239263534546, "seed": 936916416, "started_at": "2026-11-22T03:23:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.269831, 0.180353], "effect": -0.044739, "evaluated_at": "2026-11-22T03:23:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-163ad0b43591", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.724637681, 0.836601307, 0.708333333, 0.833333333], "frontier": [0.434782609, 0.790849673, 0.708333333, 0.833333333], "goal": [0.695652174, 0.790849673, 0.708333333, 0.833333333], "hypdisc": [0.579710145, 0.790849673, 0.666666667, 0.833333333], "infogain": [0.710144928, 0.790849673, 0.708333333, 0.833333333], "prederr": [0.710144928, 0.790849673, 0.708333333, 0.833333333], "random": [0.710144928, 0.777777778, 0.875, 0.583333333], "risk": [0.434782609, 0.790849673, 0.708333333, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "277c196e376c521bc711cd0e1d9bb8c277f15fcd6eefa60dfa712c63b40fa608", "primary_control": "random", "result_hash": "99bb6b7b21d0faf5927d7896f8d40c7b144b1389b0e3e28822cfa4d16e1333ab", "signature": "0d0bf4d6bf914838e729ec21b2225557e1b185ecc1cc08ec0fea9dc1bb41dcf09edd81dc2713160d1d81b1f45cec7c70cf6568f887b7f839ecc61c326cb0e603", "state": "VERIFIED", "summary": {"count": 0.775726, "frontier": 0.691825, "goal": 0.757042, "hypdisc": 0.71764, "infogain": 0.760665, "prederr": 0.760665, "random": 0.736564, "risk": 0.691825}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:23:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:14Z", "cpu_seconds": 7.385521847999996, "created_at": "2026-11-22T03:22:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-212e9e2ec1f5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "1e0992263a5449d14c62647b8eb09f27efcd2acbf9b98b4502900b2afdfedded", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-212e9e2ec1f5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:56Z", "seed": 64822091, "treatment": "frontier"}, "priority": 2.0098632561380536, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3151205dc015389cb2505273d0c5472a8017a634943cc0e351572ac8acc49251", "run_seconds": 8.851520299911499, "seed": 64822091, "started_at": "2026-11-22T03:23:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014786, 0.125374], "effect": 0.055294, "evaluated_at": "2026-11-22T03:23:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-212e9e2ec1f5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.530980392, 0.590588235, 0.542745098, 0.591372549], "frontier": [0.569411765, 0.708235294, 0.700392157, 0.629803922], "goal": [0.562352941, 0.708235294, 0.61254902, 0.691764706], "hypdisc": [0.582745098, 0.671372549, 0.680784314, 0.676862745], "infogain": [0.569411765, 0.708235294, 0.700392157, 0.629803922], "prederr": [0.569411765, 0.708235294, 0.700392157, 0.629803922], "random": [0.519215686, 0.576470588, 0.621960784, 0.669019608], "risk": [0.569411765, 0.708235294, 0.700392157, 0.629803922]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "1e0992263a5449d14c62647b8eb09f27efcd2acbf9b98b4502900b2afdfedded", "primary_control": "random", "result_hash": "3151205dc015389cb2505273d0c5472a8017a634943cc0e351572ac8acc49251", "signature": "d6a9e5ea998464b5329147dbf88afae6fda398b06f567a3f73e193fa09bc18fbfa30e96d10cf314362c37b0c9edb3fadb0ea259d62b1467d02f34dbfb4fb4608", "state": "VERIFIED", "summary": {"count": 0.563922, "frontier": 0.651961, "goal": 0.643725, "hypdisc": 0.652941, "infogain": 0.651961, "prederr": 0.651961, "random": 0.596667, "risk": 0.651961}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:23:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:18Z", "cpu_seconds": 3.5562322339999923, "created_at": "2026-11-22T03:22:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ea4da2ea533f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 10, "n": 4}, "prereg_hash": "346daba05aa374635ff312ca07231750a47476a442f1fb19047543db180dbbb6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ea4da2ea533f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:22:56Z", "seed": 1471113933, "treatment": "frontier"}, "priority": 2.0091595250532115, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "266339552dc078dcc691bdf537124804bfa3e0d63825169b53f414c8757442b9", "run_seconds": 4.880016088485718, "seed": 1471113933, "started_at": "2026-11-22T03:23:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.101107, 0.056961], "effect": -0.022073, "evaluated_at": "2026-11-22T03:23:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ea4da2ea533f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.121084337, 0.285842294, 0.505862069, 0.302083333], "frontier": [0.121987952, 0.286290323, 0.506551724, 0.288541667], "goal": [0.121987952, 0.232526882, 0.470689655, 0.2453125], "hypdisc": [0.224096386, 0.202508961, 0.499310345, 0.2421875], "infogain": [0.105421687, 0.290322581, 0.506551724, 0.304166667], "prederr": [0.127710843, 0.269713262, 0.506551724, 0.228645833], "random": [0.261445783, 0.288530466, 0.496896552, 0.244791667], "risk": [0.121987952, 0.286290323, 0.506551724, 0.288541667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "346daba05aa374635ff312ca07231750a47476a442f1fb19047543db180dbbb6", "primary_control": "random", "result_hash": "266339552dc078dcc691bdf537124804bfa3e0d63825169b53f414c8757442b9", "signature": "f381f1ac1e0d3fec92326d8ec2de877feb5bca1fc611296d1203e5b67ffa9281f35b095dbea217688b68008cb39e060745f5cfacb21cee66931f76a7bd675401", "state": "VERIFIED", "summary": {"count": 0.303718, "frontier": 0.300843, "goal": 0.267629, "hypdisc": 0.292026, "infogain": 0.301616, "prederr": 0.283155, "random": 0.322916, "risk": 0.300843}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:23:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:58Z", "cpu_seconds": 2.370430674000005, "created_at": "2026-11-22T03:22:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f32dc6d3e733", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "adba7f39400e54f36b26c61cdc243718c57dcd8fea6516479fc145e37bd23814", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f32dc6d3e733", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:22:56Z", "seed": 625484558, "treatment": "full"}, "priority": 2.008891211472745, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f7b6c00c7ca713dda4c3fea795cf452eca8d2b87d9b10f959ec07f9145180fc5", "run_seconds": 3.4902002811431885, "seed": 625484558, "started_at": "2026-11-22T03:23:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.233549, 0.516451], "effect": 0.375, "evaluated_at": "2026-11-22T03:24:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f32dc6d3e733", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.5, 0.25, 0.25], "fixed_policy": [0.75, 1.0, 0.75, 1.0], "full": [1.0, 0.75, 0.5, 0.75], "no_world_model": [0.25, 0.25, 0.25, 0.25], "random": [0.75, 0.5, 0.5, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "adba7f39400e54f36b26c61cdc243718c57dcd8fea6516479fc145e37bd23814", "primary_control": "e141_best", "result_hash": "f7b6c00c7ca713dda4c3fea795cf452eca8d2b87d9b10f959ec07f9145180fc5", "signature": "611265c3bfdc992aaaf99c19fc7d3c0a6c969c976185d3873f918f250c50627f491db5d38239283e5cc628b20de8282f9e5012b975ca85c882e8b5decde75506", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.875, "full": 0.75, "no_world_model": 0.25, "random": 0.625}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:24:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:07Z", "cpu_seconds": 0.7399493899999925, "created_at": "2026-11-22T03:22:56Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a2c237ad3d82", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "prereg_hash": "1e5c89779ce15a2dc47b3438f077402e7e9d6c35e40e3e48b4d90cf7f125c874", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a2c237ad3d82", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:22:56Z", "seed": 1497917348, "treatment": "gated"}, "priority": 2.0098184411045588, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "123e4bc50e9ff6e181461232f170cca2f8950fced968439d865dc4fa75436469", "run_seconds": 0.770524263381958, "seed": 1497917348, "started_at": "2026-11-22T03:23:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.033913, 0.104565], "effect": 0.035326, "evaluated_at": "2026-11-22T03:23:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a2c237ad3d82", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.65, 0.716666667, 0.380434783, 0.616666667], "gated": [0.65, 0.716666667, 0.52173913, 0.616666667], "naive": [0.366666667, 0.666666667, 0.456521739, 0.583333333], "scratch": [0.65, 0.716666667, 0.380434783, 0.616666667], "ungated": [0.366666667, 0.733333333, 0.565217391, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "1e5c89779ce15a2dc47b3438f077402e7e9d6c35e40e3e48b4d90cf7f125c874", "primary_control": "scratch", "result_hash": "123e4bc50e9ff6e181461232f170cca2f8950fced968439d865dc4fa75436469", "signature": "af9794171d9ce7d49c3f67f658b6d6c3ab66d4a452170ef65d556a445900934c45c57f52fd01ec36b7a9fb3367dc53b7339c7c89fc8c4475a26c8f498ea8d409", "state": "VERIFIED", "summary": {"compressed": 0.590942, "gated": 0.626268, "naive": 0.518297, "scratch": 0.590942, "ungated": 0.541304}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:23:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:28Z", "cpu_seconds": 1.2544359660000026, "created_at": "2026-11-22T03:23:13Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-69a53c3f3495", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 10, "n": 4}, "prereg_hash": "f00c73d838891d3789069efa3349114614f5706c111e984a34eb4b447a2122df", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-69a53c3f3495", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:23:13Z", "seed": 463519772, "treatment": "full"}, "priority": 2.0092881985005984, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1e3a76a77ab7b93238027c796ad2312368cf194d8ddd3fd248a067b7767fb693", "run_seconds": 1.9608979225158691, "seed": 463519772, "started_at": "2026-11-22T03:23:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:23:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-69a53c3f3495", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 1.0, 0.0], "fixed_policy": [1.0, 0.0, 0.0, 0.0], "full": [0.5, 1.0, 1.0, 0.0], "no_world_model": [0.5, 1.0, 1.0, 0.0], "random": [0.75, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f00c73d838891d3789069efa3349114614f5706c111e984a34eb4b447a2122df", "primary_control": "e141_best", "result_hash": "1e3a76a77ab7b93238027c796ad2312368cf194d8ddd3fd248a067b7767fb693", "signature": "1a6a82e3c3c2f9f928ecbd8780b85bf21c01aab1a946c192245393efcbc242b6657d981ede82e1be11f871a34c36fd164647cab5fd0de0dd6607fe32e1901c05", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.25, "full": 0.625, "no_world_model": 0.625, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:23:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:24Z", "cpu_seconds": 0.21578605799999195, "created_at": "2026-11-22T03:23:13Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-953e47f32aef", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "3fc8e8de8e497feb2154961cbed3dbe9a4dd79929327da1a5b17e922202a0b2d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-953e47f32aef", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:23:13Z", "seed": 246991302, "treatment": "full"}, "priority": 2.009641044495176, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c687188747cd08a1f1d50e0f02a9397da7e5ec3cee1c495e4387dc5ad904f078", "run_seconds": 0.9234657287597656, "seed": 246991302, "started_at": "2026-11-22T03:23:23Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:23:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-953e47f32aef", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.5, 1.0], "fixed_policy": [0.25, 0.0, 0.25, 0.0], "full": [0.75, 1.0, 0.5, 1.0], "no_world_model": [0.75, 1.0, 0.25, 1.0], "random": [0.75, 1.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3fc8e8de8e497feb2154961cbed3dbe9a4dd79929327da1a5b17e922202a0b2d", "primary_control": "e141_best", "result_hash": "c687188747cd08a1f1d50e0f02a9397da7e5ec3cee1c495e4387dc5ad904f078", "signature": "ef35e6467b2a1eda245e67f75cbbded2c62e9bc3f80a94d419fd34e87c3b287a77742a4f203021c8a2300aca7899f6a5e5a907974561fae7654eb49d1ecc880a", "state": "PARTIAL", "summary": {"e141_best": 0.8125, "fixed_policy": 0.125, "full": 0.8125, "no_world_model": 0.75, "random": 0.8125}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:23:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:21Z", "cpu_seconds": 1.5341620389999946, "created_at": "2026-11-22T03:23:13Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f4539f4922a2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "c98d0b1682f3f53ccf666debdef5c8cd6c73cc9dfa30b00cab6b5826bb858dcb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f4539f4922a2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:23:13Z", "seed": 1225643, "treatment": "full"}, "priority": 2.0099700931609794, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d4b392c76827cda156e62ba455bc282b0657f5e1892e9f559b07e20bc958abac", "run_seconds": 1.9120869636535645, "seed": 1225643, "started_at": "2026-11-22T03:23:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002149, 0.261299], "effect": 0.129575, "evaluated_at": "2026-11-22T03:23:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f4539f4922a2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.735294118, 0.533333333, 0.5, 0.766666667], "frozen": [0.284313725, 0.225, 0.314814815, 0.316666667], "full": [0.764705882, 0.8, 0.722222222, 0.766666667], "retrieval": [0.764705882, 0.8, 0.703703704, 0.766666667], "rules_nodetect": [0.862745098, 0.816666667, 0.740740741, 0.766666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c98d0b1682f3f53ccf666debdef5c8cd6c73cc9dfa30b00cab6b5826bb858dcb", "primary_control": "facts", "result_hash": "d4b392c76827cda156e62ba455bc282b0657f5e1892e9f559b07e20bc958abac", "signature": "de9dad7aeeb124be09b70398ef17a5fcec118553287f6532ef54ab671ed9f8f91845c5b9fb10ea4db2a79d3961f6d9415c3bfb0397affe962a56d4a22b93c408", "state": "VERIFIED", "summary": {"facts": 0.633824, "frozen": 0.285199, "full": 0.763399, "retrieval": 0.758769, "rules_nodetect": 0.796705}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:23:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:29Z", "cpu_seconds": 0.40513950199999726, "created_at": "2026-11-22T03:23:13Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-736b57296668", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "prereg_hash": "fae558d2b8f8ebf9f9ca3d55e8a313cb26d21babcd4c93a4fba7f33acbf65d0e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-736b57296668", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:23:13Z", "seed": 1075019066, "treatment": "gated"}, "priority": 2.0089220066745206, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "2a1bb373c5f691f20e1bf86a2f5f1a5c65c92920b4042954d768032391adee8a", "run_seconds": 0.7140114307403564, "seed": 1075019066, "started_at": "2026-11-22T03:23:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01, 0.030833], "effect": 0.010417, "evaluated_at": "2026-11-22T03:23:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-736b57296668", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.708333333, 0.541666667, 0.5, 0.75], "gated": [0.708333333, 0.541666667, 0.541666667, 0.75], "naive": [0.104166667, 0.25, 0.4375, 0.291666667], "scratch": [0.708333333, 0.541666667, 0.5, 0.75], "ungated": [0.5, 0.25, 0.4375, 0.291666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "fae558d2b8f8ebf9f9ca3d55e8a313cb26d21babcd4c93a4fba7f33acbf65d0e", "primary_control": "scratch", "result_hash": "2a1bb373c5f691f20e1bf86a2f5f1a5c65c92920b4042954d768032391adee8a", "signature": "5ef5f4248346063bc6593b400b68e5c21375c3deccfbb5a53caaa8d628282049b08891477b896b964f624464677554cc5cdf159aa683693c51e242dcd0cc6d08", "state": "VERIFIED", "summary": {"compressed": 0.625, "gated": 0.635417, "naive": 0.270833, "scratch": 0.625, "ungated": 0.369792}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:23:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:26Z", "cpu_seconds": 6.174653433000003, "created_at": "2026-11-22T03:23:13Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-005a2bf5a890", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7e6d0500cf3bc20caa100dbf80b98e8a8720e54a8510d5d311c24ecf47a255ec", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-005a2bf5a890", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:23:13Z", "seed": 642331684, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "b6d6879088892bf312aed4af5ba6260c5ee892ce16bb42cd3375646dce21ba6a", "run_seconds": 9.099040985107422, "seed": 642331684, "started_at": "2026-11-22T03:23:17Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.079138, 0.012471], "effect": -0.033333, "evaluated_at": "2026-11-22T03:23:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-005a2bf5a890", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.622144543444593, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0], "root": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7e6d0500cf3bc20caa100dbf80b98e8a8720e54a8510d5d311c24ecf47a255ec", "primary_control": "reset", "result_hash": "b6d6879088892bf312aed4af5ba6260c5ee892ce16bb42cd3375646dce21ba6a", "signature": "373fd3f1c3b128d500d9016c85501619f56b9c514accee10224d88215ef6101bf01c2b8b80b02908095627d2ae161cc373423475c46a021d9c4b7f20722db300", "state": "VERIFIED", "summary": {"inherit": 0.7, "reset": 0.7333333333333333, "root": 0.45}, "ties": 58, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:23:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:49Z", "cpu_seconds": 11.11023691199999, "created_at": "2026-11-22T03:23:29Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-27c7d8e7d46e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 7, "n": 4}, "prereg_hash": "6882fbc5ccc8aec786f4582b2eb2a8faf2ae3dbac4545306370fbe0cef96365d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-27c7d8e7d46e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:23:29Z", "seed": 1256247896, "treatment": "frontier"}, "priority": 2.0093922172122833, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "bbccafad14305fddbad569e805af2030dce6f1329e63529cfede620ce0ee6f0b", "run_seconds": 14.054667234420776, "seed": 1256247896, "started_at": "2026-11-22T03:23:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.047681, 0.058488], "effect": 0.005403, "evaluated_at": "2026-11-22T03:23:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-27c7d8e7d46e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.41957672, 0.422941176, 1.0, 0.564673913], "frontier": [0.285185185, 0.440588235, 1.0, 0.588043478], "goal": [0.274603175, 0.355294118, 1.0, 0.52173913], "hypdisc": [0.350793651, 0.386470588, 1.0, 0.539673913], "infogain": [0.285185185, 0.440588235, 1.0, 0.588043478], "prederr": [0.285185185, 0.440588235, 1.0, 0.588043478], "random": [0.353968254, 0.388235294, 1.0, 0.55], "risk": [0.285185185, 0.440588235, 1.0, 0.588043478]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6882fbc5ccc8aec786f4582b2eb2a8faf2ae3dbac4545306370fbe0cef96365d", "primary_control": "random", "result_hash": "bbccafad14305fddbad569e805af2030dce6f1329e63529cfede620ce0ee6f0b", "signature": "ca8f3529ca1cf1580bdcc545496f11e5a8ee9013da5c52f1548a7e07d502e53bcef9394c017c42835b117e33e8b6a20e244bc294c05605eaa644a038a8b35408", "state": "VERIFIED", "summary": {"count": 0.601798, "frontier": 0.578454, "goal": 0.537909, "hypdisc": 0.569235, "infogain": 0.578454, "prederr": 0.578454, "random": 0.573051, "risk": 0.578454}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:23:51Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:23:29Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_goal", "fence": 3, "id": "E141-ow_goal-391d90e9c77b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 4, "n": 4}, "prereg_hash": "14a970a8468bf5d4d4f0417b90c8189f500cb6f12a8f64eb0b8a03975f164c8a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-391d90e9c77b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:23:29Z", "seed": 942528102, "treatment": "full"}, "priority": 2.00923469650721, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 942528102, "started_at": "2026-11-22T03:23:50Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:23:35Z", "cpu_seconds": 1.5990113030000117, "created_at": "2026-11-22T03:23:29Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-76195dc0f64c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 4, "n": 4}, "prereg_hash": "51291a33f3811e310a7246e8afcdd757851501b6349ab3c02ac4e51ecc43b5a1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-76195dc0f64c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:23:29Z", "seed": 893236351, "treatment": "full"}, "priority": 2.009904583458844, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a33b426438ee262b6b2ae7fc09d04bc3cd5232e874f2db630e3817657f080c17", "run_seconds": 2.189711093902588, "seed": 893236351, "started_at": "2026-11-22T03:23:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.043117, 0.406535], "effect": 0.224826, "evaluated_at": "2026-11-22T03:23:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-76195dc0f64c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.565972222, 0.701388889, 0.833333333], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 0.930555556, 0.944444444, 0.902777778], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "51291a33f3811e310a7246e8afcdd757851501b6349ab3c02ac4e51ecc43b5a1", "primary_control": "facts", "result_hash": "a33b426438ee262b6b2ae7fc09d04bc3cd5232e874f2db630e3817657f080c17", "signature": "3e9b6861980e26f395fe4ab35affde0a2e754015128f513eb164f6075fa02177b26d92526d8b6179251d04e17188e8cd62c36b070d3df60d8ed5fc41c2a5ed03", "state": "VERIFIED", "summary": {"facts": 0.775174, "frozen": 0.0, "full": 1.0, "retrieval": 0.944444, "rules_nodetect": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:23:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:45Z", "cpu_seconds": 3.7758601669999905, "created_at": "2026-11-22T03:23:29Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a5df668d89f2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "b17645f0204daa8b8537f54c2f93006a3f246f808beca7c9c8890d35233a5f1a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a5df668d89f2", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:23:29Z", "seed": 1096313125, "treatment": "gated"}, "priority": 2.009379995672842, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f688f4af2e0fef464a077492f21ab8da76678f594b840e8af987d0585a057710", "run_seconds": 5.017835855484009, "seed": 1096313125, "started_at": "2026-11-22T03:23:40Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:23:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a5df668d89f2", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.245977011, 0.85, 0.165656566, 0.32183908], "gated": [0.245977011, 0.85, 0.165656566, 0.32183908], "naive": [0.16091954, 0.35, 0.155555556, 0.15862069], "scratch": [0.245977011, 0.85, 0.165656566, 0.32183908], "ungated": [0.172413793, 0.35, 0.117171717, 0.15862069]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b17645f0204daa8b8537f54c2f93006a3f246f808beca7c9c8890d35233a5f1a", "primary_control": "scratch", "result_hash": "f688f4af2e0fef464a077492f21ab8da76678f594b840e8af987d0585a057710", "signature": "e634c962af316112a4fa9a444e6ef39c6d1aaf776cf99f240d87408084bf51e8f5dd52dd22836b45d40759daaa6da0a88e9068aa73549cd32334444dab17560c", "state": "PARTIAL", "summary": {"compressed": 0.395868, "gated": 0.395868, "naive": 0.206274, "scratch": 0.395868, "ungated": 0.199552}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:23:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:41Z", "cpu_seconds": 7.369429904, "created_at": "2026-11-22T03:23:29Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-7c7a42649181", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "fdef6adf195280d3aafbd3ae971bef8de78675b07df5e1edd64a35d5c8cc56bd", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-7c7a42649181", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:23:29Z", "seed": 54274446, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "1ae75786d1a1de5d65b8715c2e23d9eb960fe607dc5e28659e956823ac8ae117", "run_seconds": 9.800989866256714, "seed": 54274446, "started_at": "2026-11-22T03:23:31Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 4, "reset": 4}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:23:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-7c7a42649181", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fdef6adf195280d3aafbd3ae971bef8de78675b07df5e1edd64a35d5c8cc56bd", "primary_control": "reset", "result_hash": "1ae75786d1a1de5d65b8715c2e23d9eb960fe607dc5e28659e956823ac8ae117", "signature": "c82ae94cf32edff2b52f9cc84ebad6b922d89dab7c792d7463bcb3e2484493ba7801ffb5a619f8f0ee48b90c04b5877fd87cff72179257c09efe023c474a0905", "state": "PARTIAL", "summary": {"inherit": 0.8, "reset": 0.8, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:23:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:03Z", "cpu_seconds": 0.6077418550000004, "created_at": "2026-11-22T03:23:46Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4f03c8016ae3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "3b3a4b3ef460a00e58a561da06b82962d479c023cec1936384ed03fb9f56fd38", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4f03c8016ae3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:23:46Z", "seed": 792810147, "treatment": "full"}, "priority": 2.0086739662966884, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d34e64f8bc85853459e481253c35b4c04e9c504d391072cb6143c2b2a1439350", "run_seconds": 0.7044327259063721, "seed": 792810147, "started_at": "2026-11-22T03:25:02Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:25:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4f03c8016ae3", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.0, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 0.0, 0.0, 0.0], "no_world_model": [0.0, 0.0, 0.0, 0.0], "random": [0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3b3a4b3ef460a00e58a561da06b82962d479c023cec1936384ed03fb9f56fd38", "primary_control": "e141_best", "result_hash": "d34e64f8bc85853459e481253c35b4c04e9c504d391072cb6143c2b2a1439350", "signature": "bac4e24dd1aa6c9c911bd8e488f41187b11c6d9fe62470f6f8e19013964fa02c036f734f4857a555caf9755ab1b34a8f9c7d6872b442189d91891b56caf4e602", "state": "PARTIAL", "summary": {"e141_best": 0.0, "fixed_policy": 0.0, "full": 0.0, "no_world_model": 0.0, "random": 0.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:25:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:26Z", "cpu_seconds": 7.312297815999997, "created_at": "2026-11-22T03:23:46Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-732dab49d5e4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "b9b27e929731765aeea8ddedab690930d86624ffef50fd9cfbf5cea4f18f82e1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-732dab49d5e4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:23:46Z", "seed": 1657577920, "treatment": "full"}, "priority": 2.008767232026275, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ba242fcd7ebc8171e080d84083a079d52cd4928a861b9e25ba64886189ccc33a", "run_seconds": 8.775305271148682, "seed": 1657577920, "started_at": "2026-11-22T03:24:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.605917, 0.230917], "effect": -0.1875, "evaluated_at": "2026-11-22T03:24:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-732dab49d5e4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.0, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 0.0, 0.5, 0.25], "no_world_model": [0.0, 0.0, 0.75, 0.25], "random": [1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "b9b27e929731765aeea8ddedab690930d86624ffef50fd9cfbf5cea4f18f82e1", "primary_control": "e141_best", "result_hash": "ba242fcd7ebc8171e080d84083a079d52cd4928a861b9e25ba64886189ccc33a", "signature": "a441f7f108733669efb789735cce4c5aa7ccb7f8eb27dd2ad74c79427dc734a34f362bdd36e970f9a52eec5cbbe1b8d1778722f48c2589152e9ee130890e430d", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.0, "full": 0.1875, "no_world_model": 0.25, "random": 0.5}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:25:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:46Z", "cpu_seconds": 0.9679570370000006, "created_at": "2026-11-22T03:23:46Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a69a7d1d993e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 2, "n": 4}, "prereg_hash": "ee8c40ee4cf6735e498e1514bb49a8f6daf1390d2edbafb75a4f6394459afaf4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a69a7d1d993e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:23:46Z", "seed": 1328958822, "treatment": "full"}, "priority": 2.008190146300255, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "613b95f4a6e34f80f864f7e511fbe0f5e54362cd77419993a66e8ec0846a2b84", "run_seconds": 1.3250486850738525, "seed": 1328958822, "started_at": "2026-11-22T03:28:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.102348, 0.289248], "effect": 0.195798, "evaluated_at": "2026-11-22T03:28:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a69a7d1d993e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.198434442, 0.177690802, 0.606299213, 0.178473581], "frozen": [0.055968689, 0.034050881, 0.050393701, 0.030919765], "full": [0.342465753, 0.342074364, 0.743307087, 0.516242661], "retrieval": [0.40704501, 0.368688845, 0.803149606, 0.504892368], "rules_nodetect": [0.342465753, 0.342074364, 0.743307087, 0.516242661]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "ee8c40ee4cf6735e498e1514bb49a8f6daf1390d2edbafb75a4f6394459afaf4", "primary_control": "facts", "result_hash": "613b95f4a6e34f80f864f7e511fbe0f5e54362cd77419993a66e8ec0846a2b84", "signature": "08720780e632ab4f5ee3f44c11c2ad3c20f91a148235e9b837e371648d8985b2a365fca4db4cd192e0f4daa3edf02f89f8f002fdb2af53a2c8ad099d7ff7af06", "state": "VERIFIED", "summary": {"facts": 0.290225, "frozen": 0.042833, "full": 0.486022, "retrieval": 0.520944, "rules_nodetect": 0.486022}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:28:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:23:53Z", "cpu_seconds": 6.585276809999996, "created_at": "2026-11-22T03:23:46Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a61f40606e1a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6576b87debb6f36ea6d1befb9912f319b128e38e031fb3c85d00fc8a66940881", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a61f40606e1a", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:23:46Z", "seed": 85453230, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "e5d44a7fdc1f6b30c3fde392546de155b41c88972df9e9abc28cacb62b4f4475", "run_seconds": 7.235680103302002, "seed": 85453230, "started_at": "2026-11-22T03:23:46Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:23:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a61f40606e1a", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.42816996189423545, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6576b87debb6f36ea6d1befb9912f319b128e38e031fb3c85d00fc8a66940881", "primary_control": "reset", "result_hash": "e5d44a7fdc1f6b30c3fde392546de155b41c88972df9e9abc28cacb62b4f4475", "signature": "df3de1c02eb7b8986addae45c43eae63966a895f8e11ffbd897194c7558acb639b985e6e308f2a51cb8d758a108b90964419f53ad5b945d4f4bb2d592d6e660a", "state": "PARTIAL", "summary": {"inherit": 0.55, "reset": 0.55, "root": 0.26666666666666666}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:24:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:23Z", "cpu_seconds": 5.951604822, "created_at": "2026-11-22T03:24:02Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-04d7d3933165", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7d795b680e1a7282a49126558d824afa8076745b34112a0823a08a409d47fb0e", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-04d7d3933165", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:24:02Z", "seed": 974945954, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "6f204be2a962342d63845d7a4cff8903949cac4e83756823410ef3608810e133", "run_seconds": 7.49304986000061, "seed": 974945954, "started_at": "2026-11-22T03:24:15Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 4}, "better": "higher", "ci95": [-0.211226, 0.011226], "effect": -0.1, "evaluated_at": "2026-11-22T03:24:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-04d7d3933165", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6063091044991454, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0085, "losses": 9, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.981", "prereg_hash": "7d795b680e1a7282a49126558d824afa8076745b34112a0823a08a409d47fb0e", "primary_control": "reset", "result_hash": "6f204be2a962342d63845d7a4cff8903949cac4e83756823410ef3608810e133", "signature": "a53bbebb2e8fed984c4531acf3f52e9adf4201c1737be2b994a459888f96838d5b66c62ce1dd5ffdb7dc0c5b0b94b82f5f2f16dca5ebf74e72e49178e572f609", "state": "VERIFIED", "summary": {"inherit": 0.7666666666666667, "reset": 0.8666666666666667, "root": 0.26666666666666666}, "ties": 48, "treatment": "inherit", "wins": 3}, "verified_at": "2026-11-22T03:24:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:10Z", "cpu_seconds": 5.624584873000003, "created_at": "2026-11-22T03:24:02Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-167095a04cc6", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7e77961853c8ca259c3ca7ca1c8857aa669b15552dabe81790f57297eba3cdf4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-167095a04cc6", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:24:02Z", "seed": 748081296, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "3dc6aae18c9c2fde6bd82fb7c4aaf4b58ae2f8081a65d81829cfb9def37c3143", "run_seconds": 6.879366874694824, "seed": 748081296, "started_at": "2026-11-22T03:24:03Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:24:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-167095a04cc6", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "7e77961853c8ca259c3ca7ca1c8857aa669b15552dabe81790f57297eba3cdf4", "primary_control": "reset", "result_hash": "3dc6aae18c9c2fde6bd82fb7c4aaf4b58ae2f8081a65d81829cfb9def37c3143", "signature": "109f68819fe354a53a69665e1fa87c3812a55de53184f88f2276944a401f2ae9a2ebc231c863ed46abf6e90f69df0af422b82f3e3c379b36322a1c2867a07d08", "state": "PARTIAL", "summary": {"inherit": 0.7333333333333333, "reset": 0.7333333333333333, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:24:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:06Z", "cpu_seconds": 2.1472004860000027, "created_at": "2026-11-22T03:24:03Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7e29b1fa55b3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 6, "n": 4}, "prereg_hash": "b36050c2fb61022a3915ed1bba169022257e97376f68cf17a41474c08638f063", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7e29b1fa55b3", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:24:03Z", "seed": 376025422, "treatment": "frontier"}, "priority": 2.0094975699034783, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "96b9d2771c42b0044bd59a313503fa34947652239425461d72171c969cef7b87", "run_seconds": 3.0532352924346924, "seed": 376025422, "started_at": "2026-11-22T03:24:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026108, 0.165379], "effect": 0.069635, "evaluated_at": "2026-11-22T03:24:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7e29b1fa55b3", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.6546875, 0.33, 0.9125], "frontier": [1.0, 0.60625, 0.493333333, 0.9875], "goal": [0.975, 0.6140625, 0.493333333, 0.925], "hypdisc": [1.0, 0.63125, 0.5675, 0.93125], "infogain": [0.975, 0.61875, 0.489166667, 0.975], "prederr": [0.925, 0.5546875, 0.1275, 0.2125], "random": [0.975, 0.653125, 0.336666667, 0.84375], "risk": [1.0, 0.60625, 0.493333333, 0.9875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "b36050c2fb61022a3915ed1bba169022257e97376f68cf17a41474c08638f063", "primary_control": "random", "result_hash": "96b9d2771c42b0044bd59a313503fa34947652239425461d72171c969cef7b87", "signature": "70cdf8cec19e7070c29f03970a5f19629563761b306b13f1a8523ba0f0b86a6d8a37b01abf62d039036c94ab7f30af1154e5c846def0523b91554c6f3d1f0809", "state": "VERIFIED", "summary": {"count": 0.724297, "frontier": 0.771771, "goal": 0.751849, "hypdisc": 0.7825, "infogain": 0.764479, "prederr": 0.454922, "random": 0.702135, "risk": 0.771771}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:24:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:12Z", "cpu_seconds": 0.9158250520000024, "created_at": "2026-11-22T03:24:03Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8f609aa327cf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 10, "n": 4}, "prereg_hash": "07a24ec3db5c5f1bcbbeb878d2f14b99d19d59b466262d6db376e22f8a598b07", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8f609aa327cf", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:24:03Z", "seed": 1294399538, "treatment": "full"}, "priority": 2.0094104585476025, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5b6c1ad5a307b773cfe14ef80947fff1ad592f2f51e896fc00ca3ac5859894c8", "run_seconds": 1.065086841583252, "seed": 1294399538, "started_at": "2026-11-22T03:24:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.010446, 0.282407], "effect": 0.135981, "evaluated_at": "2026-11-22T03:24:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8f609aa327cf", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.725, 0.875, 0.355034722, 0.49375], "frozen": [0.488970588, 0.75, 0.291666667, 0.40625], "full": [0.775, 0.875, 0.692708333, 0.65], "retrieval": [0.818382353, 0.875, 0.75, 0.778125], "rules_nodetect": [0.775, 0.875, 0.315972222, 0.65]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "07a24ec3db5c5f1bcbbeb878d2f14b99d19d59b466262d6db376e22f8a598b07", "primary_control": "facts", "result_hash": "5b6c1ad5a307b773cfe14ef80947fff1ad592f2f51e896fc00ca3ac5859894c8", "signature": "dece876e4c8c98fa85cfa59e5e53fda460b4f902f27d1af32db85a140b5a29132cef377a85c506612c06d565bd2adf86b4bebcb6842af66249d7fc6f76ec2306", "state": "VERIFIED", "summary": {"facts": 0.612196, "frozen": 0.484222, "full": 0.748177, "retrieval": 0.805377, "rules_nodetect": 0.653993}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:24:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:52Z", "cpu_seconds": 0.6822483610000001, "created_at": "2026-11-22T03:24:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b4535be7fa4c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "c51f6f13d4a188558340420190fad8ad0ec42071b470a69dc2fce9a64733acf0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b4535be7fa4c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:24:19Z", "seed": 2015497424, "treatment": "frontier"}, "priority": 2.0093808708929166, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ef4acdf27620196b19493765182908fba1c796775aaa67cdb88b1788811bec24", "run_seconds": 0.752589225769043, "seed": 2015497424, "started_at": "2026-11-22T03:24:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.070345, 0.007845], "effect": -0.03125, "evaluated_at": "2026-11-22T03:25:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b4535be7fa4c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 0.916666667, 0.916666667, 1.0], "goal": [1.0, 0.916666667, 0.9375, 0.958333333], "hypdisc": [1.0, 0.958333333, 0.916666667, 1.0], "infogain": [1.0, 0.916666667, 0.916666667, 1.0], "prederr": [1.0, 0.916666667, 0.916666667, 1.0], "random": [1.0, 1.0, 0.958333333, 1.0], "risk": [1.0, 0.916666667, 0.916666667, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c51f6f13d4a188558340420190fad8ad0ec42071b470a69dc2fce9a64733acf0", "primary_control": "random", "result_hash": "ef4acdf27620196b19493765182908fba1c796775aaa67cdb88b1788811bec24", "signature": "7d955a5f0cd3725a4324fd84ad111e98576464f24e6abd525c182588eb9e92d32d079ef85a94af64fa867f0137044a184a917081cb8a06c927962eb40927c107", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.958333, "goal": 0.953125, "hypdisc": 0.96875, "infogain": 0.958333, "prederr": 0.958333, "random": 0.989583, "risk": 0.958333}, "ties": 2, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:25:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:46Z", "cpu_seconds": 1.391503101, "created_at": "2026-11-22T03:24:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-51c43adb94b6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 8, "n": 4}, "prereg_hash": "442ba8bffb7d1d2b5804dc01cfffe5bbd49096f26f3add767a33dafb5dd4507a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-51c43adb94b6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:24:19Z", "seed": 1255134238, "treatment": "full"}, "priority": 2.009709601004586, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c727f9b92228027b160362251df2990f1521736333d14d477d2ddeec8c52d78d", "run_seconds": 1.4483819007873535, "seed": 1255134238, "started_at": "2026-11-22T03:24:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024461, 0.165789], "effect": 0.070664, "evaluated_at": "2026-11-22T03:25:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-51c43adb94b6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.509547739, 1.0, 0.494711538, 1.0], "frozen": [0.361809045, 0.6, 0.423076923, 0.65], "full": [0.675376884, 1.0, 0.636538462, 0.975], "retrieval": [0.84120603, 1.0, 0.835576923, 0.975], "rules_nodetect": [0.675376884, 1.0, 0.636538462, 0.975]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "442ba8bffb7d1d2b5804dc01cfffe5bbd49096f26f3add767a33dafb5dd4507a", "primary_control": "facts", "result_hash": "c727f9b92228027b160362251df2990f1521736333d14d477d2ddeec8c52d78d", "signature": "1d17a55e7ae1f1158658701ab60dca0002b089fc7b3bc7045ad7020ed231de3f1177cd99471e77a9dcd403000c3dd7b72ff603352414fe2eaec15f42ee255d06", "state": "VERIFIED", "summary": {"facts": 0.751065, "frozen": 0.508721, "full": 0.821729, "retrieval": 0.912946, "rules_nodetect": 0.821729}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:25:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:31Z", "cpu_seconds": 0.4486976740000017, "created_at": "2026-11-22T03:24:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b4f1ab5fa942", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "faf99777dc368ba15f7321c6785757894faa084640e1d8ae1a4292e7e204d7ad", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b4f1ab5fa942", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:24:19Z", "seed": 810130487, "treatment": "full"}, "priority": 2.009861972797093, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4cb19ea0dacb9390fc97eb644fb886a8715c55a8fb119491b185f40cbd852ceb", "run_seconds": 0.8402180671691895, "seed": 810130487, "started_at": "2026-11-22T03:24:31Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:25:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b4f1ab5fa942", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.975], "frozen": [1.0, 0.0], "full": [1.0, 0.991667], "retrieval": [1.0, 0.995833], "rules_nodetect": [1.0, 0.991667]}, "prereg_hash": "faf99777dc368ba15f7321c6785757894faa084640e1d8ae1a4292e7e204d7ad", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "4cb19ea0dacb9390fc97eb644fb886a8715c55a8fb119491b185f40cbd852ceb", "signature": "55aef7a9c8b94eb3eb1316b1402bca9c315af297805bd34c404edefb2912fb1d6e08214414826e0238f3277eb927d34ff95234c4e30a167773b11d6a13cd3e0d", "state": "REJECTED", "summary": {"facts": 0.975, "frozen": 0.0, "full": 0.991667, "retrieval": 0.995833, "rules_nodetect": 0.991667}}, "verified_at": "2026-11-22T03:25:01Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:26:05Z", "cpu_seconds": 8.30602892, "created_at": "2026-11-22T03:24:19Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-8debf6d1d370", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "da086c501fd856dfda23399c564e5a688d932cb04fccd06686e2b74ab10cf7d4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-8debf6d1d370", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:24:19Z", "seed": 1884968919, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "148afd6fc11853f083967e8ec45e6608a89778a88be7bedc7a4d2f8f8a4ecb8d", "run_seconds": 10.459118604660034, "seed": 1884968919, "started_at": "2026-11-22T03:24:28Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:26:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-8debf6d1d370", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "da086c501fd856dfda23399c564e5a688d932cb04fccd06686e2b74ab10cf7d4", "primary_control": "reset", "result_hash": "148afd6fc11853f083967e8ec45e6608a89778a88be7bedc7a4d2f8f8a4ecb8d", "signature": "5e198dcd3f1b00caa066dd901c0c1d936f1ac7b4dea0ca5f35ab90b0805987ff658872cd8c32c55e20fb149191a567e4e238f0b6d07a898c1190ef48c2f3e200", "state": "PARTIAL", "summary": {"inherit": 0.5, "reset": 0.5, "root": 0.2833333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:26:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:39Z", "cpu_seconds": 1.8421745559999998, "created_at": "2026-11-22T03:24:36Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-39ca32b32345", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "83b53b5c70f255549f588ad30156e189c6c1fba9207a25e6182494aa36e74014", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-39ca32b32345", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:24:36Z", "seed": 797237232, "treatment": "iw_temperature"}, "priority": 2.0098707848542956, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-a8918dbdc7b4", "result_hash": "ed9a8bb23141e78e3394b52e00dcc0ccbddda99c5ec92aed1146b8f819eda3cb", "run_seconds": 3.029026985168457, "seed": 797237232, "started_at": "2026-11-22T03:24:36Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015192, 0.004091], "effect": -0.00555, "evaluated_at": "2026-11-22T03:25:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-39ca32b32345", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.301217028, 0.309764138, 0.316833748, 0.311745962, 0.258128417, 0.298173435, 0.214375673, 0.254282363, 0.300392763, 0.208271975, 0.269340552, 0.25568261, 0.300207522, 0.27941512, 0.315145725, 0.21810746], "raw": [0.20771586, 0.277466077, 0.316833748, 0.300897425, 0.236206637, 0.28481541, 0.188957148, 0.269938662, 0.276490395, 0.190012505, 0.231161192, 0.24134953, 0.300207522, 0.251230552, 0.283670758, 0.185399628], "temperature": [0.246292975, 0.294897105, 0.326375115, 0.311745962, 0.258128417, 0.298173435, 0.200721645, 0.286131455, 0.302072475, 0.190012505, 0.269340552, 0.26364653, 0.312276282, 0.269526677, 0.307538625, 0.185399628]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0936, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.806", "prereg_hash": "83b53b5c70f255549f588ad30156e189c6c1fba9207a25e6182494aa36e74014", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-a8918dbdc7b4", "result_hash": "ed9a8bb23141e78e3394b52e00dcc0ccbddda99c5ec92aed1146b8f819eda3cb", "signature": "22a51aa8d6d28f8b4aead31b558b031c784958b0b7bf1bcd87118db681a0ac8ecaca73be388b8d9dab476abac7b731038bf8673efacf5dab45e16d091e69d201", "state": "VERIFIED", "summary": {"iw_temperature": 0.275693, "raw": 0.252647, "temperature": 0.270142}, "ties": 4, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:25:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:11Z", "cpu_seconds": 7.991401074999999, "created_at": "2026-11-22T03:24:36Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-eb828def08c3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "468e0370989da2ed8e3ff0b389e623cd392fd1ce8d38d1d5b51a6abe5ac10911", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-eb828def08c3", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:24:36Z", "seed": 1857828464, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "ec1afbd61118c5f5cb0316381243103ba10154e156ccca1d7dba6f2d386f03bf", "run_seconds": 11.420547008514404, "seed": 1857828464, "started_at": "2026-11-22T03:24:59Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:25:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-eb828def08c3", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.42471987874860984, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "468e0370989da2ed8e3ff0b389e623cd392fd1ce8d38d1d5b51a6abe5ac10911", "primary_control": "reset", "result_hash": "ec1afbd61118c5f5cb0316381243103ba10154e156ccca1d7dba6f2d386f03bf", "signature": "dab86d6e5b5a4b3c6ec407b645e1795a2969499af37f1e69e3df6fd13c931e7f58d3a7e352757994eae355db71f556e5f71a6ec6f2a6ac60922a9a465789290b", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:25:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:02Z", "cpu_seconds": 1.68658607, "created_at": "2026-11-22T03:24:52Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-8d6f55a73443", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "c88950ad324b355fd1bc9059329bbf527fff049aa8c89d8de0f63adfce0c75ab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-8d6f55a73443", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:24:52Z", "seed": 908229379, "treatment": "iw_temperature"}, "priority": 2.009705396989309, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-9b3a759205e8", "result_hash": "49c8195206b81be465b45089568fe62f5abea1ad27d62ef9cff5521aee9c5444", "run_seconds": 3.0720272064208984, "seed": 908229379, "started_at": "2026-11-22T03:24:59Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.025479, 0.010715], "effect": -0.007382, "evaluated_at": "2026-11-22T03:25:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-8d6f55a73443", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.280708165, 0.26387959, 0.236685903, 0.349117067, 0.3307384, 0.258949983, 0.302150415, 0.26309348, 0.216918593, 0.18686589, 0.266111197, 0.31432301, 0.299744907, 0.141636968, 0.164216382, 0.27121235], "raw": [0.14876377, 0.235838972, 0.22204971, 0.307830278, 0.299051788, 0.272682637, 0.230662767, 0.254223162, 0.18273376, 0.164336905, 0.234296542, 0.30477984, 0.290036958, 0.249492198, 0.193979855, 0.259724653], "temperature": [0.203015173, 0.224134215, 0.236685903, 0.340910082, 0.309399133, 0.209277173, 0.262239975, 0.26309348, 0.239233737, 0.19924483, 0.255115245, 0.31432301, 0.280101408, 0.172071517, 0.24817944, 0.27121235]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "c88950ad324b355fd1bc9059329bbf527fff049aa8c89d8de0f63adfce0c75ab", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-9b3a759205e8", "result_hash": "49c8195206b81be465b45089568fe62f5abea1ad27d62ef9cff5521aee9c5444", "signature": "f9a0d5cf383bf11cd86f67c2310ea72a83f5910ae5467169bf4b8aa0c01bd566bdd894630e1ca922d05d08a4de5e34b54d08f2bcb5639a25395dc685f9d4a407", "state": "VERIFIED", "summary": {"iw_temperature": 0.259147, "raw": 0.240655, "temperature": 0.251765}, "ties": 4, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:25:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:24:57Z", "cpu_seconds": 0.6621496769999995, "created_at": "2026-11-22T03:24:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-9d363dd0f5ed", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "75034dffc64464c140c3edd3e93709172a157c1f64e5856e882b2a0055955661", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-9d363dd0f5ed", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:24:52Z", "seed": 1265085923, "treatment": "gated"}, "priority": 2.0099913869935624, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "33dcfc4b7ecc3ef2a9ac9d0c74b58ec07680823d0d98a7ace4d40eb662b0168a", "run_seconds": 0.6915850639343262, "seed": 1265085923, "started_at": "2026-11-22T03:24:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001843, 0.023436], "effect": 0.010796, "evaluated_at": "2026-11-22T03:25:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-9d363dd0f5ed", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.684210526, 0.743589744, 0.536585366, 0.666666667], "gated": [0.684210526, 0.743589744, 0.536585366, 0.666666667], "naive": [0.666666667, 0.641025641, 0.463414634, 0.47008547], "scratch": [0.666666667, 0.717948718, 0.536585366, 0.666666667], "ungated": [0.719298246, 0.487179487, 0.536585366, 0.666666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "75034dffc64464c140c3edd3e93709172a157c1f64e5856e882b2a0055955661", "primary_control": "scratch", "result_hash": "33dcfc4b7ecc3ef2a9ac9d0c74b58ec07680823d0d98a7ace4d40eb662b0168a", "signature": "e5f790f6129cff0f8367909281aca47db35b4e98c5468a9009dd1fd19b55a531519b6690cc556843dda80950fc9722b422c615f9c955760571cb8c0b71493104", "state": "VERIFIED", "summary": {"compressed": 0.657763, "gated": 0.657763, "naive": 0.560298, "scratch": 0.646967, "ungated": 0.602432}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:25:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:11Z", "cpu_seconds": 0.4409173259999992, "created_at": "2026-11-22T03:25:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-91d44d23b086", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "b38bf56f0788be30b9d8b83f130a46b062acf0a052243d94efcc4b97d2c27bb0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-91d44d23b086", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:25:09Z", "seed": 1272289978, "treatment": "frontier"}, "priority": 2.0086444661303857, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f656794e5cc3443db40974686562169eb245beebd6d74677a0940aa6c781dcfd", "run_seconds": 0.6432571411132812, "seed": 1272289978, "started_at": "2026-11-22T03:26:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.031864, 0.187832], "effect": 0.109848, "evaluated_at": "2026-11-22T03:26:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-91d44d23b086", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.984848485, 1.0, 1.0, 0.96969697], "frontier": [0.833333333, 1.0, 1.0, 1.0], "goal": [0.651515152, 0.757575758, 0.893939394, 0.818181818], "hypdisc": [0.924242424, 0.848484848, 0.939393939, 0.803030303], "infogain": [0.833333333, 1.0, 1.0, 1.0], "prederr": [0.833333333, 1.0, 1.0, 1.0], "random": [0.833333333, 0.818181818, 0.893939394, 0.848484848], "risk": [0.833333333, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b38bf56f0788be30b9d8b83f130a46b062acf0a052243d94efcc4b97d2c27bb0", "primary_control": "random", "result_hash": "f656794e5cc3443db40974686562169eb245beebd6d74677a0940aa6c781dcfd", "signature": "f92ab5fba64fe899bdfead07f015ab6604422162219e6cee6a85d33173b58ec8c16633908d4ce49612418043edcc72ea0d07d03e9c1a5a3fbbf1980c3683cc09", "state": "VERIFIED", "summary": {"count": 0.988636, "frontier": 0.958333, "goal": 0.780303, "hypdisc": 0.878788, "infogain": 0.958333, "prederr": 0.958333, "random": 0.848485, "risk": 0.958333}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:26:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:17Z", "cpu_seconds": 1.7122526249999996, "created_at": "2026-11-22T03:25:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f860e9f914ec", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "11f72e462d13f22149597a2608af0b375679d1dcd007d3f50c715ab9bafb4902", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f860e9f914ec", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:25:09Z", "seed": 728634809, "treatment": "frontier"}, "priority": 2.00969819497477, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "124d9e4e221b452bb29cfc1135a903d292614db1c208cb4aba6934bf75d07f63", "run_seconds": 1.7520594596862793, "seed": 728634809, "started_at": "2026-11-22T03:25:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.027308, 0.48321], "effect": 0.255259, "evaluated_at": "2026-11-22T03:25:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f860e9f914ec", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.585185185, 0.782608696, 0.925, 0.713043478], "frontier": [1.0, 0.773913043, 1.0, 1.0], "goal": [0.940740741, 0.756521739, 1.0, 1.0], "hypdisc": [0.814814815, 0.713043478, 0.975, 0.773913043], "infogain": [1.0, 0.773913043, 1.0, 1.0], "prederr": [1.0, 0.773913043, 1.0, 1.0], "random": [0.474074074, 0.756521739, 0.8875, 0.634782609], "risk": [1.0, 0.773913043, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "11f72e462d13f22149597a2608af0b375679d1dcd007d3f50c715ab9bafb4902", "primary_control": "random", "result_hash": "124d9e4e221b452bb29cfc1135a903d292614db1c208cb4aba6934bf75d07f63", "signature": "1dd5626f07b0008c571a9e67e5541881d5eb2965816367dfdab99882c9600e86fb8659802b9ce2381400e625ac429edc7f011ee579a2035a682610236e3da70a", "state": "VERIFIED", "summary": {"count": 0.751459, "frontier": 0.943478, "goal": 0.924316, "hypdisc": 0.819193, "infogain": 0.943478, "prederr": 0.943478, "random": 0.68822, "risk": 0.943478}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:25:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:24Z", "cpu_seconds": 1.2782539579999987, "created_at": "2026-11-22T03:25:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f8e295b5d8bd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "f8cd0afdd80d66b337b87967128bc1de58e077d3c2e4a301cc681d6c0c52338b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f8e295b5d8bd", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:25:09Z", "seed": 1372775234, "treatment": "frontier"}, "priority": 2.0095113225321213, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b54790dffc69a439b2ea77908b31d0f9c7bf1e4d31febf00f9fc6c440a67f76a", "run_seconds": 1.3392627239227295, "seed": 1372775234, "started_at": "2026-11-22T03:25:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.102277, 0.234324], "effect": 0.168301, "evaluated_at": "2026-11-22T03:25:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f8e295b5d8bd", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.732026144, 0.823529412, 0.784313725, 0.797385621], "frontier": [0.901960784, 0.921568627, 0.888888889, 0.803921569], "goal": [0.777777778, 0.843137255, 0.790849673, 0.732026144], "hypdisc": [0.77124183, 0.843137255, 0.751633987, 0.732026144], "infogain": [0.901960784, 0.921568627, 0.888888889, 0.803921569], "prederr": [0.901960784, 0.921568627, 0.888888889, 0.803921569], "random": [0.692810458, 0.679738562, 0.777777778, 0.692810458], "risk": [0.901960784, 0.921568627, 0.888888889, 0.803921569]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f8cd0afdd80d66b337b87967128bc1de58e077d3c2e4a301cc681d6c0c52338b", "primary_control": "random", "result_hash": "b54790dffc69a439b2ea77908b31d0f9c7bf1e4d31febf00f9fc6c440a67f76a", "signature": "90f0897f2e449b6b98fcffcddf168b90dafdef995c61ade759ffe24501739b0b0499ae58444b64bd34b2050eb055e96b47a3b6b73603259248170c3d0c5ac80f", "state": "VERIFIED", "summary": {"count": 0.784314, "frontier": 0.879085, "goal": 0.785948, "hypdisc": 0.77451, "infogain": 0.879085, "prederr": 0.879085, "random": 0.710784, "risk": 0.879085}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:25:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:06Z", "cpu_seconds": 0.433781427, "created_at": "2026-11-22T03:25:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0fcb5056af4a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "c93a91fbc7304089656f5abd07a27741b8cd6ff4a4ff74b0e9fab74764e9dc0c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0fcb5056af4a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:25:09Z", "seed": 1984234070, "treatment": "full"}, "priority": 2.0079559413574604, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7f0ce4f32a830f6ec29f05e3113efff42bd035cc807d4fa21fcf9957a8acf0b4", "run_seconds": 1.3757143020629883, "seed": 1984234070, "started_at": "2026-11-22T03:30:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.097842, 0.154723], "effect": 0.126282, "evaluated_at": "2026-11-22T03:30:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0fcb5056af4a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.620155039, 0.555555556, 0.295454545, 0.529411765], "frozen": [0.108527132, 0.091503268, 0.098484848, 0.091503268], "full": [0.76744186, 0.692810458, 0.378787879, 0.666666667], "retrieval": [0.790697674, 0.784313725, 0.5, 0.718954248], "rules_nodetect": [0.511627907, 0.490196078, 0.174242424, 0.54248366]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "c93a91fbc7304089656f5abd07a27741b8cd6ff4a4ff74b0e9fab74764e9dc0c", "primary_control": "facts", "result_hash": "7f0ce4f32a830f6ec29f05e3113efff42bd035cc807d4fa21fcf9957a8acf0b4", "signature": "9716ba1268dcb65ac2652c71a2657f9d633ae9ebcbbc7e4898404f06c033948b37414d1dc4a4c6cd38e66370c1e99e46ce8c8729d6c589309050ec64df8fc90f", "state": "VERIFIED", "summary": {"facts": 0.500144, "frozen": 0.097505, "full": 0.626427, "retrieval": 0.698491, "rules_nodetect": 0.429638}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:30:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:29Z", "cpu_seconds": 0.3961483330000011, "created_at": "2026-11-22T03:25:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-30f17e04fc6b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "c0e38d5a72a96c345c5cf34e8ee6635ee55a3cd9a10d46bf707772d2334b1b58", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-30f17e04fc6b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:25:09Z", "seed": 661929835, "treatment": "full"}, "priority": 2.009077656020325, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "513888a05059b280263584300933cad28142694623742194f306e3fe3ecbcb13", "run_seconds": 0.5247170925140381, "seed": 661929835, "started_at": "2026-11-22T03:25:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00913, 0.028152], "effect": 0.009511, "evaluated_at": "2026-11-22T03:25:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-30f17e04fc6b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.994565217, 1.0, 1.0, 0.842391304], "frozen": [0.016304348, 0.016304348, 0.010869565, 0.016304348], "full": [0.994565217, 1.0, 1.0, 0.880434783], "retrieval": [0.983695652, 1.0, 1.0, 0.880434783], "rules_nodetect": [0.994565217, 1.0, 1.0, 0.880434783]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "c0e38d5a72a96c345c5cf34e8ee6635ee55a3cd9a10d46bf707772d2334b1b58", "primary_control": "facts", "result_hash": "513888a05059b280263584300933cad28142694623742194f306e3fe3ecbcb13", "signature": "c87ba12b68fc95e2b3b9681acbf33c8d6a0a25503e9390c49dc2a78efdff22a715098e7266904ce907681c46db76a3e707545554fc3c701ad38f090e6b1da60e", "state": "VERIFIED", "summary": {"facts": 0.959239, "frozen": 0.014946, "full": 0.96875, "retrieval": 0.966033, "rules_nodetect": 0.96875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:25:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:15Z", "cpu_seconds": 1.5406420830000003, "created_at": "2026-11-22T03:25:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-bcc65026d33f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 3, "n": 4}, "prereg_hash": "aea4d4a044b90bca50a2160d0db99d27ce255b8660a54a00dc9a2952fcdd59a8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bcc65026d33f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:25:09Z", "seed": 1551327693, "treatment": "full"}, "priority": 2.008382554939591, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a1d0564a34644629f0fb7739c04279078e0827665a0bbe9c985951947387401b", "run_seconds": 2.3273136615753174, "seed": 1551327693, "started_at": "2026-11-22T03:26:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014016, 0.231561], "effect": 0.108773, "evaluated_at": "2026-11-22T03:26:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-bcc65026d33f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.609448819, 0.996825397, 0.994285714, 0.404188482], "frozen": [0.023622047, 0.111111111, 0.034285714, 0.016753927], "full": [0.801574803, 1.0, 0.994285714, 0.643979058], "retrieval": [0.784251969, 1.0, 0.994285714, 0.661780105], "rules_nodetect": [0.801574803, 1.0, 0.994285714, 0.643979058]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "aea4d4a044b90bca50a2160d0db99d27ce255b8660a54a00dc9a2952fcdd59a8", "primary_control": "facts", "result_hash": "a1d0564a34644629f0fb7739c04279078e0827665a0bbe9c985951947387401b", "signature": "93bffbf1b4c7d00c7352f54296b5a143bbf391b3b574e41181ddd6e461c20c6739ec6c1c898c7183df39d8f276ab1b54cfc23ae8fd254304086592f8c61ca308", "state": "VERIFIED", "summary": {"facts": 0.751187, "frozen": 0.046443, "full": 0.85996, "retrieval": 0.860079, "rules_nodetect": 0.85996}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:26:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:29Z", "cpu_seconds": 0.981861353, "created_at": "2026-11-22T03:25:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-01b290438d89", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "7317cad6b29650419623a9c164c301f00ae9cc9fa990cbdbafe86493ede3a7af", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-01b290438d89", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:25:25Z", "seed": 3898804, "treatment": "full"}, "priority": 2.0095397498894876, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9687e08cce00bc63f752bd565293e8563000b3694757e5509d3aa4736ea63610", "run_seconds": 1.149742603302002, "seed": 3898804, "started_at": "2026-11-22T03:25:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.081115, 0.114909], "effect": 0.016897, "evaluated_at": "2026-11-22T03:25:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-01b290438d89", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.916666667, 0.772727273, 0.630434783], "frozen": [0.325, 0.35, 0.386363636, 0.244565217], "full": [1.0, 0.866666667, 0.727272727, 0.793478261], "retrieval": [1.0, 0.916666667, 0.840909091, 0.85326087], "rules_nodetect": [1.0, 0.866666667, 0.727272727, 0.793478261]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "7317cad6b29650419623a9c164c301f00ae9cc9fa990cbdbafe86493ede3a7af", "primary_control": "facts", "result_hash": "9687e08cce00bc63f752bd565293e8563000b3694757e5509d3aa4736ea63610", "signature": "9471cd51ab6884c8d9a6539e4ad7181c58aeb00510303680cbcf1ef5944cf2adc1aefba8f26235cb8f9d6adad1d541310001cddce77d7b03607c09e536d22905", "state": "VERIFIED", "summary": {"facts": 0.829957, "frozen": 0.326482, "full": 0.846854, "retrieval": 0.902709, "rules_nodetect": 0.846854}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:25:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:25:28Z", "cpu_seconds": 0.701936514, "created_at": "2026-11-22T03:25:25Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d17342c917fc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "b0b66a1c0302aefe34f5576a1098240af3aed653be4a586bcfada107cb7b3c88", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d17342c917fc", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:25:25Z", "seed": 1046862363, "treatment": "gated"}, "priority": 2.0099354324328327, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "552dcaf1952b3af2a5a41b4bf7449e8652bd91e21e6ba87886133a2fc304aa94", "run_seconds": 0.736868143081665, "seed": 1046862363, "started_at": "2026-11-22T03:25:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.066582, 0.010423], "effect": -0.02808, "evaluated_at": "2026-11-22T03:25:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d17342c917fc", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.642512077, 0.608187135, 0.593137255, 0.75], "gated": [0.642512077, 0.608187135, 0.593137255, 0.75], "naive": [0.719806763, 0.608187135, 0.681372549, 0.833333333], "scratch": [0.671497585, 0.608187135, 0.593137255, 0.833333333], "ungated": [0.362318841, 0.608187135, 0.529411765, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "b0b66a1c0302aefe34f5576a1098240af3aed653be4a586bcfada107cb7b3c88", "primary_control": "scratch", "result_hash": "552dcaf1952b3af2a5a41b4bf7449e8652bd91e21e6ba87886133a2fc304aa94", "signature": "e8303f2043933008a11a899af28feaa3f683d08a3ccd103d044023122cfc83314346fb2115657a0e7d5b691f98e26480e49a2e982ffe754ab8c8e79c10ee1501", "state": "VERIFIED", "summary": {"compressed": 0.648459, "gated": 0.648459, "naive": 0.710675, "scratch": 0.676539, "ungated": 0.583313}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:25:53Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:27:16Z", "cpu_seconds": 4.297953413, "created_at": "2026-11-22T03:25:42Z", "error": "lease expired; ", "family": "ow_explore", "fence": 2, "id": "E141-ow_explore-eee1ffb51c52", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 5, "n": 4}, "prereg_hash": "4e619d90428fc30f29104aa113ed2285ee49ec83b60afd2fc160d23c34de9cbe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-eee1ffb51c52", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:25:42Z", "seed": 327829485, "treatment": "frontier"}, "priority": 2.009821544926459, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "bbcd4ea4bb7eeeaab0090f35ccd0ce1009a9945de440d7ba084a4876d51b21bf", "run_seconds": 5.823341369628906, "seed": 327829485, "started_at": "2026-11-22T03:25:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.163209, 0.129331], "effect": -0.016939, "evaluated_at": "2026-11-22T03:27:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-eee1ffb51c52", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.873809524, 1.0, 0.395833333, 1.0], "frontier": [1.0, 0.777777778, 0.404166667, 0.991666667], "goal": [0.919047619, 0.488888889, 0.329166667, 1.0], "hypdisc": [0.921428571, 0.955555556, 0.35, 0.991666667], "infogain": [1.0, 0.777777778, 0.410416667, 0.991666667], "prederr": [1.0, 0.777777778, 0.416666667, 0.991666667], "random": [0.864285714, 1.0, 0.39375, 0.983333333], "risk": [1.0, 0.777777778, 0.404166667, 0.991666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "4e619d90428fc30f29104aa113ed2285ee49ec83b60afd2fc160d23c34de9cbe", "primary_control": "random", "result_hash": "bbcd4ea4bb7eeeaab0090f35ccd0ce1009a9945de440d7ba084a4876d51b21bf", "signature": "5106908d91c5e35158b9522a320912b09113a9a8a43b38461ad65d79f77b12655deb88358e4ca3c40782a8961b52a678aadeb536c443be57baca664439e2590a", "state": "VERIFIED", "summary": {"count": 0.817411, "frontier": 0.793403, "goal": 0.684276, "hypdisc": 0.804663, "infogain": 0.794965, "prederr": 0.796528, "random": 0.810342, "risk": 0.793403}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:27:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:06Z", "cpu_seconds": 2.1563489760000003, "created_at": "2026-11-22T03:25:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-071e8c90d1f4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "ed461d9e723299ed2acd253e4c84cb86940bc015e63ce012226e035a38e5195f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-071e8c90d1f4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:25:59Z", "seed": 500992801, "treatment": "full"}, "priority": 2.0095058530500305, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ca5a15bb01982cb4fad2cb1c6de474c7e78c349c6cbc8d1d8d351560a3a70ad2", "run_seconds": 3.0624217987060547, "seed": 500992801, "started_at": "2026-11-22T03:26:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.096482, 0.596482], "effect": 0.25, "evaluated_at": "2026-11-22T03:26:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-071e8c90d1f4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 0.25], "fixed_policy": [0.25, 0.25, 0.5, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 0.25, 1.0, 0.25], "random": [0.75, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "ed461d9e723299ed2acd253e4c84cb86940bc015e63ce012226e035a38e5195f", "primary_control": "e141_best", "result_hash": "ca5a15bb01982cb4fad2cb1c6de474c7e78c349c6cbc8d1d8d351560a3a70ad2", "signature": "4e8b7835b160929d5d954e0ff37e63bcefbc20a4e2a558953eb6a565187497ba32d2df6891a59961bfeb99ec952735d39649f3fb1b91c29402eb682b802a880f", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 0.625, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:26:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:09Z", "cpu_seconds": 0.21248243199999994, "created_at": "2026-11-22T03:25:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7fc25b88b0f8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "68fe85ac19231edaca6f40a73c2bc5b192cac7ac18cd515dbc4e3330ca3aaff3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7fc25b88b0f8", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:25:59Z", "seed": 1875710464, "treatment": "full"}, "priority": 2.009411020627136, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "39627eb150a17b4242e5dc875a1ecdd4defcba62baa90e89b350e2d69438e5ae", "run_seconds": 0.3521726131439209, "seed": 1875710464, "started_at": "2026-11-22T03:26:09Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:26:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7fc25b88b0f8", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.25, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "68fe85ac19231edaca6f40a73c2bc5b192cac7ac18cd515dbc4e3330ca3aaff3", "primary_control": "e141_best", "result_hash": "39627eb150a17b4242e5dc875a1ecdd4defcba62baa90e89b350e2d69438e5ae", "signature": "0e71f56c59ee0bfdb848f66bb1f5dd0a71b5348913224142edeec3b3f85e5524ad53850be0700c036793a80ff46f50f54bf765243428b3661904bd5bec3df60c", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:26:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:04Z", "cpu_seconds": 0.191117865, "created_at": "2026-11-22T03:25:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b0c1b8f4d0ae", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "4afbb74fccbaae33c2f8902c21b6ff94f9f2fccedf6adade913121410d38e915", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b0c1b8f4d0ae", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:25:59Z", "seed": 1729216999, "treatment": "full"}, "priority": 2.009715322545538, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b538c07b96b1587072b30df13e9a7d0bef2b297e4d79e8eb1bfc6988a4f4f075", "run_seconds": 0.2319340705871582, "seed": 1729216999, "started_at": "2026-11-22T03:26:03Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:26:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b0c1b8f4d0ae", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.75, 0.5, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4afbb74fccbaae33c2f8902c21b6ff94f9f2fccedf6adade913121410d38e915", "primary_control": "e141_best", "result_hash": "b538c07b96b1587072b30df13e9a7d0bef2b297e4d79e8eb1bfc6988a4f4f075", "signature": "f7a4e1be75520549cb796b30269d91d4adc25e4ba37e316790dcd7da51b194d70abdd240f6d76aaf94e511e9e85775e68e654a7309d30106d1ce864494065505", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:26:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:15Z", "cpu_seconds": 8.602638214, "created_at": "2026-11-22T03:25:59Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a7d4fef3051b", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "2a16058ece9d8f1bc603e16958fb893a85af26b277fe7ea3ca44ba001ac2c9aa", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a7d4fef3051b", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:25:59Z", "seed": 2012038618, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "d526201d967e643509b60d5d5a671c89bdcccd67da7ee35935e78706915f4516", "run_seconds": 11.828159809112549, "seed": 2012038618, "started_at": "2026-11-22T03:26:03Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:26:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a7d4fef3051b", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "2a16058ece9d8f1bc603e16958fb893a85af26b277fe7ea3ca44ba001ac2c9aa", "primary_control": "reset", "result_hash": "d526201d967e643509b60d5d5a671c89bdcccd67da7ee35935e78706915f4516", "signature": "ca802eba14d9118348847ccc24ff03cc012a57e85359e9ba75b6225bab3d6dab38d6de418002f74bbf9927168cdaef7c3ac24e1ae62574f9056ba419c7b5e000", "state": "PARTIAL", "summary": {"inherit": 0.5666666666666667, "reset": 0.5666666666666667, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:26:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:18Z", "cpu_seconds": 1.6325580429999995, "created_at": "2026-11-22T03:26:15Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-3497a644d062", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "prereg_hash": "30395ac21388e2556734d35af8d9cad27f89f7ac94b1068fe5de5275be49b51d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-3497a644d062", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.15}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:26:15Z", "seed": 818909080, "treatment": "iw_temperature"}, "priority": 2.009939961528928, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-64dacb30eb0e", "result_hash": "04e1e2d46f388892a45f9efdad09083ae4abe1862e1d2453a07681321291195c", "run_seconds": 2.15399169921875, "seed": 818909080, "started_at": "2026-11-22T03:26:16Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021787, -0.005342], "effect": -0.013565, "evaluated_at": "2026-11-22T03:26:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-3497a644d062", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.102894158, 0.047289935, 0.125925395, 0.1151632, 0.133697682, 0.167404585, 0.16513792, 0.101954895, 0.142084897, 0.170821302, 0.162150367, 0.123373353, 0.15252753, 0.145133435, 0.113865283, 0.138964665], "raw": [0.090292617, 0.076731945, 0.087470707, 0.09927484, 0.09106866, 0.112158312, 0.154589075, 0.123190837, 0.09792378, 0.105243173, 0.07579479, 0.075762325, 0.099570435, 0.109033903, 0.083232937, 0.12445829], "temperature": [0.106553285, 0.050891738, 0.100565857, 0.1151632, 0.11571186, 0.15303362, 0.16513792, 0.123190837, 0.114824828, 0.134524107, 0.129069587, 0.097311408, 0.1216611, 0.145133435, 0.088250582, 0.130333107]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0049, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.989", "prereg_hash": "30395ac21388e2556734d35af8d9cad27f89f7ac94b1068fe5de5275be49b51d", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-64dacb30eb0e", "result_hash": "04e1e2d46f388892a45f9efdad09083ae4abe1862e1d2453a07681321291195c", "signature": "a9704e2ae25a53eef08f90c0cf4008bddf4329a0ea7ac1fd0875bfd0d7d7f21d8b6e791b34d4802fbed1adc391d9e081e8576144e64bc8ea0cdb001e6edd7000", "state": "VERIFIED", "summary": {"iw_temperature": 0.131774, "raw": 0.100362, "temperature": 0.11821}, "ties": 3, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:26:29Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:26:15Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_explore", "fence": 3, "id": "E141-ow_explore-e1c45e3ee844", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 3, "n": 4}, "prereg_hash": "13d2ecca62bf7e06b5e859ae3063ed1b77903e13341e4bf13514866ccca691a1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e1c45e3ee844", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:26:15Z", "seed": 948975428, "treatment": "frontier"}, "priority": 2.0095124700229516, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 948975428, "started_at": "2026-11-22T03:26:25Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:26:54Z", "cpu_seconds": 0.4257516270000001, "created_at": "2026-11-22T03:26:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-699fa7388b44", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "65b795aa2621e7076b0bbf0c9ab2b062494caeb26cccd767111204469e81cf5c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-699fa7388b44", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:26:15Z", "seed": 445272714, "treatment": "full"}, "priority": 2.0093622906759987, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6358235f88774abb4e399da9d6d612ab417104ff0d9d19dfec6fbfc4c7c926a9", "run_seconds": 0.6244721412658691, "seed": 445272714, "started_at": "2026-11-22T03:26:53Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:26:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-699fa7388b44", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.75, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "65b795aa2621e7076b0bbf0c9ab2b062494caeb26cccd767111204469e81cf5c", "primary_control": "e141_best", "result_hash": "6358235f88774abb4e399da9d6d612ab417104ff0d9d19dfec6fbfc4c7c926a9", "signature": "f9abb4b2ad7a6a0fb05b478b2d2f33912aa04b4b341cf5f4d205e066686cb3f8771f9f2aff7146eaa518651a349c662e6663c9533c32c9de35d0acb34ba2950b", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:26:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:24Z", "cpu_seconds": 0.6468459800000002, "created_at": "2026-11-22T03:26:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f6630f74ab00", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "48e61cce122a01d23228344b1a083527f5159026a486e5de939d098ddc9cca7c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f6630f74ab00", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:26:15Z", "seed": 951457852, "treatment": "full"}, "priority": 2.0095704903278713, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6d4ec87470e85e67e82f68a922fdc0764bdea592bbcf5baedf6fd73c054b258b", "run_seconds": 0.7075638771057129, "seed": 951457852, "started_at": "2026-11-22T03:26:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.532902, 0.032902], "effect": -0.25, "evaluated_at": "2026-11-22T03:26:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f6630f74ab00", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 1.0, 0.5], "fixed_policy": [0.5, 1.0, 0.0, 0.5], "full": [1.0, 0.0, 0.5, 0.0], "no_world_model": [1.0, 0.0, 0.0, 0.0], "random": [0.75, 0.0, 1.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "48e61cce122a01d23228344b1a083527f5159026a486e5de939d098ddc9cca7c", "primary_control": "e141_best", "result_hash": "6d4ec87470e85e67e82f68a922fdc0764bdea592bbcf5baedf6fd73c054b258b", "signature": "c17fc5ce44630c04dfee28fb992593ef90ec63ca084db70a7fc819c74461a49f52b199a7f67ea0853a66e7f15a2572e9333905179471baad535eca3477835302", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.5, "full": 0.375, "no_world_model": 0.25, "random": 0.5}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:26:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:33Z", "cpu_seconds": 2.9651982500000003, "created_at": "2026-11-22T03:26:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-09d4ef80e35d", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "e5ab8b8c37dc45e86cf169e360e5fa0d3441f127fe3197be045f990034793c3d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-09d4ef80e35d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:26:15Z", "seed": 363815257, "treatment": "gated"}, "priority": 2.00937094490915, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9b1950e77929baf6bac035c707d6f6e8567d0adebf1176f04e73e4c2874caea2", "run_seconds": 3.5382909774780273, "seed": 363815257, "started_at": "2026-11-22T03:26:29Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:26:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-09d4ef80e35d", "family": "ow_transfer", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"compressed": [1.0, 0.478412], "gated": [1.0, 0.478412], "naive": [1.0, 0.630032], "scratch": [1.0, 0.495239], "ungated": [1.0, 0.566001]}, "prereg_hash": "e5ab8b8c37dc45e86cf169e360e5fa0d3441f127fe3197be045f990034793c3d", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "9b1950e77929baf6bac035c707d6f6e8567d0adebf1176f04e73e4c2874caea2", "signature": "f459b2a0533e6de3013f5a00751c518ad75018bf43ab64a3d6b47657e4404b2d75dee5c1810d14878bd6bef79a701d86a4639b49594be00bc1049acc6cbabb01", "state": "REJECTED", "summary": {"compressed": 0.478412, "gated": 0.478412, "naive": 0.630032, "scratch": 0.495239, "ungated": 0.566001}}, "verified_at": "2026-11-22T03:26:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:20Z", "cpu_seconds": 0.4767683439999999, "created_at": "2026-11-22T03:26:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d6c965d0522e", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "prereg_hash": "7f9c10c0326533d03c962334ff56be1caf37c73f28926f1d4b74cb76213fff8d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d6c965d0522e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:26:15Z", "seed": 306404698, "treatment": "gated"}, "priority": 2.009677345317012, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "a4dac6d9fb7552e17da453e354739a13e3c7d45a4ea47601fb5cdaf34187aa8f", "run_seconds": 0.575563907623291, "seed": 306404698, "started_at": "2026-11-22T03:26:20Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:26:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d6c965d0522e", "family": "ow_transfer", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"naive": [1.0, 0.597491], "ungated": [1.0, 0.689964]}, "prereg_hash": "7f9c10c0326533d03c962334ff56be1caf37c73f28926f1d4b74cb76213fff8d", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "a4dac6d9fb7552e17da453e354739a13e3c7d45a4ea47601fb5cdaf34187aa8f", "signature": "0a593274d03155f085577a01afe5a138339ab59ad4910f2702427b7e1c5b1ce95e2b35fc8b6d7b2a9fdd6fe54b70652227a8f103d37ac4115546cca6d4f3360f", "state": "REJECTED", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.597491, "scratch": 1.0, "ungated": 0.689964}}, "verified_at": "2026-11-22T03:26:29Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:26:32Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_explore", "fence": 3, "id": "E141-ow_explore-23e6001dc913", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 7, "n": 4}, "prereg_hash": "1f629e830b27eb209f6390929b767e155c9cd03526152efc28eff87f8dc573ad", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-23e6001dc913", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:26:32Z", "seed": 467007539, "treatment": "frontier"}, "priority": 2.009702705485772, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 467007539, "started_at": "2026-11-22T03:26:53Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:26:59Z", "cpu_seconds": 5.500688079, "created_at": "2026-11-22T03:26:32Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4201e55abb62", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 3, "n": 4}, "prereg_hash": "22471cac4547b61100f0d04a8c4493c3dd664cc5afa71601b05b2d6b71cc91b2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4201e55abb62", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:26:32Z", "seed": 747474904, "treatment": "frontier"}, "priority": 2.0098801746910744, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ecea5b94514f28618102b26a21d44ef50ff60e486f62d8360117f9cbf2ef97b8", "run_seconds": 6.177821636199951, "seed": 747474904, "started_at": "2026-11-22T03:26:53Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:27:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4201e55abb62", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "22471cac4547b61100f0d04a8c4493c3dd664cc5afa71601b05b2d6b71cc91b2", "primary_control": "random", "result_hash": "ecea5b94514f28618102b26a21d44ef50ff60e486f62d8360117f9cbf2ef97b8", "signature": "d3ed68c09b1edb9b1917aeac920cc98fe16f0cdf12405fa99ee56fd9eff81ec2b2af4e481949d96f90cc78cdc51804d55e1010ba180d4cb9e00ec1b81b258608", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:27:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:02Z", "cpu_seconds": 2.345881774, "created_at": "2026-11-22T03:26:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ac3d4ef1b639", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "17a605ccc2ecfdb12e478a56ae10b26e23c0c763190875ce3fb29be7814b68ab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ac3d4ef1b639", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:26:32Z", "seed": 1072948444, "treatment": "full"}, "priority": 2.0086713938932585, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "46ca9bf1004525e1a6718849fa413435a98150e5cbdf0734e9a21ea34de18eec", "run_seconds": 2.9128055572509766, "seed": 1072948444, "started_at": "2026-11-22T03:26:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:27:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ac3d4ef1b639", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "17a605ccc2ecfdb12e478a56ae10b26e23c0c763190875ce3fb29be7814b68ab", "primary_control": "e141_best", "result_hash": "46ca9bf1004525e1a6718849fa413435a98150e5cbdf0734e9a21ea34de18eec", "signature": "a1c240721cb3cd8ab27c628b086cc6e5e45210c72a0e441dcf23f6bc9fa1e2b63f5aa89d5bf416ffea61a2ace5120a89424ce6d82fbcea0dcd7203703aa9380a", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:27:06Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:28:00Z", "cpu_seconds": 6.630273367, "created_at": "2026-11-22T03:26:32Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-2fc7379ccc96", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6208a45887cbd9fdef9b83e8e253e2d92f784764ea0eb96eb15d212d86f91919", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-2fc7379ccc96", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:26:32Z", "seed": 282028062, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "821f4c17d8627565234f0131bb8a5b0f56e170bd9bb88dc9d68ed142a16a245a", "run_seconds": 7.730878591537476, "seed": 282028062, "started_at": "2026-11-22T03:26:38Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [0.055819, 0.310848], "effect": 0.183333, "evaluated_at": "2026-11-22T03:28:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-2fc7379ccc96", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5753623568468522, "rstep": 2}, "hypothesis_supported": true, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0], "root": [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -2.1963, "losses": 3, "n": 60, "outcome": "SUPPORTED", "p_sign_display": "0.00636", "prereg_hash": "6208a45887cbd9fdef9b83e8e253e2d92f784764ea0eb96eb15d212d86f91919", "primary_control": "reset", "result_hash": "821f4c17d8627565234f0131bb8a5b0f56e170bd9bb88dc9d68ed142a16a245a", "signature": "8492cbbb1b6f3c9c5c21aa4e13201665e8e3f19b907322d42af6d4f53c0aa0212cdb43d93a59d8daf39659c4b5f5a505efe18bf9b2c6819ef39cbba946a62200", "state": "VERIFIED", "summary": {"inherit": 0.7666666666666667, "reset": 0.5833333333333334, "root": 0.35}, "ties": 43, "treatment": "inherit", "wins": 14}, "verified_at": "2026-11-22T03:28:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:26:48Z", "cpu_seconds": 8.734610247, "created_at": "2026-11-22T03:26:32Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-9a3a22925e5e", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e09e76591f3703056076bbb3c97667a1998eaa44820c70a4b8757906a9d53745", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-9a3a22925e5e", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:26:32Z", "seed": 202334244, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "aaf13784ae152c4dfce790980f42b84fb23b41246cefd2ee6c4bb4942e93991b", "run_seconds": 13.478922367095947, "seed": 202334244, "started_at": "2026-11-22T03:26:34Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:26:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-9a3a22925e5e", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e09e76591f3703056076bbb3c97667a1998eaa44820c70a4b8757906a9d53745", "primary_control": "reset", "result_hash": "aaf13784ae152c4dfce790980f42b84fb23b41246cefd2ee6c4bb4942e93991b", "signature": "5e33f9df8ce3d1ff6be585f501c54c268126ed31a1b95be90315c01b210dc63e8045efccd0c336898e71dba6465961a0e88876d6f59cad5642793d537b13ec0e", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.5}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:26:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:22Z", "cpu_seconds": 1.6019744360000008, "created_at": "2026-11-22T03:27:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a50d25b5fc07", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 4, "n": 4}, "prereg_hash": "60865c07687c8f543ff7c2fbe15f6750634999ac5c663d2a5b51f4eb72970fc8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a50d25b5fc07", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:27:05Z", "seed": 1752760900, "treatment": "full"}, "priority": 2.0088018654124085, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "44a094ba89eed77b143f380186789fca64ba3883679f936a43a137b7d4b841b7", "run_seconds": 1.7842600345611572, "seed": 1752760900, "started_at": "2026-11-22T03:27:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.036333, 0.100459], "effect": 0.032063, "evaluated_at": "2026-11-22T03:27:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a50d25b5fc07", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.663414634, 1.0, 0.84], "frozen": [0.366666667, 0.104878049, 0.0, 0.0], "full": [0.991666667, 0.8, 1.0, 0.84], "retrieval": [0.991666667, 0.848780488, 1.0, 0.944], "rules_nodetect": [0.991666667, 0.8, 1.0, 0.84]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "60865c07687c8f543ff7c2fbe15f6750634999ac5c663d2a5b51f4eb72970fc8", "primary_control": "facts", "result_hash": "44a094ba89eed77b143f380186789fca64ba3883679f936a43a137b7d4b841b7", "signature": "eb512090be4efb581228058426528cd724d6429e05899fcbdfebf4301d65170da99e62befe0cabf47e041648e868c8236e9d1456ee8a47163f7e86ddf7f1bc04", "state": "VERIFIED", "summary": {"facts": 0.875854, "frozen": 0.117886, "full": 0.907917, "retrieval": 0.946112, "rules_nodetect": 0.907917}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:27:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:23Z", "cpu_seconds": 2.9058178209999994, "created_at": "2026-11-22T03:27:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e4c7de56605c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 3, "n": 4}, "prereg_hash": "4c3f13ac6684188568d4a968d90d16e127d02c75adb5b82a7ec359cd6b19b622", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e4c7de56605c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:27:05Z", "seed": 595269172, "treatment": "full"}, "priority": 2.0088861489082155, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b33b189cc5ae5799946e6460464ca94a24078ff1963d039c1893e3a7598507ac", "run_seconds": 3.192328453063965, "seed": 595269172, "started_at": "2026-11-22T03:27:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.123418, 0.200337], "effect": 0.161877, "evaluated_at": "2026-11-22T03:27:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e4c7de56605c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.300766284, 0.561302682, 0.530651341, 0.517241379], "frozen": [0.095785441, 0.084291188, 0.093869732, 0.093869732], "full": [0.511494253, 0.72605364, 0.687739464, 0.632183908], "retrieval": [0.555555556, 0.791187739, 0.720306513, 0.714559387], "rules_nodetect": [0.511494253, 0.72605364, 0.687739464, 0.632183908]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "4c3f13ac6684188568d4a968d90d16e127d02c75adb5b82a7ec359cd6b19b622", "primary_control": "facts", "result_hash": "b33b189cc5ae5799946e6460464ca94a24078ff1963d039c1893e3a7598507ac", "signature": "65ca4021df503d99f38a63b02c6f2f53cf8bbb76b53dfbb7077f7462d9af8478611fbc71060f101259757dbb173d1411f09a941b91db7bf6df2b35b6128ff20f", "state": "VERIFIED", "summary": {"facts": 0.47749, "frozen": 0.091954, "full": 0.639368, "retrieval": 0.695402, "rules_nodetect": 0.639368}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:27:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:15Z", "cpu_seconds": 6.995350972000001, "created_at": "2026-11-22T03:27:05Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1e20d9c1c633", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "203a5ad184e61b79c6fb812429b1d01aa300e8bf7411dbb0b4dbc83b6807a8d9", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1e20d9c1c633", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:27:05Z", "seed": 193295509, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "3bc72ec910f2251ccfdee7164662aa710440e273657289193b30361b559d0e73", "run_seconds": 8.530171632766724, "seed": 193295509, "started_at": "2026-11-22T03:27:07Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 1, "reset": 2}, "better": "higher", "ci95": [-0.341258, -0.125408], "effect": -0.233333, "evaluated_at": "2026-11-22T03:27:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1e20d9c1c633", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 14, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "203a5ad184e61b79c6fb812429b1d01aa300e8bf7411dbb0b4dbc83b6807a8d9", "primary_control": "reset", "result_hash": "3bc72ec910f2251ccfdee7164662aa710440e273657289193b30361b559d0e73", "signature": "927f79b0da979170e2fe18d725e3895f81ac68fc3a5474462076a92a3d48e5b7827bdf8897dc657e5166feefa0bf41d692f73cf9e684546a5dcf4e99814aea03", "state": "VERIFIED", "summary": {"inherit": 0.48333333333333334, "reset": 0.7166666666666667, "root": 0.3333333333333333}, "ties": 46, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:27:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:28Z", "cpu_seconds": 0.4764065120000005, "created_at": "2026-11-22T03:27:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-94b80a1c9de1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "4d921237dd93bc51c935988f4667e720be6cd0cc3525994c398983fa72a3b860", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-94b80a1c9de1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:27:22Z", "seed": 1551856099, "treatment": "full"}, "priority": 2.0099101757357407, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "505459714907c51fce21961a44abd19fe869092f6706e125c36580c3a48fc9eb", "run_seconds": 0.4800379276275635, "seed": 1551856099, "started_at": "2026-11-22T03:27:27Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:27:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-94b80a1c9de1", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4d921237dd93bc51c935988f4667e720be6cd0cc3525994c398983fa72a3b860", "primary_control": "e141_best", "result_hash": "505459714907c51fce21961a44abd19fe869092f6706e125c36580c3a48fc9eb", "signature": "13cd78ffc1ccb4a5a7581b2c020cb9c4fbece2ca383556d44d4ae7a140405c3962d61f071fef03123b9b5fbd608b222333f5aac61b8dd156a7817868470d350b", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:27:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:29Z", "cpu_seconds": 0.42432000400000014, "created_at": "2026-11-22T03:27:22Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6bd3295c970f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "b14c2921eae9bb3cde235827eeec0d84b096e4215a2a2451e576262cb53933e3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6bd3295c970f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:27:22Z", "seed": 1972556954, "treatment": "full"}, "priority": 2.0096229927587888, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "79b7fdd34420486ddfae965cb89d6ac17d8928ac97a777f4409f676bb36b57e0", "run_seconds": 0.42978477478027344, "seed": 1972556954, "started_at": "2026-11-22T03:27:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.018166, 0.219632], "effect": 0.118899, "evaluated_at": "2026-11-22T03:27:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6bd3295c970f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.910714286, 0.8, 1.0, 0.666666667], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.982142857, 0.975, 1.0, 0.895833333], "retrieval": [0.946428571, 0.9, 1.0, 0.854166667], "rules_nodetect": [0.982142857, 0.975, 1.0, 0.895833333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b14c2921eae9bb3cde235827eeec0d84b096e4215a2a2451e576262cb53933e3", "primary_control": "facts", "result_hash": "79b7fdd34420486ddfae965cb89d6ac17d8928ac97a777f4409f676bb36b57e0", "signature": "e401fa5a01c81f95d9c9526b7f2a100753d008590a75af166f6f030981c128e5535bf8a4cbe671e15e5f22457649a2f5f24d5758fc10da2c7f209ee66455b106", "state": "VERIFIED", "summary": {"facts": 0.844345, "frozen": 0.0, "full": 0.963244, "retrieval": 0.925149, "rules_nodetect": 0.963244}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:27:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:34Z", "cpu_seconds": 0.9096342149999996, "created_at": "2026-11-22T03:27:22Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-6bc75240bf73", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "prereg_hash": "7cd45fffcb5e26c039a593b4f1442c1854f87728e6543e6f499c1080d68d6510", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-6bc75240bf73", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:27:22Z", "seed": 1481063367, "treatment": "gated"}, "priority": 2.0095065816624187, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "2ed2398800a6c76eecade38471fcc22de03bf6374b94cde795a8534aa80122bb", "run_seconds": 1.2097110748291016, "seed": 1481063367, "started_at": "2026-11-22T03:27:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018387, 0.009177], "effect": -0.004605, "evaluated_at": "2026-11-22T03:27:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-6bc75240bf73", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.195121951, 0.09787234, 0.095744681, 0.202857143], "gated": [0.195121951, 0.09787234, 0.095744681, 0.202857143], "naive": [0.124390244, 0.129787234, 0.123404255, 0.168571429], "scratch": [0.195121951, 0.123404255, 0.091489362, 0.2], "ungated": [0.124390244, 0.172340426, 0.123404255, 0.18]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7cd45fffcb5e26c039a593b4f1442c1854f87728e6543e6f499c1080d68d6510", "primary_control": "scratch", "result_hash": "2ed2398800a6c76eecade38471fcc22de03bf6374b94cde795a8534aa80122bb", "signature": "0c8cb10177b4017e887d60ccf7a31745295fb286e455033bd96b59580a514744561232c1bde2dafc08c6d719f556a1808d83d2077447f6845ed8e387012f2c0c", "state": "VERIFIED", "summary": {"compressed": 0.147899, "gated": 0.147899, "naive": 0.136538, "scratch": 0.152504, "ungated": 0.150034}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:27:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:34Z", "cpu_seconds": 0.18395153599999858, "created_at": "2026-11-22T03:27:22Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-830ac39d7afe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "prereg_hash": "1075225703897d0baee062aa9428a8f5d4287637c573e25178d6bbc40c879aa6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-830ac39d7afe", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:27:22Z", "seed": 1633959545, "treatment": "gated"}, "priority": 2.00942884535975, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7289fcd11c118cdf5629344d5df8ead00e4ef217d038238da05a43afc5b6c3a3", "run_seconds": 0.36215758323669434, "seed": 1633959545, "started_at": "2026-11-22T03:27:34Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:27:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-830ac39d7afe", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.409090909, 0.621212121, 0.212121212, 0.212121212], "gated": [0.409090909, 0.621212121, 0.212121212, 0.212121212], "naive": [0.409090909, 0.621212121, 0.212121212, 0.212121212], "scratch": [0.409090909, 0.621212121, 0.212121212, 0.212121212], "ungated": [0.409090909, 0.621212121, 0.212121212, 0.212121212]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1075225703897d0baee062aa9428a8f5d4287637c573e25178d6bbc40c879aa6", "primary_control": "scratch", "result_hash": "7289fcd11c118cdf5629344d5df8ead00e4ef217d038238da05a43afc5b6c3a3", "signature": "84f9f2d03ef4fc4d801ace2bf24056c10c2e421992d1e10d73def97b037e1521bb3e6f864a84b18e886c93c97581b01cc0b8555094f084ec1aefcfa901a9ad0f", "state": "PARTIAL", "summary": {"compressed": 0.363636, "gated": 0.363636, "naive": 0.363636, "scratch": 0.363636, "ungated": 0.363636}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:27:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:57Z", "cpu_seconds": 1.083065392, "created_at": "2026-11-22T03:27:38Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a670d9745c20", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "97f00c2bada31f5ab0ff82fbeb05e50aa6e0058da9f7999bd901733d34b02209", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a670d9745c20", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:27:38Z", "seed": 1494927003, "treatment": "full"}, "priority": 2.009491742576871, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "32c34057c527d6acd6917d193b1c13814bad284d03317943f92707627190a750", "run_seconds": 1.248732328414917, "seed": 1494927003, "started_at": "2026-11-22T03:27:56Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:27:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a670d9745c20", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.75, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 0.75, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "97f00c2bada31f5ab0ff82fbeb05e50aa6e0058da9f7999bd901733d34b02209", "primary_control": "e141_best", "result_hash": "32c34057c527d6acd6917d193b1c13814bad284d03317943f92707627190a750", "signature": "280c2bbb9af4057baaa4a108c1554ff2e101b43b1ee447c20bcf2de9f783c9f64ff71a9e5e9b6f568a74f5b233527be1b9e67c248f2388eb2ac4e5a44b760707", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.9375, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:27:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:46Z", "cpu_seconds": 0.5327976470000007, "created_at": "2026-11-22T03:27:38Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0f55d0485f41", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "129838ccc2207c0a6d1bb750cc795483e52c2f095a4e6649b36976ae58c16b7c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0f55d0485f41", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:27:38Z", "seed": 1622499574, "treatment": "full"}, "priority": 2.0097936693727387, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e93da63b421277b7b5d5ad5ea0f53b8b2b632a6950269c99cde067ddfb4b5e91", "run_seconds": 0.5920064449310303, "seed": 1622499574, "started_at": "2026-11-22T03:27:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.013499, 0.101409], "effect": 0.057454, "evaluated_at": "2026-11-22T03:27:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0f55d0485f41", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.588652482, 0.609929078, 0.884615385, 0.734939759], "frozen": [0.09929078, 0.095744681, 0.102564103, 0.104417671], "full": [0.65248227, 0.726950355, 0.897435897, 0.771084337], "retrieval": [0.780141844, 0.787234043, 0.916666667, 0.907630522], "rules_nodetect": [0.65248227, 0.726950355, 0.897435897, 0.771084337]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "129838ccc2207c0a6d1bb750cc795483e52c2f095a4e6649b36976ae58c16b7c", "primary_control": "facts", "result_hash": "e93da63b421277b7b5d5ad5ea0f53b8b2b632a6950269c99cde067ddfb4b5e91", "signature": "1303d59b23026c0c027231e019156a9d49b42a250e5d8a1b87093fce9e537c37dc2311f16bbeb486c6b6cdecc6587dddf5c74ce91d956da94b5e9e07d0177005", "state": "VERIFIED", "summary": {"facts": 0.704534, "frozen": 0.100504, "full": 0.761988, "retrieval": 0.847918, "rules_nodetect": 0.761988}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:27:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:40Z", "cpu_seconds": 0.9194661710000007, "created_at": "2026-11-22T03:27:38Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6de906df314f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "da20e37595f2c78d3cf337790373c997957f00d12bc906df172ed53357139dd0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6de906df314f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:27:38Z", "seed": 203392010, "treatment": "full"}, "priority": 2.0099181597008218, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "62845c8caa43de58614000fbba2755638bd82dad68506f7ccd2d2a83aee4ec85", "run_seconds": 0.9395020008087158, "seed": 203392010, "started_at": "2026-11-22T03:27:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026957, 0.019711], "effect": -0.003623, "evaluated_at": "2026-11-22T03:27:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6de906df314f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.819444444, 0.875, 0.855072464, 0.777777778], "frozen": [0.208333333, 0.472222222, 0.31884058, 0.305555556], "full": [0.847222222, 0.847222222, 0.84057971, 0.777777778], "retrieval": [0.819444444, 0.861111111, 0.84057971, 0.777777778], "rules_nodetect": [0.583333333, 0.888888889, 0.536231884, 0.416666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "da20e37595f2c78d3cf337790373c997957f00d12bc906df172ed53357139dd0", "primary_control": "facts", "result_hash": "62845c8caa43de58614000fbba2755638bd82dad68506f7ccd2d2a83aee4ec85", "signature": "51218caef680a69c31f8a7ef03b394faa559b5992add5e8b5947c0a4cac912fd6568cb0eee97a4c005a5dd997c5a87d69f02b10e17fd08d0fe8497b4b432ee08", "state": "VERIFIED", "summary": {"facts": 0.831824, "frozen": 0.326238, "full": 0.8282, "retrieval": 0.824728, "rules_nodetect": 0.60628}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:27:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:51Z", "cpu_seconds": 0.41497799600000107, "created_at": "2026-11-22T03:27:38Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-55271ef8e3cc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "prereg_hash": "aac301c3933584f31d24e7eab4a1604afaeac3cc43ecd6fdd8ce72ccaa40750f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-55271ef8e3cc", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:27:38Z", "seed": 153667038, "treatment": "gated"}, "priority": 2.009561794751834, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "0a110808cac442ae41b9e32dbfc7244ea63c4766bc9db0031d71948ce0acf0e4", "run_seconds": 0.41677069664001465, "seed": 153667038, "started_at": "2026-11-22T03:27:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.008534, 0.116466], "effect": 0.0625, "evaluated_at": "2026-11-22T03:27:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-55271ef8e3cc", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.75, 0.616666667, 0.616666667, 0.5], "gated": [0.75, 0.683333333, 0.75, 0.55], "naive": [0.75, 0.4, 0.516666667, 0.233333333], "scratch": [0.75, 0.616666667, 0.616666667, 0.5], "ungated": [0.25, 0.5, 0.516666667, 0.233333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "aac301c3933584f31d24e7eab4a1604afaeac3cc43ecd6fdd8ce72ccaa40750f", "primary_control": "scratch", "result_hash": "0a110808cac442ae41b9e32dbfc7244ea63c4766bc9db0031d71948ce0acf0e4", "signature": "115b0a21cf89d75a8fffbe2a46a48fdb06082f472b7432805e545ee5a413977669c24675447e60c54b989f95dfde947aa02edda84a5f5955f9de50c3b3fae203", "state": "VERIFIED", "summary": {"compressed": 0.620833, "gated": 0.683333, "naive": 0.475, "scratch": 0.620833, "ungated": 0.375}, "ties": 1, "treatment": "gated", "wins": 3}, "verified_at": "2026-11-22T03:27:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:27:48Z", "cpu_seconds": 7.4274068799999995, "created_at": "2026-11-22T03:27:38Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-62f77cd9295a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f2bec5ff833b21807f913f344d57458bd70564799121c2f39e594277a33c69cd", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-62f77cd9295a", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:27:38Z", "seed": 1914707220, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a28c7a4eb1c07cb01a11ca4eba199d5f22e9b5b53735598975ce554f8dc47a84", "run_seconds": 8.334065675735474, "seed": 1914707220, "started_at": "2026-11-22T03:27:39Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.125408, 0.341258], "effect": 0.233333, "evaluated_at": "2026-11-22T03:27:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-62f77cd9295a", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": true, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -4.2144, "losses": 0, "n": 60, "outcome": "SUPPORTED", "p_sign_display": "6.1e-05", "prereg_hash": "f2bec5ff833b21807f913f344d57458bd70564799121c2f39e594277a33c69cd", "primary_control": "reset", "result_hash": "a28c7a4eb1c07cb01a11ca4eba199d5f22e9b5b53735598975ce554f8dc47a84", "signature": "538e846e7404f094cc5bf3a47672c3f9e86c7c9f05ab87f674a18542be39cbae2d07dad3c2fb123079dac8dfe9a100cc2831cfc4212ebeef886885443a29ca03", "state": "VERIFIED", "summary": {"inherit": 0.8333333333333334, "reset": 0.6, "root": 0.21666666666666667}, "ties": 46, "treatment": "inherit", "wins": 14}, "verified_at": "2026-11-22T03:27:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:31Z", "cpu_seconds": 0.356307567, "created_at": "2026-11-22T03:27:55Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5d21de13f24e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "f26ef08ca932cd4147644151c4735b677bf6d43ac35e6bfb6778485f203ef8a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5d21de13f24e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:27:55Z", "seed": 1755595649, "treatment": "frontier"}, "priority": 2.0091480398811306, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a0523b9fc8490e2d54c2d5abd4e9006bec62fb04988cb2edffe38ee558f4e55c", "run_seconds": 0.357468843460083, "seed": 1755595649, "started_at": "2026-11-22T03:28:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.085949, 0.164051], "effect": 0.125, "evaluated_at": "2026-11-22T03:28:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5d21de13f24e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.984848485, 1.0, 1.0], "frontier": [1.0, 1.0, 0.984848485, 1.0], "goal": [0.863636364, 0.787878788, 0.833333333, 0.848484848], "hypdisc": [0.863636364, 0.742424242, 0.96969697, 0.954545455], "infogain": [1.0, 1.0, 0.984848485, 1.0], "prederr": [1.0, 1.0, 0.984848485, 1.0], "random": [0.909090909, 0.878787879, 0.803030303, 0.893939394], "risk": [1.0, 1.0, 0.984848485, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f26ef08ca932cd4147644151c4735b677bf6d43ac35e6bfb6778485f203ef8a2", "primary_control": "random", "result_hash": "a0523b9fc8490e2d54c2d5abd4e9006bec62fb04988cb2edffe38ee558f4e55c", "signature": "7045b1d991bf942ca77a313a1fe7bd9b70aabb3a7bb30788e065c84e07d27c1590549d02ad24f8eacb4e138ffd47a25e3666f2646b3d3f71b341344cb9af5305", "state": "VERIFIED", "summary": {"count": 0.996212, "frontier": 0.996212, "goal": 0.833333, "hypdisc": 0.882576, "infogain": 0.996212, "prederr": 0.996212, "random": 0.871212, "risk": 0.996212}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:28:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:28Z", "cpu_seconds": 3.261524398, "created_at": "2026-11-22T03:27:55Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-eba53dced85f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "eaaa30f8c1cbf13fd62fa157b178ecce83d6a00c3029eda5c4f027c58772deb0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-eba53dced85f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:27:55Z", "seed": 787439054, "treatment": "gated"}, "priority": 2.009410513795436, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c3af08fd3420062b27f510a66e11309683495876ca0eb404fb444f7687cf11cc", "run_seconds": 3.468107223510742, "seed": 787439054, "started_at": "2026-11-22T03:28:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.061667, 0.02], "effect": -0.020833, "evaluated_at": "2026-11-22T03:28:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-eba53dced85f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.733333333, 0.875, 0.422988506, 0.143434343], "gated": [0.733333333, 0.875, 0.422988506, 0.143434343], "naive": [0.65, 0.875, 0.275862069, 0.165656566], "scratch": [0.816666667, 0.875, 0.422988506, 0.143434343], "ungated": [0.65, 0.875, 0.275862069, 0.165656566]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "eaaa30f8c1cbf13fd62fa157b178ecce83d6a00c3029eda5c4f027c58772deb0", "primary_control": "scratch", "result_hash": "c3af08fd3420062b27f510a66e11309683495876ca0eb404fb444f7687cf11cc", "signature": "8c455d8868d60f82bc1b4a757ab250c42c8e4d8a23657b666de6647c2d798c9413e401026e2e8d43ad011535b337c5eeced0f3d0a39ff5fbd21b6ac1affb3808", "state": "VERIFIED", "summary": {"compressed": 0.543689, "gated": 0.543689, "naive": 0.49163, "scratch": 0.564522, "ungated": 0.49163}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:28:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:25Z", "cpu_seconds": 1.240475788, "created_at": "2026-11-22T03:28:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6f595fc639f6", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 5, "n": 4}, "prereg_hash": "a3c46ae1fe70b4ba732e8767ec18b280750f5ca6fa52a1743d2e5b079e11b7ca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6f595fc639f6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:28:11Z", "seed": 844415047, "treatment": "full"}, "priority": 2.0098507479352503, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-5ecebb80e991", "result_hash": "ac82066d637f26917946e1175c33bd46f3f9fe3a5a0f162fa83b619024b7d892", "run_seconds": 1.254274845123291, "seed": 844415047, "started_at": "2026-11-22T03:28:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.01346, 0.224219], "effect": 0.118839, "evaluated_at": "2026-11-22T03:28:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6f595fc639f6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.414583333, 0.42278481, 0.660240964, 1.0], "frozen": [0.179166667, 0.108860759, 0.175903614, 0.0], "full": [0.4875, 0.673417722, 0.812048193, 1.0], "retrieval": [0.485416667, 0.635443038, 0.812048193, 1.0], "rules_nodetect": [0.49375, 0.673417722, 0.812048193, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "a3c46ae1fe70b4ba732e8767ec18b280750f5ca6fa52a1743d2e5b079e11b7ca", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-5ecebb80e991", "result_hash": "ac82066d637f26917946e1175c33bd46f3f9fe3a5a0f162fa83b619024b7d892", "signature": "514ab6d32be127f466ce45d908d45bc61252d730b2f85ed7e4ac38963453a5943b3ef08870ad3feccde02703a26396042c34567991a4c8e759103ff9d201c50c", "state": "VERIFIED", "summary": {"facts": 0.624402, "frozen": 0.115983, "full": 0.743241, "retrieval": 0.733227, "rules_nodetect": 0.744804}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:28:27Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:28:11Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_rsi", "fence": 3, "id": "E141-ow_rsi-9b6449172474", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "bd1f9b0270c05ec3a58c2fbba4a97eedefc1fcfb26550c79816ceb4ce2ab5a4e", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-9b6449172474", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:28:11Z", "seed": 391948795, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 391948795, "started_at": "2026-11-22T03:28:13Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:28:28Z", "cpu_seconds": 3.616166818, "created_at": "2026-11-22T03:28:11Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0731631fb0a2", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e698812fbda1198c4e8c8e0951f7850191c74250032c11b5d43317dfeded73cc", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0731631fb0a2", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:28:11Z", "seed": 596058723, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "fbbecd050989f7db4ff43da7584349b81b8fb5eefa45aa34d78e6a131541a5ec", "run_seconds": 3.833254814147949, "seed": 596058723, "started_at": "2026-11-22T03:28:24Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:28:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0731631fb0a2", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0], "reset": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e698812fbda1198c4e8c8e0951f7850191c74250032c11b5d43317dfeded73cc", "primary_control": "reset", "result_hash": "fbbecd050989f7db4ff43da7584349b81b8fb5eefa45aa34d78e6a131541a5ec", "signature": "ca20f0e50bc52c436a0e5b89285572bdd932b6b02793741a5f693f5aba5fe545ab7cd40f03e1b663093b338cb1a9ada7f9797ec622d127fd5e038d020b209401", "state": "PARTIAL", "summary": {"inherit": 0.45, "reset": 0.45, "root": 0.26666666666666666}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:28:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:40Z", "cpu_seconds": 6.896291978, "created_at": "2026-11-22T03:28:11Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-cc2e6dbcfdcf", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7360f391c215ca0ed3c32b77b280fc98bd500ffc6198955907435cad42e1683f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-cc2e6dbcfdcf", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:28:11Z", "seed": 1142665111, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "07ed6c06c52653b7bfc5d15e90eab5f69caae38065f6d3ae9908ea7d045d7923", "run_seconds": 7.365496397018433, "seed": 1142665111, "started_at": "2026-11-22T03:28:33Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.179507, 0.046174], "effect": -0.066667, "evaluated_at": "2026-11-22T03:28:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-cc2e6dbcfdcf", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.2619298960304671, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0329, "losses": 8, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.927", "prereg_hash": "7360f391c215ca0ed3c32b77b280fc98bd500ffc6198955907435cad42e1683f", "primary_control": "reset", "result_hash": "07ed6c06c52653b7bfc5d15e90eab5f69caae38065f6d3ae9908ea7d045d7923", "signature": "6f4d33e08da464d71ae6670bbdd90e57b5c7d7c15b841dfffea9126d6609e54a2b8c6633993bda84335c072b58df7f3fe6b6c78938398eb1dfa2530a0c52510b", "state": "VERIFIED", "summary": {"inherit": 0.6333333333333333, "reset": 0.7, "root": 0.3}, "ties": 48, "treatment": "inherit", "wins": 4}, "verified_at": "2026-11-22T03:28:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:33Z", "cpu_seconds": 0.5378804989999995, "created_at": "2026-11-22T03:28:28Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a8f1939fd3cc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "797aedad2f4b3653481a70bd3d5075570b320d42d2c495b3d747d6d4eb53e070", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a8f1939fd3cc", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:28:28Z", "seed": 884564674, "treatment": "full"}, "priority": 2.0090899277223073, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "61e63de934f9ebe48be31b3f97da2e69fc29593c2df2daca40733edd19178f5e", "run_seconds": 0.6129903793334961, "seed": 884564674, "started_at": "2026-11-22T03:28:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:28:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a8f1939fd3cc", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.25, 0.5, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [1.0, 0.0, 0.5, 0.0], "no_world_model": [1.0, 0.25, 0.5, 0.0], "random": [1.0, 0.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "797aedad2f4b3653481a70bd3d5075570b320d42d2c495b3d747d6d4eb53e070", "primary_control": "e141_best", "result_hash": "61e63de934f9ebe48be31b3f97da2e69fc29593c2df2daca40733edd19178f5e", "signature": "8ca45f7d5748fd07ac1831d263b38a96d4d4b0837f7cb5d785ada0ca0ddeab367f63d6ccbb442b797d4438843e05bf15489d0ecd395d6d8ff39db729ad31a10f", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.0625, "full": 0.375, "no_world_model": 0.4375, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:28:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:01Z", "cpu_seconds": 1.794771625, "created_at": "2026-11-22T03:28:44Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-836557f4e661", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "80db0092d2e047fb0711bc717424c1f4283837e645b7c0200bd4eb50a6c522c1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-836557f4e661", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:28:44Z", "seed": 1353430756, "treatment": "frontier"}, "priority": 2.009246424848861, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "187644a47d21d96a5fc347037bde5fba8bf2948a41439e7b63f44fac6e41d05c", "run_seconds": 1.8231451511383057, "seed": 1353430756, "started_at": "2026-11-22T03:28:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.328898, 0.250873], "effect": -0.039013, "evaluated_at": "2026-11-22T03:29:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-836557f4e661", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.552380952, 0.6, 0.634210526, 0.461176471], "frontier": [0.204761905, 0.540983607, 0.265789474, 0.625882353], "goal": [0.161904762, 0.514754098, 0.65, 0.435294118], "hypdisc": [0.342857143, 0.33442623, 0.665789474, 0.197647059], "infogain": [0.204761905, 0.540983607, 0.05, 0.694117647], "prederr": [0.204761905, 0.540983607, 0.128947368, 0.656470588], "random": [0.471428571, 0.504918033, 0.539473684, 0.277647059], "risk": [0.204761905, 0.540983607, 0.265789474, 0.625882353]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "80db0092d2e047fb0711bc717424c1f4283837e645b7c0200bd4eb50a6c522c1", "primary_control": "random", "result_hash": "187644a47d21d96a5fc347037bde5fba8bf2948a41439e7b63f44fac6e41d05c", "signature": "f998d4e1916e3869888d2511c7676db3ee2e9ab005e83215fabfe16c76d2d11ea6eae15be02655ced657c1100d405a0446578afbd06686c1ff85f60f77201506", "state": "VERIFIED", "summary": {"count": 0.561942, "frontier": 0.409354, "goal": 0.440488, "hypdisc": 0.38518, "infogain": 0.372466, "prederr": 0.382791, "random": 0.448367, "risk": 0.409354}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:29:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:14Z", "cpu_seconds": 5.125447218999999, "created_at": "2026-11-22T03:28:44Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8f33cde6955d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 3, "n": 4}, "prereg_hash": "ce04d96b3bdccff19f2bcb47eb8a654586198cae63ef60f2a23956fcce448a89", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8f33cde6955d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:28:44Z", "seed": 1118781263, "treatment": "frontier"}, "priority": 2.0078367906809294, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f3b8168da3c850a56b7dd86c94889e773be6f4bac579fb2c173839692312a07d", "run_seconds": 7.684702634811401, "seed": 1118781263, "started_at": "2026-11-22T03:30:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016455, 0.124048], "effect": 0.053796, "evaluated_at": "2026-11-22T03:30:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-8f33cde6955d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.524853229, 0.495499022, 1.0, 1.0], "frontier": [0.524853229, 0.495499022, 0.897142857, 1.0], "goal": [0.502935421, 0.495499022, 0.777142857, 0.994285714], "hypdisc": [0.487279843, 0.479843444, 0.88, 0.965714286], "infogain": [0.524853229, 0.495499022, 0.897142857, 1.0], "prederr": [0.524853229, 0.495499022, 0.897142857, 1.0], "random": [0.515459883, 0.506849315, 0.828571429, 0.851428571], "risk": [0.524853229, 0.495499022, 0.897142857, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "ce04d96b3bdccff19f2bcb47eb8a654586198cae63ef60f2a23956fcce448a89", "primary_control": "random", "result_hash": "f3b8168da3c850a56b7dd86c94889e773be6f4bac579fb2c173839692312a07d", "signature": "f4b0973ace49e3b26ecfe111fabb7b355aeb0bb7f67b1c75e43fe05ecb266598bd8ee03dee13cbda5b904cc0c17b75f99c2fddf351edf3f0a0afb02daddf2401", "state": "VERIFIED", "summary": {"count": 0.755088, "frontier": 0.729374, "goal": 0.692466, "hypdisc": 0.703209, "infogain": 0.729374, "prederr": 0.729374, "random": 0.675577, "risk": 0.729374}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:30:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:11Z", "cpu_seconds": 0.26321793000000016, "created_at": "2026-11-22T03:28:44Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-64bc8c258053", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "59c6a66ad66886cf119c1ad5b2c5f8a5e8f101c90abebb8f1afbe6606cfc7266", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-64bc8c258053", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:28:44Z", "seed": 792141405, "treatment": "full"}, "priority": 2.0085790122613654, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "57d3452287ccb65d20039804938705eed1b94217e78f2d45e18e869ceadd8fde", "run_seconds": 0.266002893447876, "seed": 792141405, "started_at": "2026-11-22T03:29:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:29:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-64bc8c258053", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 0.75], "fixed_policy": [0.25, 0.25, 0.75, 0.25], "full": [1.0, 1.0, 1.0, 0.75], "no_world_model": [1.0, 1.0, 1.0, 0.75], "random": [1.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "59c6a66ad66886cf119c1ad5b2c5f8a5e8f101c90abebb8f1afbe6606cfc7266", "primary_control": "e141_best", "result_hash": "57d3452287ccb65d20039804938705eed1b94217e78f2d45e18e869ceadd8fde", "signature": "a33d05456c45c31a4a37987264b41e18adeaff93ccfb15ec7e0b3d48b5657e5ac93e30b4b36332caca9f8446719aa960ade46872c04b9b88f27f468700254607", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.375, "full": 0.9375, "no_world_model": 0.9375, "random": 0.9375}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:29:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:53Z", "cpu_seconds": 2.3276666999999946, "created_at": "2026-11-22T03:28:44Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d0e864dbc3c5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "1ba701bafbce8ea59e9e80e21829b32572ed343ff63f968a797824d6796b349a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d0e864dbc3c5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:28:44Z", "seed": 396966977, "treatment": "full"}, "priority": 2.007382054604212, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8aad7a4a91b8d58969bb4dbf1bcb67092cfab22bffd4a37e327f90beb4b36a2a", "run_seconds": 3.036921977996826, "seed": 396966977, "started_at": "2026-11-22T03:36:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.17207, 0.29707], "effect": 0.0625, "evaluated_at": "2026-11-22T03:38:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d0e864dbc3c5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 0.5, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.75], "full": [0.75, 1.0, 0.5, 1.0], "no_world_model": [0.5, 0.75, 0.25, 1.0], "random": [0.5, 1.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "1ba701bafbce8ea59e9e80e21829b32572ed343ff63f968a797824d6796b349a", "primary_control": "e141_best", "result_hash": "8aad7a4a91b8d58969bb4dbf1bcb67092cfab22bffd4a37e327f90beb4b36a2a", "signature": "2b45226a8ba8af3420492b1b230b536ec32f07694f3d145e94c30d43014809a5f76be1143307ef81b46baf1589693fdcc72f440da22eef1b1d835993344b230c", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.1875, "full": 0.8125, "no_world_model": 0.625, "random": 0.75}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:38:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:59Z", "cpu_seconds": 0.18502743999999993, "created_at": "2026-11-22T03:28:44Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-fd92cf6118fb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "2568b0f9d6f3a82c63a8a9b831f3d4cdaeec9fb5621c3b31a40cc8e0531c0ac9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-fd92cf6118fb", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:28:44Z", "seed": 2113348641, "treatment": "full"}, "priority": 2.008965949095301, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1f7bd892072ba3bda1601d378d44bd96a91831c9a4113fb3a9c9be3e458b5746", "run_seconds": 0.21977806091308594, "seed": 2113348641, "started_at": "2026-11-22T03:28:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016338, 0.057752], "effect": 0.020707, "evaluated_at": "2026-11-22T03:29:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-fd92cf6118fb", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.061111111, 0.572222222, 0.55], "frozen": [1.0, 0.027777778, 0.272222222, 0.031818182], "full": [1.0, 0.066666667, 0.572222222, 0.627272727], "retrieval": [1.0, 0.261111111, 0.572222222, 0.631818182], "rules_nodetect": [1.0, 0.388888889, 0.572222222, 0.627272727]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "2568b0f9d6f3a82c63a8a9b831f3d4cdaeec9fb5621c3b31a40cc8e0531c0ac9", "primary_control": "facts", "result_hash": "1f7bd892072ba3bda1601d378d44bd96a91831c9a4113fb3a9c9be3e458b5746", "signature": "400cf98c532dd5c22f249b6377fd5fcf59fe8e667eb3b7d9fef3036dc46106deb7237f86858fbfc8f925900612bd225ed523e6a770096ea88c423dfc4e04ee04", "state": "VERIFIED", "summary": {"facts": 0.545833, "frozen": 0.332955, "full": 0.56654, "retrieval": 0.616288, "rules_nodetect": 0.647096}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:29:03Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:30:00Z", "cpu_seconds": 1.1242455370000002, "created_at": "2026-11-22T03:28:44Z", "error": "lease expired; ", "family": "ow_transfer", "fence": 2, "id": "E141-ow_transfer-841050de6c4c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "prereg_hash": "f37709998c8a8ff338ae252ecc9a29940e2d1f307e6cccef4b8026fb15fded0b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-841050de6c4c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:28:44Z", "seed": 2128778109, "treatment": "gated"}, "priority": 2.009573078965892, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "bd4d10b9e5e62a2d40a1e1247a46a88ac794cd4c7e3af4d496840dba1380b166", "run_seconds": 1.1929278373718262, "seed": 2128778109, "started_at": "2026-11-22T03:28:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023423, 0.082176], "effect": 0.029377, "evaluated_at": "2026-11-22T03:30:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-841050de6c4c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.305714286, 0.268085106, 0.144680851, 0.268085106], "gated": [0.274285714, 0.268085106, 0.144680851, 0.268085106], "naive": [0.28, 0.187234043, 0.144680851, 0.191489362], "scratch": [0.305714286, 0.187234043, 0.144680851, 0.2], "ungated": [0.262857143, 0.246808511, 0.144680851, 0.24893617]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f37709998c8a8ff338ae252ecc9a29940e2d1f307e6cccef4b8026fb15fded0b", "primary_control": "scratch", "result_hash": "bd4d10b9e5e62a2d40a1e1247a46a88ac794cd4c7e3af4d496840dba1380b166", "signature": "812aae050161715f5fe028333a9bf8c328e98a83e6b01e89dcba86f4a43108513350e48731a66858b84d2f3e9053e2aace087f8b305f78308c22751c9ebb460c", "state": "VERIFIED", "summary": {"compressed": 0.246641, "gated": 0.238784, "naive": 0.200851, "scratch": 0.209407, "ungated": 0.225821}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:30:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:28:55Z", "cpu_seconds": 7.656179397000001, "created_at": "2026-11-22T03:28:44Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-ad1a61a23880", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d9be82c0510783cd5003213a835158633c910fe11be89a2332c07e3709635ed3", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-ad1a61a23880", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:28:44Z", "seed": 1415480822, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f58e5d2a4ba38d7daa6b67d25901b631b7ff05f8809120dfc4cc0a25d02ab1e3", "run_seconds": 9.91738510131836, "seed": 1415480822, "started_at": "2026-11-22T03:28:45Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:28:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-ad1a61a23880", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.36829765087471333, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d9be82c0510783cd5003213a835158633c910fe11be89a2332c07e3709635ed3", "primary_control": "reset", "result_hash": "f58e5d2a4ba38d7daa6b67d25901b631b7ff05f8809120dfc4cc0a25d02ab1e3", "signature": "6abed8f5dd3424b84c79c4f1b47269edb045f3d18661a620acb08f420c9d900aa6338587114e27096f510db576d384a2e9cb1fe94bd68cc01992e12b9aa04e0a", "state": "PARTIAL", "summary": {"inherit": 0.7, "reset": 0.7, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:29:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:01Z", "cpu_seconds": 0.34588730599999806, "created_at": "2026-11-22T03:28:44Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-f59203dddb10", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "5693203ecc8f545563dd0b9c406585b905f68f1bd81628d42ecf607bf413a138", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-f59203dddb10", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:28:44Z", "seed": 84792163, "treatment": "frontier"}, "priority": 2.008721749473325, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-e41583d38779", "result_hash": "be85f1190b9f8d461d3544bc022d170d6939350c5298bb4774b8b18d90bf6dca", "run_seconds": 0.3863818645477295, "seed": 84792163, "started_at": "2026-11-22T03:29:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.066678, 0.125225], "effect": 0.095951, "evaluated_at": "2026-11-22T03:29:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-f59203dddb10", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.765789474, 0.800505051, 0.7725, 0.842857143, 0.783505155, 0.776515152, 0.753246753, 0.724025974, 1.0, 0.805851064, 0.838095238, 0.75], "frontier": [0.878947368, 0.893939394, 0.8675, 1.0, 0.886597938, 0.950757576, 0.902597403, 0.857142857, 1.0, 0.904255319, 0.902380952, 0.914473684], "random_walk": [0.773684211, 0.775252525, 0.805, 0.921428571, 0.832474227, 0.772727273, 0.756493506, 0.74025974, 1.0, 0.821808511, 0.854761905, 0.753289474]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "5693203ecc8f545563dd0b9c406585b905f68f1bd81628d42ecf607bf413a138", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-e41583d38779", "result_hash": "be85f1190b9f8d461d3544bc022d170d6939350c5298bb4774b8b18d90bf6dca", "signature": "f548e9a3986dc8c898f31be22a4deda740231cae881d879f889a01ac4419a6c59bf8a878360861c15f6ea4a1156cd01d1c223362d085e56615935b6e946e0806", "state": "VERIFIED", "summary": {"curiosity": 0.801074, "frontier": 0.913216, "random_walk": 0.817265}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:29:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:14Z", "cpu_seconds": 9.190821438, "created_at": "2026-11-22T03:29:01Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-549d67c50a59", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "ce338916c9cc09f60fc9410e227a1ee8b5dff820c9d1ac2bde5dfac471269766", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-549d67c50a59", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:29:01Z", "seed": 1657381890, "treatment": "frontier"}, "priority": 2.00960796004944, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "298cb315241dca35b775cdd88606880b6bc7c01fb843e965faa4f5376697956e", "run_seconds": 9.887758731842041, "seed": 1657381890, "started_at": "2026-11-22T03:29:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.300409, 0.244984], "effect": -0.027712, "evaluated_at": "2026-11-22T03:29:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-549d67c50a59", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.485026042, 0.574793875, 0.508680556, 0.437943262], "frontier": [0.40625, 0.730859835, 0.155671296, 0.147163121], "goal": [0.358723958, 0.547114252, 0.347800926, 0.180851064], "hypdisc": [0.3203125, 0.323910483, 0.530092593, 0.226950355], "infogain": [0.40625, 0.500588928, 0.566550926, 0.147163121], "prederr": [0.40625, 0.482921084, 0.486111111, 0.147163121], "random": [0.162109375, 0.550647821, 0.417824074, 0.420212766], "risk": [0.40625, 0.204946996, 0.155671296, 0.147163121]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "ce338916c9cc09f60fc9410e227a1ee8b5dff820c9d1ac2bde5dfac471269766", "primary_control": "random", "result_hash": "298cb315241dca35b775cdd88606880b6bc7c01fb843e965faa4f5376697956e", "signature": "0860e0bbe995ee1000e843cb90f9c6ca71bfd526b563491ad835a8800f100194446b232f902a48c6b885710a97c8de42db2991d46bd4b8e35a9b0afbed30640d", "state": "VERIFIED", "summary": {"count": 0.501611, "frontier": 0.359986, "goal": 0.358623, "hypdisc": 0.350316, "infogain": 0.405138, "prederr": 0.380611, "random": 0.387699, "risk": 0.228508}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:29:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:16Z", "cpu_seconds": 2.806664755, "created_at": "2026-11-22T03:29:01Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f8664def533a", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "3928467b35966b132cf268e6876704501a2475981dd68b75efdd918334538ac2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f8664def533a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:29:01Z", "seed": 430324781, "treatment": "frontier"}, "priority": 2.008087349300056, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c87da203773d7442c28afb32f6c669185aab20fbdcfc47d7afa2bb6fc0e9615e", "run_seconds": 3.218862533569336, "seed": 430324781, "started_at": "2026-11-22T03:29:13Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:29:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f8664def533a", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.501617], "frontier": [1.0, 0.480598], "goal": [1.0, 0.439938], "hypdisc": [1.0, 0.416842], "infogain": [1.0, 0.466536], "prederr": [1.0, 0.466536], "random": [1.0, 0.395698], "risk": [1.0, 0.480598]}, "prereg_hash": "3928467b35966b132cf268e6876704501a2475981dd68b75efdd918334538ac2", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "c87da203773d7442c28afb32f6c669185aab20fbdcfc47d7afa2bb6fc0e9615e", "signature": "86629ea3573093150fa2e1d4cc2b31b5dad39764033c7c3130d596a226d9ae92126ccc312ca7c70e535a7f6ca8425b1c1ee33d3563a13b49330b88eed0a22c0d", "state": "REJECTED", "summary": {"count": 0.501617, "frontier": 0.480598, "goal": 0.439938, "hypdisc": 0.416842, "infogain": 0.466536, "prederr": 0.466536, "random": 0.395698, "risk": 0.480598}}, "verified_at": "2026-11-22T03:29:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:06Z", "cpu_seconds": 0.08204079699999989, "created_at": "2026-11-22T03:29:01Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-146c16c06d9f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "c531ed94e8836bc66ba8af485e82c0d07803574211fc91ffeb849f17ec310789", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-146c16c06d9f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:29:01Z", "seed": 538905662, "treatment": "full"}, "priority": 2.0090742541337474, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "3bd3442c0c8e6c46cb95e877650d0f716fc32f3fc68f0d63ddfb6c2b473de8ee", "run_seconds": 0.09343719482421875, "seed": 538905662, "started_at": "2026-11-22T03:29:06Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:29:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-146c16c06d9f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.75, 0.5], "fixed_policy": [0.25, 0.5, 0.25, 0.25], "full": [0.75, 1.0, 0.75, 0.5], "no_world_model": [0.75, 1.0, 0.75, 0.5], "random": [0.75, 1.0, 0.75, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c531ed94e8836bc66ba8af485e82c0d07803574211fc91ffeb849f17ec310789", "primary_control": "e141_best", "result_hash": "3bd3442c0c8e6c46cb95e877650d0f716fc32f3fc68f0d63ddfb6c2b473de8ee", "signature": "dc4778612cf4dc167a485110795934360594e928f53a3c96295a4728a2393ed0c7f6a7e3f687ec1d60f70b23c8e27544edc9f7af84c1e54909443f2e7602e10c", "state": "PARTIAL", "summary": {"e141_best": 0.75, "fixed_policy": 0.3125, "full": 0.75, "no_world_model": 0.75, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:29:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:08Z", "cpu_seconds": 2.0268070840000014, "created_at": "2026-11-22T03:29:01Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8d8f245393f3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "851c4ec64e23d87f5319c34aaafe0a003403154b823fa70968b41c4fa3bcd2b0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8d8f245393f3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:29:01Z", "seed": 785639747, "treatment": "full"}, "priority": 2.0093228367773115, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "54e514b934de54d7581a653feca34af1b3318395f01570644b86cb262bf4d168", "run_seconds": 2.5356926918029785, "seed": 785639747, "started_at": "2026-11-22T03:29:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00796, 0.059394], "effect": 0.025717, "evaluated_at": "2026-11-22T03:29:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8d8f245393f3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.977777778, 0.924731183, 0.994623656], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 0.97311828, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "851c4ec64e23d87f5319c34aaafe0a003403154b823fa70968b41c4fa3bcd2b0", "primary_control": "facts", "result_hash": "54e514b934de54d7581a653feca34af1b3318395f01570644b86cb262bf4d168", "signature": "c2f954749b46620af8027747f8186699ba8bb973fed6f3bf81bfbaa2ee7cdedc5aa8bfaf67ff0e95485e7083ece6b26bbca4d9b900e9deeeee5e14bbbcabc403", "state": "VERIFIED", "summary": {"facts": 0.974283, "frozen": 0.0, "full": 1.0, "retrieval": 0.99328, "rules_nodetect": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:29:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:10Z", "cpu_seconds": 6.897539029000001, "created_at": "2026-11-22T03:29:01Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-1b0778cc89c2", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "e6b047db900404bc48375443c7ca794dd8dd0d666d46dabfc87f0d4c4a28a03d", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-1b0778cc89c2", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:29:01Z", "seed": 2136562041, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f705d976c06c7bff4e3cc8c2dd5ea989a8466eb9d98352092c4a6bc2b107fa73", "run_seconds": 7.5031819343566895, "seed": 2136562041, "started_at": "2026-11-22T03:29:02Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 1, "reset": 2}, "better": "higher", "ci95": [-0.261763, -0.07157], "effect": -0.166667, "evaluated_at": "2026-11-22T03:29:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-1b0778cc89c2", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.31422645983055986, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 10, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e6b047db900404bc48375443c7ca794dd8dd0d666d46dabfc87f0d4c4a28a03d", "primary_control": "reset", "result_hash": "f705d976c06c7bff4e3cc8c2dd5ea989a8466eb9d98352092c4a6bc2b107fa73", "signature": "6b5ecf61daac389a0d3aa0f816e93609f309f73af627190ac46adc09fb553d7a239e60388adea430c7d7ad1f550ec01a2f08bd63a23af78eaacf5efbcd45e608", "state": "VERIFIED", "summary": {"inherit": 0.6333333333333333, "reset": 0.8, "root": 0.4166666666666667}, "ties": 50, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:29:17Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:29:17Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_explore", "fence": 3, "id": "E141-ow_explore-06679e9cb136", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "c1696433051b6f554aff5785bae2cdc13033efd1d67d192d1caff9390b26fb9a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-06679e9cb136", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:29:17Z", "seed": 140266806, "treatment": "frontier"}, "priority": 2.009560006603434, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 140266806, "started_at": "2026-11-22T03:29:19Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 2, "completed_at": "2026-11-22T03:31:43Z", "cpu_seconds": 18.452551632, "created_at": "2026-11-22T03:29:17Z", "error": "lease expired; ", "family": "ow_explore", "fence": 2, "id": "E141-ow_explore-c5a2a666bb2b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "502e33e8261cf761365ca67ecb847b28bf4fcb126a21c6a3b48a14ea39f901d0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c5a2a666bb2b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:29:17Z", "seed": 1522344020, "treatment": "frontier"}, "priority": 2.0087180868501866, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "12602ba6e2be4b4f848105e79c68e68f2855c8c4f12f3e49e9a3fadbb3a1f415", "run_seconds": 23.3785240650177, "seed": 1522344020, "started_at": "2026-11-22T03:29:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.025392, 0.16635], "effect": 0.070479, "evaluated_at": "2026-11-22T03:31:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c5a2a666bb2b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.579787234, 0.420212766, 0.538043478, 0.562056738], "frontier": [0.443262411, 0.526595745, 0.449275362, 0.326241135], "goal": [0.15070922, 0.462765957, 0.429347826, 0.356382979], "hypdisc": [0.473404255, 0.239361702, 0.460144928, 0.290780142], "infogain": [0.44858156, 0.503546099, 0.490942029, 0.411347518], "prederr": [0.44858156, 0.430851064, 0.490942029, 0.395390071], "random": [0.443262411, 0.319148936, 0.449275362, 0.25177305], "risk": [0.443262411, 0.526595745, 0.449275362, 0.326241135]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "502e33e8261cf761365ca67ecb847b28bf4fcb126a21c6a3b48a14ea39f901d0", "primary_control": "random", "result_hash": "12602ba6e2be4b4f848105e79c68e68f2855c8c4f12f3e49e9a3fadbb3a1f415", "signature": "b817c2c9eece18acfc20ffe982851d27f557640d874208cda56f592da1ed8865af72baf06f85bbeaf1934de1dbce47a1110a1437a8b5d31d9b4a5d4d3bfd4a0e", "state": "VERIFIED", "summary": {"count": 0.525025, "frontier": 0.436344, "goal": 0.349801, "hypdisc": 0.365923, "infogain": 0.463604, "prederr": 0.441441, "random": 0.365865, "risk": 0.436344}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:31:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:36Z", "cpu_seconds": 6.574937749, "created_at": "2026-11-22T03:29:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2e1a28d0a8fa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 4, "n": 4}, "prereg_hash": "fb20dd442b37c66a975fbf6ccb5b617487a3d236cb0a028eb17a3c6f4183d29c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2e1a28d0a8fa", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:29:17Z", "seed": 353869478, "treatment": "full"}, "priority": 2.0094998657058194, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "203c7baa70a5c5e6fed3ac3e660fab5b5664439ad6ad43f6e3b57d174676baa2", "run_seconds": 7.785719394683838, "seed": 353869478, "started_at": "2026-11-22T03:29:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.17207, 0.29707], "effect": 0.0625, "evaluated_at": "2026-11-22T03:29:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2e1a28d0a8fa", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.75, 0.25], "fixed_policy": [0.75, 0.0, 1.0, 0.5], "full": [1.0, 1.0, 1.0, 0.0], "no_world_model": [0.5, 1.0, 0.75, 0.25], "random": [1.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "fb20dd442b37c66a975fbf6ccb5b617487a3d236cb0a028eb17a3c6f4183d29c", "primary_control": "e141_best", "result_hash": "203c7baa70a5c5e6fed3ac3e660fab5b5664439ad6ad43f6e3b57d174676baa2", "signature": "00739e65252eaa119684201f2d046e94e833dfd13bd3cf901011d4ad5a63de0c2a1387502e2159700b7c64d115d6d90a00a4ab967e32b3f888decc53d2efa207", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.5625, "full": 0.75, "no_world_model": 0.625, "random": 0.9375}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:29:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:47Z", "cpu_seconds": 0.3560251050000005, "created_at": "2026-11-22T03:29:17Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e852ca55d6a7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "252d06a6822db9abba49f790cd1354185106cc20c34b0741e814d2e677e9eee7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e852ca55d6a7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:29:17Z", "seed": 654566827, "treatment": "full"}, "priority": 2.007651226868583, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fd119f23d62c39bc591d6a254e802f0e7b1044bf214022aeb8f7f30ce9d0da32", "run_seconds": 0.35616421699523926, "seed": 654566827, "started_at": "2026-11-22T03:34:47Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:35:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e852ca55d6a7", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.986666667, 1.0, 1.0, 1.0], "frozen": [0.0, 0.0, 0.0, 0.2], "full": [0.986666667, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 1.0, 1.0], "rules_nodetect": [0.986666667, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "252d06a6822db9abba49f790cd1354185106cc20c34b0741e814d2e677e9eee7", "primary_control": "facts", "result_hash": "fd119f23d62c39bc591d6a254e802f0e7b1044bf214022aeb8f7f30ce9d0da32", "signature": "81cec10a4cc2ba3dd6a114fe81666fc4f47bc0f39e5ff0e29ae2d418c4d06b3e5a793d07a295fdd6fc33dcf4c5adfe9f72c84536ebdb726356716bef558f0007", "state": "PARTIAL", "summary": {"facts": 0.996667, "frozen": 0.05, "full": 0.996667, "retrieval": 1.0, "rules_nodetect": 0.996667}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:35:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:54Z", "cpu_seconds": 0.22094919899999999, "created_at": "2026-11-22T03:29:17Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-647850aef026", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "prereg_hash": "b30b4400d970d1246fb0f363b176c10471751b95ad93266191fa14f75fb3e758", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-647850aef026", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:29:17Z", "seed": 563816714, "treatment": "gated"}, "priority": 2.0085745350430857, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "29ad55c35d39a047077cc98ca365e169e62f7f97bbc4732693290524ef9de3df", "run_seconds": 0.41919779777526855, "seed": 563816714, "started_at": "2026-11-22T03:29:53Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:29:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-647850aef026", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.46969697, 0.46969697, 0.378787879, 0.303030303], "gated": [0.46969697, 0.46969697, 0.378787879, 0.303030303], "naive": [0.46969697, 0.46969697, 0.196969697, 0.303030303], "scratch": [0.46969697, 0.46969697, 0.378787879, 0.303030303], "ungated": [0.46969697, 0.46969697, 0.196969697, 0.303030303]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b30b4400d970d1246fb0f363b176c10471751b95ad93266191fa14f75fb3e758", "primary_control": "scratch", "result_hash": "29ad55c35d39a047077cc98ca365e169e62f7f97bbc4732693290524ef9de3df", "signature": "7eadcd52251c3167fa13c2f012ebefa90aea61af441bbba7899422f59d2155fe6a91e54c4c197fe7a156c4b5f135c49f57175f986c63a4ed9ebceea3ad17880a", "state": "PARTIAL", "summary": {"compressed": 0.405303, "gated": 0.405303, "naive": 0.359848, "scratch": 0.405303, "ungated": 0.359848}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:29:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:26Z", "cpu_seconds": 4.900037555000001, "created_at": "2026-11-22T03:29:17Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-67695f25c8e3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "02c8e58291d93c1c4d04f317ad295b803efef49cafa088bc692bafa071eda76c", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-67695f25c8e3", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:29:17Z", "seed": 381841345, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "2e33bdb42d77a03a8857beaae37986a67734b9f1ecc0dfc840079bb54d0f8700", "run_seconds": 8.46817660331726, "seed": 381841345, "started_at": "2026-11-22T03:29:18Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 1, "reset": 2}, "better": "higher", "ci95": [-0.220075, -0.046592], "effect": -0.133333, "evaluated_at": "2026-11-22T03:29:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-67695f25c8e3", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.46982582798643713, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 8, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "02c8e58291d93c1c4d04f317ad295b803efef49cafa088bc692bafa071eda76c", "primary_control": "reset", "result_hash": "2e33bdb42d77a03a8857beaae37986a67734b9f1ecc0dfc840079bb54d0f8700", "signature": "4d067090204d015f52ec64d7853982d21515dde4bda3194eb657f06927207e8a8fc4aa6d5e28566eae9112afb182fd20edadc647f1dc69e04950ab7c1db6d802", "state": "VERIFIED", "summary": {"inherit": 0.36666666666666664, "reset": 0.5, "root": 0.26666666666666666}, "ties": 52, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:29:33Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:30:36Z", "cpu_seconds": 6.6630717520000005, "created_at": "2026-11-22T03:29:17Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-da1a0f413bbe", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "93f5dc70291eff54abe5d9f5ba0e5f16da714abc640fdce5b31b55f2fc4b90fd", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-da1a0f413bbe", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:29:17Z", "seed": 1388324310, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "de930ad9d2f00383ac7fa5c14ff27c5b171b6e8d4456702816127965c9c2fa49", "run_seconds": 7.133072853088379, "seed": 1388324310, "started_at": "2026-11-22T03:29:21Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:30:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-da1a0f413bbe", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4660888272648502, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "93f5dc70291eff54abe5d9f5ba0e5f16da714abc640fdce5b31b55f2fc4b90fd", "primary_control": "reset", "result_hash": "de930ad9d2f00383ac7fa5c14ff27c5b171b6e8d4456702816127965c9c2fa49", "signature": "ee650e10ea304030e09700409b8c013e1691b49d8f04cbd0b19fcae32a1ba1ec83bcb2df5dbf73eee325c406c88342eee621ffbf8f2b56e1ba1e4e472c370201", "state": "PARTIAL", "summary": {"inherit": 0.8333333333333334, "reset": 0.8333333333333334, "root": 0.3333333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:30:44Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:31:22Z", "cpu_seconds": 3.7978036000000004, "created_at": "2026-11-22T03:29:34Z", "error": "lease expired; ", "family": "ow_explore", "fence": 2, "id": "E141-ow_explore-9cc4b1c2604c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 2, "n": 4}, "prereg_hash": "a8cc65f28ab93a4573f16f21117f7a7f89ebcfd61600826e6e5a866fc90edc3c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9cc4b1c2604c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:29:34Z", "seed": 177709602, "treatment": "frontier"}, "priority": 2.0089619062124817, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e237b620997a82afc5295db2bcc7a10b699b4ab37f8438c63c226286ad5f6d60", "run_seconds": 4.417421579360962, "seed": 177709602, "started_at": "2026-11-22T03:29:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023881, 0.140602], "effect": 0.058361, "evaluated_at": "2026-11-22T03:31:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9cc4b1c2604c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.342465753, 0.961904762, 0.505675147, 0.324853229], "frontier": [0.342465753, 1.0, 0.505675147, 0.350293542], "goal": [0.342465753, 0.844444444, 0.500587084, 0.350293542], "hypdisc": [0.326027397, 0.863492063, 0.505675147, 0.313502935], "infogain": [0.342465753, 1.0, 0.505675147, 0.350293542], "prederr": [0.342465753, 1.0, 0.505675147, 0.350293542], "random": [0.33072407, 0.815873016, 0.485322896, 0.333072407], "risk": [0.342465753, 1.0, 0.505675147, 0.350293542]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a8cc65f28ab93a4573f16f21117f7a7f89ebcfd61600826e6e5a866fc90edc3c", "primary_control": "random", "result_hash": "e237b620997a82afc5295db2bcc7a10b699b4ab37f8438c63c226286ad5f6d60", "signature": "da1bfa0ab431a7f53d603439086edd83a3daad445358448af98ba5f265a96f6b7381f03cba3d43185c27c9a6d7f7d30a526a12be3cf218e49ad660b52e2bb10e", "state": "VERIFIED", "summary": {"count": 0.533725, "frontier": 0.549609, "goal": 0.509448, "hypdisc": 0.502174, "infogain": 0.549609, "prederr": 0.549609, "random": 0.491248, "risk": 0.549609}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:31:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:54Z", "cpu_seconds": 0.589016481, "created_at": "2026-11-22T03:29:34Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3dc2ffedf7e6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "194465502627e2b38a71963628d75834ee6568f0eb85541a866f2a58ef4d04bc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3dc2ffedf7e6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:29:34Z", "seed": 989538719, "treatment": "full"}, "priority": 2.0088159233192195, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c53e54c8b8d0d70cb4b6645d852d49d619d44ae8635321e44bd85af9f1adec69", "run_seconds": 0.8894627094268799, "seed": 989538719, "started_at": "2026-11-22T03:29:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001564, 0.094748], "effect": 0.048156, "evaluated_at": "2026-11-22T03:29:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3dc2ffedf7e6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.846153846, 0.779411765, 0.927083333, 0.651785714], "frozen": [0.336538462, 0.235294118, 0.302083333, 0.357142857], "full": [0.875, 0.897058824, 0.9375, 0.6875], "retrieval": [0.884615385, 0.882352941, 0.916666667, 0.660714286], "rules_nodetect": [0.317307692, 0.25, 0.458333333, 0.589285714]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "194465502627e2b38a71963628d75834ee6568f0eb85541a866f2a58ef4d04bc", "primary_control": "facts", "result_hash": "c53e54c8b8d0d70cb4b6645d852d49d619d44ae8635321e44bd85af9f1adec69", "signature": "12123a733d0a7c084f2c69663761096bef0a13156582f9087d8df6b598e3f264f24043c3e6354c6f631e8b21f0a9c6e86393b2c6390260979c14e48c7d1d5b04", "state": "VERIFIED", "summary": {"facts": 0.801109, "frozen": 0.307765, "full": 0.849265, "retrieval": 0.836087, "rules_nodetect": 0.403732}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:29:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:29:46Z", "cpu_seconds": 0.856019383, "created_at": "2026-11-22T03:29:34Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7fd90a9e6beb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "8697c74dcf2bb175c8303ef05bd3ebbaa8bbb8c26395dabf75318ba002052d5f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7fd90a9e6beb", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:29:34Z", "seed": 48904862, "treatment": "full"}, "priority": 2.009863469775993, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5780da9fd364d8306441a382c6537abcd2906b72a6faca2f78de6d5f9c134025", "run_seconds": 1.0498740673065186, "seed": 48904862, "started_at": "2026-11-22T03:29:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018873, 0.123147], "effect": 0.052137, "evaluated_at": "2026-11-22T03:29:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7fd90a9e6beb", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.534482759, 0.982142857, 1.0, 0.657894737], "frozen": [0.327586207, 0.294642857, 0.321428571, 0.315789474], "full": [0.629310345, 0.964285714, 1.0, 0.789473684], "retrieval": [0.698275862, 0.973214286, 1.0, 0.855263158], "rules_nodetect": [0.629310345, 0.964285714, 1.0, 0.789473684]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "8697c74dcf2bb175c8303ef05bd3ebbaa8bbb8c26395dabf75318ba002052d5f", "primary_control": "facts", "result_hash": "5780da9fd364d8306441a382c6537abcd2906b72a6faca2f78de6d5f9c134025", "signature": "a4994fc7736151b40206f174d7be29fa35ae60fa9d0299f9a224e893c7bbb0f8dca8ad4fe9b016b48833e916ced3b08d2bc9d6ec0803c6866ab0c8e3d61de204", "state": "VERIFIED", "summary": {"facts": 0.79363, "frozen": 0.314862, "full": 0.845767, "retrieval": 0.881688, "rules_nodetect": 0.845767}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:29:47Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:30:47Z", "cpu_seconds": 6.654493437, "created_at": "2026-11-22T03:29:34Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-4739040c9cb3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "848b11bc0a9723f8a508b46aa4edeb5b5ae7ad71af19ed5e16c094cfb241e452", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-4739040c9cb3", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:29:34Z", "seed": 183802319, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "cb9cf7afcbb2557b31acd197816b2efb9f622df77b636f46fd8809cfdbc38381", "run_seconds": 8.325444459915161, "seed": 183802319, "started_at": "2026-11-22T03:29:36Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:30:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-4739040c9cb3", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5335096237243402, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "848b11bc0a9723f8a508b46aa4edeb5b5ae7ad71af19ed5e16c094cfb241e452", "primary_control": "reset", "result_hash": "cb9cf7afcbb2557b31acd197816b2efb9f622df77b636f46fd8809cfdbc38381", "signature": "d956068ea867b57846d73334fd30a83a9979da5aa60005fdc01a8e7f35822074c7c81a1d2ab11fb031e813a6b59332064156783b1b57646f43c562ab25d83c0a", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:30:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:00Z", "cpu_seconds": 5.9977468290000004, "created_at": "2026-11-22T03:29:51Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-712e9d573b9b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 5, "n": 4}, "prereg_hash": "23366f093a92f31722ee79793b8e2356b33639550bef434c0e9c6074b51443c8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-712e9d573b9b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:29:51Z", "seed": 1329636926, "treatment": "full"}, "priority": 2.0092531761820296, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1ce863ec6c7631c0fa4334b5a15012170c3418c651cd68722a09f21953418b28", "run_seconds": 7.146985054016113, "seed": 1329636926, "started_at": "2026-11-22T03:29:53Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:30:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-712e9d573b9b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.5, 0.0], "full": [1.0, 0.0, 1.0, 1.0], "no_world_model": [1.0, 0.0, 1.0, 1.0], "random": [1.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "23366f093a92f31722ee79793b8e2356b33639550bef434c0e9c6074b51443c8", "primary_control": "e141_best", "result_hash": "1ce863ec6c7631c0fa4334b5a15012170c3418c651cd68722a09f21953418b28", "signature": "40683bec8beec8a50a9ec2301a88176c0c3a949cfd6f81e9fcb26a371f958a3e7afe966a67b2638cf80bae7e999e488733385beaf56d10a120ff2e6a95d74602", "state": "PARTIAL", "summary": {"e141_best": 0.75, "fixed_policy": 0.1875, "full": 0.75, "no_world_model": 0.75, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:30:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:23Z", "cpu_seconds": 2.9688663880000004, "created_at": "2026-11-22T03:30:07Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5b69c406b4a5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "fda034df2ad499260e983307c1932dc632cbc7be30ac631d04fd530ff2f66baa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5b69c406b4a5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:30:07Z", "seed": 1252048370, "treatment": "frontier"}, "priority": 2.0095189222203946, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fdafed59d2973b66b128d966b00a1f0cf3184578343c85eb8d7e720058d604df", "run_seconds": 4.473731517791748, "seed": 1252048370, "started_at": "2026-11-22T03:30:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.032461, 0.101525], "effect": 0.066993, "evaluated_at": "2026-11-22T03:30:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5b69c406b4a5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.329397875, 0.524122807, 0.209405501, 0.45212766], "frontier": [0.316410862, 0.453947368, 0.265306122, 0.432624113], "goal": [0.316410862, 0.375, 0.265306122, 0.432624113], "hypdisc": [0.207792208, 0.282894737, 0.204968944, 0.427304965], "infogain": [0.468713105, 0.25877193, 0.273291925, 0.432624113], "prederr": [0.387249115, 0.168859649, 0.193433895, 0.432624113], "random": [0.227863046, 0.438596491, 0.191659272, 0.342198582], "risk": [0.316410862, 0.407894737, 0.265306122, 0.432624113]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "fda034df2ad499260e983307c1932dc632cbc7be30ac631d04fd530ff2f66baa", "primary_control": "random", "result_hash": "fdafed59d2973b66b128d966b00a1f0cf3184578343c85eb8d7e720058d604df", "signature": "d262bb9b4c6d224f3bd5cd4e85b40e7a4f5541e611283979d2363f59957467560235a2b015c88e647a591b1aaa587a5601c4c3c60211f0e86e2f990bfc569f04", "state": "VERIFIED", "summary": {"count": 0.378763, "frontier": 0.367072, "goal": 0.347335, "hypdisc": 0.28074, "infogain": 0.35835, "prederr": 0.295542, "random": 0.300079, "risk": 0.355559}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:30:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:32Z", "cpu_seconds": 0.7926057989999995, "created_at": "2026-11-22T03:30:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8a6326934a54", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "f528898569790bfe48ac30c9fa3706d74ba6b40328a87b0d6876fc55d1323fcd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8a6326934a54", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:30:07Z", "seed": 226770539, "treatment": "full"}, "priority": 2.008002093198347, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1e60a6fb9457337a15dccfbda964f16b0bcb20a5e252c720d0f6ea1c34616d9f", "run_seconds": 1.5704100131988525, "seed": 226770539, "started_at": "2026-11-22T03:32:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.370785, 0.245785], "effect": -0.0625, "evaluated_at": "2026-11-22T03:32:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8a6326934a54", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [1.0, 1.0, 0.5, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [0.25, 0.5, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "f528898569790bfe48ac30c9fa3706d74ba6b40328a87b0d6876fc55d1323fcd", "primary_control": "e141_best", "result_hash": "1e60a6fb9457337a15dccfbda964f16b0bcb20a5e252c720d0f6ea1c34616d9f", "signature": "f1adbeeb56d3f809abb3267e609f735a24b48ebaa56c6bc7f1e031862b58e62580c96b42b7197313ae3ab8f11ac88fd917928c694ed77d3c8f7888f50a4a1909", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0625, "full": 0.875, "no_world_model": 0.9375, "random": 0.6875}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:32:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:28Z", "cpu_seconds": 0.8909173720000005, "created_at": "2026-11-22T03:30:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ecd6a9473a8e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "e96c240ae5ab293f1a80473987a55e099ccfe35a9b6df7a15e81d9fd9e0914d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ecd6a9473a8e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:30:07Z", "seed": 1986524321, "treatment": "full"}, "priority": 2.008048482283825, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "03c331b0ed31678590604174689ba5f728e733efbcf6de2b34b38120183d2589", "run_seconds": 1.18048095703125, "seed": 1986524321, "started_at": "2026-11-22T03:31:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T03:31:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ecd6a9473a8e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.25, 0.25], "full": [1.0, 1.0, 1.0, 0.25], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 0.75, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e96c240ae5ab293f1a80473987a55e099ccfe35a9b6df7a15e81d9fd9e0914d2", "primary_control": "e141_best", "result_hash": "03c331b0ed31678590604174689ba5f728e733efbcf6de2b34b38120183d2589", "signature": "37cb5dd484db23f858036485708fb45c0f8447f2403357619e222e2f87da0f97c3748e3ef76f7df073671b75f42a558c06617ad4a8961f05c7cb4e69c9f6da0f", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.125, "full": 0.8125, "no_world_model": 0.9375, "random": 0.875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:31:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:21Z", "cpu_seconds": 1.4665127019999993, "created_at": "2026-11-22T03:30:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0522907313ca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "2305eaa08d14f4f7121838e072f75c1eb67dae194bee4db813282cb7c19b50b2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0522907313ca", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:30:07Z", "seed": 198063920, "treatment": "full"}, "priority": 2.0093421698680585, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "be31975cf42ccba6160132af361a06850d0ceddebbc46ebe16afa4f23e185c3e", "run_seconds": 1.920900583267212, "seed": 198063920, "started_at": "2026-11-22T03:30:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.040895, 0.013263], "effect": -0.013816, "evaluated_at": "2026-11-22T03:30:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0522907313ca", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.739473684, 0.507894737, 0.752238806, 0.763157895], "frozen": [0.110526316, 0.110526316, 0.125373134, 0.115789474], "full": [0.739473684, 0.452631579, 0.752238806, 0.763157895], "retrieval": [0.739473684, 0.481578947, 0.752238806, 0.778947368], "rules_nodetect": [0.739473684, 0.452631579, 0.752238806, 0.763157895]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2305eaa08d14f4f7121838e072f75c1eb67dae194bee4db813282cb7c19b50b2", "primary_control": "facts", "result_hash": "be31975cf42ccba6160132af361a06850d0ceddebbc46ebe16afa4f23e185c3e", "signature": "d6eed94c4b9c41cd0d9a054e4e5841f8d1dff0d1809c92266ec05a332787eae4dc6512a60be619ddc9365f419cc1d420eb70957d67275a04da382ae1b869c60e", "state": "VERIFIED", "summary": {"facts": 0.690691, "frozen": 0.115554, "full": 0.676875, "retrieval": 0.68806, "rules_nodetect": 0.676875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:30:23Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:32:20Z", "cpu_seconds": 1.865158895, "created_at": "2026-11-22T03:30:07Z", "error": "lease expired; ", "family": "ow_model", "fence": 2, "id": "E141-ow_model-3d81896ef66a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "79fdeccaee18aac4c505666c2d94c46ccc0018c358b73c9f0485c42bec7d7e8a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3d81896ef66a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:30:07Z", "seed": 1319220529, "treatment": "full"}, "priority": 2.0086263791568224, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-a56a76034979", "result_hash": "279ad872097f06b102c2b78c8f7979f3e09d5c766854d5c070f6c3d3db5adb9d", "run_seconds": 3.5997161865234375, "seed": 1319220529, "started_at": "2026-11-22T03:30:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.00096, 0.032448], "effect": 0.016704, "evaluated_at": "2026-11-22T03:32:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3d81896ef66a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.929577465, 0.314009662, 0.9342723, 0.901408451], "frozen": [0.084507042, 0.082125604, 0.084507042, 0.084507042], "full": [0.9342723, 0.352657005, 0.938967136, 0.920187793], "retrieval": [0.981220657, 0.603864734, 0.971830986, 0.929577465], "rules_nodetect": [0.9342723, 0.352657005, 0.938967136, 0.920187793]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "79fdeccaee18aac4c505666c2d94c46ccc0018c358b73c9f0485c42bec7d7e8a", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-a56a76034979", "result_hash": "279ad872097f06b102c2b78c8f7979f3e09d5c766854d5c070f6c3d3db5adb9d", "signature": "d4b714de185c8b44c86afc6692de4e0270cad38c7e8f0abe3865e197106269df33f521c75183b933e8dce6c480ff1ffebaedb283e45326c60300e73fc11e6e07", "state": "VERIFIED", "summary": {"facts": 0.769817, "frozen": 0.083912, "full": 0.786521, "retrieval": 0.871623, "rules_nodetect": 0.786521}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:32:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:22Z", "cpu_seconds": 1.781470347, "created_at": "2026-11-22T03:30:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-caaa459f75e5", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "729c1e27eb7c23b708e3b0aa9081f8899dd7a57b8926e61b1c6edcf958221923", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-caaa459f75e5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:30:07Z", "seed": 1634084978, "treatment": "full"}, "priority": 2.0094713300938394, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "bd1c2193ca03a1fbbb4b6a7c927df82c1aeac7bbf2275bcce24b4c21eed443f4", "run_seconds": 3.231520891189575, "seed": 1634084978, "started_at": "2026-11-22T03:30:18Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:30:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-caaa459f75e5", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.761051], "frozen": [1.0, 0.158554], "full": [1.0, 0.817787], "retrieval": [1.0, 0.895584], "rules_nodetect": [1.0, 0.817787]}, "prereg_hash": "729c1e27eb7c23b708e3b0aa9081f8899dd7a57b8926e61b1c6edcf958221923", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "bd1c2193ca03a1fbbb4b6a7c927df82c1aeac7bbf2275bcce24b4c21eed443f4", "signature": "832585810368535882755d3f5ec4d3d7c2d7e9a697951a48201e868a3bf3c87f817ed8e08aabbe070eaadf94d9f5e753a58c3d9c8fdf990520490988c4b33402", "state": "REJECTED", "summary": {"facts": 0.761051, "frozen": 0.158554, "full": 0.817787, "retrieval": 0.895584, "rules_nodetect": 0.817787}}, "verified_at": "2026-11-22T03:30:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:40Z", "cpu_seconds": 0.79667595, "created_at": "2026-11-22T03:30:24Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-494b1abc371d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "6611b78db74b21821dd2673c63407e72f72a637d04eb5f12f45c926e9004d5fd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-494b1abc371d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:30:24Z", "seed": 1583467468, "treatment": "full"}, "priority": 2.0089772354084885, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "693c477e329858d939b3a773f34130869f880110bea634c628ff5d4deb3af24e", "run_seconds": 1.3762807846069336, "seed": 1583467468, "started_at": "2026-11-22T03:30:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013752, 0.133944], "effect": 0.060096, "evaluated_at": "2026-11-22T03:30:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-494b1abc371d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.6, 0.633333333, 1.0], "frozen": [0.357142857, 0.25, 0.316666667, 0.336538462], "full": [1.0, 0.733333333, 0.75, 0.990384615], "retrieval": [1.0, 0.766666667, 0.833333333, 1.0], "rules_nodetect": [1.0, 0.733333333, 0.75, 0.990384615]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6611b78db74b21821dd2673c63407e72f72a637d04eb5f12f45c926e9004d5fd", "primary_control": "facts", "result_hash": "693c477e329858d939b3a773f34130869f880110bea634c628ff5d4deb3af24e", "signature": "3c174664d3e0253c01fd970942a1a7b43b7c69340436c7435d661346e4a6ea5f8bf5f1334eefefe05b929fe5008f383c7b700c473a3167b98c00fd9c22974d0f", "state": "VERIFIED", "summary": {"facts": 0.808333, "frozen": 0.315087, "full": 0.868429, "retrieval": 0.9, "rules_nodetect": 0.868429}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:30:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:27Z", "cpu_seconds": 1.257000252000001, "created_at": "2026-11-22T03:30:24Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c26b530e6eb6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "de21995f6b6cd692e4b77f2c153d5daea41d8105e8a5805bfbd90386d6c7cbeb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c26b530e6eb6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:30:24Z", "seed": 411162310, "treatment": "full"}, "priority": 2.00945694677725, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "440a90a8f6b376327de0de895f93e8ceeef1d66154439b0018b94f452e6e679e", "run_seconds": 1.7009379863739014, "seed": 411162310, "started_at": "2026-11-22T03:30:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.038614, -0.017553], "effect": -0.028084, "evaluated_at": "2026-11-22T03:30:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c26b530e6eb6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.540425532, 0.530434783, 0.465957447, 0.513043478], "frozen": [0.134042553, 0.134782609, 0.127659574, 0.136231884], "full": [0.517021277, 0.513043478, 0.423404255, 0.484057971], "retrieval": [0.555319149, 0.545652174, 0.419148936, 0.484057971], "rules_nodetect": [0.655319149, 0.647826087, 0.423404255, 0.484057971]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "de21995f6b6cd692e4b77f2c153d5daea41d8105e8a5805bfbd90386d6c7cbeb", "primary_control": "facts", "result_hash": "440a90a8f6b376327de0de895f93e8ceeef1d66154439b0018b94f452e6e679e", "signature": "657a119086263d64db3434787b59dd34d45f92ff2a7182b8ca244c0a74fcbc0b91581336f990f073c2ea66707ee7afcafabda4c412f744dcf9330d73aee9f00d", "state": "VERIFIED", "summary": {"facts": 0.512465, "frozen": 0.133179, "full": 0.484382, "retrieval": 0.501045, "rules_nodetect": 0.552652}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:30:28Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:32:19Z", "cpu_seconds": 3.3396690600000003, "created_at": "2026-11-22T03:30:24Z", "error": "lease expired; ", "family": "world_model_frontier", "fence": 2, "id": "E141-world_model_frontier-bac5dab56b05", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "acfbe65c8c2788013c845c205666b4226f7015c9cc9d9f24149b4eb8b9b7dbda", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-bac5dab56b05", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:30:24Z", "seed": 113545532, "treatment": "frontier"}, "priority": 2.009055651262769, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-073a626dac3e", "result_hash": "049a59324dc15238aa62f5061adc0c806f4ce2da75e0f36ed878b785e451f8e8", "run_seconds": 4.607100486755371, "seed": 113545532, "started_at": "2026-11-22T03:30:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.038794, 0.097646], "effect": 0.06822, "evaluated_at": "2026-11-22T03:32:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-bac5dab56b05", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.823033708, 0.801724138, 1.0, 0.881188119, 0.932432432, 0.842391304, 0.839108911, 1.0, 0.852631579, 0.849431818, 0.85483871, 1.0], "frontier": [1.0, 1.0, 1.0, 0.992574257, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.912921348, 0.824712644, 1.0, 0.891089109, 0.925675676, 0.923913043, 0.930693069, 1.0, 0.960526316, 0.911931818, 0.892473118, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "acfbe65c8c2788013c845c205666b4226f7015c9cc9d9f24149b4eb8b9b7dbda", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-073a626dac3e", "result_hash": "049a59324dc15238aa62f5061adc0c806f4ce2da75e0f36ed878b785e451f8e8", "signature": "a8fd28bd52e907d6d1b4f53d836dd17eb47b8e727cea6f23e931934095771cf569399afd0162408b52fae36733bf1b6c7f1b0bb9dce27ec594eae267847d6a06", "state": "VERIFIED", "summary": {"curiosity": 0.889732, "frontier": 0.999381, "random_walk": 0.931161}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:32:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:30:56Z", "cpu_seconds": 3.6222167159999996, "created_at": "2026-11-22T03:30:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7a97aa2c72a3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "72a4b8899c1e5ea98c246dcdf41a02488b54e26e31984888a13ae2849d352b8b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7a97aa2c72a3", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:30:41Z", "seed": 2051767168, "treatment": "frontier"}, "priority": 2.0098381003161316, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "10c9738b46ebd2bfd4ffbfe9b912e20408ce3ef23d39c9b4c32cb589daec6b6e", "run_seconds": 3.993443012237549, "seed": 2051767168, "started_at": "2026-11-22T03:30:52Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:30:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7a97aa2c72a3", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "72a4b8899c1e5ea98c246dcdf41a02488b54e26e31984888a13ae2849d352b8b", "primary_control": "random", "result_hash": "10c9738b46ebd2bfd4ffbfe9b912e20408ce3ef23d39c9b4c32cb589daec6b6e", "signature": "607ce01031ddd4d16991dde9679235880f3aa48c8d54775e7cf4919677728af36dadd8a07560adb7e59b7aa764f091e1a9744cf0e19bfc9ee848d11c75631806", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:30:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:10Z", "cpu_seconds": 4.870457771, "created_at": "2026-11-22T03:30:41Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1e082537f8d4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 7, "n": 4}, "prereg_hash": "d6d3e97b19d0d871a4aae9ca4cea9684b0ca83b03debcd96b9201306999359db", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1e082537f8d4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:30:41Z", "seed": 333257448, "treatment": "full"}, "priority": 2.0093230658145638, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2e1c9323b354362ccef97bf0485f321c335712cba0b1840b4ba9c2537e084500", "run_seconds": 5.72603964805603, "seed": 333257448, "started_at": "2026-11-22T03:31:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.230917, 0.605917], "effect": 0.1875, "evaluated_at": "2026-11-22T03:31:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1e082537f8d4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 1.0, 0.5], "fixed_policy": [0.75, 0.75, 0.75, 0.75], "full": [1.0, 1.0, 0.75, 0.75], "no_world_model": [0.25, 0.75, 0.75, 0.5], "random": [1.0, 0.75, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d6d3e97b19d0d871a4aae9ca4cea9684b0ca83b03debcd96b9201306999359db", "primary_control": "e141_best", "result_hash": "2e1c9323b354362ccef97bf0485f321c335712cba0b1840b4ba9c2537e084500", "signature": "1821e3a498b7172d27298f51ab688415c4ab59d4c92c9c52d986753687cab9705826119b9fc5ca605a8df4e9859b5f83859e6a3d22b8c4ba3e8b6b83c7a9e20a", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.75, "full": 0.875, "no_world_model": 0.5625, "random": 0.875}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:31:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:06Z", "cpu_seconds": 1.6962108959999997, "created_at": "2026-11-22T03:30:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-720329efb621", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "e660352594e7d77d5798da4c29477e93127c35ea9afbf703ef3c6b400fce7a7a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-720329efb621", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:30:41Z", "seed": 562954480, "treatment": "full"}, "priority": 2.009792269904539, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b78a09b78631c6b66fe4dd8f872b5a5471d26c84326465433a248fbdffc0fce7", "run_seconds": 2.2599098682403564, "seed": 562954480, "started_at": "2026-11-22T03:31:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013412, 0.312023], "effect": 0.149306, "evaluated_at": "2026-11-22T03:31:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-720329efb621", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.680555556, 1.0, 0.986111111, 0.736111111], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.916666667, 1.0, 0.993055556, 0.916666667], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "e660352594e7d77d5798da4c29477e93127c35ea9afbf703ef3c6b400fce7a7a", "primary_control": "facts", "result_hash": "b78a09b78631c6b66fe4dd8f872b5a5471d26c84326465433a248fbdffc0fce7", "signature": "f7ea1a1a57b39bbb11756d57bda127ab05239183233c472ff44790a14f7c75ba6475d981e5ef5644c8130d68dd7a0c1e0ffda7c0b4e33e661c7ce8544d44fb0c", "state": "VERIFIED", "summary": {"facts": 0.850694, "frozen": 0.0, "full": 1.0, "retrieval": 0.956597, "rules_nodetect": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:31:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:11Z", "cpu_seconds": 0.621201567, "created_at": "2026-11-22T03:30:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d1e9fa9eac48", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "a0a437f64edc230966c67fb90acefd0dd0eee80300500714a393edc9668e1658", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d1e9fa9eac48", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:30:41Z", "seed": 1909510458, "treatment": "full"}, "priority": 2.009243845609296, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0fc4fdaa461f867d54cc3f2db2a6063094a193c51d31cd562871729b23a1d03a", "run_seconds": 0.6920959949493408, "seed": 1909510458, "started_at": "2026-11-22T03:31:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.011341, 0.073221], "effect": 0.03094, "evaluated_at": "2026-11-22T03:31:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d1e9fa9eac48", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.787439614, 0.806451613, 0.846153846, 0.633986928], "frozen": [0.299516908, 0.35483871, 0.41025641, 0.320261438], "full": [0.787439614, 0.838709677, 0.846153846, 0.725490196], "retrieval": [0.787439614, 0.806451613, 0.846153846, 0.738562092], "rules_nodetect": [0.758454106, 0.838709677, 0.846153846, 0.816993464]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "a0a437f64edc230966c67fb90acefd0dd0eee80300500714a393edc9668e1658", "primary_control": "facts", "result_hash": "0fc4fdaa461f867d54cc3f2db2a6063094a193c51d31cd562871729b23a1d03a", "signature": "974fc933f18926461e8c8cca07eda03e7163ce155b62a2a4fd6930ea6976a3bdd0ffd6ee06b6c4d319fd9bb4a0985e0722c321542d6d7f761fbcc121b07f860b", "state": "VERIFIED", "summary": {"facts": 0.768508, "frozen": 0.346218, "full": 0.799448, "retrieval": 0.794652, "rules_nodetect": 0.815078}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:31:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:10Z", "cpu_seconds": 7.673038644, "created_at": "2026-11-22T03:30:57Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0b8aa43f1560", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f5250b6f1c669eb39bc3320de34a6de1db2a47a871d94c00e789e36ea92677b5", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0b8aa43f1560", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:30:57Z", "seed": 1009322049, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "b4530412bf68b953fa9f8fdae89aaa0387d85cff282aa5ef794bf6b6d51c19c6", "run_seconds": 8.83215069770813, "seed": 1009322049, "started_at": "2026-11-22T03:31:01Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.029275, 0.229275], "effect": 0.1, "evaluated_at": "2026-11-22T03:31:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0b8aa43f1560", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.9786, "losses": 5, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.105", "prereg_hash": "f5250b6f1c669eb39bc3320de34a6de1db2a47a871d94c00e789e36ea92677b5", "primary_control": "reset", "result_hash": "b4530412bf68b953fa9f8fdae89aaa0387d85cff282aa5ef794bf6b6d51c19c6", "signature": "627d60920f30af2b22f29aec820fdaa7667081e46afb9ffefa866cbb5a71c7483e0c47b479ee73e576ba82407df58a8ed300c76919d6188ddf9405110075ad0f", "state": "VERIFIED", "summary": {"inherit": 0.7166666666666667, "reset": 0.6166666666666667, "root": 0.2833333333333333}, "ties": 44, "treatment": "inherit", "wins": 11}, "verified_at": "2026-11-22T03:31:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:11Z", "cpu_seconds": 6.540868507999999, "created_at": "2026-11-22T03:30:57Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-c582023ce55b", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "3866ccc1cb444e41306629911de0208a92db6791d1082628b25d07d5e88bc039", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-c582023ce55b", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:30:57Z", "seed": 1389354270, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "5351935e7f3c94a48fb8894f8d5789d6f0746a101b4ce22e56192244bd415f0a", "run_seconds": 7.523513078689575, "seed": 1389354270, "started_at": "2026-11-22T03:31:04Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:31:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-c582023ce55b", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3866ccc1cb444e41306629911de0208a92db6791d1082628b25d07d5e88bc039", "primary_control": "reset", "result_hash": "5351935e7f3c94a48fb8894f8d5789d6f0746a101b4ce22e56192244bd415f0a", "signature": "a4c39b73bb1a0a718480a09dcd2fac57cdde14219b7ddf9875d74a9a5f58bab6114e1bbea7dc6d73fe5433ea4a55d7e4fe7ed66e798a2c16737a8e6ad1184600", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.3}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:31:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:15Z", "cpu_seconds": 0.17283697500000006, "created_at": "2026-11-22T03:31:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-160b4d3f97ab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "a59f246b249ef07d73529f3aec0dd916b9e7fc2e666d40b1ff983a427da74832", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-160b4d3f97ab", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:31:12Z", "seed": 2143791838, "treatment": "full"}, "priority": 2.009974693843192, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c88de47227fa634c8ed56c7e89ce1e69dd31a3564d30dff86e2c2f9fde372112", "run_seconds": 0.18891525268554688, "seed": 2143791838, "started_at": "2026-11-22T03:31:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003636, 0.011212], "effect": 0.003788, "evaluated_at": "2026-11-22T03:31:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-160b4d3f97ab", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.939393939, 1.0, 1.0], "frozen": [0.015151515, 0.03030303, 0.03030303, 0.03030303], "full": [1.0, 0.954545455, 1.0, 1.0], "retrieval": [1.0, 0.954545455, 1.0, 1.0], "rules_nodetect": [1.0, 0.954545455, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a59f246b249ef07d73529f3aec0dd916b9e7fc2e666d40b1ff983a427da74832", "primary_control": "facts", "result_hash": "c88de47227fa634c8ed56c7e89ce1e69dd31a3564d30dff86e2c2f9fde372112", "signature": "d725db1c095a132f8b957218a2d872aba1470345372a3c22ad93b3a8afd2b10736b19c77afbf7d4fb3dc7863789d3494f3f9b825991d7974488722be9f6eb300", "state": "VERIFIED", "summary": {"facts": 0.984848, "frozen": 0.026515, "full": 0.988636, "retrieval": 0.988636, "rules_nodetect": 0.988636}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:31:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:52Z", "cpu_seconds": 7.5028694950000006, "created_at": "2026-11-22T03:31:28Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-8f06e65e8a6b", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "726e1c0f92af57daa537e655b46708e7c7d0a262c699782122c5a93d7fb76635", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-8f06e65e8a6b", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:31:28Z", "seed": 425004927, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9070b4cf8c3780f587b4071992eb42622de65292168d20968403d35c74ed1ebd", "run_seconds": 9.904353141784668, "seed": 425004927, "started_at": "2026-11-22T03:31:42Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:31:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-8f06e65e8a6b", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4610856650547188, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "726e1c0f92af57daa537e655b46708e7c7d0a262c699782122c5a93d7fb76635", "primary_control": "reset", "result_hash": "9070b4cf8c3780f587b4071992eb42622de65292168d20968403d35c74ed1ebd", "signature": "3015665d5495e5c11436de92ddefba58030f2d4c96e7d3b4db1b95e6d6916abdb3b8fd04a9d4cbc305ec9fc717b27eced60dd399c642c8e36883584cb2b92204", "state": "PARTIAL", "summary": {"inherit": 0.7166666666666667, "reset": 0.7166666666666667, "root": 0.4666666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:32:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:37Z", "cpu_seconds": 5.259916284, "created_at": "2026-11-22T03:31:28Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-de12cdd29635", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "392a6796be4f1cbfa9994cb591850e0672df127d085d1731f85807f7a7c42ff5", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-de12cdd29635", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:31:28Z", "seed": 667442522, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "c0daed030eb1eae526282b9785171bf25bab167779df80d94706f2769be782eb", "run_seconds": 5.90496826171875, "seed": 667442522, "started_at": "2026-11-22T03:31:31Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:31:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-de12cdd29635", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "392a6796be4f1cbfa9994cb591850e0672df127d085d1731f85807f7a7c42ff5", "primary_control": "reset", "result_hash": "c0daed030eb1eae526282b9785171bf25bab167779df80d94706f2769be782eb", "signature": "1ef086706be15726076cdbd41323a0519d297ff6a444be34d9c0c7ff98e9e9208cebfdf0fc47586f0711793a470d9d763e21b10972290d93136d4a30b67fc80b", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:31:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:33Z", "cpu_seconds": 0.1848424780000002, "created_at": "2026-11-22T03:31:29Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3166f0ba049d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "51c4c4670b93a86a4d3a0bcbc8c2860372c1e83ee4c95b46cbdb6d11e9fc6aa3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3166f0ba049d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:31:29Z", "seed": 329166557, "treatment": "full"}, "priority": 2.009929917944211, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1b309b0805421c9043ed3dea7a7279ba0b2df96131968448ffc3c3d33214b549", "run_seconds": 0.21772432327270508, "seed": 329166557, "started_at": "2026-11-22T03:31:33Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:31:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3166f0ba049d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.5, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "51c4c4670b93a86a4d3a0bcbc8c2860372c1e83ee4c95b46cbdb6d11e9fc6aa3", "primary_control": "e141_best", "result_hash": "1b309b0805421c9043ed3dea7a7279ba0b2df96131968448ffc3c3d33214b549", "signature": "55d6c0c0eac3e4d00da0513ffec21db27752c3fa8fa2c94718c37786958dbc510399ad97b06b01534c50a91d8b4a6fbfa7e0ce1ed47676b1c2125b61d3fe5500", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:31:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:40Z", "cpu_seconds": 0.975756616, "created_at": "2026-11-22T03:31:29Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4287b43e1d4e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "af2ecabf9dfdafc786b3a938a157319085be6600d45d6425a1388dff04454e9c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4287b43e1d4e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:31:29Z", "seed": 662241576, "treatment": "full"}, "priority": 2.0098189766347825, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f98c3c3a5d07dbb27fd94506480c487e3cc4992020ec930c536f3c99f7f88c3d", "run_seconds": 1.639711856842041, "seed": 662241576, "started_at": "2026-11-22T03:31:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.091188, 0.253455], "effect": 0.172321, "evaluated_at": "2026-11-22T03:31:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4287b43e1d4e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.907142857, 0.708333333, 0.716666667, 0.833333333], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.971428571, 0.966666667, 0.925, 0.991666667], "retrieval": [0.985714286, 0.9, 0.933333333, 0.958333333], "rules_nodetect": [0.971428571, 0.966666667, 0.925, 0.991666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "af2ecabf9dfdafc786b3a938a157319085be6600d45d6425a1388dff04454e9c", "primary_control": "facts", "result_hash": "f98c3c3a5d07dbb27fd94506480c487e3cc4992020ec930c536f3c99f7f88c3d", "signature": "41a055a9165ad8d9d2dda559d3937ee1853d98aedb5d1dc4ca33d421869a5580308647d926c0a65dd0c74211be4fb2aa66e578b1b339058309f86d65c7df7c08", "state": "VERIFIED", "summary": {"facts": 0.791369, "frozen": 0.0, "full": 0.96369, "retrieval": 0.944345, "rules_nodetect": 0.96369}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:31:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:49Z", "cpu_seconds": 2.6401889710000006, "created_at": "2026-11-22T03:31:29Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-a786a9528d86", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "c41dd6146141216f57bf5b046d2e7aeb81af222b9ec5f1b2c6e15e968f864bf4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-a786a9528d86", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:31:29Z", "seed": 348607642, "treatment": "frontier"}, "priority": 2.0097941086577937, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-7256712810e8", "result_hash": "bcf46da7f8d6de331effbf85fd307e87da81630b55026b297bfa186259c55c0b", "run_seconds": 4.30633544921875, "seed": 348607642, "started_at": "2026-11-22T03:31:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.016786, 0.077562], "effect": 0.047174, "evaluated_at": "2026-11-22T03:31:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-a786a9528d86", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.925233645, 0.789855072, 0.845360825, 1.0, 1.0, 0.862903226, 0.789893617, 1.0, 1.0, 1.0, 0.804123711, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.920560748, 0.894927536, 0.920103093, 1.0, 1.0, 0.943548387, 0.904255319, 1.0, 1.0, 1.0, 0.850515464, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.8062, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.0156", "prereg_hash": "c41dd6146141216f57bf5b046d2e7aeb81af222b9ec5f1b2c6e15e968f864bf4", "primary_control": "random_walk", "replicated": false, "replication_of": "E141-world_model_frontier-7256712810e8", "result_hash": "bcf46da7f8d6de331effbf85fd307e87da81630b55026b297bfa186259c55c0b", "signature": "8aa146f21856193eebbd23be0cbcb38f0f6bad887a52d21cdf74da9c480ced69091ec487e631b5c666953f9a7f99c032bcb05adee510870821edc171de3b1a07", "state": "VERIFIED", "summary": {"curiosity": 0.918114, "frontier": 1.0, "random_walk": 0.952826}, "ties": 6, "treatment": "frontier", "wins": 6}, "verified_at": "2026-11-22T03:31:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:55Z", "cpu_seconds": 0.9758951570000001, "created_at": "2026-11-22T03:31:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2bcb7ebd44fc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "78470092814c2d86f783e3db69080236b765eda89b373bc7cee9ddd63d2a6e6d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2bcb7ebd44fc", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:31:45Z", "seed": 1979294926, "treatment": "full"}, "priority": 2.0097634902825305, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d98ff3f9520a12041a018d9959c12da999fd96790cfb749d0ada25c95f68d990", "run_seconds": 1.1665115356445312, "seed": 1979294926, "started_at": "2026-11-22T03:31:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.155861, 1.094139], "effect": 0.625, "evaluated_at": "2026-11-22T03:32:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2bcb7ebd44fc", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 0.5, 0.0], "fixed_policy": [0.25, 0.25, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 0.5, 0.75], "random": [0.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "78470092814c2d86f783e3db69080236b765eda89b373bc7cee9ddd63d2a6e6d", "primary_control": "e141_best", "result_hash": "d98ff3f9520a12041a018d9959c12da999fd96790cfb749d0ada25c95f68d990", "signature": "864f65d1388ef01cf79454f926fb7d55b29691c50681e1a2fc14f1ebfcbe6284b974741c4928e03afd2e045799cbf6b6d0725d47cdba7dc4fe7e3f25d5d8f900", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 0.6875, "random": 0.5}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:32:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:02Z", "cpu_seconds": 1.2465253780000012, "created_at": "2026-11-22T03:31:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7b2f4419251e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "18359d77f9cd606dcddd0aeebce762033b66c4ae7b7629f9c227461428ea69fc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7b2f4419251e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:31:45Z", "seed": 226574783, "treatment": "full"}, "priority": 2.009221232733356, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "23b0b294831cd50a963e66c767aaccb138cd94fd3ccc521cc839f1c0556a6258", "run_seconds": 2.373969793319702, "seed": 226574783, "started_at": "2026-11-22T03:32:00Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7b2f4419251e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "18359d77f9cd606dcddd0aeebce762033b66c4ae7b7629f9c227461428ea69fc", "primary_control": "e141_best", "result_hash": "23b0b294831cd50a963e66c767aaccb138cd94fd3ccc521cc839f1c0556a6258", "signature": "0502490e3918e9332cc1cef192d752fb50f22ef3d418d4324df272c664514f1227e2bb80d89d9da87de58fd02402248f9b114352412557d9de4f8bc8df0bf40e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:32:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:31:49Z", "cpu_seconds": 0.3056041950000008, "created_at": "2026-11-22T03:31:45Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f26a07e68523", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "8c0044f8a48e6487fbeef069d51f1e0889dad66996b04f62811fbc9df8a2d4ec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f26a07e68523", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:31:45Z", "seed": 1231441675, "treatment": "full"}, "priority": 2.0099094341193693, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a6c39cbf0b68225718aa501924b58f1dec9afb709df46039233341d0add00cf3", "run_seconds": 0.3717811107635498, "seed": 1231441675, "started_at": "2026-11-22T03:31:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.056659, 0.160528], "effect": 0.108594, "evaluated_at": "2026-11-22T03:31:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f26a07e68523", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.9, 0.9625, 0.8125, 0.828125], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.96875, 0.96875], "retrieval": [0.975, 0.9875, 0.984375, 0.9375], "rules_nodetect": [1.0, 1.0, 0.96875, 0.96875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "8c0044f8a48e6487fbeef069d51f1e0889dad66996b04f62811fbc9df8a2d4ec", "primary_control": "facts", "result_hash": "a6c39cbf0b68225718aa501924b58f1dec9afb709df46039233341d0add00cf3", "signature": "a31113663b90bb330f0b8005442cf53a87b3458f389034f6dff2c43ff49e72cbf146c242b131cae09867360cf4e59fd09027cc08f6c1f870181456909acc9707", "state": "VERIFIED", "summary": {"facts": 0.875781, "frozen": 0.0, "full": 0.984375, "retrieval": 0.971094, "rules_nodetect": 0.984375}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:31:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:33Z", "cpu_seconds": 1.5428999459999986, "created_at": "2026-11-22T03:32:02Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-a62b6ee4a184", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "4389e6dfbe38a141e52770240b9763a1aedb1499e13aab1b808845b56947e8b7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-a62b6ee4a184", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:32:02Z", "seed": 939686757, "treatment": "iw_temperature"}, "priority": 2.0085223432019723, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-03a1cb12fe67", "result_hash": "073ea5392b21780fe6fcf8a9603112d6abc4e4664511f2f8027e43d763e430f3", "run_seconds": 2.4564497470855713, "seed": 939686757, "started_at": "2026-11-22T03:32:31Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017475, 0.015742], "effect": -0.000866, "evaluated_at": "2026-11-22T03:32:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-a62b6ee4a184", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.28392898, 0.219396228, 0.222856315, 0.133106048, 0.204124462, 0.189099313, 0.233532138, 0.262861282, 0.21443101, 0.26177777, 0.197707127, 0.30564979, 0.23516586, 0.275448417, 0.317000155, 0.243984042], "raw": [0.273676252, 0.225619975, 0.155283395, 0.182099745, 0.239309427, 0.189099313, 0.258180855, 0.238776215, 0.165564738, 0.308327888, 0.197707127, 0.22392137, 0.22671515, 0.23799151, 0.295887545, 0.191951192], "temperature": [0.28392898, 0.20729877, 0.222856315, 0.160775447, 0.256509265, 0.21755422, 0.258180855, 0.233293652, 0.22942025, 0.308327888, 0.183502982, 0.260505793, 0.23516586, 0.262621635, 0.30138954, 0.164876877]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1491, "losses": 7, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.709", "prereg_hash": "4389e6dfbe38a141e52770240b9763a1aedb1499e13aab1b808845b56947e8b7", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-03a1cb12fe67", "result_hash": "073ea5392b21780fe6fcf8a9603112d6abc4e4664511f2f8027e43d763e430f3", "signature": "a677fcd26e4dfc8e190217763816f3f8fe90e9107dae00ab344dab52e9e897f938c5595d8672987077e047fbc52b96be75e328f960afa555bc7017a397913704", "state": "VERIFIED", "summary": {"iw_temperature": 0.237504, "raw": 0.225632, "temperature": 0.236638}, "ties": 3, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:32:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:10Z", "cpu_seconds": 0.883557995, "created_at": "2026-11-22T03:32:02Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e1c8e6d2f565", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "f37c47f02c34f4533a33662bd5f500711841ebe2078f09581bedde6fb822ba3a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e1c8e6d2f565", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:32:02Z", "seed": 1375652811, "treatment": "full"}, "priority": 2.0093981368390947, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5a660726fdbd0fa4f7e2eb61dbc380572f69ee2dc05ca490e3a97fd2caaad8ba", "run_seconds": 1.0250861644744873, "seed": 1375652811, "started_at": "2026-11-22T03:32:09Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e1c8e6d2f565", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.5, 0.0, 0.25], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [0.5, 0.5, 0.75, 0.75], "random": [0.75, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f37c47f02c34f4533a33662bd5f500711841ebe2078f09581bedde6fb822ba3a", "primary_control": "e141_best", "result_hash": "5a660726fdbd0fa4f7e2eb61dbc380572f69ee2dc05ca490e3a97fd2caaad8ba", "signature": "7bff32c232ab5177fc4885aa965846f1d38a4234916a0f26773f55aeae656e293e2928d7777567d5849b8e136943c7c585f31406b7068a2bcfb279ed6e81ec03", "state": "PARTIAL", "summary": {"e141_best": 0.9375, "fixed_policy": 0.1875, "full": 0.9375, "no_world_model": 0.625, "random": 0.9375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:32:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:11Z", "cpu_seconds": 1.969064703, "created_at": "2026-11-22T03:32:02Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8104d328b1ef", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "prereg_hash": "0395222cbf22e579783e801dd7f7a64e4ee4a9093d1a3725d51e34abf4e3273b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8104d328b1ef", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:32:02Z", "seed": 290864797, "treatment": "gated"}, "priority": 2.0097647598151935, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "1954397a98f49b5e8588cc38dea23e9e976151ade9171fb8e6e7a00b3a60cd33", "run_seconds": 2.593536853790283, "seed": 290864797, "started_at": "2026-11-22T03:32:08Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-8104d328b1ef", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.616666667, 0.75, 0.733333333, 0.8], "gated": [0.616666667, 0.75, 0.733333333, 0.8], "naive": [0.3, 0.416666667, 0.35, 0.2], "scratch": [0.616666667, 0.75, 0.733333333, 0.8], "ungated": [0.566666667, 0.6, 0.55, 0.283333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0395222cbf22e579783e801dd7f7a64e4ee4a9093d1a3725d51e34abf4e3273b", "primary_control": "scratch", "result_hash": "1954397a98f49b5e8588cc38dea23e9e976151ade9171fb8e6e7a00b3a60cd33", "signature": "6cef4e6f97d5453a57bbaaf376300a6ec2220d3454f1f28fea472d8c72d36720725bb54f4cd324b1614e742d2986bebcc94b89f6add0a09da9896328421cae02", "state": "PARTIAL", "summary": {"compressed": 0.725, "gated": 0.725, "naive": 0.316667, "scratch": 0.725, "ungated": 0.5}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:32:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:11Z", "cpu_seconds": 7.665648600000001, "created_at": "2026-11-22T03:32:02Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-148addb390ce", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "043b20054285614e063f64bf4143a54f2b2a21a5fdb642293ac35f14f79106ba", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-148addb390ce", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:32:02Z", "seed": 942093336, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "366f7927d81e66a23be89a78807a71d1ca1195188d6cb283283d6f6c39b8d217", "run_seconds": 8.835033416748047, "seed": 942093336, "started_at": "2026-11-22T03:32:02Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-148addb390ce", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "043b20054285614e063f64bf4143a54f2b2a21a5fdb642293ac35f14f79106ba", "primary_control": "reset", "result_hash": "366f7927d81e66a23be89a78807a71d1ca1195188d6cb283283d6f6c39b8d217", "signature": "c5bf7c1969ab025ed0a8233158bec469844a0ca3ad37c0de83cca580aa703d5f088c9082417a783cd9af0c90c34caa4dd8fd62e21a754d3634d861eb2bb99e0e", "state": "PARTIAL", "summary": {"inherit": 0.5166666666666667, "reset": 0.5166666666666667, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:32:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:19Z", "cpu_seconds": 8.762997705, "created_at": "2026-11-22T03:32:02Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-dc084b91bf6d", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "01c19b7489ca70635e11ab104f8ea9bf34c41c79e9733e9ddd0c10157617b66c", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-dc084b91bf6d", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:32:02Z", "seed": 1507748818, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "5cdd0265430f9b11e429ff9da80ebff723e1cf3d1b0671d022565ae0edc11bbe", "run_seconds": 10.855977773666382, "seed": 1507748818, "started_at": "2026-11-22T03:32:08Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-dc084b91bf6d", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0], "root": [0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "01c19b7489ca70635e11ab104f8ea9bf34c41c79e9733e9ddd0c10157617b66c", "primary_control": "reset", "result_hash": "5cdd0265430f9b11e429ff9da80ebff723e1cf3d1b0671d022565ae0edc11bbe", "signature": "97a27f98cfb4b42b3d0fca00e46d4d11fd6509a438c719d3cb77794c5632559c2bf14667b5f5d1e38359c8c20823f9614e52d632c111952bc80bf14ab426b409", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:32:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:35Z", "cpu_seconds": 3.378652721, "created_at": "2026-11-22T03:32:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d34af031c057", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "8211f2e5f795d241a5d36b05c480f3a3bba1506c605e5683331b29e35088f680", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d34af031c057", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:32:19Z", "seed": 1538976769, "treatment": "frontier"}, "priority": 2.0087425822484515, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "bf42240ef396623b11364c04c6649d300ef44549d326e8567e32d9e55169d2ba", "run_seconds": 4.593533992767334, "seed": 1538976769, "started_at": "2026-11-22T03:32:30Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d34af031c057", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8211f2e5f795d241a5d36b05c480f3a3bba1506c605e5683331b29e35088f680", "primary_control": "random", "result_hash": "bf42240ef396623b11364c04c6649d300ef44549d326e8567e32d9e55169d2ba", "signature": "be8d4144b7e4db358b2af6428f5ff15f67f5cbc08d28eedad4ae045efb5e3e7083c4b1a84b1e4e03b18cfb602e5b73e3cb0e07432a24f98698177aab11ad430c", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:32:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:26Z", "cpu_seconds": 1.106822845, "created_at": "2026-11-22T03:32:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-55e6be2ad704", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "45bd179d60e01166dd5e02bbe5a39e9d82b19fe805a06e0cf371181323a4f718", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-55e6be2ad704", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:32:19Z", "seed": 1389631692, "treatment": "full"}, "priority": 2.009726529279195, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7decb8d55d62b8239cad1bd8f06e5b25620bd1582f437f73961887aea0a5299b", "run_seconds": 1.238062858581543, "seed": 1389631692, "started_at": "2026-11-22T03:32:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.230917, 0.605917], "effect": 0.1875, "evaluated_at": "2026-11-22T03:32:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-55e6be2ad704", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 0.25, 1.0], "fixed_policy": [0.0, 0.0, 1.0, 0.25], "full": [0.75, 1.0, 1.0, 0.75], "no_world_model": [0.75, 0.5, 0.25, 0.0], "random": [0.5, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "45bd179d60e01166dd5e02bbe5a39e9d82b19fe805a06e0cf371181323a4f718", "primary_control": "e141_best", "result_hash": "7decb8d55d62b8239cad1bd8f06e5b25620bd1582f437f73961887aea0a5299b", "signature": "a41bd51f513d985647e1967f0dc1e871140c5b3730df1f184f2e449a7c76740da1f0900678ed69a6439244a665e8eff99612ebe8dd93561e422ac9dddfc3a307", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.3125, "full": 0.875, "no_world_model": 0.375, "random": 0.875}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:32:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:25Z", "cpu_seconds": 0.6677121980000003, "created_at": "2026-11-22T03:32:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c79d51c846c3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "08bc614a2b07e57817667598744fa1d63f5aa200de6c23f1c977e2eb9d1a7545", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c79d51c846c3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:32:19Z", "seed": 868609120, "treatment": "full"}, "priority": 2.009781300148337, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e3bcb26db97a15d0289a823329df310375200142f6678ba6c333151e9c850af8", "run_seconds": 0.784787654876709, "seed": 868609120, "started_at": "2026-11-22T03:32:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.042439, 0.073256], "effect": 0.015409, "evaluated_at": "2026-11-22T03:32:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c79d51c846c3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.18, 0.54893617, 0.709090909, 0.672727273], "frozen": [0.13, 0.127659574, 0.109090909, 0.109090909], "full": [0.125, 0.538297872, 0.781818182, 0.727272727], "retrieval": [0.08, 0.557446809, 0.727272727, 0.690909091], "rules_nodetect": [0.0625, 0.525531915, 0.272727273, 0.090909091]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "08bc614a2b07e57817667598744fa1d63f5aa200de6c23f1c977e2eb9d1a7545", "primary_control": "facts", "result_hash": "e3bcb26db97a15d0289a823329df310375200142f6678ba6c333151e9c850af8", "signature": "7f4e86dcb1d6b860f7679a6e7d9a55480a83268e674073201b168f9cb57dee2aff5400f721f87c1db6fdc2b374c26e5028fb2b6b01db269d930e495a9a3f4201", "state": "VERIFIED", "summary": {"facts": 0.527689, "frozen": 0.11896, "full": 0.543097, "retrieval": 0.513907, "rules_nodetect": 0.237917}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:32:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:26Z", "cpu_seconds": 0.875573428, "created_at": "2026-11-22T03:32:19Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-bda8fea23a83", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "93830a488735a7641edb7fd8dbabc71b44e090d13cc79d5c950263a7f52dd908", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-bda8fea23a83", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:32:19Z", "seed": 2011354510, "treatment": "gated"}, "priority": 2.0087484348508178, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "02af447331fb06fb9b9f3cc6774bab8b4c0d44c2d7f9dc2addec4d8b7b885208", "run_seconds": 0.9728968143463135, "seed": 2011354510, "started_at": "2026-11-22T03:32:25Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-bda8fea23a83", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.712820513, 0.846153846, 0.742857143, 0.820512821], "gated": [0.712820513, 0.846153846, 0.742857143, 0.820512821], "naive": [0.692307692, 0.846153846, 0.704761905, 0.846153846], "scratch": [0.712820513, 0.846153846, 0.742857143, 0.820512821], "ungated": [0.692307692, 0.846153846, 0.704761905, 0.846153846]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "93830a488735a7641edb7fd8dbabc71b44e090d13cc79d5c950263a7f52dd908", "primary_control": "scratch", "result_hash": "02af447331fb06fb9b9f3cc6774bab8b4c0d44c2d7f9dc2addec4d8b7b885208", "signature": "031cb35e350df8bed11769ced6ca74b608c8348eae041299b3a4017974f2e1c025a8e1e4b802346e312b4c20b84cb3796f7c6835dbbbbfe8cbdf945c19a97e09", "state": "PARTIAL", "summary": {"compressed": 0.780586, "gated": 0.780586, "naive": 0.772344, "scratch": 0.780586, "ungated": 0.772344}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:32:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:45Z", "cpu_seconds": 1.6347215549999996, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-c3866ea5bf79", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "bcbaa22ee1746df1a5cb611577e1f7b19261dbb85d498fa7c0032bce4221bb77", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-c3866ea5bf79", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:32:35Z", "seed": 1136469186, "treatment": "iw_temperature"}, "priority": 2.009161930373275, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-96b14b5a0efa", "result_hash": "627fe133baeed03c1098df932e632865ce8bf0b01dbd80a4420bfdcf1487d252", "run_seconds": 1.8137905597686768, "seed": 1136469186, "started_at": "2026-11-22T03:32:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.012559, 0.013958], "effect": 0.000699, "evaluated_at": "2026-11-22T03:32:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-c3866ea5bf79", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.29792968, 0.280957085, 0.271842933, 0.168122892, 0.217851902, 0.23380611, 0.180094525, 0.268175977, 0.275553432, 0.27720574, 0.279815085, 0.30274204, 0.178670572, 0.257708985, 0.247090737, 0.234898748], "raw": [0.28727078, 0.274858965, 0.197268573, 0.139506668, 0.25074468, 0.204352112, 0.160010382, 0.276583973, 0.293419545, 0.237821278, 0.26590895, 0.290897868, 0.178670572, 0.253951247, 0.1468907, 0.265942707], "temperature": [0.28727078, 0.279269655, 0.271842933, 0.175112295, 0.259125777, 0.23380611, 0.160010382, 0.25937479, 0.300764503, 0.258568937, 0.284756465, 0.302121423, 0.201200358, 0.253951247, 0.176088772, 0.280392797]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "bcbaa22ee1746df1a5cb611577e1f7b19261dbb85d498fa7c0032bce4221bb77", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-96b14b5a0efa", "result_hash": "627fe133baeed03c1098df932e632865ce8bf0b01dbd80a4420bfdcf1487d252", "signature": "99bbc2ba9b381adf99c945fb4183d8c097e9366d8861beb39fab1751bc93891f8bc2982bd75627601dafc9ee8dc707878ae54c720fe5330dddae6ab012bcb208", "state": "VERIFIED", "summary": {"iw_temperature": 0.248279, "raw": 0.232756, "temperature": 0.248979}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:32:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:42Z", "cpu_seconds": 1.835550435, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-f3e91ca60c6f", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "prereg_hash": "6856e04e5fa9e4d6a9b42039327c5801f94277a801bd69836fd4bc7901164c63", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-f3e91ca60c6f", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 0.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:32:35Z", "seed": 348988462, "treatment": "iw_temperature"}, "priority": 2.0097045588521394, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-ebbd80053a51", "result_hash": "87e61f651caf6097e8c03d5db5c1debac0258382bb68eaeae08a038250a0f43a", "run_seconds": 2.0384023189544678, "seed": 348988462, "started_at": "2026-11-22T03:32:40Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.019849, -0.001361], "effect": -0.010605, "evaluated_at": "2026-11-22T03:32:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-f3e91ca60c6f", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.046837733, 0.110320645, 0.0345208, 0.034992125, 0.037415768, 0.05858758, 0.042286887, 0.075697915, 0.091245082, 0.06407463, 0.054310115, 0.079908138, 0.036466192, 0.072598255, 0.040534655, 0.067681132], "raw": [0.048666387, 0.06202321, 0.04749438, 0.034992125, 0.056470665, 0.045879365, 0.065087198, 0.05853922, 0.042588575, 0.034962635, 0.047687613, 0.079908138, 0.057848262, 0.072598255, 0.05810474, 0.02580452], "temperature": [0.046837733, 0.06202321, 0.029221038, 0.034992125, 0.044931722, 0.044036877, 0.042286887, 0.035059402, 0.040690645, 0.056317855, 0.047687613, 0.079908138, 0.038908027, 0.07897952, 0.028230955, 0.067681132]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "6856e04e5fa9e4d6a9b42039327c5801f94277a801bd69836fd4bc7901164c63", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-ebbd80053a51", "result_hash": "87e61f651caf6097e8c03d5db5c1debac0258382bb68eaeae08a038250a0f43a", "signature": "881ff37b3b199557934152d5a3ee3c8367396cf559346dc494bdce2bd4c7dde3badd70c8846e2c6049f3fbd4772d789963fdca80cb37501e35ac64475d583908", "state": "VERIFIED", "summary": {"iw_temperature": 0.059217, "raw": 0.052416, "temperature": 0.048612}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:32:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:52Z", "cpu_seconds": 1.227363829999998, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1a615abda78c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "f29db256d597e9b4a679c75e86423fc9da307a4dead39ab9cec3be4fb1298b27", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1a615abda78c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:32:35Z", "seed": 922999324, "treatment": "frontier"}, "priority": 2.008749818650621, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c8f4eb09baba344443540dcc439935604249289d61b9a9d6219b36c1812df2cc", "run_seconds": 1.7693400382995605, "seed": 922999324, "started_at": "2026-11-22T03:32:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.101994, 0.006298], "effect": -0.047848, "evaluated_at": "2026-11-22T03:33:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1a615abda78c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.698717949, 1.0, 0.719047619], "frontier": [1.0, 0.724358974, 1.0, 0.728571429], "goal": [1.0, 0.846153846, 1.0, 0.728571429], "hypdisc": [1.0, 0.647435897, 1.0, 0.666666667], "infogain": [1.0, 0.75, 1.0, 0.704761905], "prederr": [1.0, 0.551282051, 1.0, 0.642857143], "random": [1.0, 0.820512821, 1.0, 0.823809524], "risk": [1.0, 0.698717949, 1.0, 0.728571429]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "f29db256d597e9b4a679c75e86423fc9da307a4dead39ab9cec3be4fb1298b27", "primary_control": "random", "result_hash": "c8f4eb09baba344443540dcc439935604249289d61b9a9d6219b36c1812df2cc", "signature": "28fa326abe6013be5f80f66cd06fdb2b6d59e16c664caf063afa4d4ea2ae5f79bb15a139810397243b6b01401d34766c679bb1fd7bbec97babe223107e9e7505", "state": "VERIFIED", "summary": {"count": 0.854441, "frontier": 0.863233, "goal": 0.893681, "hypdisc": 0.828526, "infogain": 0.86369, "prederr": 0.798535, "random": 0.911081, "risk": 0.856822}, "ties": 2, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:33:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:58Z", "cpu_seconds": 5.536719451, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-08393b9c09a9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 7, "n": 4}, "prereg_hash": "c10faf3451f78f3e90b35a90e370f06d612ae9ad1db8c6c99b69b718087b70e1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-08393b9c09a9", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:32:35Z", "seed": 1575673658, "treatment": "full"}, "priority": 2.0088057569246858, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "800d465664a3a77808dc27ed856e94f642ae11086b10131287b132eed5b074fd", "run_seconds": 8.70703911781311, "seed": 1575673658, "started_at": "2026-11-22T03:32:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.04552, 0.207186], "effect": 0.080833, "evaluated_at": "2026-11-22T03:33:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-08393b9c09a9", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.886666667, 1.0, 0.9, 0.488888889], "frozen": [0.0, 0.0, 0.366666667, 0.248888889], "full": [0.888888889, 1.0, 0.95, 0.76], "retrieval": [0.968888889, 1.0, 0.958333333, 0.896296296], "rules_nodetect": [0.888888889, 1.0, 0.95, 0.76]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c10faf3451f78f3e90b35a90e370f06d612ae9ad1db8c6c99b69b718087b70e1", "primary_control": "facts", "result_hash": "800d465664a3a77808dc27ed856e94f642ae11086b10131287b132eed5b074fd", "signature": "ce9e9ae66e8bc4dad5796d56d9a12097260923809a0479a430f176b291cf6a1e72fddf35a9f6927d44eae8c6f4d3ded95044827250fc5e5cbb3ac65dd892380e", "state": "VERIFIED", "summary": {"facts": 0.818889, "frozen": 0.153889, "full": 0.899722, "retrieval": 0.95588, "rules_nodetect": 0.899722}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:33:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:38Z", "cpu_seconds": 0.5289905350000002, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d8d06782315c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "78e97446dcba71561d771d8e950fc1e748e36131155e24a6b085928b1fa7ff9c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d8d06782315c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:32:35Z", "seed": 1657248633, "treatment": "gated"}, "priority": 2.0099672128053436, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "51dadad913feb3e5abc66dc493623d94764a88aefcfdbaf29ac38b5b7ea6f365", "run_seconds": 0.5295689105987549, "seed": 1657248633, "started_at": "2026-11-22T03:32:37Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:32:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d8d06782315c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.888888889, 0.731481481, 0.748792271, 0.625], "gated": [0.888888889, 0.731481481, 0.748792271, 0.625], "naive": [0.888888889, 0.675925926, 0.729468599, 0.625], "scratch": [0.888888889, 0.731481481, 0.748792271, 0.625], "ungated": [0.888888889, 0.675925926, 0.768115942, 0.625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "78e97446dcba71561d771d8e950fc1e748e36131155e24a6b085928b1fa7ff9c", "primary_control": "scratch", "result_hash": "51dadad913feb3e5abc66dc493623d94764a88aefcfdbaf29ac38b5b7ea6f365", "signature": "98f2df94fe294199538f7a6c1a422251c80910ea8d912d7c323ae6018799377d8051fe58b1c6e48b01b7ed3b9db5a1a8670f7aaaf46b12e297dfd4ddbbc76302", "state": "PARTIAL", "summary": {"compressed": 0.748541, "gated": 0.748541, "naive": 0.729821, "scratch": 0.748541, "ungated": 0.739483}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:32:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:45Z", "cpu_seconds": 6.777800354000002, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-10d798d7ba4d", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "47e2d5563a2df571a6d53f88dcadcd63869b4308bab5a3a96253b965d6cd0491", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-10d798d7ba4d", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:32:35Z", "seed": 1028887225, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "96e05bb3284b225a693ca8154b80236a417cdc71f991e4e2cbaf6094d2aff742", "run_seconds": 7.241187810897827, "seed": 1028887225, "started_at": "2026-11-22T03:32:38Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.085224, 0.185224], "effect": 0.05, "evaluated_at": "2026-11-22T03:32:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-10d798d7ba4d", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.7335081628675542, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.5023, "losses": 7, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.315", "prereg_hash": "47e2d5563a2df571a6d53f88dcadcd63869b4308bab5a3a96253b965d6cd0491", "primary_control": "reset", "result_hash": "96e05bb3284b225a693ca8154b80236a417cdc71f991e4e2cbaf6094d2aff742", "signature": "55460445d70ad931abba2d7ade6229d75a8f38eb0dd5fb77031ba60c8a80985cc24686cee85da70f01eeb65f46c7904d9cfd63343a0e6be4747643c9097d1a06", "state": "VERIFIED", "summary": {"inherit": 0.65, "reset": 0.6, "root": 0.3333333333333333}, "ties": 43, "treatment": "inherit", "wins": 10}, "verified_at": "2026-11-22T03:32:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:45Z", "cpu_seconds": 6.700644711999999, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-df239511aeb8", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "57099406ac906024415588c98d4ed11cac125d70a2ab960b96525e74194a60d7", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-df239511aeb8", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:32:35Z", "seed": 689472424, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "5187932276d983f34be51e5a38c78b62fe356161a8ed3a382c5bd2af3b32a3a2", "run_seconds": 7.026494026184082, "seed": 689472424, "started_at": "2026-11-22T03:32:38Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.213004, 0.079671], "effect": -0.066667, "evaluated_at": "2026-11-22T03:32:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-df239511aeb8", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 3}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0], "reset": [1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0], "root": [1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0613, "losses": 12, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.868", "prereg_hash": "57099406ac906024415588c98d4ed11cac125d70a2ab960b96525e74194a60d7", "primary_control": "reset", "result_hash": "5187932276d983f34be51e5a38c78b62fe356161a8ed3a382c5bd2af3b32a3a2", "signature": "88d72624855d944954d5f7465b59e4dbe55c75034e478db89a80c619dd003f8809f96e045c1b5fe3d1cdb47de7e8d31d56a5e35e07b2dd5de27a82bd8980ec08", "state": "VERIFIED", "summary": {"inherit": 0.5, "reset": 0.5666666666666667, "root": 0.3}, "ties": 40, "treatment": "inherit", "wins": 8}, "verified_at": "2026-11-22T03:33:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:32:47Z", "cpu_seconds": 0.3868023239999996, "created_at": "2026-11-22T03:32:35Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-1772a5f176ba", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "8813ab78048866e43e0a4f9af85b5fe6f537a4844950140aece1d5b7767c80be", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-1772a5f176ba", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:32:35Z", "seed": 308821791, "treatment": "frontier"}, "priority": 2.0090960726429836, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-38290d04ea05", "result_hash": "8c1154a5f2325154f62233e36620411f3657611935b6541a3fc98212b5d7dc26", "run_seconds": 0.4439988136291504, "seed": 308821791, "started_at": "2026-11-22T03:32:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.080912, 0.178162], "effect": 0.129537, "evaluated_at": "2026-11-22T03:33:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-1772a5f176ba", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.840909091, 1.0, 1.0, 0.9125, 0.921568627, 0.765306122, 0.787037037, 0.846590909, 1.0, 1.0, 0.879807692, 0.881578947], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.958333333, 1.0, 1.0, 1.0], "random_walk": [0.831818182, 1.0, 1.0, 0.891666667, 0.803921569, 0.785714286, 0.814814815, 0.761363636, 0.791666667, 1.0, 0.841346154, 0.881578947]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -2.7093, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.00195", "prereg_hash": "8813ab78048866e43e0a4f9af85b5fe6f537a4844950140aece1d5b7767c80be", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-38290d04ea05", "result_hash": "8c1154a5f2325154f62233e36620411f3657611935b6541a3fc98212b5d7dc26", "signature": "3c556c402df5664831c33fd11bd4229d6a62001619786733a2f43973d2b245615d0b8d15543c7bc39e10da19af0c6cd0c852bd90ba21b52327153e94fa881d0d", "state": "VERIFIED", "summary": {"curiosity": 0.902942, "frontier": 0.996528, "random_walk": 0.866991}, "ties": 3, "treatment": "frontier", "wins": 9}, "verified_at": "2026-11-22T03:33:02Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:36:38Z", "cpu_seconds": 72.44246673299999, "created_at": "2026-11-22T03:32:52Z", "error": "lease expired; ", "family": "ow_explore", "fence": 2, "id": "E141-ow_explore-645e39630e5a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 7, "n": 4}, "prereg_hash": "f9862de55d198db8272b59bd11f284c305227bce7339ca4a9a1e3d671ab92e1c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-645e39630e5a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:32:52Z", "seed": 186249620, "treatment": "frontier"}, "priority": 2.0097163407051246, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "2da42e22139af342abdb3e27ddf263e16fc53b71b383fd8e969b3aee4ffdf7ba", "run_seconds": 88.10758113861084, "seed": 186249620, "started_at": "2026-11-22T03:32:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006033, 0.210166], "effect": 0.102067, "evaluated_at": "2026-11-22T03:38:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-645e39630e5a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.495466667, 0.658, 0.6712, 0.712], "frontier": [0.749333333, 0.6496, 0.784, 0.787333333], "goal": [0.6784, 0.5516, 0.7208, 0.763333333], "hypdisc": [0.662933333, 0.67, 0.7256, 0.706], "infogain": [0.749333333, 0.6496, 0.784, 0.787333333], "prederr": [0.749333333, 0.6496, 0.784, 0.787333333], "random": [0.491733333, 0.6524, 0.7112, 0.706666667], "risk": [0.749333333, 0.6496, 0.784, 0.787333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "f9862de55d198db8272b59bd11f284c305227bce7339ca4a9a1e3d671ab92e1c", "primary_control": "random", "result_hash": "2da42e22139af342abdb3e27ddf263e16fc53b71b383fd8e969b3aee4ffdf7ba", "signature": "dfb8dfa1185f83173294ca3b43866a2361652bc8ff3ee28662e134eea2da50420bfe1637729470838ddf80da9c12cc51c4643dec16fa2710b8534add97ec3f03", "state": "VERIFIED", "summary": {"count": 0.634167, "frontier": 0.742567, "goal": 0.678533, "hypdisc": 0.691133, "infogain": 0.742567, "prederr": 0.742567, "random": 0.6405, "risk": 0.742567}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:32Z", "cpu_seconds": 4.283414248, "created_at": "2026-11-22T03:32:52Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-650964647eb2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 2, "n": 4}, "prereg_hash": "18a5ccff11e7a46ce9d3481a6f1e5da828d4b964e82ab4ad6876535c68237a94", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-650964647eb2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:32:52Z", "seed": 744002201, "treatment": "frontier"}, "priority": 2.008760367726987, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "cc82f6028edcd1a6cfe181d6fe863c071d692105f47023f1ff4bcaee9ea63d8b", "run_seconds": 4.4096410274505615, "seed": 744002201, "started_at": "2026-11-22T03:34:28Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:34:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-650964647eb2", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "18a5ccff11e7a46ce9d3481a6f1e5da828d4b964e82ab4ad6876535c68237a94", "primary_control": "random", "result_hash": "cc82f6028edcd1a6cfe181d6fe863c071d692105f47023f1ff4bcaee9ea63d8b", "signature": "001a09c77c96e9a686059584a0de0bc87ee9bd948e86d101c4174ad031dc537b9fcea4b7aef4a4809be65dad8d306605089b020bde6aacafff8f40cad68bdf01", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:34:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:04Z", "cpu_seconds": 0.3735733159999999, "created_at": "2026-11-22T03:32:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f57036e0d736", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "1006ac24e8b984a74988133d96ee4a0926198183ea6a867a88eb0d65a204593d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f57036e0d736", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:32:52Z", "seed": 737476940, "treatment": "full"}, "priority": 2.009303696515475, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "054e0cd8a1377e2c1c34845d3a37aabb9b515233dce065e9722ca6800bfa6169", "run_seconds": 0.4271843433380127, "seed": 737476940, "started_at": "2026-11-22T03:33:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.144083, 0.980917], "effect": 0.5625, "evaluated_at": "2026-11-22T03:33:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f57036e0d736", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 0.25, 0.5], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 0.25, 1.0, 1.0], "no_world_model": [0.75, 0.25, 0.5, 0.75], "random": [1.0, 0.25, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "1006ac24e8b984a74988133d96ee4a0926198183ea6a867a88eb0d65a204593d", "primary_control": "e141_best", "result_hash": "054e0cd8a1377e2c1c34845d3a37aabb9b515233dce065e9722ca6800bfa6169", "signature": "22cee65dffd7ca0944f2c4c0bcd3a1e2a8e9ffe458974debe65703c19cc06411f5ecb2352f7643767cca61427abc96648b8e18779673f27f345b7e8d56febd02", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 0.8125, "no_world_model": 0.5625, "random": 0.8125}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:33:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:56Z", "cpu_seconds": 0.8553440360000018, "created_at": "2026-11-22T03:32:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0e2f75474b74", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "5376013defddc787d2c4c65e28dc510e344247d8a97e4dc05ec9b806233f7ac0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0e2f75474b74", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:32:52Z", "seed": 1452789600, "treatment": "full"}, "priority": 2.0089936803259416, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3ec49964d37097bd11a3108bd977ae2af3dc620df530edfbfc391d392ef899eb", "run_seconds": 0.9858624935150146, "seed": 1452789600, "started_at": "2026-11-22T03:33:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001709, 0.128487], "effect": 0.065098, "evaluated_at": "2026-11-22T03:33:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0e2f75474b74", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.501960784, 0.520784314, 0.518431373, 0.552156863], "frozen": [0.380392157, 0.390588235, 0.390588235, 0.437647059], "full": [0.511372549, 0.563137255, 0.568627451, 0.710588235], "retrieval": [0.814901961, 0.824313725, 0.820392157, 0.828235294], "rules_nodetect": [0.511372549, 0.563137255, 0.568627451, 0.710588235]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "5376013defddc787d2c4c65e28dc510e344247d8a97e4dc05ec9b806233f7ac0", "primary_control": "facts", "result_hash": "3ec49964d37097bd11a3108bd977ae2af3dc620df530edfbfc391d392ef899eb", "signature": "ae51b69750305806318807a0478a42b82dcae8fd127aa712819dc498e88266611216a0674019163dd64978ff750b6b95bddc09c25d9c0a424909b92328899604", "state": "VERIFIED", "summary": {"facts": 0.523333, "frozen": 0.399804, "full": 0.588431, "retrieval": 0.821961, "rules_nodetect": 0.588431}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:33:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:09Z", "cpu_seconds": 1.2541228490000016, "created_at": "2026-11-22T03:32:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-20b5773af8e8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 10, "n": 4}, "prereg_hash": "f9bed9acf8f4475412c2b6c98519473ab1b404a0907bbb659018b4469703b055", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-20b5773af8e8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:32:52Z", "seed": 723435529, "treatment": "full"}, "priority": 2.0091049690137437, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e1ebe1ee14e56e236a5b46661767cc04c5a40685fd793fa1aa7d016b38385c27", "run_seconds": 1.2552146911621094, "seed": 723435529, "started_at": "2026-11-22T03:33:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.105434, 0.237361], "effect": 0.171397, "evaluated_at": "2026-11-22T03:33:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-20b5773af8e8", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.091666667, 0.074911348, 0.117831541, 0.154317549], "frozen": [0.043055556, 0.048315603, 0.026433692, 0.03454039], "full": [0.219444444, 0.204343972, 0.387992832, 0.312534819], "retrieval": [0.229166667, 0.175975177, 0.355286738, 0.316155989], "rules_nodetect": [0.158333333, 0.207446809, 0.298387097, 0.312534819]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f9bed9acf8f4475412c2b6c98519473ab1b404a0907bbb659018b4469703b055", "primary_control": "facts", "result_hash": "e1ebe1ee14e56e236a5b46661767cc04c5a40685fd793fa1aa7d016b38385c27", "signature": "58913bf287050490f631ef77acf3171022c8ba519d9711c1da0406d97a8f98d7802aa160ada6bc9ef15ceaba207851a6e0c7ebeeba562b50e507e9ed8111bb00", "state": "VERIFIED", "summary": {"facts": 0.109682, "frozen": 0.038086, "full": 0.281079, "retrieval": 0.269146, "rules_nodetect": 0.244176}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:33:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:03Z", "cpu_seconds": 3.2077351159999985, "created_at": "2026-11-22T03:32:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9580d2b91ab7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "116f58df8fa40226b07f35acb40a3a38999632a77b901f7df947786591876c2e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9580d2b91ab7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:32:52Z", "seed": 1613277921, "treatment": "full"}, "priority": 2.0093056923769, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1269e90bcceb2a9c5d759c51822871bc4818284a8b52326dea8329fb15b66785", "run_seconds": 5.455507755279541, "seed": 1613277921, "started_at": "2026-11-22T03:32:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023644, 0.206621], "effect": 0.091489, "evaluated_at": "2026-11-22T03:33:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9580d2b91ab7", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.10013089, 0.149553571, 1.0, 0.519230769], "frozen": [0.036649215, 0.073660714, 0.0, 0.038461538], "full": [0.305628272, 0.329241071, 1.0, 0.5], "retrieval": [0.337696335, 0.399553571, 1.0, 0.509615385], "rules_nodetect": [0.305628272, 0.329241071, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "116f58df8fa40226b07f35acb40a3a38999632a77b901f7df947786591876c2e", "primary_control": "facts", "result_hash": "1269e90bcceb2a9c5d759c51822871bc4818284a8b52326dea8329fb15b66785", "signature": "4d5c1f9021a873ceb1c5824c90b223e33d29ea0a593cae5443cd1c6c8ec383fc1e20ae32c9e1bca1536821c9da6e6ebf875fd8e67b95160e9553a2c020b4ec03", "state": "VERIFIED", "summary": {"facts": 0.442229, "frozen": 0.037193, "full": 0.533717, "retrieval": 0.561716, "rules_nodetect": 0.533717}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:33:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:33Z", "cpu_seconds": 13.207714655999998, "created_at": "2026-11-22T03:33:08Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d8e52a86380e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 6, "n": 4}, "prereg_hash": "5ce14dcc4473c1e0a9bdbc489ea61d8cd0bdeebd279489abf2be1780f563bda4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d8e52a86380e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:33:08Z", "seed": 1285303073, "treatment": "frontier"}, "priority": 2.009524947605083, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fbccb014bc1a85c4982c841941571b61ac6758b83f5d7d5b6ab0aa5f95cfab9b", "run_seconds": 17.97467541694641, "seed": 1285303073, "started_at": "2026-11-22T03:33:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.049366, 0.179759], "effect": 0.065196, "evaluated_at": "2026-11-22T03:33:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d8e52a86380e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.370833333, 0.539534884, 0.541176471, 0.233333333], "frontier": [0.252083333, 0.606976744, 0.64, 0.397777778], "goal": [0.316666667, 0.534883721, 0.585882353, 0.353333333], "hypdisc": [0.2625, 0.553488372, 0.541176471, 0.284444444], "infogain": [0.289583333, 0.611627907, 0.602352941, 0.395555556], "prederr": [0.275, 0.43255814, 0.510588235, 0.262222222], "random": [0.34375, 0.520930233, 0.564705882, 0.206666667], "risk": [0.252083333, 0.606976744, 0.64, 0.397777778]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "5ce14dcc4473c1e0a9bdbc489ea61d8cd0bdeebd279489abf2be1780f563bda4", "primary_control": "random", "result_hash": "fbccb014bc1a85c4982c841941571b61ac6758b83f5d7d5b6ab0aa5f95cfab9b", "signature": "9329b26b00cc84c78efd60f418ac6d275ca8da903032ddd30686d7cee874b4026343e06b3de95d3c5208c6cc0f9fd1ea1f80aabc22caca13ea5e26a8399edc00", "state": "VERIFIED", "summary": {"count": 0.42122, "frontier": 0.474209, "goal": 0.447692, "hypdisc": 0.410402, "infogain": 0.47478, "prederr": 0.370092, "random": 0.409013, "risk": 0.474209}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:33:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:34Z", "cpu_seconds": 3.777515418, "created_at": "2026-11-22T03:33:08Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e4ab21eab960", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "1ab6ecf29f01cb04858fa202bdfc72048131b1fc8e5e4e5e8cec76b38defbe00", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e4ab21eab960", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:33:08Z", "seed": 1944105640, "treatment": "frontier"}, "priority": 2.008483331688768, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b042760a11c9ccf1e7135d1d16a3f32676ea37aa2bcd359e43904256d1c3378b", "run_seconds": 4.041800260543823, "seed": 1944105640, "started_at": "2026-11-22T03:34:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.239914, 0.113542], "effect": -0.063186, "evaluated_at": "2026-11-22T03:35:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e4ab21eab960", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.334677419, 0.75, 0.312827225, 0.378255208], "frontier": [0.347446237, 0.31, 0.296465969, 0.326171875], "goal": [0.341397849, 0.34, 0.298429319, 0.321614583], "hypdisc": [0.301075269, 0.45, 0.298429319, 0.315755208], "infogain": [0.340053763, 0.31, 0.296465969, 0.348307292], "prederr": [0.35483871, 0.31, 0.296465969, 0.340494792], "random": [0.289650538, 0.64, 0.310209424, 0.29296875], "risk": [0.347446237, 0.31, 0.296465969, 0.326171875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "1ab6ecf29f01cb04858fa202bdfc72048131b1fc8e5e4e5e8cec76b38defbe00", "primary_control": "random", "result_hash": "b042760a11c9ccf1e7135d1d16a3f32676ea37aa2bcd359e43904256d1c3378b", "signature": "aa067784e558dc47d097bac48ea089e6c4db7f69eae31ba693627217df3075ea201969ba78a6805f87a6da52cfd8d2c44b04dac475e4f268ede08348068bab02", "state": "VERIFIED", "summary": {"count": 0.44394, "frontier": 0.320021, "goal": 0.32536, "hypdisc": 0.341315, "infogain": 0.323707, "prederr": 0.32545, "random": 0.383207, "risk": 0.320021}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:35:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:10Z", "cpu_seconds": 1.123086551, "created_at": "2026-11-22T03:33:08Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-eddc1095265a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 4, "n": 4}, "prereg_hash": "501dda3420086afd0598d9a3e46f6b0d97ee81eae839ee5b14ce9d1b9025a9c9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-eddc1095265a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:33:08Z", "seed": 663563196, "treatment": "full"}, "priority": 2.0099225341786777, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5095a64cbf11e6ab60979c75e0969c264617a51056c8ee938f710323cf42ef09", "run_seconds": 1.162984848022461, "seed": 663563196, "started_at": "2026-11-22T03:33:09Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:33:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-eddc1095265a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.75, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "501dda3420086afd0598d9a3e46f6b0d97ee81eae839ee5b14ce9d1b9025a9c9", "primary_control": "e141_best", "result_hash": "5095a64cbf11e6ab60979c75e0969c264617a51056c8ee938f710323cf42ef09", "signature": "635c64e9690ba963c6a588dc2691847cda2a87eac905ae1119c15ed2041397501d2b941dc82501524006ada02c63ba7b275cdcc592463fee9f0e1b74da0f2f00", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.6875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:33:14Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:33:08Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_rsi", "fence": 3, "id": "E141-ow_rsi-e71828b3dfcb", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "b2d66603aabd5c054c8c8b5308b8c6541cf2ef08e889a4b02a74b915290494c7", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-e71828b3dfcb", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:33:08Z", "seed": 1227117159, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 1227117159, "started_at": "2026-11-22T03:33:13Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:35:05Z", "cpu_seconds": 0.48241369600000006, "created_at": "2026-11-22T03:33:08Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-0100f887352b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "prereg_hash": "94c601317b2d0e148925a50304a52f2f5be8c4ac4aca226bf70626563e4dac6c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-0100f887352b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:33:08Z", "seed": 2064793285, "treatment": "gated"}, "priority": 2.0075218448817242, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "4f66788fa0579eec54c787555c2555ff290329754e4575201e023b5ed74fe0d9", "run_seconds": 0.587428092956543, "seed": 2064793285, "started_at": "2026-11-22T03:35:04Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:35:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-0100f887352b", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.529100529, 0.548387097, 0.698924731, 0.553763441], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.447089947, 0.548387097, 0.85483871, 0.85483871]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "94c601317b2d0e148925a50304a52f2f5be8c4ac4aca226bf70626563e4dac6c", "primary_control": "scratch", "result_hash": "4f66788fa0579eec54c787555c2555ff290329754e4575201e023b5ed74fe0d9", "signature": "e1371c36c2d7e84a695cd0bfe01f788dd9039adae79a5d23a747b79ab4444b845896bd8559f219a345528b8551d953c2e989145e8365894308fc89e3f16f8a01", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.582544, "scratch": 1.0, "ungated": 0.676289}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:35:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:41Z", "cpu_seconds": 6.491064708000003, "created_at": "2026-11-22T03:33:08Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-2e45f4770ca3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "8a06f84555a972f44e5661e4678d54c3f6375815fc80c24f55848d258e9c5332", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-2e45f4770ca3", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:33:08Z", "seed": 1515052661, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "5dd750af9de51c539625916239ade8b3bc11bb725c117226f47b545806aec311", "run_seconds": 8.563137769699097, "seed": 1515052661, "started_at": "2026-11-22T03:33:33Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.005613, 0.105613], "effect": 0.05, "evaluated_at": "2026-11-22T03:33:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-2e45f4770ca3", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0], "root": [0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "8a06f84555a972f44e5661e4678d54c3f6375815fc80c24f55848d258e9c5332", "primary_control": "reset", "result_hash": "5dd750af9de51c539625916239ade8b3bc11bb725c117226f47b545806aec311", "signature": "346360d98d0adf2feb18b5f4f2cac4f26eee017326804ea5063f45a2701b45069a9e1c522176e6ba43960a7429c85fceed42b253e9931d5201cab12b6263240c", "state": "VERIFIED", "summary": {"inherit": 0.6, "reset": 0.55, "root": 0.3333333333333333}, "ties": 57, "treatment": "inherit", "wins": 3}, "verified_at": "2026-11-22T03:33:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:28Z", "cpu_seconds": 9.674822527, "created_at": "2026-11-22T03:33:08Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d0a4431d9dd0", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "26a6e10df8cfe14f36cd9afdbd4c850ec251309699e262485f095602055a94e1", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d0a4431d9dd0", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:33:08Z", "seed": 1347828864, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "34e40a5239391cdc8cc11de32aaa8a1fec0aa0e85fd36c93363e05a79653228e", "run_seconds": 13.013352394104004, "seed": 1347828864, "started_at": "2026-11-22T03:33:15Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 4, "reset": 4}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:33:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d0a4431d9dd0", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "26a6e10df8cfe14f36cd9afdbd4c850ec251309699e262485f095602055a94e1", "primary_control": "reset", "result_hash": "34e40a5239391cdc8cc11de32aaa8a1fec0aa0e85fd36c93363e05a79653228e", "signature": "53e704e01f1c0655c978a4fa13d3c17b2ae0f571cedbaedd2a2b6a3f3e6b0c385093ab5d5e7140c63a0ebb4b039e93f69a38144dcbac52ac34eff7c4e0770a08", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:33:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:39Z", "cpu_seconds": 0.2796025980000003, "created_at": "2026-11-22T03:33:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3e401a927b40", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "34d9526794ad494e24138d39dc604febd3d6ae04f2e6abac9c055d9453c749f7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3e401a927b40", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:33:25Z", "seed": 733007918, "treatment": "full"}, "priority": 2.0096346075147364, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f7f784aa94da29728e69e2102e30c3b5c369400826bcde39d45329058a1ebca8", "run_seconds": 0.41236376762390137, "seed": 733007918, "started_at": "2026-11-22T03:33:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.076782, 0.248218], "effect": 0.1625, "evaluated_at": "2026-11-22T03:33:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3e401a927b40", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.75, 0.854166667, 0.833333333, 0.625], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.895833333, 0.916666667, 1.0, 0.9], "retrieval": [0.854166667, 0.916666667, 0.916666667, 0.8], "rules_nodetect": [0.291666667, 0.333333333, 0.583333333, 0.275]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "34d9526794ad494e24138d39dc604febd3d6ae04f2e6abac9c055d9453c749f7", "primary_control": "facts", "result_hash": "f7f784aa94da29728e69e2102e30c3b5c369400826bcde39d45329058a1ebca8", "signature": "d5b477269da50f7799c8e92a6f75fcda29518eb47c7ba7ba3b4d911c574d56c92001c773653d59012de3e859a089109518884f418f519769f183558e1683040c", "state": "VERIFIED", "summary": {"facts": 0.765625, "frozen": 0.0, "full": 0.928125, "retrieval": 0.871875, "rules_nodetect": 0.370833}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:33:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:59Z", "cpu_seconds": 9.071991324000003, "created_at": "2026-11-22T03:33:41Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-28b855f4023c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6cf94bba7073e94ec8c745cab5b0aa2c56d0489c870f1a02f200475250f3d73f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-28b855f4023c", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:33:41Z", "seed": 1359604755, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "2c2b3abcb529d89b25223294e03fd3d0d21f3d50e52e27704d472875cf783099", "run_seconds": 12.915909767150879, "seed": 1359604755, "started_at": "2026-11-22T03:33:46Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 4, "reset": 2}, "better": "higher", "ci95": [-0.24073, 0.007397], "effect": -0.116667, "evaluated_at": "2026-11-22T03:34:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-28b855f4023c", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.413967182891189, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0], "root": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0077, "losses": 11, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.982", "prereg_hash": "6cf94bba7073e94ec8c745cab5b0aa2c56d0489c870f1a02f200475250f3d73f", "primary_control": "reset", "result_hash": "2c2b3abcb529d89b25223294e03fd3d0d21f3d50e52e27704d472875cf783099", "signature": "aa99cf113302cce22c2792a293a945f8671d8c1ab3dff5c0429a94a75bfddef0eddfbd719c9619716fe4d668db0111b6c9bd05e6c592bc973d482b2492512b05", "state": "VERIFIED", "summary": {"inherit": 0.35, "reset": 0.4666666666666667, "root": 0.2}, "ties": 45, "treatment": "inherit", "wins": 4}, "verified_at": "2026-11-22T03:34:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:50Z", "cpu_seconds": 0.32734529000000023, "created_at": "2026-11-22T03:33:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0f7ec36edc2e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "a2c9f7de2a378748ab1c36f1a526bcdc5fbe661228c1cecea82afb8217720353", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0f7ec36edc2e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:33:42Z", "seed": 523619277, "treatment": "frontier"}, "priority": 2.0094402416667037, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "0d7a513dfb25cdc51cac93b2e09bb1c3dc3be685737ef69ff26221bb312ab0c5", "run_seconds": 0.6777901649475098, "seed": 523619277, "started_at": "2026-11-22T03:33:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.205413, 0.243451], "effect": 0.224432, "evaluated_at": "2026-11-22T03:33:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0f7ec36edc2e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.954545455, 0.977272727, 0.954545455, 0.965909091], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.738636364, 0.659090909, 0.818181818, 0.795454545], "hypdisc": [0.806818182, 0.761363636, 0.738636364, 0.829545455], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.784090909, 0.75, 0.772727273, 0.795454545], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a2c9f7de2a378748ab1c36f1a526bcdc5fbe661228c1cecea82afb8217720353", "primary_control": "random", "result_hash": "0d7a513dfb25cdc51cac93b2e09bb1c3dc3be685737ef69ff26221bb312ab0c5", "signature": "19d8a6d03484cf758cdaac336c8192a44cb64bf7f9bc92d66a038595610c6617ed09c8adab088c56a7073acbf204760d718b5cc6a7e18843d6ac4fd8945c5c0a", "state": "VERIFIED", "summary": {"count": 0.963068, "frontier": 1.0, "goal": 0.752841, "hypdisc": 0.784091, "infogain": 1.0, "prederr": 1.0, "random": 0.775568, "risk": 1.0}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:33:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:33:44Z", "cpu_seconds": 0.33104300300000133, "created_at": "2026-11-22T03:33:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7212b313d19d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "0aa9c5bb1abc5848c8e0b904fd2649ccb86c586f1f82be0a85c31d06848b2054", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7212b313d19d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:33:42Z", "seed": 1226161871, "treatment": "full"}, "priority": 2.009888635603092, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a7b5cbf166c34041bcf6f80870f72fc6a7605ec71be348b472c4d28635de89e9", "run_seconds": 0.3964662551879883, "seed": 1226161871, "started_at": "2026-11-22T03:33:44Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:33:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7212b313d19d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.75, 0.25, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0aa9c5bb1abc5848c8e0b904fd2649ccb86c586f1f82be0a85c31d06848b2054", "primary_control": "e141_best", "result_hash": "a7b5cbf166c34041bcf6f80870f72fc6a7605ec71be348b472c4d28635de89e9", "signature": "480efe8d64557025d31501e489fbf27aa041da0ccfce0d1e0644618b7fe0e8e0aaf537f4b48ee7f084db793719f8441b0015f91ac6c2bac9d290c1bfa8947a08", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5625, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:33:52Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:35:19Z", "cpu_seconds": 1.161343071999994, "created_at": "2026-11-22T03:33:58Z", "error": "lease expired; ", "family": "ow_explore", "fence": 2, "id": "E141-ow_explore-e760181092de", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "8a3ece7356317fb0f5790d63884e34f5a3ac35d83e16cad0a0a5132e60beb4bc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e760181092de", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:33:58Z", "seed": 1759214990, "treatment": "frontier"}, "priority": 2.0092783958837503, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "432dca71cec5f93aabfae029dabf4e5b97ada7238de199e4dc5c44eb05bed717", "run_seconds": 1.4192399978637695, "seed": 1759214990, "started_at": "2026-11-22T03:34:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.319823, -0.04824], "effect": -0.184031, "evaluated_at": "2026-11-22T03:37:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e760181092de", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.866071429, 0.76, 0.613636364, 0.875], "frontier": [0.5625, 0.45, 0.295454545, 0.875], "goal": [0.678571429, 0.42, 0.4, 0.875], "hypdisc": [0.8125, 0.546666667, 0.436363636, 0.875], "infogain": [0.580357143, 0.546666667, 0.609090909, 0.875], "prederr": [0.428571429, 0.726666667, 0.427272727, 0.875], "random": [0.758928571, 0.653333333, 0.631818182, 0.875], "risk": [0.5625, 0.45, 0.295454545, 0.875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "8a3ece7356317fb0f5790d63884e34f5a3ac35d83e16cad0a0a5132e60beb4bc", "primary_control": "random", "result_hash": "432dca71cec5f93aabfae029dabf4e5b97ada7238de199e4dc5c44eb05bed717", "signature": "0a98467621e019cc2458162e768b2cdd8e3e5c1dba0132b2bd736f5ceb910af90389d2b0557ab305ed7fe506a4aa449a8c6edb2e60f565c14a177f346ebcd808", "state": "VERIFIED", "summary": {"count": 0.778677, "frontier": 0.545739, "goal": 0.593393, "hypdisc": 0.667633, "infogain": 0.652779, "prederr": 0.614378, "random": 0.72977, "risk": 0.545739}, "ties": 1, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:37:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:25Z", "cpu_seconds": 0.993177312, "created_at": "2026-11-22T03:33:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-119ed3de0355", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 4, "n": 4}, "prereg_hash": "972a2e356c361449d68af5b1050d84335827fc341c5ff400ad986ace37694539", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-119ed3de0355", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:33:58Z", "seed": 1335079600, "treatment": "full"}, "priority": 2.0088845545968645, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "093a214d125703242fc133d0460a8d3d718339c6634104c60411ce4cc7343542", "run_seconds": 1.1675622463226318, "seed": 1335079600, "started_at": "2026-11-22T03:34:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.009929, 0.126408], "effect": 0.068169, "evaluated_at": "2026-11-22T03:34:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-119ed3de0355", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.534210526, 1.0, 0.524166667, 0.528333333], "frozen": [0.380701754, 0.61, 0.42, 0.350833333], "full": [0.571052632, 1.0, 0.639166667, 0.649166667], "retrieval": [0.831578947, 1.0, 0.791666667, 0.835], "rules_nodetect": [0.571052632, 1.0, 0.639166667, 0.649166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "972a2e356c361449d68af5b1050d84335827fc341c5ff400ad986ace37694539", "primary_control": "facts", "result_hash": "093a214d125703242fc133d0460a8d3d718339c6634104c60411ce4cc7343542", "signature": "1ea345056ffb53b0a5555e5b14639c5e3e927b970b5319908eace5c6c081be336669b987361c0c8baec80b9230f0f60dcbb74858b5b3235bef18431c432be307", "state": "VERIFIED", "summary": {"facts": 0.646678, "frozen": 0.440384, "full": 0.714846, "retrieval": 0.864561, "rules_nodetect": 0.714846}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:34:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:34Z", "cpu_seconds": 2.9110700389999997, "created_at": "2026-11-22T03:33:58Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-faf68abf9b57", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "prereg_hash": "8163b9c2e330a11cb923d7e98b40e11d4029c9262ec6e6471e0f08fc51d2ddf8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-faf68abf9b57", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:33:58Z", "seed": 1991107979, "treatment": "gated"}, "priority": 2.008452487290847, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "4a8cde64a66dd447263d40369a9d0b8092d50377d6794701c180ad29e86cee2e", "run_seconds": 3.0734493732452393, "seed": 1991107979, "started_at": "2026-11-22T03:34:31Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:34:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-faf68abf9b57", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.272015656, 0.27446184, 0.545275591, 0.511252446], "gated": [0.272015656, 0.27446184, 0.545275591, 0.511252446], "naive": [0.279843444, 0.145303327, 0.480314961, 0.293542074], "scratch": [0.272015656, 0.27446184, 0.545275591, 0.511252446], "ungated": [0.279843444, 0.065068493, 0.539370079, 0.200587084]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8163b9c2e330a11cb923d7e98b40e11d4029c9262ec6e6471e0f08fc51d2ddf8", "primary_control": "scratch", "result_hash": "4a8cde64a66dd447263d40369a9d0b8092d50377d6794701c180ad29e86cee2e", "signature": "7c289b25f8b6c53e9d8d9d18233f8674bc97290f3a1f26a1629b501a54a226b65e5365d5fa0d382a96e98aac819a69911ea940395c301798a14cc0ef88f2bc07", "state": "PARTIAL", "summary": {"compressed": 0.400751, "gated": 0.400751, "naive": 0.299751, "scratch": 0.400751, "ungated": 0.271217}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:35:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:11Z", "cpu_seconds": 5.444566795999997, "created_at": "2026-11-22T03:33:58Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-9591cc1283b5", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "54e73b404faeec1a36f3b23a80054118f3910de5de0c4549e3aa13c42db7a7d8", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-9591cc1283b5", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:33:58Z", "seed": 1772559704, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "85858e5c2677b69438d7b865271db98d3b5bf20b9bafd0998ff6d75152db79b3", "run_seconds": 6.631812572479248, "seed": 1772559704, "started_at": "2026-11-22T03:34:04Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:34:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-9591cc1283b5", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "54e73b404faeec1a36f3b23a80054118f3910de5de0c4549e3aa13c42db7a7d8", "primary_control": "reset", "result_hash": "85858e5c2677b69438d7b865271db98d3b5bf20b9bafd0998ff6d75152db79b3", "signature": "4729a818c8055f7c8e5f4afac9891dae5435aa7823615f21d6be4b40e7fcc56587ce7dbddd199de595f48d82e7e87c35cebbfeb2a49fe0440cc1652a075f510a", "state": "PARTIAL", "summary": {"inherit": 0.7333333333333333, "reset": 0.7333333333333333, "root": 0.3333333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:34:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:26Z", "cpu_seconds": 1.553731086, "created_at": "2026-11-22T03:34:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c4ae545942c2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "c5130a9d6ab04a4020d46ed796a682e073268f96689da6f7f353c0d6c78dda9a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c4ae545942c2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:34:15Z", "seed": 420843843, "treatment": "full"}, "priority": 2.009373523230458, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "234e274a5de081264a1ebe891ff8effe155ea0a15be324ac5c5d54396d2e5cf6", "run_seconds": 1.6614160537719727, "seed": 420843843, "started_at": "2026-11-22T03:34:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.370785, 0.245785], "effect": -0.0625, "evaluated_at": "2026-11-22T03:34:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c4ae545942c2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.75, 1.0], "fixed_policy": [0.5, 0.0, 0.0, 0.25], "full": [1.0, 0.0, 1.0, 0.5], "no_world_model": [1.0, 0.0, 1.0, 1.0], "random": [1.0, 0.25, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "c5130a9d6ab04a4020d46ed796a682e073268f96689da6f7f353c0d6c78dda9a", "primary_control": "e141_best", "result_hash": "234e274a5de081264a1ebe891ff8effe155ea0a15be324ac5c5d54396d2e5cf6", "signature": "9355125d08dc85d25a39cbaff7617460c8b9c1a7fd2f9ce422425f8f8a70968f405de9195623032fd8ce0e32960c30758e55114eb9865ad7ce67f0fe5b08f805", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.1875, "full": 0.625, "no_world_model": 0.75, "random": 0.8125}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:34:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:23Z", "cpu_seconds": 7.184399146000004, "created_at": "2026-11-22T03:34:15Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-4b4bbe9a8fd3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "bfadd4f4adf33273e134e6c5ebc852fea97b78ecfcfc1b3c39723e201ac469c2", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-4b4bbe9a8fd3", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:34:15Z", "seed": 1677598428, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "338261ad96929c6d8c9e948f7e97f2e021e8859c7586d0f0f3693e1f35203b14", "run_seconds": 7.211380481719971, "seed": 1677598428, "started_at": "2026-11-22T03:34:16Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:34:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-4b4bbe9a8fd3", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0], "reset": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "bfadd4f4adf33273e134e6c5ebc852fea97b78ecfcfc1b3c39723e201ac469c2", "primary_control": "reset", "result_hash": "338261ad96929c6d8c9e948f7e97f2e021e8859c7586d0f0f3693e1f35203b14", "signature": "cfe4831ab1af122550efab9c6952bda9aaa9e6bcca2d160af93c75e6fc6d93a07666222314278c559408b50be747bc2112170ccfbbab461512bc82ce57959f0f", "state": "PARTIAL", "summary": {"inherit": 0.55, "reset": 0.55, "root": 0.23333333333333334}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:34:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:42Z", "cpu_seconds": 4.169307416999999, "created_at": "2026-11-22T03:34:31Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-474ef48093b5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 9, "n": 4}, "prereg_hash": "477d190729b71f1cebd780ef187d1d93eea9a9ccb82bb82267e1c692c225ab25", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-474ef48093b5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:34:31Z", "seed": 768786637, "treatment": "frontier"}, "priority": 2.009968519470899, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b15c80d3c495c548a5ec4166722ed9a99560eee5f80114c465d6720a6ce288eb", "run_seconds": 4.618969678878784, "seed": 768786637, "started_at": "2026-11-22T03:34:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.029415, 0.090377], "effect": 0.059896, "evaluated_at": "2026-11-22T03:35:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-474ef48093b5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.839583333, 0.650833333, 0.666666667, 0.635], "frontier": [0.922916667, 0.695, 0.690833333, 0.691666667], "goal": [0.697916667, 0.659166667, 0.568333333, 0.663333333], "hypdisc": [0.860416667, 0.688333333, 0.5975, 0.599166667], "infogain": [0.922916667, 0.695, 0.690833333, 0.691666667], "prederr": [0.922916667, 0.695, 0.690833333, 0.691666667], "random": [0.8625, 0.610833333, 0.675, 0.6125], "risk": [0.922916667, 0.695, 0.690833333, 0.691666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "477d190729b71f1cebd780ef187d1d93eea9a9ccb82bb82267e1c692c225ab25", "primary_control": "random", "result_hash": "b15c80d3c495c548a5ec4166722ed9a99560eee5f80114c465d6720a6ce288eb", "signature": "68c4501418dd0c426a0ddfab73937fc0bed4b1476bba8fd28bcc7c32b58c8a28945efb0c6c8835d2fe767cba57694b325d2d0e1e81ab5f5f028ad846a4a88d09", "state": "VERIFIED", "summary": {"count": 0.698021, "frontier": 0.750104, "goal": 0.647188, "hypdisc": 0.686354, "infogain": 0.750104, "prederr": 0.750104, "random": 0.690208, "risk": 0.750104}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:35:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:42Z", "cpu_seconds": 2.059106743, "created_at": "2026-11-22T03:34:31Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b86cb911ddb3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 4, "n": 4}, "prereg_hash": "167ac485e592841a1f3979ea5f0af38cf6eff6678762aab65cdf112c2f1bd24f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b86cb911ddb3", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:34:31Z", "seed": 1833634686, "treatment": "frontier"}, "priority": 2.008675239599585, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c2d5aef46ba10a79c7d2c3a1ba966dd032d28252b086639a9b857abea5f643c2", "run_seconds": 2.326401472091675, "seed": 1833634686, "started_at": "2026-11-22T03:34:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024237, 0.095461], "effect": 0.035612, "evaluated_at": "2026-11-22T03:35:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b86cb911ddb3", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.5515625, 0.646875, 0.542708333, 0.55703125], "frontier": [0.5078125, 0.651041667, 0.641666667, 0.60390625], "goal": [0.5078125, 0.723958333, 0.623958333, 0.60390625], "hypdisc": [0.48125, 0.666666667, 0.598958333, 0.465625], "infogain": [0.5078125, 0.651041667, 0.641666667, 0.60390625], "prederr": [0.5078125, 0.651041667, 0.641666667, 0.60390625], "random": [0.54453125, 0.638541667, 0.5375, 0.54140625], "risk": [0.5078125, 0.651041667, 0.641666667, 0.60390625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "167ac485e592841a1f3979ea5f0af38cf6eff6678762aab65cdf112c2f1bd24f", "primary_control": "random", "result_hash": "c2d5aef46ba10a79c7d2c3a1ba966dd032d28252b086639a9b857abea5f643c2", "signature": "4e2c1d16e67af9b9feab1034b3c8830f586e0b7187a4aef5352fbeccd97b2a677a52e239e4f7acc9682db34f4203cc4c06fce7d47d34177a6f259d3ebc46420e", "state": "VERIFIED", "summary": {"count": 0.574544, "frontier": 0.601107, "goal": 0.614909, "hypdisc": 0.553125, "infogain": 0.601107, "prederr": 0.601107, "random": 0.565495, "risk": 0.601107}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:35:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:46Z", "cpu_seconds": 0.9800589520000003, "created_at": "2026-11-22T03:34:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-283d71b3a883", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "bee07baee2fb482814abdb5a3c17627a53432a859f7ec1a96059702202203c09", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-283d71b3a883", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:34:31Z", "seed": 457060349, "treatment": "full"}, "priority": 2.00863064603544, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f63c5e32c1cd30ccf49cf0b290ea1ceed7db16c49973eff8898c271cb3cd4cb1", "run_seconds": 1.0291688442230225, "seed": 457060349, "started_at": "2026-11-22T03:34:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:35:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-283d71b3a883", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 1.0, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 1.0, 1.0, 0.25], "no_world_model": [0.0, 1.0, 1.0, 0.25], "random": [0.0, 0.75, 1.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "bee07baee2fb482814abdb5a3c17627a53432a859f7ec1a96059702202203c09", "primary_control": "e141_best", "result_hash": "f63c5e32c1cd30ccf49cf0b290ea1ceed7db16c49973eff8898c271cb3cd4cb1", "signature": "0de66d770d0f28c82ad1a45e351e2185ac6ce019036d5511ca16b8f8574edac824a064fe6b17d4d3253867439562caf1cd76adff187a68ef723bb357829c6c0b", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.0, "full": 0.5625, "no_world_model": 0.5625, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:35:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:47Z", "cpu_seconds": 0.3984146260000001, "created_at": "2026-11-22T03:34:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-319799ffc555", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "565791a4bab817c3f495358295079587153ffef7be86b9b54603e6b1cfa50e08", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-319799ffc555", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:34:31Z", "seed": 1688176322, "treatment": "full"}, "priority": 2.008398466091198, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b7884f6776fe627c5ba6c9c13b4b43e7e7fbaf248218f6d42edc610e168a3ac9", "run_seconds": 0.39876580238342285, "seed": 1688176322, "started_at": "2026-11-22T03:34:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.346482, 0.346482], "effect": 0.0, "evaluated_at": "2026-11-22T03:35:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-319799ffc555", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.25, 0.25], "fixed_policy": [0.25, 0.5, 0.0, 0.25], "full": [0.0, 0.5, 0.0, 0.0], "no_world_model": [0.25, 0.0, 0.0, 0.0], "random": [0.0, 0.5, 0.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "565791a4bab817c3f495358295079587153ffef7be86b9b54603e6b1cfa50e08", "primary_control": "e141_best", "result_hash": "b7884f6776fe627c5ba6c9c13b4b43e7e7fbaf248218f6d42edc610e168a3ac9", "signature": "3a8aa82504f49eb8c6daf885a657c73e3aa587e73a8ca7d4e53c9c3a63ffc7abda16d529f8da5e65b24134f3ca1efb954b5baab89c634794d87507862ebc5404", "state": "VERIFIED", "summary": {"e141_best": 0.125, "fixed_policy": 0.25, "full": 0.125, "no_world_model": 0.0625, "random": 0.1875}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:35:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:40Z", "cpu_seconds": 0.9220793970000001, "created_at": "2026-11-22T03:34:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-389d01fce9c4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "c94d0416ac0caac701f5fb6440eaba23b101ee33e772a535ece92d585ae9bc67", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-389d01fce9c4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:34:31Z", "seed": 1513337073, "treatment": "full"}, "priority": 2.0090969853739473, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8624e6bc7eda55225592d16bbea9dbfc7bbe9055977677b154a59100f6576060", "run_seconds": 1.3446142673492432, "seed": 1513337073, "started_at": "2026-11-22T03:34:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.13, 0.62], "effect": 0.375, "evaluated_at": "2026-11-22T03:35:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-389d01fce9c4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.75, 0.5, 0.5], "fixed_policy": [0.0, 0.25, 0.0, 0.5], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [0.75, 0.75, 0.75, 0.5], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c94d0416ac0caac701f5fb6440eaba23b101ee33e772a535ece92d585ae9bc67", "primary_control": "e141_best", "result_hash": "8624e6bc7eda55225592d16bbea9dbfc7bbe9055977677b154a59100f6576060", "signature": "792e90670ec84b578e618969b04ebf44b9649f2c4dba13c2b78bb91a872099f6df18ecfefc54a6e218e41ebbaa36d66ecc2270ac4a8accca6136f4590a65180e", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.1875, "full": 0.9375, "no_world_model": 0.6875, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:35:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:41Z", "cpu_seconds": 6.528843128, "created_at": "2026-11-22T03:34:31Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d88bca306c67", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "95fcbb544d12ea2ec8c7895c7ee3183fe04ac7140f999fe07957f86ee520159b", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d88bca306c67", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:34:31Z", "seed": 2046797573, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f08e891331a51b0106156f3e04ac84ba2200ec71840306cca3ea7a5b3d3d2fd9", "run_seconds": 7.157632350921631, "seed": 2046797573, "started_at": "2026-11-22T03:34:34Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:35:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d88bca306c67", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.45435727662294906, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "95fcbb544d12ea2ec8c7895c7ee3183fe04ac7140f999fe07957f86ee520159b", "primary_control": "reset", "result_hash": "f08e891331a51b0106156f3e04ac84ba2200ec71840306cca3ea7a5b3d3d2fd9", "signature": "79a0f650f7dea8d609b9da35a7f20845369249300de10355ed49413e8f4bf40ee887e1da095cd5f2647c3c577245f17eea7e5402abbd63c42d1a1ba31e95540e", "state": "PARTIAL", "summary": {"inherit": 0.6666666666666666, "reset": 0.6666666666666666, "root": 0.4}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:35:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:58Z", "cpu_seconds": 0.9646521440000004, "created_at": "2026-11-22T03:34:48Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0a8d66e4cf42", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "3ba049d893a00c8aa4609565c1e0f3381bddcf47439e6a25a465e553e15b1df9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0a8d66e4cf42", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:34:48Z", "seed": 1686014138, "treatment": "frontier"}, "priority": 2.008746847035142, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b1f4629b1b0e2bcd6bf74045ed6517feac7d3b959325e8665d3df322f22bd589", "run_seconds": 0.9953234195709229, "seed": 1686014138, "started_at": "2026-11-22T03:34:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.331655, 0.138681], "effect": -0.096487, "evaluated_at": "2026-11-22T03:35:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0a8d66e4cf42", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.678571429, 0.666666667, 0.695652174, 0.658333333], "frontier": [0.396825397, 0.888888889, 0.739130435, 0.383333333], "goal": [0.71031746, 0.851851852, 0.492753623, 0.495833333], "hypdisc": [0.281746032, 0.888888889, 0.623188406, 0.416666667], "infogain": [0.484126984, 0.888888889, 0.797101449, 0.666666667], "prederr": [0.722222222, 0.703703704, 0.536231884, 0.541666667], "random": [0.726190476, 0.666666667, 0.797101449, 0.604166667], "risk": [0.396825397, 0.888888889, 0.739130435, 0.383333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "3ba049d893a00c8aa4609565c1e0f3381bddcf47439e6a25a465e553e15b1df9", "primary_control": "random", "result_hash": "b1f4629b1b0e2bcd6bf74045ed6517feac7d3b959325e8665d3df322f22bd589", "signature": "7e246ee394e84dbee49e1af808fe2b3c2cf6ab148c86b8b0f05fb5e2971eb6feede5d9c1caa766326b2e1088305d6dcff1bd60402d1214d0c918132e10c6d60f", "state": "VERIFIED", "summary": {"count": 0.674806, "frontier": 0.602045, "goal": 0.637689, "hypdisc": 0.552622, "infogain": 0.709196, "prederr": 0.625956, "random": 0.698531, "risk": 0.602045}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:35:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:52Z", "cpu_seconds": 0.9379708670000007, "created_at": "2026-11-22T03:34:48Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-bd69b96e9731", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "168cfed9a22e9c60d4f5612bd2ae15bc1dc70456ee452aff1324cd6990189c2d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-bd69b96e9731", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:34:48Z", "seed": 774626182, "treatment": "frontier"}, "priority": 2.009498158881635, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a61a5814038acbc5807387bcf86a4d14ec4d9555b69dbc2b17586db2ea4a6c51", "run_seconds": 1.0915772914886475, "seed": 774626182, "started_at": "2026-11-22T03:34:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.147208, 0.292415], "effect": 0.072604, "evaluated_at": "2026-11-22T03:35:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-bd69b96e9731", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.622807018, 0.743589744, 0.737373737, 0.818181818], "frontier": [0.552631579, 0.615384615, 0.515151515, 0.878787879], "goal": [0.631578947, 0.666666667, 0.676767677, 0.878787879], "hypdisc": [0.421052632, 0.692307692, 0.717171717, 0.878787879], "infogain": [0.526315789, 0.692307692, 0.757575758, 0.878787879], "prederr": [0.543859649, 0.794871795, 0.717171717, 0.878787879], "random": [0.438596491, 0.641025641, 0.676767677, 0.515151515], "risk": [0.552631579, 0.615384615, 0.515151515, 0.878787879]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "168cfed9a22e9c60d4f5612bd2ae15bc1dc70456ee452aff1324cd6990189c2d", "primary_control": "random", "result_hash": "a61a5814038acbc5807387bcf86a4d14ec4d9555b69dbc2b17586db2ea4a6c51", "signature": "e495656142bec7ad079189eb9a6db5677d889a28bbeaeeb3e3c624aa8851578b070aa0828440b4aad4c01c480a276360db7027edb190e351f84937304102af0d", "state": "VERIFIED", "summary": {"count": 0.730488, "frontier": 0.640489, "goal": 0.71345, "hypdisc": 0.67733, "infogain": 0.713747, "prederr": 0.733673, "random": 0.567885, "risk": 0.640489}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:35:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:58Z", "cpu_seconds": 5.3383747270000015, "created_at": "2026-11-22T03:34:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ce264e440433", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 7, "n": 4}, "prereg_hash": "8bab29b7bc990b196c4fcaa1c198722119dae3315154a3ebc3303edf4f68a385", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ce264e440433", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:34:48Z", "seed": 1152998858, "treatment": "full"}, "priority": 2.0089498193451156, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5e1c8cf9ba837a100825d8a3782499e1a14332ee7b8be43fb8f336ea85e74e3c", "run_seconds": 5.623344898223877, "seed": 1152998858, "started_at": "2026-11-22T03:34:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.12, 0.37], "effect": 0.125, "evaluated_at": "2026-11-22T03:35:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ce264e440433", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.5], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "8bab29b7bc990b196c4fcaa1c198722119dae3315154a3ebc3303edf4f68a385", "primary_control": "e141_best", "result_hash": "5e1c8cf9ba837a100825d8a3782499e1a14332ee7b8be43fb8f336ea85e74e3c", "signature": "5d5dd7717d87ed0f2d0d4684605ead6c8b3a3aa8306d2618782df511a7a445f65e5250024b4c876356b6bf78bc34c88271973ad1984a0d284aebc2624b69bc04", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:35:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:59Z", "cpu_seconds": 1.378421164999999, "created_at": "2026-11-22T03:34:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d0b03c5a944f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "1dac29c670d5d24ed19b81f9d60edec4ca6f6a2d031d68003fd19116ce040d08", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d0b03c5a944f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:34:48Z", "seed": 531330069, "treatment": "full"}, "priority": 2.0085891547757893, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f6e934648ab1027576b80aae1e14b7460e90a5af8f49363b273bd9c9e3c44f41", "run_seconds": 1.408789873123169, "seed": 531330069, "started_at": "2026-11-22T03:34:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T03:35:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d0b03c5a944f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.25, 0.75], "fixed_policy": [1.0, 1.0, 0.75, 1.0], "full": [1.0, 1.0, 0.5, 1.0], "no_world_model": [1.0, 1.0, 0.5, 0.5], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "1dac29c670d5d24ed19b81f9d60edec4ca6f6a2d031d68003fd19116ce040d08", "primary_control": "e141_best", "result_hash": "f6e934648ab1027576b80aae1e14b7460e90a5af8f49363b273bd9c9e3c44f41", "signature": "cc728acbbe920cd9493ca55e7d01d18116e1a474fb051dc6ec322201932ff3b39a464feaf06087cdf75a1ce290a3a11d2e6c717adae5c7b4bd9fbfb9e54e2b04", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.9375, "full": 0.875, "no_world_model": 0.75, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:35:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:06Z", "cpu_seconds": 2.2880641739999987, "created_at": "2026-11-22T03:34:48Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a2ee17f85135", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 2, "n": 4}, "prereg_hash": "8d536f509f3490611d2979f26e63b2b863b81289e92c6da4f25497b843dcce66", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a2ee17f85135", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:34:48Z", "seed": 2046867761, "treatment": "full"}, "priority": 2.00841485876545, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3c65f938713ae3de40b6c7da9d122cf014223408d9281e7b207e13804c493ac4", "run_seconds": 2.5791568756103516, "seed": 2046867761, "started_at": "2026-11-22T03:35:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00208, 0.074618], "effect": 0.036269, "evaluated_at": "2026-11-22T03:35:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a2ee17f85135", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.935483871, 1.0, 0.994708995, 0.924731183], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.983870968, 1.0, 1.0, 0.962365591], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "8d536f509f3490611d2979f26e63b2b863b81289e92c6da4f25497b843dcce66", "primary_control": "facts", "result_hash": "3c65f938713ae3de40b6c7da9d122cf014223408d9281e7b207e13804c493ac4", "signature": "3bc4fd1157ae565ddb1dc95dc222a5ba0456a70c5da12f5dd13d4e392d92794720b2e2ce5de9a445081a4f5164ae4d128abda502782418d3cdc8628433932e07", "state": "VERIFIED", "summary": {"facts": 0.963731, "frozen": 0.0, "full": 1.0, "retrieval": 0.986559, "rules_nodetect": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:35:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:53Z", "cpu_seconds": 0.5808503040000002, "created_at": "2026-11-22T03:34:48Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-fac03d9d6548", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "cbb4252237fad0b7d4dc4681138d07825119cc9ba99241f467c3c0d651aea873", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-fac03d9d6548", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:34:48Z", "seed": 687039967, "treatment": "full"}, "priority": 2.0092885344402007, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fa177e487969a93afee5da7adaf8085642ab9de8a22961c506117de296c1b84f", "run_seconds": 0.7195219993591309, "seed": 687039967, "started_at": "2026-11-22T03:34:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004561, 0.108691], "effect": 0.052065, "evaluated_at": "2026-11-22T03:35:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-fac03d9d6548", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.705882353, 0.888888889, 0.777777778, 0.46969697], "frozen": [0.091503268, 0.091503268, 0.08496732, 0.098484848], "full": [0.745098039, 0.895424837, 0.803921569, 0.606060606], "retrieval": [0.790849673, 0.967320261, 0.862745098, 0.78030303], "rules_nodetect": [0.745098039, 0.895424837, 0.803921569, 0.606060606]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "cbb4252237fad0b7d4dc4681138d07825119cc9ba99241f467c3c0d651aea873", "primary_control": "facts", "result_hash": "fa177e487969a93afee5da7adaf8085642ab9de8a22961c506117de296c1b84f", "signature": "14c65b483f525faed884f9fc5226fdb875c75d985affeb6d8bc2c47ec3a8bf7ae487acd3a31621a118e19e87c0ed901cd4282188121472f64d020a34cd51a407", "state": "VERIFIED", "summary": {"facts": 0.710561, "frozen": 0.091615, "full": 0.762626, "retrieval": 0.850305, "rules_nodetect": 0.762626}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:35:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:34:56Z", "cpu_seconds": 6.2502015140000005, "created_at": "2026-11-22T03:34:48Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-7a69bca579f0", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7efb3ba6e2f89201b573bbe1100a1bc0c92f8d64e981a7eacb5aad4d63c43e6e", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-7a69bca579f0", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:34:48Z", "seed": 1714008078, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "fa930b8cce714aa2b118e7806acfb09e26d269964f0c38da644d76559c56b345", "run_seconds": 6.613815546035767, "seed": 1714008078, "started_at": "2026-11-22T03:34:49Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.231794, -0.034873], "effect": -0.133333, "evaluated_at": "2026-11-22T03:35:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-7a69bca579f0", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.325200327405344, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0004, "losses": 9, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.999", "prereg_hash": "7efb3ba6e2f89201b573bbe1100a1bc0c92f8d64e981a7eacb5aad4d63c43e6e", "primary_control": "reset", "result_hash": "fa930b8cce714aa2b118e7806acfb09e26d269964f0c38da644d76559c56b345", "signature": "89eeb5833020a813fa816d3c9b3ea2ef33cbe6a67279a499c21d8536d10d50125f78a7633b2e2eb2a93e43511b8debbab026875214490a160b10f884a3b59101", "state": "VERIFIED", "summary": {"inherit": 0.45, "reset": 0.5833333333333334, "root": 0.31666666666666665}, "ties": 50, "treatment": "inherit", "wins": 1}, "verified_at": "2026-11-22T03:35:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:00Z", "cpu_seconds": 0.26319668399999996, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0edb2a8c82ba", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "bc3036d8142ed0f327aeb7cc0f2a8dbb312b8b28883b71ee5c32829630feb7b3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0edb2a8c82ba", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:35:05Z", "seed": 700642856, "treatment": "frontier"}, "priority": 2.0064724728228414, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "cc55906fc57ea3da6986a72bb5cf65c9a5629fe68b8c3175592d0ac0f7a505fa", "run_seconds": 0.27912235260009766, "seed": 700642856, "started_at": "2026-11-22T03:38:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.098298, 0.22746], "effect": 0.162879, "evaluated_at": "2026-11-22T03:38:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0edb2a8c82ba", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.651515152, 0.46969697, 0.515151515, 0.515151515], "frontier": [0.712121212, 0.818181818, 0.863636364, 0.848484848], "goal": [0.560606061, 0.378787879, 0.742424242, 0.575757576], "hypdisc": [0.636363636, 0.621212121, 0.772727273, 0.53030303], "infogain": [0.727272727, 0.727272727, 0.863636364, 0.803030303], "prederr": [0.727272727, 0.772727273, 0.863636364, 0.803030303], "random": [0.606060606, 0.606060606, 0.757575758, 0.621212121], "risk": [0.712121212, 0.818181818, 0.863636364, 0.848484848]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "bc3036d8142ed0f327aeb7cc0f2a8dbb312b8b28883b71ee5c32829630feb7b3", "primary_control": "random", "result_hash": "cc55906fc57ea3da6986a72bb5cf65c9a5629fe68b8c3175592d0ac0f7a505fa", "signature": "24e6d226700c60a998989129e54e4bc611176da3352007e356a8e369c8ab3b760ae3692f4ebfd74f65f168400e12b988e8f7c24f51ff164ad2828340998cfc00", "state": "VERIFIED", "summary": {"count": 0.537879, "frontier": 0.810606, "goal": 0.564394, "hypdisc": 0.640152, "infogain": 0.780303, "prederr": 0.791667, "random": 0.647727, "risk": 0.810606}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:38:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:35Z", "cpu_seconds": 47.786252055, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b5afbe457432", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 8, "n": 4}, "prereg_hash": "ea1dee476a6c29caf72192f85bf40d2d0ef1ec53d2734dbe75aab95468232be5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b5afbe457432", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:35:05Z", "seed": 1243466993, "treatment": "frontier"}, "priority": 2.008970050948748, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b901a40e307cec94c8c08f25fd381c5acd9b67ae16903adc8886b5b108bdace3", "run_seconds": 59.16632699966431, "seed": 1243466993, "started_at": "2026-11-22T03:35:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018654, 0.214318], "effect": 0.097832, "evaluated_at": "2026-11-22T03:38:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b5afbe457432", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.526182965, 0.689166667, 0.551612903, 0.43492823], "frontier": [0.548264984, 0.695, 0.669892473, 0.580382775], "goal": [0.689589905, 0.674166667, 0.546236559, 0.532057416], "hypdisc": [0.631545741, 0.671666667, 0.697849462, 0.409569378], "infogain": [0.548264984, 0.695, 0.669892473, 0.580382775], "prederr": [0.548264984, 0.695, 0.669892473, 0.580382775], "random": [0.623974763, 0.550833333, 0.540322581, 0.38708134], "risk": [0.548264984, 0.695, 0.669892473, 0.580382775]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "ea1dee476a6c29caf72192f85bf40d2d0ef1ec53d2734dbe75aab95468232be5", "primary_control": "random", "result_hash": "b901a40e307cec94c8c08f25fd381c5acd9b67ae16903adc8886b5b108bdace3", "signature": "d15bc2509e6b231847c13969d34ed907e5b9459a42781fe7eab6e71941920d5f9acc277eeae5f3ae8132debdc781883e8778f15c46cb2943cbf0b3f4a528e007", "state": "VERIFIED", "summary": {"count": 0.550473, "frontier": 0.623385, "goal": 0.610513, "hypdisc": 0.602658, "infogain": 0.623385, "prederr": 0.623385, "random": 0.525553, "risk": 0.623385}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:38:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:22Z", "cpu_seconds": 0.622511767000006, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e6513c5b2750", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "5a9b5285348e063ce3f13820d8d39aee35aba460719ef6e28ae5408934aca056", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e6513c5b2750", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:35:05Z", "seed": 1369801092, "treatment": "full"}, "priority": 2.0080307574841423, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bdc539c65202006c140ecaaaf4391e322756f74a0279e42b8b20fe1e20e7ccce", "run_seconds": 0.8535702228546143, "seed": 1369801092, "started_at": "2026-11-22T03:36:21Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:38:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e6513c5b2750", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.375], "fixed_policy": [1.0, 0.0], "full": [1.0, 0.4375], "no_world_model": [1.0, 0.3125], "random": [1.0, 0.4375]}, "prereg_hash": "5a9b5285348e063ce3f13820d8d39aee35aba460719ef6e28ae5408934aca056", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "bdc539c65202006c140ecaaaf4391e322756f74a0279e42b8b20fe1e20e7ccce", "signature": "56e71ec43c7e524ba2e473ef1a7b3854f7f91ec23fb35612f57b04c28248a7a23e0d9e79d42c22c328a046c14b0de89257a52b7b6da9ff1acfa4965018e98008", "state": "REJECTED", "summary": {"e141_best": 0.375, "fixed_policy": 0.0, "full": 0.4375, "no_world_model": 0.3125, "random": 0.4375}}, "verified_at": "2026-11-22T03:38:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:15Z", "cpu_seconds": 3.0999530679999996, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-018159a912fc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "af3abe8080020ae652157cfec76c4eaa4bf6e36e06b940a0479a9691e2cb2772", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-018159a912fc", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:35:05Z", "seed": 1232714671, "treatment": "full"}, "priority": 2.0093402075880737, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "46914cc7d2d444a53ae0aa9248904ef060c62924746b83e7fd260dfe39b4de7b", "run_seconds": 3.560791015625, "seed": 1232714671, "started_at": "2026-11-22T03:35:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037902, 0.034422], "effect": -0.00174, "evaluated_at": "2026-11-22T03:37:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-018159a912fc", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.520661157, 0.737179487, 0.191489362, 0.471354167], "frozen": [0.067493113, 0.153846154, 0.156028369, 0.154947917], "full": [0.550964187, 0.760683761, 0.180851064, 0.421223958], "retrieval": [0.578512397, 0.752136752, 0.163120567, 0.44921875], "rules_nodetect": [0.55785124, 0.760683761, 0.180851064, 0.421223958]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "af3abe8080020ae652157cfec76c4eaa4bf6e36e06b940a0479a9691e2cb2772", "primary_control": "facts", "result_hash": "46914cc7d2d444a53ae0aa9248904ef060c62924746b83e7fd260dfe39b4de7b", "signature": "4c90edcb70e3df6081591f922da42e115203e9894520d1481897285f9fc50a16598459f4c70a12109ed335dad73ae1864640e59646430c769a0cf0162a4afc0a", "state": "VERIFIED", "summary": {"facts": 0.480171, "frozen": 0.133079, "full": 0.478431, "retrieval": 0.485747, "rules_nodetect": 0.480153}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:37:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:16Z", "cpu_seconds": 0.6258149439999983, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-52baef8f9c64", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "0778f74998fbee4bb80e0c675fd13970f97c14fbcf7443b8c4dcdc4095412dba", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-52baef8f9c64", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:35:05Z", "seed": 1725055815, "treatment": "full"}, "priority": 2.0081767157742987, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ea45c74630c1e98d2d7fcff375bf8d797bf795ebdb5bbb4ec5190bcd6924d429", "run_seconds": 0.6923418045043945, "seed": 1725055815, "started_at": "2026-11-22T03:36:15Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-52baef8f9c64", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.761904762, 0.846153846, 0.851851852, 0.851851852], "frozen": [0.428571429, 0.358974359, 0.444444444, 0.444444444], "full": [0.761904762, 0.846153846, 0.851851852, 0.851851852], "retrieval": [0.761904762, 0.846153846, 0.851851852, 0.851851852], "rules_nodetect": [0.571428571, 0.384615385, 0.666666667, 0.62962963]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0778f74998fbee4bb80e0c675fd13970f97c14fbcf7443b8c4dcdc4095412dba", "primary_control": "facts", "result_hash": "ea45c74630c1e98d2d7fcff375bf8d797bf795ebdb5bbb4ec5190bcd6924d429", "signature": "0f80830ec0b428f646b5f2ad334f094d773bdf633ee1699afe7c2c394c7f08fb1df0b3e094844a8f2408774538f8b0d905b7cd54c6a1fb3c300b5a887843ae02", "state": "PARTIAL", "summary": {"facts": 0.827941, "frozen": 0.419109, "full": 0.827941, "retrieval": 0.827941, "rules_nodetect": 0.563085}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:38:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:41Z", "cpu_seconds": 0.6869772600000061, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6380482e7fab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "445993909b9ad6d7bfac14b7df7c211bf3cc6dacceb9ea1420b96a76d1b88b91", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6380482e7fab", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:35:05Z", "seed": 1278941458, "treatment": "full"}, "priority": 2.0078999785178144, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c7180552398b39dcfa8ceeda99bec12d75ec00eca35540128cfb9edc02fc57c4", "run_seconds": 0.7255673408508301, "seed": 1278941458, "started_at": "2026-11-22T03:36:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.012928, 0.020123], "effect": 0.003598, "evaluated_at": "2026-11-22T03:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6380482e7fab", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.768707483, 0.775641026, 0.874074074, 0.722222222], "frozen": [0.095238095, 0.083333333, 0.081481481, 0.090277778], "full": [0.795918367, 0.762820513, 0.874074074, 0.722222222], "retrieval": [0.80952381, 0.775641026, 0.888888889, 0.819444444], "rules_nodetect": [0.795918367, 0.762820513, 0.874074074, 0.722222222]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "445993909b9ad6d7bfac14b7df7c211bf3cc6dacceb9ea1420b96a76d1b88b91", "primary_control": "facts", "result_hash": "c7180552398b39dcfa8ceeda99bec12d75ec00eca35540128cfb9edc02fc57c4", "signature": "0c25fa457994e7789724563c852bd180f85afc14163af72551ee9ae1e538c5004ef0c9a4d380039fde2a87b37c78808bad14a8f0821feb5a60639f16b1aff303", "state": "VERIFIED", "summary": {"facts": 0.785161, "frozen": 0.087583, "full": 0.788759, "retrieval": 0.823375, "rules_nodetect": 0.788759}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:12Z", "cpu_seconds": 1.4258419180000033, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7961d778db7e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "2611c353000b1edef0002437d6b7cf88f2f375e8552485a36ba06b79d3b5cb9b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7961d778db7e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:35:05Z", "seed": 2082357475, "treatment": "full"}, "priority": 2.0093817350053165, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "750391da9f49254d54341ad723f83bf436cf7dcccdffdd3a922ca7e32548f697", "run_seconds": 1.47251558303833, "seed": 2082357475, "started_at": "2026-11-22T03:35:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.046413, 0.23689], "effect": 0.141652, "evaluated_at": "2026-11-22T03:37:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7961d778db7e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.241758242, 0.964285714, 0.1, 0.162429379], "frozen": [0.052884615, 0.0, 0.0625, 0.042372881], "full": [0.407967033, 0.964285714, 0.3203125, 0.342514124], "retrieval": [0.450549451, 0.964285714, 0.303125, 0.359463277], "rules_nodetect": [0.407967033, 0.964285714, 0.3203125, 0.342514124]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "2611c353000b1edef0002437d6b7cf88f2f375e8552485a36ba06b79d3b5cb9b", "primary_control": "facts", "result_hash": "750391da9f49254d54341ad723f83bf436cf7dcccdffdd3a922ca7e32548f697", "signature": "4868ad9fa1883c8de93d5d7a5b900377d32c15a333ecd1892902799f95b91e2c782480c2daa3c3da21e015bdb58e7aa3866a434d609d106fea23353b4561fc0a", "state": "VERIFIED", "summary": {"facts": 0.367118, "frozen": 0.039439, "full": 0.50877, "retrieval": 0.519356, "rules_nodetect": 0.50877}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:37:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:58Z", "cpu_seconds": 0.32123531700000285, "created_at": "2026-11-22T03:35:05Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-ecacb98a7eb5", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "prereg_hash": "6a6120563892dd9be5c0922daa1b4b8bce9607d60cbf756ae5e2a5318010a912", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-ecacb98a7eb5", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 0.5, "n": 12, "size": 12, "teleporters": 0, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:35:05Z", "seed": 1737135475, "treatment": "frontier"}, "priority": 2.0068534694250286, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-92ea3ced8150", "result_hash": "d82260c8912667179b8b0982463b5a8e0ba7aa5f78622c44418efc91d358c816", "run_seconds": 0.3464930057525635, "seed": 1737135475, "started_at": "2026-11-22T03:36:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.055574, 0.104662], "effect": 0.080118, "evaluated_at": "2026-11-22T03:38:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-ecacb98a7eb5", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.746323529, 0.825268817, 0.85, 0.825757576, 0.833333333, 0.835164835, 1.0, 0.819277108, 0.796116505, 0.834158416, 0.800505051, 0.829081633], "frontier": [0.926470588, 0.903225806, 0.9, 0.891414141, 0.90625, 0.928571429, 1.0, 0.924698795, 0.893203883, 0.908415842, 0.886363636, 0.910714286], "random_walk": [0.746323529, 0.809139785, 0.836842105, 0.840909091, 0.838541667, 0.824175824, 1.0, 0.813253012, 0.834951456, 0.844059406, 0.787878788, 0.841836735]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.3113, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "6a6120563892dd9be5c0922daa1b4b8bce9607d60cbf756ae5e2a5318010a912", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-92ea3ced8150", "result_hash": "d82260c8912667179b8b0982463b5a8e0ba7aa5f78622c44418efc91d358c816", "signature": "731e63c17dc0efacc2079fa86cc61c5a703931d74f6c096ecafd3eb83d123861f2475f7039a0bb7c174a2bbb2eb02da275b08d7dab3d567b9a82a1951794f204", "state": "VERIFIED", "summary": {"curiosity": 0.832916, "frontier": 0.914944, "random_walk": 0.834826}, "ties": 1, "treatment": "frontier", "wins": 11}, "verified_at": "2026-11-22T03:38:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:24Z", "cpu_seconds": 2.949102203999999, "created_at": "2026-11-22T03:35:20Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4a841933e1c2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "653c99c0cb5d04f70f78c5c9b721a1fcd2f3e123b9727430a0d42a3175fe4602", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4a841933e1c2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:35:20Z", "seed": 1509237331, "treatment": "frontier"}, "priority": 2.0099025789769476, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1a0bc3b8d4c57770d30a71bb53c6d0f420089825442ea48faf20f6d55dee13d8", "run_seconds": 3.534580945968628, "seed": 1509237331, "started_at": "2026-11-22T03:35:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.02891, 0.170413], "effect": 0.070752, "evaluated_at": "2026-11-22T03:37:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4a841933e1c2", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.641176471, 0.678431373, 0.62745098, 0.638562092], "frontier": [0.691503268, 0.643137255, 0.735947712, 0.695424837], "goal": [0.691503268, 0.643137255, 0.697385621, 0.695424837], "hypdisc": [0.633333333, 0.550980392, 0.723529412, 0.67124183], "infogain": [0.691503268, 0.643137255, 0.735947712, 0.695424837], "prederr": [0.691503268, 0.643137255, 0.735947712, 0.695424837], "random": [0.562091503, 0.68496732, 0.55620915, 0.679738562], "risk": [0.691503268, 0.643137255, 0.735947712, 0.695424837]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "653c99c0cb5d04f70f78c5c9b721a1fcd2f3e123b9727430a0d42a3175fe4602", "primary_control": "random", "result_hash": "1a0bc3b8d4c57770d30a71bb53c6d0f420089825442ea48faf20f6d55dee13d8", "signature": "492979830930fd7bdf0eabbf3a31f37d65835d7c1290028020c2a4bb42ccf71845552517b35df23bca6a4eb272a5300596c2169edecd345a6131f9d6a4d58e08", "state": "VERIFIED", "summary": {"count": 0.646405, "frontier": 0.691503, "goal": 0.681863, "hypdisc": 0.644771, "infogain": 0.691503, "prederr": 0.691503, "random": 0.620752, "risk": 0.691503}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:37:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:30Z", "cpu_seconds": 0.7449514740000005, "created_at": "2026-11-22T03:35:20Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c17b9fc30382", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "0aa6c1a93bb7514605ea6cca83f156d71e00da03066bc398300237e20241b13e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c17b9fc30382", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:35:20Z", "seed": 1286109378, "treatment": "full"}, "priority": 2.009559583104927, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "47e9afe8c3822e633f5602803d035395be829144dbc7500c4914990f5e64feef", "run_seconds": 1.0905792713165283, "seed": 1286109378, "started_at": "2026-11-22T03:35:29Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:37:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c17b9fc30382", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.5, 0.0, 0.0], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [0.75, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0aa6c1a93bb7514605ea6cca83f156d71e00da03066bc398300237e20241b13e", "primary_control": "e141_best", "result_hash": "47e9afe8c3822e633f5602803d035395be829144dbc7500c4914990f5e64feef", "signature": "ee975bd15e3ced4c6c679bbea8bdb53aaf34b55667a9d59480c9a53b6e1ee124147d2b6cb030a56a74cfc270f0d6a3a7a3be170e90be6a78b8a88d00e6a26402", "state": "PARTIAL", "summary": {"e141_best": 0.9375, "fixed_policy": 0.125, "full": 0.9375, "no_world_model": 0.9375, "random": 0.9375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:37:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:31Z", "cpu_seconds": 6.142913293999996, "created_at": "2026-11-22T03:35:20Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-692ad001f5eb", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "66c4e73d974d51e0ddc6e3b6ac1a9a1755a033826e349330f7c79d7ccb6863c7", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-692ad001f5eb", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:35:20Z", "seed": 1109588335, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "4c944e83468b0bbdb1ce0a50d6f26b2a369129db6a423cb113c10fa899085efb", "run_seconds": 7.645839214324951, "seed": 1109588335, "started_at": "2026-11-22T03:35:24Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.331651, -0.101683], "effect": -0.216667, "evaluated_at": "2026-11-22T03:37:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-692ad001f5eb", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5630689393486074, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 14, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "66c4e73d974d51e0ddc6e3b6ac1a9a1755a033826e349330f7c79d7ccb6863c7", "primary_control": "reset", "result_hash": "4c944e83468b0bbdb1ce0a50d6f26b2a369129db6a423cb113c10fa899085efb", "signature": "ed7a602b61a4389c5ba64e0c9b9555053c10942c61c10a2448dd37752a3be83aa63d4ff20f576477eaf7a040a0ddcddfe0179b77f4b22a0df4cafc399117410d", "state": "VERIFIED", "summary": {"inherit": 0.5666666666666667, "reset": 0.7833333333333333, "root": 0.3333333333333333}, "ties": 45, "treatment": "inherit", "wins": 1}, "verified_at": "2026-11-22T03:37:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:00Z", "cpu_seconds": 0.5700734059999917, "created_at": "2026-11-22T03:35:36Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0f4f8c44f978", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "79a4614eb32b8f3ec7230fb15358d6fc86048f54a59a739723ad8bebca2cc37f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0f4f8c44f978", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:35:36Z", "seed": 286232671, "treatment": "full"}, "priority": 2.0068051759849515, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2010eb28a05e767599d4d8278a3679bc28a13a8c799e454cbd6fdf3156fb3de8", "run_seconds": 0.9751579761505127, "seed": 286232671, "started_at": "2026-11-22T03:36:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.08997, 0.129185], "effect": 0.019608, "evaluated_at": "2026-11-22T03:38:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0f4f8c44f978", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.503267974, 0.450980392, 0.294573643, 0.39869281], "frozen": [0.183006536, 0.39869281, 0.294573643, 0.405228758], "full": [0.68627451, 0.405228758, 0.294573643, 0.339869281], "retrieval": [0.62745098, 0.37254902, 0.286821705, 0.385620915], "rules_nodetect": [0.908496732, 0.816993464, 0.294573643, 0.431372549]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "79a4614eb32b8f3ec7230fb15358d6fc86048f54a59a739723ad8bebca2cc37f", "primary_control": "facts", "result_hash": "2010eb28a05e767599d4d8278a3679bc28a13a8c799e454cbd6fdf3156fb3de8", "signature": "76db7787d9d8b5c0dc018ef5456125c617628f01fa9f1dafac499e69032f806cd0374cafa0d308a0311329ef040bb7b9ddeaaae9428ff723e2365adf7f3af708", "state": "VERIFIED", "summary": {"facts": 0.411879, "frozen": 0.320375, "full": 0.431487, "retrieval": 0.418111, "rules_nodetect": 0.612859}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:38:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:50Z", "cpu_seconds": 1.0226976150000127, "created_at": "2026-11-22T03:35:36Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-13b523654aa1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "182f5e5cbefe77c3a593533104b5d5d53a06845b9fb1d3b14c8e29db977c40f6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-13b523654aa1", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:35:36Z", "seed": 1934807529, "treatment": "full"}, "priority": 2.009133893303528, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "150b3ff9ccd6d156f3ada91dac4ce6a661706406b780bc3d64ebd2b0b68bc201", "run_seconds": 1.3355915546417236, "seed": 1934807529, "started_at": "2026-11-22T03:35:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.068904, 0.016807], "effect": -0.026049, "evaluated_at": "2026-11-22T03:38:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-13b523654aa1", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.154255319, 0.450980392, 0.453900709, 0.345360825], "frozen": [0.106382979, 0.111764706, 0.177304965, 0.113402062], "full": [0.191489362, 0.419607843, 0.395390071, 0.293814433], "retrieval": [0.177304965, 0.431372549, 0.411347518, 0.350515464], "rules_nodetect": [0.725177305, 0.419607843, 0.395390071, 0.293814433]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "182f5e5cbefe77c3a593533104b5d5d53a06845b9fb1d3b14c8e29db977c40f6", "primary_control": "facts", "result_hash": "150b3ff9ccd6d156f3ada91dac4ce6a661706406b780bc3d64ebd2b0b68bc201", "signature": "fab29d90cd3f29dd8b1b002e86cba2ebd0fda94dabde3e1c128470ea9cd7ddf882860ec662256f6051ad142fbded109408bba8059443f0539e902b3dcb8a1a05", "state": "VERIFIED", "summary": {"facts": 0.351124, "frozen": 0.127214, "full": 0.325075, "retrieval": 0.342635, "rules_nodetect": 0.458497}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:38:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:54Z", "cpu_seconds": 0.7216371450000025, "created_at": "2026-11-22T03:35:36Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1ae849c9d6fe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "52d9af9ac159943c30c729e7e5d54089f29bc372df00702f16a819458fc7861f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1ae849c9d6fe", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:35:36Z", "seed": 1616129556, "treatment": "full"}, "priority": 2.007105785669866, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "211e3822eab826727b14638fb77387a73e355af6dbe7f4ca829c80dfeefe701d", "run_seconds": 0.7598559856414795, "seed": 1616129556, "started_at": "2026-11-22T03:36:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.008243, 0.062712], "effect": 0.027235, "evaluated_at": "2026-11-22T03:38:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1ae849c9d6fe", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.57254902, 0.882352941, 0.634920635, 0.469387755], "frozen": [0.352941176, 0.352941176, 0.37037037, 0.307823129], "full": [0.580392157, 0.882352941, 0.656084656, 0.549319728], "retrieval": [0.6, 0.882352941, 0.682539683, 0.579931973], "rules_nodetect": [0.764705882, 0.882352941, 0.814814815, 0.721088435]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "52d9af9ac159943c30c729e7e5d54089f29bc372df00702f16a819458fc7861f", "primary_control": "facts", "result_hash": "211e3822eab826727b14638fb77387a73e355af6dbe7f4ca829c80dfeefe701d", "signature": "0a6bdcf0f2669aed81e2f1da497c2893b0b9ecd25291ab847afc5264d5a150f2ad275046f19599710de96e34bd4944d52a7e2a7a739f437adff279e615f3d603", "state": "VERIFIED", "summary": {"facts": 0.639803, "frozen": 0.346019, "full": 0.667037, "retrieval": 0.686206, "rules_nodetect": 0.795741}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:38:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:35:43Z", "cpu_seconds": 6.039726346999998, "created_at": "2026-11-22T03:35:36Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-87a044ed3b34", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "2ac1d9d63d6ecb347d7ef9da32ccd99226d8fbfefe057894107239404c05ce01", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-87a044ed3b34", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:35:36Z", "seed": 136721400, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "6ad867896fee7ff55f64ca8bb508b08d0ba6542c79ace5827a68b377ff5414df", "run_seconds": 7.050152540206909, "seed": 136721400, "started_at": "2026-11-22T03:35:36Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:37:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-87a044ed3b34", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.8024281161945168, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "2ac1d9d63d6ecb347d7ef9da32ccd99226d8fbfefe057894107239404c05ce01", "primary_control": "reset", "result_hash": "6ad867896fee7ff55f64ca8bb508b08d0ba6542c79ace5827a68b377ff5414df", "signature": "1a62b3cc5f63ef2c3c50d9d3285b31be71946d81219df29d79dfc200e9efecb56e82bc12cdff7c5b3c7ab94a4a8eef2650f16d77edb0beef9420924a06af5b03", "state": "PARTIAL", "summary": {"inherit": 0.75, "reset": 0.75, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:38:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:04Z", "cpu_seconds": 7.0188751499999995, "created_at": "2026-11-22T03:35:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7ffd5c093c52", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "36180960ff5cd5343bd43f6519da8fdd804264c3e7e62b73800b4726c4903606", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7ffd5c093c52", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:35:53Z", "seed": 1505874087, "treatment": "frontier"}, "priority": 2.009218196521832, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b78c2bcfa1a76aa063748a656b684997d502b6ecf9d65c2aefbf54ef10a8e172", "run_seconds": 9.638897895812988, "seed": 1505874087, "started_at": "2026-11-22T03:35:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.03972, 0.048511], "effect": 0.004396, "evaluated_at": "2026-11-22T03:38:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7ffd5c093c52", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.535598706, 0.418154762, 0.431547619, 0.564724919], "frontier": [0.469255663, 0.430059524, 0.398809524, 0.485436893], "goal": [0.469255663, 0.380952381, 0.398809524, 0.511326861], "hypdisc": [0.472491909, 0.421130952, 0.334821429, 0.475728155], "infogain": [0.443365696, 0.400297619, 0.392857143, 0.203883495], "prederr": [0.51618123, 0.168154762, 0.392857143, 0.553398058], "random": [0.530744337, 0.389880952, 0.380952381, 0.464401294], "risk": [0.194174757, 0.430059524, 0.398809524, 0.485436893]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "36180960ff5cd5343bd43f6519da8fdd804264c3e7e62b73800b4726c4903606", "primary_control": "random", "result_hash": "b78c2bcfa1a76aa063748a656b684997d502b6ecf9d65c2aefbf54ef10a8e172", "signature": "e1b70395499f5cec4cd86440368b0553b1008d6d8142ddcb22dec8590fb790935cb42cdd04d0e6b76df5440a07ab3afc05f781146bdbe7a481df61a476ac6f00", "state": "VERIFIED", "summary": {"count": 0.487507, "frontier": 0.44589, "goal": 0.440086, "hypdisc": 0.426043, "infogain": 0.360101, "prederr": 0.407648, "random": 0.441495, "risk": 0.37712}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:38:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:10Z", "cpu_seconds": 0.43930082800000037, "created_at": "2026-11-22T03:36:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8f8eeb7e674f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "07e079e09d08602825ed8445a667f09fab024f1330c23e396a7830a02d7d65e8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8f8eeb7e674f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:36:09Z", "seed": 1166910435, "treatment": "frontier"}, "priority": 2.0099231188753244, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "02c5ffb42ae4648a25ac6e91b1df7d84af6071791954273896286c5bb8590a44", "run_seconds": 0.720064640045166, "seed": 1166910435, "started_at": "2026-11-22T03:36:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.094467, 0.132806], "effect": 0.113636, "evaluated_at": "2026-11-22T03:38:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-8f8eeb7e674f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.984848485, 1.0, 1.0, 1.0], "frontier": [0.96969697, 1.0, 1.0, 1.0], "goal": [0.909090909, 0.833333333, 0.878787879, 0.893939394], "hypdisc": [0.848484848, 0.924242424, 0.909090909, 0.954545455], "infogain": [0.96969697, 1.0, 1.0, 1.0], "prederr": [0.96969697, 1.0, 1.0, 1.0], "random": [0.833333333, 0.909090909, 0.878787879, 0.893939394], "risk": [0.96969697, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "07e079e09d08602825ed8445a667f09fab024f1330c23e396a7830a02d7d65e8", "primary_control": "random", "result_hash": "02c5ffb42ae4648a25ac6e91b1df7d84af6071791954273896286c5bb8590a44", "signature": "43ae22353699237291e2672daf8293b130a3fdc934aea160857aa576feddf02985ee7d50f55380374bdf47b92e2db1f5a32041494b23fb04f69b7de2b584ee06", "state": "VERIFIED", "summary": {"count": 0.996212, "frontier": 0.992424, "goal": 0.878788, "hypdisc": 0.909091, "infogain": 0.992424, "prederr": 0.992424, "random": 0.878788, "risk": 0.992424}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:38:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:40Z", "cpu_seconds": 4.456667259, "created_at": "2026-11-22T03:36:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-db66d9464cc5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "d7d1bc2e25fe6d69f3e4ede1155da238bb88ae2e6e27d5d6a064600fcc60a70d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-db66d9464cc5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:36:26Z", "seed": 1659057152, "treatment": "frontier"}, "priority": 2.0080527221093383, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "43c722618fd6c3371643b85bd13ba56e9c52415f1ff6326ff16b320a79c44da6", "run_seconds": 6.068053245544434, "seed": 1659057152, "started_at": "2026-11-22T03:36:33Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-db66d9464cc5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d7d1bc2e25fe6d69f3e4ede1155da238bb88ae2e6e27d5d6a064600fcc60a70d", "primary_control": "random", "result_hash": "43c722618fd6c3371643b85bd13ba56e9c52415f1ff6326ff16b320a79c44da6", "signature": "4db26c8dd55b59a8482da0a2f1ae3df2ac1b25b30129d7949a93a4fc1cf51b9c8a5b1713a18790f4926427fee79e6029b3a7ff797f261ae2beeb64250ff34106", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:28Z", "cpu_seconds": 0.6073588000000001, "created_at": "2026-11-22T03:36:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3c9a93570695", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "803c7f338821b1158b1a14c0fe068c5d07bed63655322943e18a3c455827111a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3c9a93570695", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:36:26Z", "seed": 1001956415, "treatment": "full"}, "priority": 2.0090301393963834, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4011d5b9374d344c30f8675743e42d3d840670c0759eeaeb46beac611dce2c7f", "run_seconds": 1.6596403121948242, "seed": 1001956415, "started_at": "2026-11-22T03:36:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.011046, 0.086475], "effect": 0.04876, "evaluated_at": "2026-11-22T03:38:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3c9a93570695", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.23255814, 0.68627451, 0.878205128, 0.85620915], "frozen": [0.100775194, 0.08496732, 0.08974359, 0.091503268], "full": [0.310077519, 0.77124183, 0.884615385, 0.882352941], "retrieval": [0.426356589, 0.830065359, 0.923076923, 0.934640523], "rules_nodetect": [0.310077519, 0.77124183, 0.884615385, 0.882352941]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "803c7f338821b1158b1a14c0fe068c5d07bed63655322943e18a3c455827111a", "primary_control": "facts", "result_hash": "4011d5b9374d344c30f8675743e42d3d840670c0759eeaeb46beac611dce2c7f", "signature": "cd2605ca63dedb0fea4c1d67daa53b2f0b1367402e1017b295782fefd0dcf889c43309f4567cc11500075a24250a973b285aa63480b7efc03f29f12d52affb01", "state": "VERIFIED", "summary": {"facts": 0.663312, "frozen": 0.091747, "full": 0.712072, "retrieval": 0.778535, "rules_nodetect": 0.712072}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:38:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:57Z", "cpu_seconds": 0.37622768199999257, "created_at": "2026-11-22T03:36:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-99a83c17894b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "0dc07aeb7409c8b9e56811fa2eeccbe3709795d047e2f6e969555f20b35ae6dc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-99a83c17894b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:36:26Z", "seed": 1195686118, "treatment": "full"}, "priority": 2.0070841351162954, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1071115b9c53520cf53ae0bedd0c203457b4f78c379d25362688fe55b547b42c", "run_seconds": 0.4420952796936035, "seed": 1195686118, "started_at": "2026-11-22T03:36:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014116, 0.153699], "effect": 0.069792, "evaluated_at": "2026-11-22T03:38:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-99a83c17894b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.825, 1.0, 1.0, 0.854166667], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 0.958333333], "retrieval": [0.925, 1.0, 1.0, 0.916666667], "rules_nodetect": [1.0, 1.0, 1.0, 0.958333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "0dc07aeb7409c8b9e56811fa2eeccbe3709795d047e2f6e969555f20b35ae6dc", "primary_control": "facts", "result_hash": "1071115b9c53520cf53ae0bedd0c203457b4f78c379d25362688fe55b547b42c", "signature": "7b25a6c7baef1a5d9ec42b25cbe075ce867c90874bf54c013f75e14f086c4f04b08dc0ac3ce26609b017ee6f4a5e59f3f2695b47b73913dc7aa500d8e1610f0a", "state": "VERIFIED", "summary": {"facts": 0.919792, "frozen": 0.0, "full": 0.989583, "retrieval": 0.960417, "rules_nodetect": 0.989583}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:38:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:45Z", "cpu_seconds": 1.7637012140000081, "created_at": "2026-11-22T03:36:43Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-8c9324b74d1c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "prereg_hash": "420e18aa8a53fa01878b8967c7d07c746d7c927ad38f75df7ba865b51fe3e970", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-8c9324b74d1c", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.0, "target_noise": 0.15}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:36:43Z", "seed": 998530416, "treatment": "iw_temperature"}, "priority": 2.0099362294106733, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-1f64331acd99", "result_hash": "55a7324b3b2ccf90e7408be8d505cded212aa7f20d97def6fcc8db311f20770f", "run_seconds": 1.8972399234771729, "seed": 998530416, "started_at": "2026-11-22T03:36:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.015427, 0.000761], "effect": -0.007333, "evaluated_at": "2026-11-22T03:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-8c9324b74d1c", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.173308535, 0.173667022, 0.137084485, 0.103188632, 0.128201505, 0.112919543, 0.113808825, 0.143852705, 0.116950673, 0.146296647, 0.21461861, 0.136624858, 0.164822592, 0.113539242, 0.11713232, 0.171167195], "raw": [0.14699504, 0.140977375, 0.118397353, 0.090613175, 0.086242782, 0.09530014, 0.113808825, 0.12965484, 0.092858993, 0.076431438, 0.11700563, 0.12955926, 0.156285985, 0.114220405, 0.085095638, 0.15824328], "temperature": [0.164963105, 0.148940105, 0.149076905, 0.090613175, 0.09831142, 0.119994635, 0.117607035, 0.12965484, 0.116950673, 0.141828167, 0.165499488, 0.141401453, 0.168420505, 0.122186975, 0.100944732, 0.17346906]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1572, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.696", "prereg_hash": "420e18aa8a53fa01878b8967c7d07c746d7c927ad38f75df7ba865b51fe3e970", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-1f64331acd99", "result_hash": "55a7324b3b2ccf90e7408be8d505cded212aa7f20d97def6fcc8db311f20770f", "signature": "1388bb554c2532475a14e00c8e6554f10d670496694203ebd2c25eb6c27a58a4a0187448874385ae3b5778a5b84c9f9ab1a5abc9acfb023c34ddd25b5aa38e0d", "state": "VERIFIED", "summary": {"iw_temperature": 0.141699, "raw": 0.115731, "temperature": 0.134366}, "ties": 1, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:51Z", "cpu_seconds": 4.7380572299999955, "created_at": "2026-11-22T03:36:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ed666253b152", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "3d40fc40c0999308ed233ed83216cb16dfd164dd91efbbeeb1b513a599b57c82", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ed666253b152", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:36:43Z", "seed": 1574220505, "treatment": "frontier"}, "priority": 2.0092590891321516, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9577ce825b8239d87758d9008380c630a41ebb2908b16a734743651e10a243b2", "run_seconds": 5.827779054641724, "seed": 1574220505, "started_at": "2026-11-22T03:36:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0844, 0.468912], "effect": 0.192256, "evaluated_at": "2026-11-22T03:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ed666253b152", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.898148148, 0.751173709, 0.701492537, 0.802816901], "frontier": [0.731481481, 0.948356808, 0.940298507, 0.676056338], "goal": [0.810185185, 0.708920188, 0.925373134, 0.666666667], "hypdisc": [0.791666667, 0.779342723, 0.63681592, 0.755868545], "infogain": [0.731481481, 0.948356808, 0.940298507, 0.676056338], "prederr": [0.731481481, 0.948356808, 0.940298507, 0.676056338], "random": [0.75, 0.685446009, 0.378109453, 0.713615023], "risk": [0.731481481, 0.948356808, 0.940298507, 0.676056338]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "3d40fc40c0999308ed233ed83216cb16dfd164dd91efbbeeb1b513a599b57c82", "primary_control": "random", "result_hash": "9577ce825b8239d87758d9008380c630a41ebb2908b16a734743651e10a243b2", "signature": "7821c6e4772408b68272db7548ef43a334cc96a5e1223d6f4bfd76a670e4cfa01e710e1432874c860b225d055d9d8b790dec2c63c4b526e19874c7012aeeb105", "state": "VERIFIED", "summary": {"count": 0.788408, "frontier": 0.824048, "goal": 0.777786, "hypdisc": 0.740923, "infogain": 0.824048, "prederr": 0.824048, "random": 0.631793, "risk": 0.824048}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:38:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:36:49Z", "cpu_seconds": 3.2502112170000004, "created_at": "2026-11-22T03:36:43Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-af5505692987", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "6bfd32a337f6eabddb02ed3b5790c1f87938cce08e97c9a6847cdbcbd5b4ce28", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-af5505692987", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:36:43Z", "seed": 507544250, "treatment": "gated"}, "priority": 2.0096228002903627, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "e75bd6a17827096a567baf6a850394c524b6190ddc0dfbe581cd975143d6b76f", "run_seconds": 4.056628704071045, "seed": 507544250, "started_at": "2026-11-22T03:36:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0074, 0.0024], "effect": -0.0025, "evaluated_at": "2026-11-22T03:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-af5505692987", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.401785714, 0.5703125, 0.684210526, 0.44], "gated": [0.401785714, 0.5703125, 0.684210526, 0.44], "naive": [0.705357143, 0.7734375, 0.710526316, 0.52], "scratch": [0.401785714, 0.5703125, 0.684210526, 0.45], "ungated": [0.625, 0.7734375, 0.710526316, 0.46]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "6bfd32a337f6eabddb02ed3b5790c1f87938cce08e97c9a6847cdbcbd5b4ce28", "primary_control": "scratch", "result_hash": "e75bd6a17827096a567baf6a850394c524b6190ddc0dfbe581cd975143d6b76f", "signature": "96c82a159092f6e19db6dae98ebba74353f0cf48dadf3edc8daafdc6237f8e9c742d8266a5103180fd6042b9a215b2dedb2f5a561592aead279c294c7c655a04", "state": "VERIFIED", "summary": {"compressed": 0.524077, "gated": 0.524077, "naive": 0.67733, "scratch": 0.526577, "ungated": 0.642241}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:11Z", "cpu_seconds": 1.650487176, "created_at": "2026-11-22T03:37:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-76e75cc7ed85", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "f863f9e258ce87ec649cc71a118c578bc7061ce92233ed35b212c40054ebad57", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-76e75cc7ed85", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:37:00Z", "seed": 1206284499, "treatment": "frontier"}, "priority": 2.0081635977252694, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e86fa397a04df414700262fae8a47e687a5de71f5317d96f6d6a20999793f3af", "run_seconds": 2.2678964138031006, "seed": 1206284499, "started_at": "2026-11-22T03:37:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.022505, 0.297756], "effect": 0.160131, "evaluated_at": "2026-11-22T03:38:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-76e75cc7ed85", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.77124183, 0.803921569, 0.77124183, 0.777777778], "frontier": [0.660130719, 0.947712418, 0.901960784, 0.908496732], "goal": [0.738562092, 0.810457516, 0.758169935, 0.732026144], "hypdisc": [0.797385621, 0.816993464, 0.823529412, 0.68627451], "infogain": [0.660130719, 0.947712418, 0.901960784, 0.908496732], "prederr": [0.660130719, 0.947712418, 0.901960784, 0.908496732], "random": [0.705882353, 0.679738562, 0.679738562, 0.712418301], "risk": [0.660130719, 0.947712418, 0.901960784, 0.908496732]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "f863f9e258ce87ec649cc71a118c578bc7061ce92233ed35b212c40054ebad57", "primary_control": "random", "result_hash": "e86fa397a04df414700262fae8a47e687a5de71f5317d96f6d6a20999793f3af", "signature": "c2c8094f7d485be8d43d7de195e119d9f6406f58c709c2af2efe4ed2ed913a856b01d20a3157e942d48711890d96122b18a2a014e7b6deae28e1482fe80e1002", "state": "VERIFIED", "summary": {"count": 0.781046, "frontier": 0.854575, "goal": 0.759804, "hypdisc": 0.781046, "infogain": 0.854575, "prederr": 0.854575, "random": 0.694444, "risk": 0.854575}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:38:15Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:39:08Z", "cpu_seconds": 33.495580796000006, "created_at": "2026-11-22T03:37:00Z", "error": "lease expired; ", "family": "ow_goal", "fence": 2, "id": "E141-ow_goal-c2a940d67696", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 2, "n": 4}, "prereg_hash": "f67e2bb19e645c3bc89ef51165974de44ba60b51ed61575ee7a378c036f1ffdf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c2a940d67696", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:37:00Z", "seed": 2006686810, "treatment": "full"}, "priority": 2.0076326543686944, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ad39c1ed9ca13fc12fe4519c40e5e90e08091c59db58b55e926d5bcab9f38926", "run_seconds": 40.43615937232971, "seed": 2006686810, "started_at": "2026-11-22T03:37:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.18, 0.555], "effect": 0.1875, "evaluated_at": "2026-11-22T03:39:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c2a940d67696", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 0.75, 0.0, 1.0], "no_world_model": [0.0, 0.5, 0.0, 1.0], "random": [0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f67e2bb19e645c3bc89ef51165974de44ba60b51ed61575ee7a378c036f1ffdf", "primary_control": "e141_best", "result_hash": "ad39c1ed9ca13fc12fe4519c40e5e90e08091c59db58b55e926d5bcab9f38926", "signature": "5bb1210cf242f2c598bb36c0f6abd1cc662aaa5e91f949f6769ff071f4cc012f9439f9186536af315e332d19ea914d52e2b6491f910abab20402d2c8f6964b0c", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0, "full": 0.4375, "no_world_model": 0.375, "random": 0.25}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:40:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:16Z", "cpu_seconds": 0.8587573070000001, "created_at": "2026-11-22T03:37:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2ec1382d193c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "bef33cac7912ed1bbaac2608fdf288f4920296be1b14093c15891c76a0b39e1f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2ec1382d193c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:37:00Z", "seed": 392862974, "treatment": "full"}, "priority": 2.0077575408587713, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "283e326d054867e55e4364171461353f1ae98e8b32189c246b4cb405dc798d7a", "run_seconds": 0.8636114597320557, "seed": 392862974, "started_at": "2026-11-22T03:37:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013957, 0.065869], "effect": 0.025956, "evaluated_at": "2026-11-22T03:38:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2ec1382d193c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.946428571, 0.432692308, 0.958333333, 0.980769231], "frozen": [0.321428571, 0.346153846, 0.229166667, 0.365384615], "full": [0.982142857, 0.509615385, 0.96875, 0.961538462], "retrieval": [0.982142857, 0.605769231, 0.958333333, 0.980769231], "rules_nodetect": [0.982142857, 0.509615385, 0.96875, 0.961538462]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "bef33cac7912ed1bbaac2608fdf288f4920296be1b14093c15891c76a0b39e1f", "primary_control": "facts", "result_hash": "283e326d054867e55e4364171461353f1ae98e8b32189c246b4cb405dc798d7a", "signature": "350fac4b1b54edd1d4c066ebae94cee08bdf2c8fecb6a79e06a876955f3e6975228082c90ab12009ca6d9b9f8c4b672397a04dd8c2c0e6e9603f62dffe61dc07", "state": "VERIFIED", "summary": {"facts": 0.829556, "frozen": 0.315533, "full": 0.855512, "retrieval": 0.881754, "rules_nodetect": 0.855512}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:38:15Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:38:10Z", "cpu_seconds": 0.9959171329999998, "created_at": "2026-11-22T03:37:00Z", "error": "lease expired; ", "family": "ow_model", "fence": 2, "id": "E141-ow_model-7129d6e24b8b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "400e1ce0cfd61d3104d237490128a82ae3af33692ba9f30aeee5cc5f66d20aa9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7129d6e24b8b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:37:00Z", "seed": 1186187767, "treatment": "full"}, "priority": 2.0091508939626928, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1969cef90c290e1dcc974fcddb1227707ae53a47d551261089ccdea97146b510", "run_seconds": 1.1219453811645508, "seed": 1186187767, "started_at": "2026-11-22T03:37:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023093, 0.098276], "effect": 0.037592, "evaluated_at": "2026-11-22T03:38:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7129d6e24b8b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.96, 0.7, 0.381578947, 0.934782609], "frozen": [0.24, 0.22, 0.289473684, 0.293478261], "full": [0.95, 0.79, 0.473684211, 0.913043478], "retrieval": [0.94, 0.84, 0.657894737, 0.913043478], "rules_nodetect": [0.95, 0.79, 0.473684211, 0.913043478]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "400e1ce0cfd61d3104d237490128a82ae3af33692ba9f30aeee5cc5f66d20aa9", "primary_control": "facts", "result_hash": "1969cef90c290e1dcc974fcddb1227707ae53a47d551261089ccdea97146b510", "signature": "51ccb143f67a438efece5958ed99e011594308cc15300d1a411eeccb66d603b894573f2c583add6f95c5c904774f961e84281e0c4621934a2e7b213187728b00", "state": "VERIFIED", "summary": {"facts": 0.74409, "frozen": 0.260738, "full": 0.781682, "retrieval": 0.837735, "rules_nodetect": 0.781682}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:38:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:10Z", "cpu_seconds": 1.371153821, "created_at": "2026-11-22T03:37:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b3fc1969d6b4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "be188c5b928c65841e4febd9f106ca153199796964e274e9bf72dcd55d150322", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b3fc1969d6b4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:37:00Z", "seed": 1413772052, "treatment": "full"}, "priority": 2.009031626102921, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3f113e7bd43af3af14e8ca3533ccee36afc9764b33384487f8fdf4c93b777e89", "run_seconds": 2.0383405685424805, "seed": 1413772052, "started_at": "2026-11-22T03:37:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.040028, 0.209671], "effect": 0.12485, "evaluated_at": "2026-11-22T03:38:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b3fc1969d6b4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.654117647, 0.817777778, 0.66, 0.609302326], "frozen": [0.101176471, 0.0, 0.38, 0.1], "full": [0.802352941, 0.817777778, 0.86, 0.760465116], "retrieval": [0.814117647, 0.937777778, 0.86, 0.837209302], "rules_nodetect": [0.802352941, 0.817777778, 0.86, 0.760465116]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "be188c5b928c65841e4febd9f106ca153199796964e274e9bf72dcd55d150322", "primary_control": "facts", "result_hash": "3f113e7bd43af3af14e8ca3533ccee36afc9764b33384487f8fdf4c93b777e89", "signature": "949ade4b20d1ae8df8a468bf6717324cea8001890219e6fe75ef4865230106043160ba8182876f0953bdbb56ae568e6267356971218477cbdea55ca8799ece0e", "state": "VERIFIED", "summary": {"facts": 0.685299, "frozen": 0.145294, "full": 0.810149, "retrieval": 0.862276, "rules_nodetect": 0.810149}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:38:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:10Z", "cpu_seconds": 1.5201355250000002, "created_at": "2026-11-22T03:37:00Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1e1a213ff520", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "prereg_hash": "0d1b17a30bf50aef3a4b546dd4dfb0898b20de84e2f9475c282a7607a5fe4776", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1e1a213ff520", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:37:00Z", "seed": 644180272, "treatment": "gated"}, "priority": 2.0082742233616027, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "751e732a6237a432e6aafa7bfa5804ba1c7b2a6717c8e9875d6b156f949d6a81", "run_seconds": 2.2246179580688477, "seed": 644180272, "started_at": "2026-11-22T03:37:08Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1e1a213ff520", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.708333333, 0.167676768, 0.165656566, 0.187878788], "gated": [0.708333333, 0.167676768, 0.165656566, 0.187878788], "naive": [0.483333333, 0.123232323, 0.088888889, 0.161616162], "scratch": [0.708333333, 0.167676768, 0.165656566, 0.187878788], "ungated": [0.55, 0.123232323, 0.088888889, 0.137373737]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0d1b17a30bf50aef3a4b546dd4dfb0898b20de84e2f9475c282a7607a5fe4776", "primary_control": "scratch", "result_hash": "751e732a6237a432e6aafa7bfa5804ba1c7b2a6717c8e9875d6b156f949d6a81", "signature": "45e2e5802a5bd5fd30e601af04c94f713793d6d5b8defbf856adee74f75971dc084914dca468671e3a925941cc368561e69d6f699cba0a6c9bea557844dd1907", "state": "PARTIAL", "summary": {"compressed": 0.307386, "gated": 0.307386, "naive": 0.214268, "scratch": 0.307386, "ungated": 0.224874}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:38:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:18Z", "cpu_seconds": 2.5016242220000002, "created_at": "2026-11-22T03:37:00Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2e475c4e3e29", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lights|lights|1|9"}, "prereg_hash": "02b82551a09d5393737b976c194e88fdea69f92b0418d3a5a72efefe950fd901", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2e475c4e3e29", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lights|lights|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:37:00Z", "seed": 928251708, "treatment": "gated"}, "priority": 2.007368965979331, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "bcc013229913e7c9f8f4b0bda22b572029204af699b179308ef073473803b43f", "run_seconds": 2.848236322402954, "seed": 928251708, "started_at": "2026-11-22T03:37:16Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2e475c4e3e29", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.388888889, 0.263888889, 0.513888889, 0.256944444], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.548611111, 0.569444444, 0.763888889, 0.555555556]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "02b82551a09d5393737b976c194e88fdea69f92b0418d3a5a72efefe950fd901", "primary_control": "scratch", "result_hash": "bcc013229913e7c9f8f4b0bda22b572029204af699b179308ef073473803b43f", "signature": "fcdf10f0ae3e15697a0691b7b3152b9af581c0d910de80d9e75dcaadddf7bac681e75a6b12b35be87fc1e5cebe3ba87d3279320c4f9fd99f85deda7987ad3b00", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.355903, "scratch": 1.0, "ungated": 0.609375}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:38:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:03Z", "cpu_seconds": 0.666215176999998, "created_at": "2026-11-22T03:37:00Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-597c5710aa11", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "b273a8c4244bab8ea2c0d4b564997d35cb358b2de57c98f429425fd53a03713e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-597c5710aa11", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:37:00Z", "seed": 434748040, "treatment": "gated"}, "priority": 2.0097307971198757, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "95bdfbe7a60615de899acb73f781fade58e0db011333ddd0387ec2d66e1cbcc8", "run_seconds": 0.844069242477417, "seed": 434748040, "started_at": "2026-11-22T03:37:02Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-597c5710aa11", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.163398693, 0.281045752, 0.286821705, 0.449612403], "gated": [0.163398693, 0.281045752, 0.286821705, 0.449612403], "naive": [0.137254902, 0.14379085, 0.23255814, 0.15503876], "scratch": [0.163398693, 0.281045752, 0.286821705, 0.449612403], "ungated": [0.22875817, 0.14379085, 0.23255814, 0.449612403]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b273a8c4244bab8ea2c0d4b564997d35cb358b2de57c98f429425fd53a03713e", "primary_control": "scratch", "result_hash": "95bdfbe7a60615de899acb73f781fade58e0db011333ddd0387ec2d66e1cbcc8", "signature": "d71d5744111e2cb551a2fe12dd2148bb7e1a9fd78d8cb46d3eb438a851db01fdbb6111f6a3fa2ba71e68340dcebf8efd3d325ec0541bd9d2e3d1b045870bb809", "state": "PARTIAL", "summary": {"compressed": 0.29522, "gated": 0.29522, "naive": 0.167161, "scratch": 0.29522, "ungated": 0.26368}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:38:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:01Z", "cpu_seconds": 1.499562013, "created_at": "2026-11-22T03:37:16Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-fbf2e77db9df", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "prereg_hash": "9c2cdef09bfffcfb1e3a242e0e5ca307d69f764cac457fd8134807b2265d26bf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-fbf2e77db9df", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:37:16Z", "seed": 224975560, "treatment": "iw_temperature"}, "priority": 2.0069079585959213, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-77c6ee597499", "result_hash": "1ce9d86749446ff3938e8a982067060ae26a5fb3028218486c98ad04edddc8ab", "run_seconds": 1.6144492626190186, "seed": 224975560, "started_at": "2026-11-22T03:37:59Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020615, 0.007198], "effect": -0.006708, "evaluated_at": "2026-11-22T03:38:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-fbf2e77db9df", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.18333078, 0.23032546, 0.28866757, 0.237596075, 0.27938968, 0.234512712, 0.229412898, 0.213528922, 0.246282543, 0.26169332, 0.24707077, 0.235625798, 0.302569205, 0.180094105, 0.266646345, 0.248478843], "raw": [0.139231168, 0.251511388, 0.28866757, 0.23231459, 0.241167872, 0.212186703, 0.214275515, 0.193927727, 0.217904307, 0.292308715, 0.289167528, 0.167683085, 0.252622148, 0.201426155, 0.266646345, 0.234163297], "temperature": [0.130662667, 0.23032546, 0.277489035, 0.224256728, 0.2262581, 0.234512712, 0.236269443, 0.193927727, 0.246282543, 0.285086285, 0.252173735, 0.19891031, 0.275752798, 0.22433603, 0.307483925, 0.234163297]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0622, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.867", "prereg_hash": "9c2cdef09bfffcfb1e3a242e0e5ca307d69f764cac457fd8134807b2265d26bf", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-77c6ee597499", "result_hash": "1ce9d86749446ff3938e8a982067060ae26a5fb3028218486c98ad04edddc8ab", "signature": "21f8cae6389d4b728949db4b74192ccacca17abd96716b2fc793a298992e2ffee662cb6239d10d262e556fbdc3f829ca06ea6b33acd06999dc073c0adeba2904", "state": "VERIFIED", "summary": {"iw_temperature": 0.242827, "raw": 0.23095, "temperature": 0.236118}, "ties": 3, "treatment": "iw_temperature", "wins": 5}, "verified_at": "2026-11-22T03:38:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:48Z", "cpu_seconds": 0.8153238679999999, "created_at": "2026-11-22T03:37:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-000cefc551ca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "bdf734c8edcdfda9788db70b70ad3e814f32e35817047842567d4617e8ac1f80", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-000cefc551ca", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:37:16Z", "seed": 612462645, "treatment": "frontier"}, "priority": 2.0078446322354586, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6a2d5c6f85d18732faae1392d18dc58c9ba3da58c3ed5e726f0aa6b5638eebfa", "run_seconds": 0.9516854286193848, "seed": 612462645, "started_at": "2026-11-22T03:37:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.271776, 0.167334], "effect": -0.052221, "evaluated_at": "2026-11-22T03:38:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-000cefc551ca", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.431372549, 0.416666667, 0.404255319, 0.68627451], "frontier": [0.424836601, 0.439393939, 0.361702128, 0.339869281], "goal": [0.424836601, 0.424242424, 0.390070922, 0.607843137], "hypdisc": [0.411764706, 0.416666667, 0.404255319, 0.745098039], "infogain": [0.411764706, 0.439393939, 0.425531915, 0.692810458], "prederr": [0.411764706, 0.424242424, 0.361702128, 0.673202614], "random": [0.254901961, 0.409090909, 0.411347518, 0.699346405], "risk": [0.424836601, 0.431818182, 0.411347518, 0.339869281]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "bdf734c8edcdfda9788db70b70ad3e814f32e35817047842567d4617e8ac1f80", "primary_control": "random", "result_hash": "6a2d5c6f85d18732faae1392d18dc58c9ba3da58c3ed5e726f0aa6b5638eebfa", "signature": "8fc3fc882d7db81127d4b4757876a164c1404b40bfd97651a5d882c270361062cc38b4c1cce75ab154dfc9aca42a2d33637d06284e9a712e7363ce95a4fee301", "state": "VERIFIED", "summary": {"count": 0.484642, "frontier": 0.39145, "goal": 0.461748, "hypdisc": 0.494446, "infogain": 0.492375, "prederr": 0.467728, "random": 0.443672, "risk": 0.401968}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:38:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:26Z", "cpu_seconds": 2.0775651130000004, "created_at": "2026-11-22T03:37:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5678b22e38b4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "29af5d0826a5b559401ee6781f6dfde5dacc0a510ec319f03a62fefd07903cb5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5678b22e38b4", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:37:16Z", "seed": 928729395, "treatment": "frontier"}, "priority": 2.0098523690052614, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "acb9bf46648f959bad01ea23f5e744aae4a6eabc55a853be764f11715ea3be6d", "run_seconds": 2.897099494934082, "seed": 928729395, "started_at": "2026-11-22T03:37:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.029135, 0.175276], "effect": 0.073071, "evaluated_at": "2026-11-22T03:38:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5678b22e38b4", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.906976744, 0.682170543, 0.765151515, 0.705426357], "frontier": [0.984496124, 0.658914729, 0.78030303, 0.496124031], "goal": [0.790697674, 0.573643411, 0.712121212, 0.604651163], "hypdisc": [0.775193798, 0.635658915, 0.696969697, 0.596899225], "infogain": [0.984496124, 0.658914729, 0.78030303, 0.496124031], "prederr": [0.984496124, 0.658914729, 0.78030303, 0.496124031], "random": [0.782945736, 0.620155039, 0.681818182, 0.542635659], "risk": [0.984496124, 0.658914729, 0.78030303, 0.496124031]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "29af5d0826a5b559401ee6781f6dfde5dacc0a510ec319f03a62fefd07903cb5", "primary_control": "random", "result_hash": "acb9bf46648f959bad01ea23f5e744aae4a6eabc55a853be764f11715ea3be6d", "signature": "fd5dbde7016111f68df0d431bbc6857d77d3504c065818bbec541928fd8ec6f99006eacf330956671d9e3edbe52de080cdb54a8b63bad2df178691933819a701", "state": "VERIFIED", "summary": {"count": 0.764931, "frontier": 0.729959, "goal": 0.670278, "hypdisc": 0.67618, "infogain": 0.729959, "prederr": 0.729959, "random": 0.656889, "risk": 0.729959}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:38:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:55Z", "cpu_seconds": 1.6295732939999998, "created_at": "2026-11-22T03:37:16Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5409700e7710", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "553c2c67b5b690424e11b9c00d437b8516635fc4791551901489c963b45f62c7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5409700e7710", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:37:16Z", "seed": 733703863, "treatment": "full"}, "priority": 2.0072772682005384, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e0c5116ccf20b5f7a2254d9269dc290fdc80782780cd953f39ec39dd57b839fd", "run_seconds": 1.6971979141235352, "seed": 733703863, "started_at": "2026-11-22T03:37:53Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5409700e7710", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.75, 1.0], "fixed_policy": [0.25, 0.25, 0.0, 0.75], "full": [0.75, 1.0, 0.75, 1.0], "no_world_model": [0.0, 1.0, 0.0, 1.0], "random": [0.0, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "553c2c67b5b690424e11b9c00d437b8516635fc4791551901489c963b45f62c7", "primary_control": "e141_best", "result_hash": "e0c5116ccf20b5f7a2254d9269dc290fdc80782780cd953f39ec39dd57b839fd", "signature": "df553271d024f9b4fdc3f45d8e1ac2fa542bbf6c97564fc4b1f365c615b982b283b7304065b9991b1acbac31172bcbabc18d617e709758d3e9633cb8480cb401", "state": "PARTIAL", "summary": {"e141_best": 0.875, "fixed_policy": 0.3125, "full": 0.875, "no_world_model": 0.5, "random": 0.6875}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:38:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:57Z", "cpu_seconds": 2.5288464740000003, "created_at": "2026-11-22T03:37:16Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9811b95f9dd1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 10, "n": 4}, "prereg_hash": "a3b3fd1f9850988e7c12652c0db3dba244844474aa6871c499bd15264d8856c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9811b95f9dd1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:37:16Z", "seed": 1426086613, "treatment": "full"}, "priority": 2.0072705966543873, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "735445faf6cf54180ead1c2e0535d360d483a4251eae27e013e0be82ca224d21", "run_seconds": 2.5608842372894287, "seed": 1426086613, "started_at": "2026-11-22T03:37:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:38:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9811b95f9dd1", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.25, 1.0], "fixed_policy": [0.0, 0.0, 0.5, 0.0], "full": [1.0, 1.0, 0.5, 1.0], "no_world_model": [1.0, 1.0, 0.25, 1.0], "random": [1.0, 1.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a3b3fd1f9850988e7c12652c0db3dba244844474aa6871c499bd15264d8856c3", "primary_control": "e141_best", "result_hash": "735445faf6cf54180ead1c2e0535d360d483a4251eae27e013e0be82ca224d21", "signature": "1a83180e66cef8edc971dd417dc38a4e371db6375f0353418da22acf0e66a117c9547a0ee6aa0156d2c95bba48840f83e00582d89fc8e03070149e9f114bba02", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.125, "full": 0.875, "no_world_model": 0.8125, "random": 0.875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:38:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:21Z", "cpu_seconds": 2.0134115159999997, "created_at": "2026-11-22T03:37:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2d79b27ad79d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "a5264335fff2d816c406d993e84ea5499e9f3c0cf1811a861dab0d5f4040792f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2d79b27ad79d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:37:16Z", "seed": 1874224301, "treatment": "full"}, "priority": 2.0047241807251446, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "dd665087d3d08da810b88612a24fd81576665d11b6e44edcd47421e3594094c4", "run_seconds": 2.462195873260498, "seed": 1874224301, "started_at": "2026-11-22T03:38:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.124893, 0.200627], "effect": 0.16276, "evaluated_at": "2026-11-22T03:38:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2d79b27ad79d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.80952381, 0.802083333, 0.892857143, 0.833333333], "frozen": [0.0, 0.0, 0.142857143, 0.142857143], "full": [0.994047619, 0.994791667, 1.0, 1.0], "retrieval": [0.970238095, 0.90625, 0.971428571, 0.916666667], "rules_nodetect": [0.357142857, 0.411458333, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a5264335fff2d816c406d993e84ea5499e9f3c0cf1811a861dab0d5f4040792f", "primary_control": "facts", "result_hash": "dd665087d3d08da810b88612a24fd81576665d11b6e44edcd47421e3594094c4", "signature": "2992b125826831e1c227f0b27f79c389769febf231fba47371cb7655c2e1587c2d7e0e321be3099a27832da8ef2da7fb9a9937847546ed3ac1ebfcbb246f1200", "state": "VERIFIED", "summary": {"facts": 0.834449, "frozen": 0.071429, "full": 0.99721, "retrieval": 0.941146, "rules_nodetect": 0.69215}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:38:35Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:38:40Z", "cpu_seconds": 3.945868802, "created_at": "2026-11-22T03:37:16Z", "error": "lease expired; ", "family": "ow_model", "fence": 2, "id": "E141-ow_model-3cc6331c04a3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 7, "n": 4}, "prereg_hash": "fce21adca7b5242ec37876181126e1e24b017c6f39dc119a12d1388d10af9587", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3cc6331c04a3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:37:16Z", "seed": 1279535753, "treatment": "full"}, "priority": 2.007873459628769, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1cbf94313cb13b4c96d82e717f8a1ab36a0796b69eee69b62bf918b668f16296", "run_seconds": 5.813216924667358, "seed": 1279535753, "started_at": "2026-11-22T03:37:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.06069, 0.287705], "effect": 0.174197, "evaluated_at": "2026-11-22T03:38:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3cc6331c04a3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.1046875, 0.300555556, 0.173913043, 0.398333333], "frozen": [0.053125, 0.047777778, 0.057971014, 0.052222222], "full": [0.451041667, 0.425555556, 0.304347826, 0.493333333], "retrieval": [0.473958333, 0.5, 0.373188406, 0.546111111], "rules_nodetect": [0.451041667, 0.425555556, 0.304347826, 0.493333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "fce21adca7b5242ec37876181126e1e24b017c6f39dc119a12d1388d10af9587", "primary_control": "facts", "result_hash": "1cbf94313cb13b4c96d82e717f8a1ab36a0796b69eee69b62bf918b668f16296", "signature": "943d3d8a491fa0629078ae8d7591cea64a1a8725b13fc30e4417eec7563bb19e313ece8d08dbdf7d64eb8cabaf119d71e993f880477bd1f4bc788ccabedcdc00", "state": "VERIFIED", "summary": {"facts": 0.244372, "frozen": 0.052774, "full": 0.41857, "retrieval": 0.473314, "rules_nodetect": 0.41857}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:38:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:27Z", "cpu_seconds": 1.264221645, "created_at": "2026-11-22T03:37:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-be89c283c3be", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "b7f1df39cd9aa0f2dae0dd824d79ed7adf4416178813c2752b42689ebb3bc95c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-be89c283c3be", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:37:16Z", "seed": 321339025, "treatment": "full"}, "priority": 2.0094513815772888, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "623f127bc806a40bda04085bb5bd2750f4cd3fbe2bf6df4593cba2ce3c57371a", "run_seconds": 1.7041995525360107, "seed": 321339025, "started_at": "2026-11-22T03:37:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.116251, 0.626804], "effect": 0.371528, "evaluated_at": "2026-11-22T03:38:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-be89c283c3be", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.708333333, 0.513888889, 0.336805556, 0.9375], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.993055556, 1.0, 0.996527778, 0.993055556], "retrieval": [0.909722222, 0.909722222, 0.833333333, 0.986111111], "rules_nodetect": [0.173611111, 0.395833333, 0.524305556, 0.388888889]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "b7f1df39cd9aa0f2dae0dd824d79ed7adf4416178813c2752b42689ebb3bc95c", "primary_control": "facts", "result_hash": "623f127bc806a40bda04085bb5bd2750f4cd3fbe2bf6df4593cba2ce3c57371a", "signature": "b8a1c765be09d5d819cab606ba69d6f90342cedeef4d9d48441fc8409de7835588e159f6f7146f8ec213c21dde3dfdb4758eb9d003e5955d051c6e6eb4217103", "state": "VERIFIED", "summary": {"facts": 0.624132, "frozen": 0.0, "full": 0.99566, "retrieval": 0.909722, "rules_nodetect": 0.37066}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:38:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:49Z", "cpu_seconds": 1.7565197130000003, "created_at": "2026-11-22T03:37:47Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-fa260300ce6e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "f142834156c8e7b612365d2277976f4c0aec1223a2dbe7faf4d78f07f5b28305", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-fa260300ce6e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:37:47Z", "seed": 633353582, "treatment": "frontier"}, "priority": 2.0086092327024887, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "283c042aa8ab8f95c8573d368bee8e55bac9a5ef2fe5a567c2b514f97b5a0c9d", "run_seconds": 2.0320119857788086, "seed": 633353582, "started_at": "2026-11-22T03:37:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.04177, 0.14496], "effect": 0.051595, "evaluated_at": "2026-11-22T03:38:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-fa260300ce6e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.903846154, 1.0, 0.974137931], "frontier": [0.847222222, 0.961538462, 1.0, 1.0], "goal": [0.694444444, 0.865384615, 1.0, 1.0], "hypdisc": [0.958333333, 0.903846154, 1.0, 1.0], "infogain": [0.847222222, 0.961538462, 1.0, 1.0], "prederr": [0.847222222, 0.961538462, 1.0, 1.0], "random": [0.902777778, 0.846153846, 1.0, 0.853448276], "risk": [0.847222222, 0.961538462, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f142834156c8e7b612365d2277976f4c0aec1223a2dbe7faf4d78f07f5b28305", "primary_control": "random", "result_hash": "283c042aa8ab8f95c8573d368bee8e55bac9a5ef2fe5a567c2b514f97b5a0c9d", "signature": "a719ad96b65d0515c6ac52e74af1bffc5d48b6cc39c047d0cd1dce0a442eaa702cddd95ebafd1b320ae035a01645f5d63a20e4fbae33d11230df82abb54e6e0f", "state": "VERIFIED", "summary": {"count": 0.969496, "frontier": 0.95219, "goal": 0.889957, "hypdisc": 0.965545, "infogain": 0.95219, "prederr": 0.95219, "random": 0.900595, "risk": 0.95219}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:38:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:48Z", "cpu_seconds": 0.681463159, "created_at": "2026-11-22T03:37:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-cdf4f9164161", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "f24a019061bcbdd4e278fc3903b364f6b03b8567a15413591ae983cc746bf7d1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-cdf4f9164161", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:37:47Z", "seed": 1213092200, "treatment": "full"}, "priority": 2.0078729023007122, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a2be518c64588134cef3e643346886539b26c1ada6eb4ad0e6fbaa4854cd34cf", "run_seconds": 0.7345948219299316, "seed": 1213092200, "started_at": "2026-11-22T03:37:47Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-cdf4f9164161", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.25, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 0.25, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f24a019061bcbdd4e278fc3903b364f6b03b8567a15413591ae983cc746bf7d1", "primary_control": "e141_best", "result_hash": "a2be518c64588134cef3e643346886539b26c1ada6eb4ad0e6fbaa4854cd34cf", "signature": "0b7d1bfa879d25cc8a805db7270ab8f3c7b11daa7b34fc697d9053a519d7e73b0eba4c1e193517158a253d21d52bb4a74b3fdc557a69bf3df6441670a5394a02", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 1.0, "random": 0.8125}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:38:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:37:54Z", "cpu_seconds": 0.687295977, "created_at": "2026-11-22T03:37:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-49d1c386fa6e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 9, "n": 4}, "prereg_hash": "2079d66fb7bd069cfabf2e3e80e98e46a37321f9b7ed63d52d4a4b66358f55e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-49d1c386fa6e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:37:47Z", "seed": 322016916, "treatment": "full"}, "priority": 2.0075873361700034, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "87346a2980ac7dac921bdc22475622f7ad5b572c97c77ba2ca9cd28be52f9565", "run_seconds": 0.6954751014709473, "seed": 322016916, "started_at": "2026-11-22T03:37:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.042668, 0.178322], "effect": 0.110495, "evaluated_at": "2026-11-22T03:38:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-49d1c386fa6e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.461666667, 0.5325, 0.6296875, 0.423958333], "frozen": [0.386666667, 0.384166667, 0.39375, 0.3125], "full": [0.556666667, 0.588333333, 0.709375, 0.635416667], "retrieval": [0.784166667, 0.805, 0.896875, 0.820833333], "rules_nodetect": [0.556666667, 0.588333333, 0.709375, 0.635416667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2079d66fb7bd069cfabf2e3e80e98e46a37321f9b7ed63d52d4a4b66358f55e2", "primary_control": "facts", "result_hash": "87346a2980ac7dac921bdc22475622f7ad5b572c97c77ba2ca9cd28be52f9565", "signature": "e9f639316cd93347b1a789d0bd99c2d87df7661ffa08fc2c1af5b930695f960922117186d541ca7d55fce00b1894d713935a40155de3e9b6b5c732fbdf74bf08", "state": "VERIFIED", "summary": {"facts": 0.511953, "frozen": 0.369271, "full": 0.622448, "retrieval": 0.826719, "rules_nodetect": 0.622448}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:38:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:12Z", "cpu_seconds": 5.324161791000001, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4eaf9398b910", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "2b14944947625f6b22b3aa0566f729eb43acd5b9152f9944e4b61240b217ced8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4eaf9398b910", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:38:02Z", "seed": 764635690, "treatment": "frontier"}, "priority": 2.008170747476011, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "50c4979228ba37da034a82a7b502ac162a8bb8c59c74774cb8badb74357d39c7", "run_seconds": 5.968979358673096, "seed": 764635690, "started_at": "2026-11-22T03:38:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.157565, 0.170761], "effect": 0.006598, "evaluated_at": "2026-11-22T03:38:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4eaf9398b910", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.418032787, 0.046938776, 0.465384615, 0.588118812], "frontier": [0.155737705, 0.648979592, 0.382692308, 0.558415842], "goal": [0.140983607, 0.304081633, 0.305769231, 0.162376238], "hypdisc": [0.281967213, 0.312244898, 0.276923077, 0.40990099], "infogain": [0.155737705, 0.657142857, 0.382692308, 0.562376238], "prederr": [0.155737705, 0.657142857, 0.382692308, 0.562376238], "random": [0.391803279, 0.502040816, 0.338461538, 0.487128713], "risk": [0.155737705, 0.665306122, 0.382692308, 0.558415842]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "2b14944947625f6b22b3aa0566f729eb43acd5b9152f9944e4b61240b217ced8", "primary_control": "random", "result_hash": "50c4979228ba37da034a82a7b502ac162a8bb8c59c74774cb8badb74357d39c7", "signature": "17bf1ebd78579c04cdae48f302b06510f7c00c48a777e81cd85eb0164f85871cbfe71a3391370d40bc34bab42350039683862142de76527c61b2871b4eb77b09", "state": "VERIFIED", "summary": {"count": 0.379619, "frontier": 0.436456, "goal": 0.228303, "hypdisc": 0.320259, "infogain": 0.439487, "prederr": 0.439487, "random": 0.429859, "risk": 0.440538}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:38:26Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:39:46Z", "cpu_seconds": 18.379552165, "created_at": "2026-11-22T03:38:02Z", "error": "lease expired; ", "family": "ow_explore", "fence": 2, "id": "E141-ow_explore-9cf421dbe986", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 8, "n": 4}, "prereg_hash": "48b9a1526492d2e0320d66d3b8bdacbffe8879824546160a61d1b68b0574afc7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9cf421dbe986", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:38:02Z", "seed": 1500439193, "treatment": "frontier"}, "priority": 2.006264815824821, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "61f2d5c333de23b3e78b2b5012e4611ef3a883b912fe223adb3b63cad160fdae", "run_seconds": 21.027230501174927, "seed": 1500439193, "started_at": "2026-11-22T03:38:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00011, 0.114713], "effect": 0.057301, "evaluated_at": "2026-11-22T03:40:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9cf421dbe986", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.976470588, 0.630508475, 0.913333333], "frontier": [0.96, 1.0, 0.747457627, 0.966666667], "goal": [0.973333333, 0.994117647, 0.608474576, 0.953333333], "hypdisc": [0.933333333, 0.994117647, 0.588135593, 0.933333333], "infogain": [0.96, 1.0, 0.747457627, 0.966666667], "prederr": [0.96, 1.0, 0.747457627, 0.966666667], "random": [0.973333333, 0.952941176, 0.618644068, 0.9], "risk": [0.96, 1.0, 0.747457627, 0.966666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "48b9a1526492d2e0320d66d3b8bdacbffe8879824546160a61d1b68b0574afc7", "primary_control": "random", "result_hash": "61f2d5c333de23b3e78b2b5012e4611ef3a883b912fe223adb3b63cad160fdae", "signature": "e098cd56634138225da1804329a36ced9a9a3283f73c6e97ec5fcdd83f981fcccf7941964df920fdc0300fa0ac8ff06c531feed576d66e453e8b6d2090e01a04", "state": "VERIFIED", "summary": {"count": 0.880078, "frontier": 0.918531, "goal": 0.882315, "hypdisc": 0.86223, "infogain": 0.918531, "prederr": 0.918531, "random": 0.86123, "risk": 0.918531}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:40:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:44Z", "cpu_seconds": 2.5852513659999943, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e35ef565205d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "2a7bf2d174a768dd9f4a5fb9dc5bd979f93a21243213ab7a61bf3d92bfbc2ba3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e35ef565205d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:38:02Z", "seed": 1887698045, "treatment": "frontier"}, "priority": 2.0046930753062315, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9c6233caecd1a033d4842881a5847f8475986b6c954ac59fd2ce2baa075c9cf4", "run_seconds": 3.0587897300720215, "seed": 1887698045, "started_at": "2026-11-22T03:39:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014653, -0.004444], "effect": -0.009549, "evaluated_at": "2026-11-22T03:40:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e35ef565205d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.996527778, 0.986111111, 0.986111111, 0.993055556], "goal": [0.993055556, 0.972222222, 0.986111111, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [0.996527778, 0.986111111, 0.986111111, 0.993055556], "prederr": [0.996527778, 0.986111111, 1.0, 0.993055556], "random": [1.0, 1.0, 1.0, 1.0], "risk": [0.996527778, 0.986111111, 0.986111111, 0.993055556]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2a7bf2d174a768dd9f4a5fb9dc5bd979f93a21243213ab7a61bf3d92bfbc2ba3", "primary_control": "random", "result_hash": "9c6233caecd1a033d4842881a5847f8475986b6c954ac59fd2ce2baa075c9cf4", "signature": "3defb9c52771619c17973ffaeddf9a70fdd3f65f3a3f838a51d077ff9e8669e3429d4a3b5819b949f3269bda3b21e51937608b30c50e66f946ea6d6e92d5fe0a", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.990451, "goal": 0.987847, "hypdisc": 1.0, "infogain": 0.990451, "prederr": 0.993924, "random": 1.0, "risk": 0.990451}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:40:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:59Z", "cpu_seconds": 2.4086994779999955, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-12d38824163c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "d1ab58a0480c96e2ad37e398e66ae19d26d7d07a515aac69981def3b43a6b22b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-12d38824163c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:38:02Z", "seed": 1509683047, "treatment": "full"}, "priority": 2.0041847735525247, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-f4539f4922a2", "result_hash": "c187ad5883cf191d504fc9ffb10f32887bd9761b3454ca99d661e2b150548edd", "run_seconds": 2.5736277103424072, "seed": 1509683047, "started_at": "2026-11-22T03:39:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016406, 0.055144], "effect": 0.019369, "evaluated_at": "2026-11-22T03:40:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-12d38824163c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.814814815, 0.714285714, 0.784313725, 0.912280702], "frozen": [0.240740741, 0.289115646, 0.284313725, 0.289473684], "full": [0.888888889, 0.717687075, 0.784313725, 0.912280702], "retrieval": [0.888888889, 0.714285714, 0.784313725, 0.912280702], "rules_nodetect": [0.888888889, 0.721088435, 0.784313725, 0.912280702]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "d1ab58a0480c96e2ad37e398e66ae19d26d7d07a515aac69981def3b43a6b22b", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-f4539f4922a2", "result_hash": "c187ad5883cf191d504fc9ffb10f32887bd9761b3454ca99d661e2b150548edd", "signature": "3ac7a7af705d7c8955d82693b64e810d92db8306b8cd6937c1384ae00d7e9f666775a7d9c8f58ed97181540bde0a6bbc0458aebb551b1152c2e3e058111faf0e", "state": "VERIFIED", "summary": {"facts": 0.806424, "frozen": 0.275911, "full": 0.825793, "retrieval": 0.824942, "rules_nodetect": 0.826643}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:40:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:13Z", "cpu_seconds": 1.050818735, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5763cf9bf46e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "97a4f88ab3079b65c5c770cf1c725abce83b3a3a5167e5f3359ab0d5f08f3797", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5763cf9bf46e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:38:02Z", "seed": 1480017231, "treatment": "full"}, "priority": 2.006475560638749, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5798f6b5c0e1e758c04a1e702ead150e03c93223eb86bf7937d3706b972914c9", "run_seconds": 1.157679796218872, "seed": 1480017231, "started_at": "2026-11-22T03:38:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.016864, 0.059465], "effect": 0.038165, "evaluated_at": "2026-11-22T03:38:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5763cf9bf46e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.25862069, 0.682352941, 0.724637681, 0.463157895], "frozen": [0.131034483, 0.141176471, 0.07826087, 0.084210526], "full": [0.282758621, 0.752941176, 0.753623188, 0.492105263], "retrieval": [0.362068966, 0.729411765, 0.750724638, 0.518421053], "rules_nodetect": [0.282758621, 0.752941176, 0.753623188, 0.492105263]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "97a4f88ab3079b65c5c770cf1c725abce83b3a3a5167e5f3359ab0d5f08f3797", "primary_control": "facts", "result_hash": "5798f6b5c0e1e758c04a1e702ead150e03c93223eb86bf7937d3706b972914c9", "signature": "8584a92b189fa32219a91f43dafa65fbc76ae4cf48abd1b38b4fc2fcbdecfad3180a2899b9e5568dc4b4fe242a31caa9767baa2a03ebb8c6dcf2d967bfe5fd02", "state": "VERIFIED", "summary": {"facts": 0.532192, "frozen": 0.108671, "full": 0.570357, "retrieval": 0.590157, "rules_nodetect": 0.570357}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:38:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:04Z", "cpu_seconds": 1.6151536210000002, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-65b40dc90f47", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "0ca56c750540b58794f282e034dc548dd3f13303f00860699c4ff10efd6bc783", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-65b40dc90f47", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:38:02Z", "seed": 859716009, "treatment": "full"}, "priority": 2.009475422396442, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e97e92edb189c532f49d5a60b0e0d721490156def2bc3a8042a388a855ddd3cf", "run_seconds": 1.697206735610962, "seed": 859716009, "started_at": "2026-11-22T03:38:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.198033, 0.222949], "effect": 0.210491, "evaluated_at": "2026-11-22T03:38:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-65b40dc90f47", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.092987805, 0.096518987, 0.096698113, 0.145833333], "frozen": [0.036585366, 0.011075949, 0.039308176, 0.033333333], "full": [0.304878049, 0.318037975, 0.312893082, 0.338194444], "retrieval": [0.31402439, 0.314873418, 0.296383648, 0.372222222], "rules_nodetect": [0.304878049, 0.318037975, 0.312893082, 0.338194444]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "0ca56c750540b58794f282e034dc548dd3f13303f00860699c4ff10efd6bc783", "primary_control": "facts", "result_hash": "e97e92edb189c532f49d5a60b0e0d721490156def2bc3a8042a388a855ddd3cf", "signature": "1f556e0872187a78f2c0891dc26596b1b7b646ad7ed5053128c20c3a91c007bbfe6daaa9caeeb1c2064cc8eddd100c839c15ee9772b643c7b9dd9ebecf1c930a", "state": "VERIFIED", "summary": {"facts": 0.10801, "frozen": 0.030076, "full": 0.318501, "retrieval": 0.324376, "rules_nodetect": 0.318501}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:38:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:17Z", "cpu_seconds": 0.6460786390000006, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-7e57e73b508e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "2a878b61e8a21ee81711be8b3135ae310c8af393f1e5c6d367a057865cd1f3bc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-7e57e73b508e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:38:02Z", "seed": 624724714, "treatment": "gated"}, "priority": 2.0052677838446216, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "42ef4d0e2ea342b65e8f1c0e6bfc44fc2e9bf594e8a0449c28e4858a24b8f24e", "run_seconds": 0.6462697982788086, "seed": 624724714, "started_at": "2026-11-22T03:38:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001818, 0.005606], "effect": 0.001894, "evaluated_at": "2026-11-22T03:38:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-7e57e73b508e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.441860465, 0.5, 0.375, 0.431818182], "gated": [0.441860465, 0.5, 0.375, 0.431818182], "naive": [0.441860465, 0.5, 0.375, 0.333333333], "scratch": [0.441860465, 0.5, 0.375, 0.424242424], "ungated": [0.465116279, 0.572916667, 0.28125, 0.431818182]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "2a878b61e8a21ee81711be8b3135ae310c8af393f1e5c6d367a057865cd1f3bc", "primary_control": "scratch", "result_hash": "42ef4d0e2ea342b65e8f1c0e6bfc44fc2e9bf594e8a0449c28e4858a24b8f24e", "signature": "8d12f26abce1a92374daf6ec71acd801fb8add26e7ae63cc1e0775c457c8973d556c4b853c12b4e4ecbcd0699b240d3dd2ea103ab9a1a605f93ff2177b1d9202", "state": "VERIFIED", "summary": {"compressed": 0.43717, "gated": 0.43717, "naive": 0.412548, "scratch": 0.435276, "ungated": 0.437775}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:38:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:07Z", "cpu_seconds": 1.1918968129999996, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e5abbec4198a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "5c86d2efd7373cc381dcf2c9fdd5afa41189da3ff0a25baa7ef87b4156904171", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e5abbec4198a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:38:02Z", "seed": 2121307021, "treatment": "gated"}, "priority": 2.009053169757781, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7cba96f618d39898cd9f3e56ac6b4fa9d8169cc7cda849125816e750e4d2dc0d", "run_seconds": 1.6173102855682373, "seed": 2121307021, "started_at": "2026-11-22T03:38:05Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-e5abbec4198a", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.230769231, 0.168421053, 0.207058824, 0.096470588], "gated": [0.230769231, 0.168421053, 0.207058824, 0.096470588], "naive": [0.233333333, 0.136842105, 0.145882353, 0.068235294], "scratch": [0.230769231, 0.168421053, 0.207058824, 0.096470588], "ungated": [0.230769231, 0.131578947, 0.145882353, 0.117647059]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "5c86d2efd7373cc381dcf2c9fdd5afa41189da3ff0a25baa7ef87b4156904171", "primary_control": "scratch", "result_hash": "7cba96f618d39898cd9f3e56ac6b4fa9d8169cc7cda849125816e750e4d2dc0d", "signature": "42e0025af1930b46c100ec19c6b1d153f32dea824f720c34190255073fbbae4fa2d61ff1408b62c0f8706aa80d9562b0ea8d307afccb891398d2f645530a1c0a", "state": "PARTIAL", "summary": {"compressed": 0.17568, "gated": 0.17568, "naive": 0.146073, "scratch": 0.17568, "ungated": 0.156469}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:38:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:13Z", "cpu_seconds": 6.767509211999999, "created_at": "2026-11-22T03:38:02Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-66f6de1f9ec7", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "b417e95fbbb485a35a78c2cea43402dabdf894e2013a5d7a5de23bb5ed1b3812", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-66f6de1f9ec7", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:38:02Z", "seed": 421089676, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "5e2f47069a683fd9910510b04baaca18a6a91ec5709a6b12d16d6dd24076e8bd", "run_seconds": 7.5797741413116455, "seed": 421089676, "started_at": "2026-11-22T03:38:05Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-66f6de1f9ec7", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0], "root": [1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b417e95fbbb485a35a78c2cea43402dabdf894e2013a5d7a5de23bb5ed1b3812", "primary_control": "reset", "result_hash": "5e2f47069a683fd9910510b04baaca18a6a91ec5709a6b12d16d6dd24076e8bd", "signature": "8472e599ec366e3e0fb066ce67157b175e0a13e636fdcd54c02879352e9920ee8ef98fed79fac9d49995aa15340a4208a1d6a77a2b84fdd7863bbe8107ea8d09", "state": "PARTIAL", "summary": {"inherit": 0.7166666666666667, "reset": 0.7166666666666667, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:38:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:30Z", "cpu_seconds": 1.637481045, "created_at": "2026-11-22T03:38:18Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-6143f6c4fa92", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "1db225b5633d7e11c379decca51b40d4d19f91884ba8cfc460adc7e7a97b230b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-6143f6c4fa92", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:38:18Z", "seed": 1697318628, "treatment": "iw_temperature"}, "priority": 2.0080996217977902, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-dd183dae9f61", "result_hash": "12d421db5cf4c7c444401d9ae8f57ae94b8ac01401ad7995ac4fb2eefae7997d", "run_seconds": 2.0521175861358643, "seed": 1697318628, "started_at": "2026-11-22T03:38:27Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.033248, 0.002791], "effect": -0.015228, "evaluated_at": "2026-11-22T03:38:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-6143f6c4fa92", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.309408198, 0.27097289, 0.293913272, 0.234670672, 0.20067104, 0.28833532, 0.303171545, 0.31421454, 0.251129475, 0.2819962, 0.33243242, 0.25159056, 0.18419192, 0.242842225, 0.265303482, 0.236339118], "raw": [0.293243908, 0.260556447, 0.221061135, 0.234670672, 0.225009678, 0.229582778, 0.18825453, 0.283801328, 0.273041865, 0.268594285, 0.283262497, 0.251013645, 0.160403587, 0.214203705, 0.211052472, 0.252142667], "temperature": [0.309408198, 0.227010818, 0.262299085, 0.27263959, 0.251535157, 0.237329625, 0.235325235, 0.30311637, 0.251129475, 0.281433465, 0.319221415, 0.253354258, 0.160403587, 0.27387057, 0.187851397, 0.191598905]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0126, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.971", "prereg_hash": "1db225b5633d7e11c379decca51b40d4d19f91884ba8cfc460adc7e7a97b230b", "primary_control": "temperature", "replicated": false, "replication_of": "E141-calibration_iw-dd183dae9f61", "result_hash": "12d421db5cf4c7c444401d9ae8f57ae94b8ac01401ad7995ac4fb2eefae7997d", "signature": "36c12dbfd4b5a2171bdbf39085fc8a902c6c5bce0e1d1e166505ee033f651b32c60e4c629cbffeaa996202822f96cf332bad40e7ed8f4cd55c8768e7b1eb5805", "state": "VERIFIED", "summary": {"iw_temperature": 0.266324, "raw": 0.240618, "temperature": 0.251095}, "ties": 2, "treatment": "iw_temperature", "wins": 4}, "verified_at": "2026-11-22T03:38:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:53Z", "cpu_seconds": 12.864827505000001, "created_at": "2026-11-22T03:38:18Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-73e69bf2a327", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "81b89c36e3e3519a4fa3f505a0824755a53126fd89ba5bae0aa93ea5ba6335db", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-73e69bf2a327", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:38:18Z", "seed": 1131583184, "treatment": "frontier"}, "priority": 2.0067603266490286, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-549d67c50a59", "result_hash": "3849a4b97c09984c9cd9fe90312527b0524d554b16ecff8e17b2c9e18ffe6435", "run_seconds": 18.180306434631348, "seed": 1131583184, "started_at": "2026-11-22T03:38:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185086, 0.127793], "effect": -0.028646, "evaluated_at": "2026-11-22T03:38:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-73e69bf2a327", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.739316239, 0.252314815, 0.924242424, 0.425795053], "frontier": [0.696581197, 0.62037037, 0.651515152, 0.409893993], "goal": [0.702991453, 0.186921296, 0.803030303, 0.393992933], "hypdisc": [0.752136752, 0.248263889, 0.848484848, 0.314487633], "infogain": [0.696581197, 0.307291667, 0.651515152, 0.438751472], "prederr": [0.696581197, 0.288773148, 0.651515152, 0.438751472], "random": [0.668803419, 0.508101852, 0.909090909, 0.406949352], "risk": [0.696581197, 0.62037037, 0.651515152, 0.409893993]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "81b89c36e3e3519a4fa3f505a0824755a53126fd89ba5bae0aa93ea5ba6335db", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-549d67c50a59", "result_hash": "3849a4b97c09984c9cd9fe90312527b0524d554b16ecff8e17b2c9e18ffe6435", "signature": "cae03169add78fc92991ae62822a014d1c42fcfb387ecf01e9e9a3b1a6bb33ec766a6baa8391cb9c97fae3d73fae181018d52fd04e9cfa6d647dafedc19fce0d", "state": "VERIFIED", "summary": {"count": 0.585417, "frontier": 0.59459, "goal": 0.521734, "hypdisc": 0.540843, "infogain": 0.523535, "prederr": 0.518905, "random": 0.623236, "risk": 0.59459}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:39:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:54Z", "cpu_seconds": 1.3589837630000003, "created_at": "2026-11-22T03:38:18Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b97afc1a4af7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "404d7a4aa9f2d21e443ddbe90c6e4b3b290bee3016aa5c01abc135ce9cf8896d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b97afc1a4af7", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:38:18Z", "seed": 1092354803, "treatment": "full"}, "priority": 2.00665836909334, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7e36bb67458da2ee15462c95fcc65f22a6b2608f7a3567ca66c3ec2c88547787", "run_seconds": 1.4806921482086182, "seed": 1092354803, "started_at": "2026-11-22T03:38:52Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:39:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b97afc1a4af7", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 0.5, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "404d7a4aa9f2d21e443ddbe90c6e4b3b290bee3016aa5c01abc135ce9cf8896d", "primary_control": "e141_best", "result_hash": "7e36bb67458da2ee15462c95fcc65f22a6b2608f7a3567ca66c3ec2c88547787", "signature": "1d95d46db7df9e5b03c5cc7bea6bd6b9215d8fe438b66207d5b572c2188ef8a1ac2a5ea0b45dfdef9d302775f692e4054079e90fbadda267e15b7961b71e5d02", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:39:00Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:39:32Z", "cpu_seconds": 0.16935058299999994, "created_at": "2026-11-22T03:38:18Z", "error": "lease expired; ", "family": "ow_model", "fence": 2, "id": "E141-ow_model-9b95ca3108a1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "13e31511e46e206f3b9925037603a5a7e5e7ac4ae5272d25e620235e09ab22d6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9b95ca3108a1", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:38:18Z", "seed": 1284030189, "treatment": "full"}, "priority": 2.0096260611826926, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9d369dcf6a83f7fc422a41d71e75b5defd9be5fa5603113bcf3a2a16b131d044", "run_seconds": 0.2447648048400879, "seed": 1284030189, "started_at": "2026-11-22T03:38:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00214, 0.019185], "effect": 0.008523, "evaluated_at": "2026-11-22T03:40:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9b95ca3108a1", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.909090909, 0.988636364, 1.0, 1.0], "frozen": [0.034090909, 0.034090909, 0.034090909, 0.011363636], "full": [0.931818182, 1.0, 1.0, 1.0], "retrieval": [0.920454545, 0.977272727, 1.0, 1.0], "rules_nodetect": [0.931818182, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "13e31511e46e206f3b9925037603a5a7e5e7ac4ae5272d25e620235e09ab22d6", "primary_control": "facts", "result_hash": "9d369dcf6a83f7fc422a41d71e75b5defd9be5fa5603113bcf3a2a16b131d044", "signature": "69666a88cae365588e402823cea96955be32f5bf07622d351ed2d705c2e539074bbbd6d7d1b8eec5c5c83e12e0b5619505a0d89b6a39d498e2644a5120841b01", "state": "VERIFIED", "summary": {"facts": 0.974432, "frozen": 0.028409, "full": 0.982955, "retrieval": 0.974432, "rules_nodetect": 0.982955}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:40:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:28Z", "cpu_seconds": 0.596652478, "created_at": "2026-11-22T03:38:18Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-85a9bb504cc1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "prereg_hash": "59cc99f21f8cd20e58d904799ef439e6179c8d7c42c280c2be44cd0b65fb29ed", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-85a9bb504cc1", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:38:18Z", "seed": 178286404, "treatment": "gated"}, "priority": 2.0089474695085827, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "8cfbe35696d8bc9813c365111b5b35ccce3129543a2c78e54ad9a00839489456", "run_seconds": 0.7236452102661133, "seed": 178286404, "started_at": "2026-11-22T03:38:27Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-85a9bb504cc1", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [1.0, 0.917989418, 0.924731183, 1.0], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.924731183, 0.917989418, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "59cc99f21f8cd20e58d904799ef439e6179c8d7c42c280c2be44cd0b65fb29ed", "primary_control": "scratch", "result_hash": "8cfbe35696d8bc9813c365111b5b35ccce3129543a2c78e54ad9a00839489456", "signature": "01850c5850a23dd21e2f88a865c04e486c90e5898574539793535b05643cceefbad3ff1b55e9308b1d681acd7477573ff1ab686e115b555a9e017dbd35639e06", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.96068, "scratch": 1.0, "ungated": 0.96068}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:38:35Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:40:01Z", "cpu_seconds": 9.447795603000003, "created_at": "2026-11-22T03:38:18Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-cfe7713ae446", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "95357eaf68371b657529033cb9f09a11bfc68848bc8f57f7a58ef8f3d503aa80", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-cfe7713ae446", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:38:18Z", "seed": 140559219, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "350c57f4117620b0a3ce01c759434bc0fa2eab928aa64e5fc0f50f2e95019555", "run_seconds": 10.831275939941406, "seed": 140559219, "started_at": "2026-11-22T03:38:22Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:40:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-cfe7713ae446", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4680355741070644, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "95357eaf68371b657529033cb9f09a11bfc68848bc8f57f7a58ef8f3d503aa80", "primary_control": "reset", "result_hash": "350c57f4117620b0a3ce01c759434bc0fa2eab928aa64e5fc0f50f2e95019555", "signature": "7013158136da517e9bfee67cbf599f0705bd157f1750c3837783745b4caa6da354559988806163164f4e022f93ce7b5edb9dde918f85ec19f31a8c9ae4582e08", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.3}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:40:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:38Z", "cpu_seconds": 0.5151363270000004, "created_at": "2026-11-22T03:38:35Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b4217f65de1f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "a64d82c27d994b1695cc385980c428d19f75d21ebbc594a342868d489265b090", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b4217f65de1f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:38:35Z", "seed": 1781039303, "treatment": "frontier"}, "priority": 2.0048650000489827, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "270f5750977294ab5eeef290649c56f97083eb143b4974d8fe3db7204ca813ae", "run_seconds": 0.6191725730895996, "seed": 1781039303, "started_at": "2026-11-22T03:39:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.101775, 0.30348], "effect": 0.100852, "evaluated_at": "2026-11-22T03:40:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b4217f65de1f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.463636364, 1.0, 0.945454545, 0.977272727], "frontier": [0.181818182, 1.0, 0.972727273, 1.0], "goal": [0.390909091, 1.0, 0.65, 0.954545455], "hypdisc": [0.381818182, 1.0, 0.686363636, 0.664772727], "infogain": [0.418181818, 1.0, 0.972727273, 1.0], "prederr": [0.409090909, 1.0, 0.972727273, 1.0], "random": [0.318181818, 1.0, 0.654545455, 0.778409091], "risk": [0.181818182, 1.0, 0.972727273, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a64d82c27d994b1695cc385980c428d19f75d21ebbc594a342868d489265b090", "primary_control": "random", "result_hash": "270f5750977294ab5eeef290649c56f97083eb143b4974d8fe3db7204ca813ae", "signature": "558aee92db95d1b636e9e8d6d75b21ae032e43d017e26bbe3450bbf18b24939c9fa1e7a55ab081dd5bfb9f178be80484c6848913b27dd328eb274876ea7be309", "state": "VERIFIED", "summary": {"count": 0.846591, "frontier": 0.788636, "goal": 0.748864, "hypdisc": 0.683239, "infogain": 0.847727, "prederr": 0.845455, "random": 0.687784, "risk": 0.788636}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:40:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:15Z", "cpu_seconds": 6.767145466000002, "created_at": "2026-11-22T03:38:35Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c7d41448397d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "d1c08518f98833419ba8e627dbc95e5c714dd2827bea0639793d6c70bd6dc1ff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c7d41448397d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:38:35Z", "seed": 679279670, "treatment": "frontier"}, "priority": 2.005885108236318, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "dfaafe66286508adaf41b145e542f19f275cf8fe57c8ba7e5f3720abc9d16676", "run_seconds": 7.393033504486084, "seed": 679279670, "started_at": "2026-11-22T03:39:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.071289, 0.312676], "effect": 0.191982, "evaluated_at": "2026-11-22T03:40:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c7d41448397d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.806060606, 0.787878788, 0.77979798, 1.0], "frontier": [0.816161616, 0.824242424, 0.82020202, 0.975], "goal": [0.482828283, 0.503030303, 0.565656566, 0.958333333], "hypdisc": [0.521212121, 0.509090909, 0.488888889, 0.983333333], "infogain": [0.816161616, 0.824242424, 0.82020202, 0.975], "prederr": [0.816161616, 0.824242424, 0.82020202, 0.975], "random": [0.553535354, 0.58989899, 0.557575758, 0.966666667], "risk": [0.816161616, 0.824242424, 0.82020202, 0.975]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "d1c08518f98833419ba8e627dbc95e5c714dd2827bea0639793d6c70bd6dc1ff", "primary_control": "random", "result_hash": "dfaafe66286508adaf41b145e542f19f275cf8fe57c8ba7e5f3720abc9d16676", "signature": "1c17942af1b745baf0a5083dffc0dff39676d19fc69a14daf029ce1295b2e67bafe20580a9ef5ef0611581946ee19f955bb865881afa0288db17cff6c24ef20d", "state": "VERIFIED", "summary": {"count": 0.843434, "frontier": 0.858902, "goal": 0.627462, "hypdisc": 0.625631, "infogain": 0.858902, "prederr": 0.858902, "random": 0.666919, "risk": 0.858902}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:40:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:05Z", "cpu_seconds": 0.2684627640000006, "created_at": "2026-11-22T03:38:35Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d7430a92b76e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "3c64debbd106159e9049f5fd9bb391a3d26a1fd0454dcec2d0bca22c327a280e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d7430a92b76e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:38:35Z", "seed": 634756407, "treatment": "full"}, "priority": 2.006354707690123, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "972fd824f46f07ba35838e405950f96e87dd7d53c2ea496a1b0a3de6619be503", "run_seconds": 0.29502224922180176, "seed": 634756407, "started_at": "2026-11-22T03:39:05Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:39:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d7430a92b76e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3c64debbd106159e9049f5fd9bb391a3d26a1fd0454dcec2d0bca22c327a280e", "primary_control": "e141_best", "result_hash": "972fd824f46f07ba35838e405950f96e87dd7d53c2ea496a1b0a3de6619be503", "signature": "044f63cd9f75df82d9ba4998404516c1921a50df6104d82aadde89bf7753dbc988382e28f8ceb48f3b94d0645038ed8e7296a6edc8350a2178c7c85fca6a9f05", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:39:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:17Z", "cpu_seconds": 1.2076659540000003, "created_at": "2026-11-22T03:38:35Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e6764134b81d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "425e513498864155688af6357b67009a87e2992e42069d0ca4127f1b17fbb3ca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e6764134b81d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:38:35Z", "seed": 229858198, "treatment": "full"}, "priority": 2.0011504427401063, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2286ee93e257c261fde0716870f0a28c85957a6a760e14e6e5f51c06904b78fc", "run_seconds": 1.2367477416992188, "seed": 229858198, "started_at": "2026-11-22T03:43:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.24, 0.74], "effect": 0.25, "evaluated_at": "2026-11-22T03:43:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e6764134b81d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.0, 1.0], "fixed_policy": [0.75, 0.5, 0.75, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "425e513498864155688af6357b67009a87e2992e42069d0ca4127f1b17fbb3ca", "primary_control": "e141_best", "result_hash": "2286ee93e257c261fde0716870f0a28c85957a6a760e14e6e5f51c06904b78fc", "signature": "ca63f91c623573ceca507f0097a0c039e4ce488ce49de3ce4dd7aca510ccb5c52e17638a2f0212072a46e235cb8af44401576485acd73110fcfffb7f4831610b", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:43:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:59Z", "cpu_seconds": 0.6749457489999955, "created_at": "2026-11-22T03:38:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-199a92b9d5b6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "5bb7b2940c62c8924cc95f68f4cf98f87471ed135d62f4635f3f416c0c00311f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-199a92b9d5b6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:38:35Z", "seed": 701828910, "treatment": "full"}, "priority": 2.003010733250623, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "15167e75b6c0b7e4753b7ccb911299156ab7b079e5bc2b50a2d9be748e8cf226", "run_seconds": 0.7975153923034668, "seed": 701828910, "started_at": "2026-11-22T03:39:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.045552, 0.156165], "effect": 0.100858, "evaluated_at": "2026-11-22T03:40:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-199a92b9d5b6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.557480315, 0.560784314, 0.543529412, 0.523921569], "frozen": [0.418110236, 0.390588235, 0.381176471, 0.381176471], "full": [0.633070866, 0.6, 0.661960784, 0.694117647], "retrieval": [0.848031496, 0.84627451, 0.838431373, 0.834509804], "rules_nodetect": [0.633070866, 0.6, 0.661960784, 0.694117647]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "5bb7b2940c62c8924cc95f68f4cf98f87471ed135d62f4635f3f416c0c00311f", "primary_control": "facts", "result_hash": "15167e75b6c0b7e4753b7ccb911299156ab7b079e5bc2b50a2d9be748e8cf226", "signature": "9cfbaa0928fe7a5e9a8d12f41048d9f43746c4473eb649bc42da0616ebe551194010cb3bdc79a61d163be17a0cb7b088d3ba382ec1f1b2575c4716d2405e8704", "state": "VERIFIED", "summary": {"facts": 0.546429, "frozen": 0.392763, "full": 0.647287, "retrieval": 0.841812, "rules_nodetect": 0.647287}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:40:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:47Z", "cpu_seconds": 1.70988849, "created_at": "2026-11-22T03:38:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-aabf5555895a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 3, "n": 4}, "prereg_hash": "cd3e56935c6876ad5c15ada07403fc8214358b0eb10434e537e84736ac129646", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-aabf5555895a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:38:35Z", "seed": 1706491698, "treatment": "full"}, "priority": 2.009379697271175, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fd73c20de5ba358c26681a71976580fbbd549df7255bdeef7e361a95210f54ff", "run_seconds": 2.579739570617676, "seed": 1706491698, "started_at": "2026-11-22T03:38:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006389, 0.101627], "effect": 0.047619, "evaluated_at": "2026-11-22T03:38:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-aabf5555895a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.909297052, 0.900226757, 1.0], "frozen": [0.0, 0.0, 0.0, 0.142857143], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 0.970521542, 0.977324263, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "cd3e56935c6876ad5c15ada07403fc8214358b0eb10434e537e84736ac129646", "primary_control": "facts", "result_hash": "fd73c20de5ba358c26681a71976580fbbd549df7255bdeef7e361a95210f54ff", "signature": "a278222cb3821caeb43908d2367e06ee01b103673f2e012392757b963f05dbd41930b46215f6c6af77b8ff4fc96939d67fe6bb1063976253bb2c593d1f03f70a", "state": "VERIFIED", "summary": {"facts": 0.952381, "frozen": 0.035714, "full": 1.0, "retrieval": 0.986961, "rules_nodetect": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:38:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:38:50Z", "cpu_seconds": 8.269215952, "created_at": "2026-11-22T03:38:35Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-058936272812", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "0ae278e93e0016dea9124e5b3df0a0d31aa11daeef33848378df83bfe7c2e56e", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-058936272812", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:38:35Z", "seed": 1151883082, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "fc9645a0201b1c599500273295bf480344a1818d9033270ce3363411e1e74c65", "run_seconds": 11.685781002044678, "seed": 1151883082, "started_at": "2026-11-22T03:38:38Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:38:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-058936272812", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0ae278e93e0016dea9124e5b3df0a0d31aa11daeef33848378df83bfe7c2e56e", "primary_control": "reset", "result_hash": "fc9645a0201b1c599500273295bf480344a1818d9033270ce3363411e1e74c65", "signature": "0e43467412909a769c8b3fb64c7d93954e30cf4388638ddb9f1ecbbcfeac99434288cd0cd0c230947408daa1d7876438dd6c1ca92bb2e6c035deae16f81a110e", "state": "PARTIAL", "summary": {"inherit": 0.6333333333333333, "reset": 0.6333333333333333, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:38:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:03Z", "cpu_seconds": 3.924584241, "created_at": "2026-11-22T03:38:53Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-079b72650971", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "prereg_hash": "658742116605946868997069f076de0aedd3dd4f07af91378e47e32ef248a6e9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-079b72650971", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:38:53Z", "seed": 2102275557, "treatment": "gated"}, "priority": 2.007595753298418, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "a91f816f28d9155f6aece81d4b25e8319be747ea9cbacc4fe09c0725aa92d360", "run_seconds": 4.777888059616089, "seed": 2102275557, "started_at": "2026-11-22T03:38:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007579, 0.023368], "effect": 0.007895, "evaluated_at": "2026-11-22T03:39:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-079b72650971", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.445783133, 0.151111111, 0.326315789, 0.733333333], "gated": [0.445783133, 0.151111111, 0.357894737, 0.733333333], "naive": [0.33253012, 0.1, 0.294736842, 0.7], "scratch": [0.445783133, 0.151111111, 0.326315789, 0.733333333], "ungated": [0.33253012, 0.126666667, 0.294736842, 0.691666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "658742116605946868997069f076de0aedd3dd4f07af91378e47e32ef248a6e9", "primary_control": "scratch", "result_hash": "a91f816f28d9155f6aece81d4b25e8319be747ea9cbacc4fe09c0725aa92d360", "signature": "27d780a3aa87844420fdcf196ebbfb0d0041315fda8493a998e45342a7817f835a77cc6a82c52d3ddff95f226a4a677c843ce2d26cb275317c3e8f4d6d8d8c08", "state": "VERIFIED", "summary": {"compressed": 0.414136, "gated": 0.422031, "naive": 0.356817, "scratch": 0.414136, "ungated": 0.3614}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:39:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:00Z", "cpu_seconds": 0.3474266549999996, "created_at": "2026-11-22T03:38:53Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-728e23d87813", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "prereg_hash": "e8c2bdc68f0925263165cc0ee54efde1be8456a9c8479baf03c5c670cebe2765", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-728e23d87813", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 1.0, "n": 12, "size": 9, "teleporters": 0, "wall_density": 0.15}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:38:53Z", "seed": 1187295191, "treatment": "frontier"}, "priority": 2.007111982756252, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-e169bdfc5bb3", "result_hash": "1eaaf9b9f5af95f4d73fb0191b509a02db32093597377f9692d87769ba284f51", "run_seconds": 0.6097445487976074, "seed": 1187295191, "started_at": "2026-11-22T03:38:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.073522, 0.111316], "effect": 0.092419, "evaluated_at": "2026-11-22T03:39:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-728e23d87813", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.958333333, 0.875, 0.899253731, 0.895522388, 0.87890625, 0.940140845, 0.919117647, 0.922794118, 0.888059701, 0.908088235, 0.924242424, 0.920289855], "frontier": [0.986111111, 0.988636364, 0.996268657, 0.992537313, 1.0, 0.992957746, 1.0, 0.981617647, 1.0, 0.992647059, 0.988636364, 0.996376812], "random_walk": [0.940972222, 0.837121212, 0.888059701, 0.921641791, 0.859375, 0.933098592, 0.900735294, 0.930147059, 0.884328358, 0.908088235, 0.890151515, 0.913043478]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "e8c2bdc68f0925263165cc0ee54efde1be8456a9c8479baf03c5c670cebe2765", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-e169bdfc5bb3", "result_hash": "1eaaf9b9f5af95f4d73fb0191b509a02db32093597377f9692d87769ba284f51", "signature": "083f61fd136c74fccb8ad564a30d1a6df7e0ca2ae6a20882056591535e28174dcc95b021c0573fb0bf92c22cd14f7edb3607cb2f86e797433b9b259fecce210e", "state": "VERIFIED", "summary": {"curiosity": 0.910812, "frontier": 0.992982, "random_walk": 0.900564}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:39:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:54Z", "cpu_seconds": 0.8662782579999977, "created_at": "2026-11-22T03:39:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9ddd6186d903", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "d363962995b6f89ed6f3e0f39eb9cfe86d1492b4428cb481114cf7068c344511", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9ddd6186d903", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:39:09Z", "seed": 1756363803, "treatment": "frontier"}, "priority": 2.0044592149978606, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a385b5ba872f6b13ee48208a65fad22a30d3bba94b78ba970eb7c113a99d3a63", "run_seconds": 1.4853217601776123, "seed": 1756363803, "started_at": "2026-11-22T03:39:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.0285, 0.06525], "effect": 0.046875, "evaluated_at": "2026-11-22T03:40:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9ddd6186d903", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.9875, 1.0, 0.9875, 1.0], "frontier": [1.0, 1.0, 0.9875, 0.9875], "goal": [0.9375, 0.925, 0.85, 0.9875], "hypdisc": [0.9875, 0.9875, 0.95, 1.0], "infogain": [1.0, 1.0, 0.9875, 0.9875], "prederr": [1.0, 1.0, 0.9875, 0.9875], "random": [0.9375, 0.9625, 0.925, 0.9625], "risk": [1.0, 1.0, 0.9875, 0.9875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "d363962995b6f89ed6f3e0f39eb9cfe86d1492b4428cb481114cf7068c344511", "primary_control": "random", "result_hash": "a385b5ba872f6b13ee48208a65fad22a30d3bba94b78ba970eb7c113a99d3a63", "signature": "981ae8bef59a923ae774a44916ba02cb5e55142205154d71c1c8b6f079c2577cb9cecd33aa3f46ca8f614d27a8930e713bf9f96fd18ba1f7d9fd4e41bbe41f0d", "state": "VERIFIED", "summary": {"count": 0.99375, "frontier": 0.99375, "goal": 0.925, "hypdisc": 0.98125, "infogain": 0.99375, "prederr": 0.99375, "random": 0.946875, "risk": 0.99375}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:40:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:18Z", "cpu_seconds": 4.110534280000003, "created_at": "2026-11-22T03:39:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9e4f52a00efd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 4, "n": 4}, "prereg_hash": "7dfee651df3d370998067351c7105ccb2621995fab46a7b8b660f714ca04ce14", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9e4f52a00efd", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:39:09Z", "seed": 1723430991, "treatment": "frontier"}, "priority": 2.0096300430643175, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "aa0a1c56613f1ef785da3e0bbbd828ea4914724ebc6767964bcbb167d6e1c5d0", "run_seconds": 4.565491437911987, "seed": 1723430991, "started_at": "2026-11-22T03:39:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016996, 0.270107], "effect": 0.126556, "evaluated_at": "2026-11-22T03:40:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9e4f52a00efd", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.651111111, 1.0, 0.84], "frontier": [0.983333333, 0.788888889, 1.0, 0.84], "goal": [0.966666667, 0.415555556, 0.95, 0.84], "hypdisc": [0.966666667, 0.442222222, 0.991666667, 0.558], "infogain": [0.983333333, 0.788888889, 1.0, 0.84], "prederr": [0.983333333, 0.788888889, 1.0, 0.84], "random": [0.991666667, 0.526666667, 0.991666667, 0.596], "risk": [0.983333333, 0.788888889, 1.0, 0.84]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "7dfee651df3d370998067351c7105ccb2621995fab46a7b8b660f714ca04ce14", "primary_control": "random", "result_hash": "aa0a1c56613f1ef785da3e0bbbd828ea4914724ebc6767964bcbb167d6e1c5d0", "signature": "842bb316a385e4ef35dc3c137830153ef4c092a538d153924e30657b52686bf486cee6d79881375d20b14927b98c813e4d8428489d00f554c13d2828af1a360c", "state": "VERIFIED", "summary": {"count": 0.872778, "frontier": 0.903056, "goal": 0.793056, "hypdisc": 0.739639, "infogain": 0.903056, "prederr": 0.903056, "random": 0.7765, "risk": 0.903056}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:40:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:21Z", "cpu_seconds": 0.8487456979999983, "created_at": "2026-11-22T03:39:09Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-68bc326fe53d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "fc14f36dc07fc54d6c7068050bad2a487dce7785882d92650c0082a6cc8f4dfb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-68bc326fe53d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:39:09Z", "seed": 1520722351, "treatment": "full"}, "priority": 2.00809508658609, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "99dbf2bcea8357a94d75b8885058e918550292ca7f6a591debb788123c069716", "run_seconds": 0.8554604053497314, "seed": 1520722351, "started_at": "2026-11-22T03:39:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.370785, 0.245785], "effect": -0.0625, "evaluated_at": "2026-11-22T03:40:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-68bc326fe53d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 0.0, 1.0], "random": [0.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "fc14f36dc07fc54d6c7068050bad2a487dce7785882d92650c0082a6cc8f4dfb", "primary_control": "e141_best", "result_hash": "99dbf2bcea8357a94d75b8885058e918550292ca7f6a591debb788123c069716", "signature": "8c8789b2d5d3ec7d0c47a045ca4812a3e64d0174a7d6289175c92200d3c5e9f635a1b9b425662efcb67a0fcecc9aab094332a8fa76fa717f47c988e13373b20e", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.0625, "full": 0.75, "no_world_model": 0.5, "random": 0.5}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:40:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:23Z", "cpu_seconds": 0.5940251330000024, "created_at": "2026-11-22T03:39:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e5b6d35cfb25", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "84479e501d3a0a6121ea7984a3ad722433aecd91d9c449fd2568ee1ab1446767", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e5b6d35cfb25", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:39:09Z", "seed": 1738158951, "treatment": "full"}, "priority": 2.0079160126851616, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "41354048f2d2cee8cf8c9f298b99fff8a57a49fb18e54340479276f94c275989", "run_seconds": 0.5959997177124023, "seed": 1738158951, "started_at": "2026-11-22T03:39:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.036565, 0.157946], "effect": 0.06069, "evaluated_at": "2026-11-22T03:40:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e5b6d35cfb25", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.566666667, 0.916666667, 0.398305085, 0.78125], "frozen": [0.333333333, 0.5, 0.251412429, 0.34375], "full": [0.65, 0.916666667, 0.588983051, 0.75], "retrieval": [0.7, 0.916666667, 0.624293785, 0.75], "rules_nodetect": [0.65, 0.916666667, 0.728813559, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "84479e501d3a0a6121ea7984a3ad722433aecd91d9c449fd2568ee1ab1446767", "primary_control": "facts", "result_hash": "41354048f2d2cee8cf8c9f298b99fff8a57a49fb18e54340479276f94c275989", "signature": "b6da1d963c940c223363c2e287f6ee38baad50766568e07cc6cd7175f3706e5f039fd90f8d3504dd56c5b243322a455a99f43ad725205aa795db40fd029a8304", "state": "VERIFIED", "summary": {"facts": 0.665722, "frozen": 0.357124, "full": 0.726412, "retrieval": 0.74774, "rules_nodetect": 0.76137}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:40:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:20Z", "cpu_seconds": 8.647739323, "created_at": "2026-11-22T03:39:09Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-b72eadff2165", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "222f2a9a80b3b2677b12d42de6341e25ca66c440dfe5ebb3825bfd4fd6ad5d02", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-b72eadff2165", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:39:09Z", "seed": 405662485, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "7e13ff89f9cb3000f2a203b9429050e6025298c01ee50a8e6101e0182c0a0b43", "run_seconds": 9.657337188720703, "seed": 405662485, "started_at": "2026-11-22T03:39:10Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:40:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-b72eadff2165", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.44595151374396713, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0], "reset": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "222f2a9a80b3b2677b12d42de6341e25ca66c440dfe5ebb3825bfd4fd6ad5d02", "primary_control": "reset", "result_hash": "7e13ff89f9cb3000f2a203b9429050e6025298c01ee50a8e6101e0182c0a0b43", "signature": "912524e78fe0d8cb531c040e48f600deba601704a9d611dd56a8f964cf4a60488d49e402fc1f9b11a7ad1548088588eda00f16139b5d3b2b51c96bb52e88750a", "state": "PARTIAL", "summary": {"inherit": 0.55, "reset": 0.55, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:40:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:36Z", "cpu_seconds": 1.4280323119999991, "created_at": "2026-11-22T03:39:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a8cd53fbe1ec", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "b1b44367be495baf8f86b3fd814a17cee8361d125af645ae3fbf1ea150f2e576", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a8cd53fbe1ec", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:39:26Z", "seed": 676615039, "treatment": "frontier"}, "priority": 2.0086012998290834, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4fbbfc262c295c302ac2d51385b8d3092230363dfd5ab1d66092eb57e6dfee55", "run_seconds": 1.7631778717041016, "seed": 676615039, "started_at": "2026-11-22T03:39:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.023676, 0.259359], "effect": 0.141517, "evaluated_at": "2026-11-22T03:40:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a8cd53fbe1ec", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.536842105, 0.302564103, 0.131506849, 0.337142857], "frontier": [0.678947368, 0.241025641, 0.394520548, 0.422857143], "goal": [0.607894737, 0.394871795, 0.194520548, 0.422857143], "hypdisc": [0.55, 0.31025641, 0.161643836, 0.348571429], "infogain": [0.678947368, 0.215384615, 0.221917808, 0.308571429], "prederr": [0.678947368, 0.31025641, 0.276712329, 0.405714286], "random": [0.386842105, 0.182051282, 0.210958904, 0.391428571], "risk": [0.678947368, 0.274358974, 0.394520548, 0.377142857]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "b1b44367be495baf8f86b3fd814a17cee8361d125af645ae3fbf1ea150f2e576", "primary_control": "random", "result_hash": "4fbbfc262c295c302ac2d51385b8d3092230363dfd5ab1d66092eb57e6dfee55", "signature": "f37962e94cc9cc16ccec5ea30434c259498b7161421dfd17153d053afb4f10aee629703e48c2ab3027e3bbee7e73df023d9b4ba2c523624f7434af12bf322a0a", "state": "VERIFIED", "summary": {"count": 0.327014, "frontier": 0.434338, "goal": 0.405036, "hypdisc": 0.342618, "infogain": 0.356205, "prederr": 0.417908, "random": 0.29282, "risk": 0.431242}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:40:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:27Z", "cpu_seconds": 0.8854502029999978, "created_at": "2026-11-22T03:39:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ace95d1b3071", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "2a0df899d56e5fd7dac5048447b792a40e17abc123bcfca099450c87f7afe65f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ace95d1b3071", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:39:26Z", "seed": 117364220, "treatment": "full"}, "priority": 2.009124453030201, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d60b51c9383103faab5e25bd317003af6c63bb279a83ae22c9cd07f26e41ec4f", "run_seconds": 0.9985699653625488, "seed": 117364220, "started_at": "2026-11-22T03:39:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.191294, 0.441294], "effect": 0.125, "evaluated_at": "2026-11-22T03:40:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ace95d1b3071", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.25, 0.75, 1.0], "fixed_policy": [0.5, 0.0, 0.25, 0.0], "full": [0.25, 0.75, 1.0, 1.0], "no_world_model": [0.25, 0.75, 1.0, 1.0], "random": [0.25, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "2a0df899d56e5fd7dac5048447b792a40e17abc123bcfca099450c87f7afe65f", "primary_control": "e141_best", "result_hash": "d60b51c9383103faab5e25bd317003af6c63bb279a83ae22c9cd07f26e41ec4f", "signature": "be2b10787b6c3220ee27487eb7421009b45649b85d34c88ae564e6642025e8bbca13cf065b075403069f03cdea0c8a9578c38449bed2d649d4d9f99a3bc0dc03", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.1875, "full": 0.75, "no_world_model": 0.75, "random": 0.5625}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:40:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:40:47Z", "cpu_seconds": 68.06292423000001, "created_at": "2026-11-22T03:39:26Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-209917e2335a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "fa3259b9f25e3bc06c4279781409be5844b6620a10d1f6f070fd672dfd01d8b7", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-209917e2335a", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:39:26Z", "seed": 1538527755, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "bdbfdb5efdbbeb5fc9147b10806f2591193ea07d6ed4193c004ece297066b8c3", "run_seconds": 79.72629642486572, "seed": 1538527755, "started_at": "2026-11-22T03:39:27Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 1, "inherit_mem": 1, "inherit_nomem": 1, "random_search": 1, "reset_mem": 1, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [-0.02861, 0.082432], "effect": 0.026911, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:40:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-209917e2335a", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 10, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 1.073, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.390625, 1.0, 0.95, 0.401515152, 0.86328125, 0.375, 0.946969697, 0.208333333, 0.495934959, 0.431992337], "compressed": [0.390625, 1.0, 0.9375, 0.401515152, 0.86328125, 0.375, 0.946969697, 0.208333333, 0.495934959, 0.713601533], "inherit_mem": [0.390625, 1.0, 0.9375, 0.401515152, 0.86328125, 0.375, 0.946969697, 0.208333333, 0.495934959, 0.713601533], "inherit_nomem": [0.3984375, 1.0, 0.9375, 0.15530303, 0.85859375, 0.3671875, 0.946969697, 0.504464286, 0.495934959, 0.585727969], "random_search": [0.421875, 1.0, 0.95, 0.40530303, 0.86171875, 0.375, 0.992424242, 0.411458333, 0.495934959, 0.594348659], "reset_mem": [0.390625, 1.0, 0.9375, 0.401515152, 0.86328125, 0.375, 0.946969697, 0.208333333, 0.495934959, 0.713601533], "retrieval": [0.390625, 1.0, 0.95, 0.401515152, 0.86328125, 0.375, 0.946969697, 0.208333333, 0.495934959, 0.431992337], "rootcause": [0.390625, 1.0, 0.9375, 0.401515152, 0.86328125, 0.375, 0.946969697, 0.208333333, 0.495934959, 0.713601533], "xfer_mem": [0.3984375, 1.0, 0.9375, 0.15530303, 0.85859375, 0.3671875, 0.946969697, 0.504464286, 0.495934959, 0.585727969]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "per_generation_sealed": [0.606365, 0.633276, 0.633276, 0.633276], "prereg_hash": "fa3259b9f25e3bc06c4279781409be5844b6620a10d1f6f070fd672dfd01d8b7", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "inherit_nomem", "random_search"], "result_hash": "bdbfdb5efdbbeb5fc9147b10806f2591193ea07d6ed4193c004ece297066b8c3", "signature": "ad913239f39294ad598ab1061664d3bc2706b0add55c28ee625006332499b4f82c81b023bd5f9812d6bf98a3bbf541e46cd33b1b5aaa48a8cb93202d9312430d", "state": "VERIFIED", "summary": {"G0": 0.606365, "compressed": 0.633276, "inherit_mem": 0.633276, "inherit_nomem": 0.625012, "random_search": 0.650806, "reset_mem": 0.633276, "retrieval": 0.606365, "rootcause": 0.633276, "xfer_mem": 0.625012}, "ties": 8, "treatment": "inherit_mem", "wins": 1}, "verified_at": "2026-11-22T03:41:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:29Z", "cpu_seconds": 0.8055488830000002, "created_at": "2026-11-22T03:39:26Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-4c6efa740bb6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "0c41691274335e9f965386d976de86ee11f76b87c5f923c02435b26cd7e34a29", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-4c6efa740bb6", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:39:26Z", "seed": 2077669718, "treatment": "gated"}, "priority": 2.008729927297835, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "abd221edc76ec4015caf833b70a10738f4fd0ad1f43864fcefb952498b2fc78f", "run_seconds": 0.943789005279541, "seed": 2077669718, "started_at": "2026-11-22T03:39:28Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:40:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-4c6efa740bb6", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.629166667, 0.585833333, 0.579166667, 0.496875], "gated": [0.629166667, 0.585833333, 0.579166667, 0.496875], "naive": [0.601666667, 0.4625, 0.446875, 0.455208333], "scratch": [0.629166667, 0.585833333, 0.579166667, 0.496875], "ungated": [0.601666667, 0.524166667, 0.579166667, 0.495833333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0c41691274335e9f965386d976de86ee11f76b87c5f923c02435b26cd7e34a29", "primary_control": "scratch", "result_hash": "abd221edc76ec4015caf833b70a10738f4fd0ad1f43864fcefb952498b2fc78f", "signature": "b9837af3bf297485451e3a4f1abf185d24bef77dcf179138f69590867c1f3af022267c04a5683926e4d32fa0d24d5f62277b578e6c1f2b23d13856bb9712700a", "state": "PARTIAL", "summary": {"compressed": 0.57276, "gated": 0.57276, "naive": 0.491563, "scratch": 0.57276, "ungated": 0.550208}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:40:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:51Z", "cpu_seconds": 1.9880654089999936, "created_at": "2026-11-22T03:39:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5a8ec6c8822e", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "c2e03cb6c009f6f35d3360e1d002661223fd4cb1ad71be9007287aed8bbc8526", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5a8ec6c8822e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:39:42Z", "seed": 597959262, "treatment": "frontier"}, "priority": 2.0061381332392347, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-e3763ea9e75a", "result_hash": "9f9c7213ebca0de40b8c920b5f4aa03ee98b9ccdbd7aa16c163c0f8fdf762d5d", "run_seconds": 2.113677501678467, "seed": 597959262, "started_at": "2026-11-22T03:39:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.030446, 0.144297], "effect": 0.056925, "evaluated_at": "2026-11-22T03:40:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5a8ec6c8822e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.376190476, 0.331578947, 0.436065574, 0.506666667], "frontier": [0.285714286, 0.407894737, 0.550819672, 0.525333333], "goal": [0.233333333, 0.265789474, 0.445901639, 0.365333333], "hypdisc": [0.238095238, 0.313157895, 0.26557377, 0.357333333], "infogain": [0.276190476, 0.031578947, 0.57704918, 0.493333333], "prederr": [0.180952381, 0.197368421, 0.452459016, 0.144], "random": [0.352380952, 0.278947368, 0.436065574, 0.474666667], "risk": [0.285714286, 0.407894737, 0.550819672, 0.525333333]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "c2e03cb6c009f6f35d3360e1d002661223fd4cb1ad71be9007287aed8bbc8526", "primary_control": "random", "replicated": false, "replication_of": "E141-ow_explore-e3763ea9e75a", "result_hash": "9f9c7213ebca0de40b8c920b5f4aa03ee98b9ccdbd7aa16c163c0f8fdf762d5d", "signature": "72b22bfa61359214012d6c146ce10f50d351cbf27c7428e01b56a46d9fd80a5245d7d6db84f0cf9732d9ed0d1679c9c7a70c0d72803f7c9f939e32ed2012f805", "state": "VERIFIED", "summary": {"count": 0.412625, "frontier": 0.442441, "goal": 0.327589, "hypdisc": 0.29354, "infogain": 0.344538, "prederr": 0.243695, "random": 0.385515, "risk": 0.442441}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:40:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:39:47Z", "cpu_seconds": 3.7920624099999998, "created_at": "2026-11-22T03:39:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2ba7b0973139", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "dbab412ef5e6b63daa68e630209e8c1600cf8771ba97b663adae599cce34dc15", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2ba7b0973139", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:39:42Z", "seed": 353632585, "treatment": "full"}, "priority": 2.006274083020649, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d3a0f56695359684810dd5be85aa76da555ed5ae60ae367f0d846b004da20853", "run_seconds": 4.18156361579895, "seed": 353632585, "started_at": "2026-11-22T03:39:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.370785, 0.245785], "effect": -0.0625, "evaluated_at": "2026-11-22T03:40:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2ba7b0973139", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.75, 0.5, 0.0], "fixed_policy": [0.25, 0.5, 0.75, 0.25], "full": [0.0, 0.75, 0.75, 0.0], "no_world_model": [0.0, 1.0, 0.5, 0.0], "random": [0.75, 0.75, 0.75, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "dbab412ef5e6b63daa68e630209e8c1600cf8771ba97b663adae599cce34dc15", "primary_control": "e141_best", "result_hash": "d3a0f56695359684810dd5be85aa76da555ed5ae60ae367f0d846b004da20853", "signature": "d48b06515c8ac0708847a1252940880448f1d5ce2aaec77c624da7bfe6e6cd2f59d1ebe171749f695c3e99c35e0ffe35566bb1ca7030bac53a52f071585c2408", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.4375, "full": 0.375, "no_world_model": 0.375, "random": 0.5625}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:40:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:59Z", "cpu_seconds": 0.669586764, "created_at": "2026-11-22T03:39:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2fe24421eb48", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "8b805686a5b8472e6596cf298f6e85dd4cb622a1859463a3577fd74393a594ed", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2fe24421eb48", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:39:42Z", "seed": 512705938, "treatment": "full"}, "priority": 2.001261427528285, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "139a2c01931af8306d64f2815ac0b1a24c953cb86acc8e58783e7c798e2b8fab", "run_seconds": 0.6955909729003906, "seed": 512705938, "started_at": "2026-11-22T03:42:59Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:43:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2fe24421eb48", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.25, 1.0, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8b805686a5b8472e6596cf298f6e85dd4cb622a1859463a3577fd74393a594ed", "primary_control": "e141_best", "result_hash": "139a2c01931af8306d64f2815ac0b1a24c953cb86acc8e58783e7c798e2b8fab", "signature": "ce56d966527ff251bc755eaffe7df6e626838e39218425dd8a70477d0017b12a1836e0812e7a9814d75dfc3ca50fbe6d07cc19edd4440660bbc5705fa063a70f", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.75, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:43:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:09Z", "cpu_seconds": 1.455552649, "created_at": "2026-11-22T03:39:42Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-45313e49a784", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "85a2dfb6ceccfa6e52c06bad39e2536db31df7916430e083277e259aa4277145", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-45313e49a784", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:39:42Z", "seed": 1629836656, "treatment": "full"}, "priority": 2.0027081189915763, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "45d9e84db89b8f174a8ba51d68b81bd80b30b200e693a5fb316adebd4724080a", "run_seconds": 2.558288812637329, "seed": 1629836656, "started_at": "2026-11-22T03:42:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.043621, 0.315813], "effect": 0.179717, "evaluated_at": "2026-11-22T03:42:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-45313e49a784", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.329545455, 0.465517241, 0.146511628, 0.45], "frozen": [0.034090909, 0.043103448, 0.077325581, 0.0], "full": [0.505681818, 0.465517241, 0.351744186, 0.7875], "retrieval": [0.522727273, 0.465517241, 0.415697674, 0.725], "rules_nodetect": [0.505681818, 0.465517241, 0.351744186, 0.7875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "85a2dfb6ceccfa6e52c06bad39e2536db31df7916430e083277e259aa4277145", "primary_control": "facts", "result_hash": "45d9e84db89b8f174a8ba51d68b81bd80b30b200e693a5fb316adebd4724080a", "signature": "36e1e1919fa2f93924ed1a72c79797252f5ee86c42f14d5e260f0f76690dcddd79a1decae574be3e01b7b1b72df8d2ff34f9da245d51686c97cd443d943f3d00", "state": "VERIFIED", "summary": {"facts": 0.347894, "frozen": 0.03863, "full": 0.527611, "retrieval": 0.532236, "rules_nodetect": 0.527611}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:42:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:40:16Z", "cpu_seconds": 7.358326185999999, "created_at": "2026-11-22T03:39:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-83ba763aa292", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "8e2870285eaa0da77cd8b7605203097242bb17f093ffd1ed9a92f1d7dd04dc3f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-83ba763aa292", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:39:59Z", "seed": 1396770524, "treatment": "frontier"}, "priority": 2.0064149329521093, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b81aaa10c235351d0fb4e8d14ffd8c53e6cb739e50152eb0351ee87b892be862", "run_seconds": 10.918313026428223, "seed": 1396770524, "started_at": "2026-11-22T03:40:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00289, 0.037871], "effect": 0.01749, "evaluated_at": "2026-11-22T03:40:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-83ba763aa292", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.388943249, 1.0, 0.420254403, 0.430528376], "frontier": [0.396771037, 1.0, 0.420254403, 0.430528376], "goal": [0.409491194, 0.933333333, 0.420254403, 0.400195695], "hypdisc": [0.380136986, 0.983333333, 0.363502935, 0.36741683], "infogain": [0.396771037, 1.0, 0.420254403, 0.430528376], "prederr": [0.396771037, 1.0, 0.420254403, 0.430528376], "random": [0.3962818, 1.0, 0.378180039, 0.403131115], "risk": [0.396771037, 1.0, 0.420254403, 0.430528376]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "8e2870285eaa0da77cd8b7605203097242bb17f093ffd1ed9a92f1d7dd04dc3f", "primary_control": "random", "result_hash": "b81aaa10c235351d0fb4e8d14ffd8c53e6cb739e50152eb0351ee87b892be862", "signature": "f4212764f231ff0af4d534cb288097c66c9ffac7b8ef47fdb059534a07395d03e980495a42c0cf8c96205d32c15d4d161e11909383fb890fcdef17e2e103a102", "state": "VERIFIED", "summary": {"count": 0.559932, "frontier": 0.561888, "goal": 0.540819, "hypdisc": 0.523598, "infogain": 0.561888, "prederr": 0.561888, "random": 0.544398, "risk": 0.561888}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:40:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:40:09Z", "cpu_seconds": 3.658755874999997, "created_at": "2026-11-22T03:39:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b8b4d5687711", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "f0b30821ec45e2afb81a1208760fa76f4da63bdb39c25bafd3a05306702901b0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b8b4d5687711", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:39:59Z", "seed": 1348902422, "treatment": "frontier"}, "priority": 2.007718426127717, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "47ec123cda9488e52377585f4a8b4b0b3aeeb9929c2d8b16ddb7775c69476ff6", "run_seconds": 5.8359010219573975, "seed": 1348902422, "started_at": "2026-11-22T03:40:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.070429, 0.278762], "effect": 0.104167, "evaluated_at": "2026-11-22T03:40:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b8b4d5687711", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.539215686, 0.656862745, 0.696078431, 0.715686275], "frontier": [0.779411765, 0.696078431, 0.367647059, 0.81372549], "goal": [0.75, 0.573529412, 0.387254902, 0.598039216], "hypdisc": [0.602941176, 0.558823529, 0.647058824, 0.637254902], "infogain": [0.779411765, 0.696078431, 0.367647059, 0.81372549], "prederr": [0.779411765, 0.696078431, 0.367647059, 0.81372549], "random": [0.534313725, 0.534313725, 0.524509804, 0.647058824], "risk": [0.779411765, 0.696078431, 0.367647059, 0.81372549]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "f0b30821ec45e2afb81a1208760fa76f4da63bdb39c25bafd3a05306702901b0", "primary_control": "random", "result_hash": "47ec123cda9488e52377585f4a8b4b0b3aeeb9929c2d8b16ddb7775c69476ff6", "signature": "4e16cd52dfe231fb1b464942de9d49bb1aab04d88e3143567e1cab0dce37b40d6796c286a72ae7febdcddeb66eb995f3e2399e2b4c7fddfea62524985f87d30e", "state": "VERIFIED", "summary": {"count": 0.651961, "frontier": 0.664216, "goal": 0.577206, "hypdisc": 0.61152, "infogain": 0.664216, "prederr": 0.664216, "random": 0.560049, "risk": 0.664216}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:40:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:40:09Z", "cpu_seconds": 1.320207805999999, "created_at": "2026-11-22T03:39:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-da3f6fd3578f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "8014112fe27231ba5094f75118e54ccafaff328fe52e49c29b5c2179a82947b4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-da3f6fd3578f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:39:59Z", "seed": 203332074, "treatment": "frontier"}, "priority": 2.0053125604101742, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "35a45012e99c3ca40c2d2f852abbe79751485a3eb0fccaeac4c7e5a66b5b4a81", "run_seconds": 2.307866334915161, "seed": 203332074, "started_at": "2026-11-22T03:40:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.233248, 0.052693], "effect": -0.090278, "evaluated_at": "2026-11-22T03:40:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-da3f6fd3578f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.805555556, 0.875, 0.875, 0.733333333], "frontier": [0.472222222, 0.875, 0.625, 0.733333333], "goal": [0.805555556, 0.875, 0.736111111, 0.733333333], "hypdisc": [0.777777778, 0.833333333, 0.625, 0.733333333], "infogain": [0.805555556, 0.875, 0.625, 0.733333333], "prederr": [0.805555556, 0.875, 0.625, 0.733333333], "random": [0.777777778, 0.875, 0.680555556, 0.733333333], "risk": [0.472222222, 0.875, 0.625, 0.733333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "8014112fe27231ba5094f75118e54ccafaff328fe52e49c29b5c2179a82947b4", "primary_control": "random", "result_hash": "35a45012e99c3ca40c2d2f852abbe79751485a3eb0fccaeac4c7e5a66b5b4a81", "signature": "5f5dc3627ca268955c5cdd6ab264703683b731d09813faf9cfc271c9cb745255b5a67b953e2b03319f46360a878921fdec709f881599a046a2310d0b79341b06", "state": "VERIFIED", "summary": {"count": 0.822222, "frontier": 0.676389, "goal": 0.7875, "hypdisc": 0.742361, "infogain": 0.759722, "prederr": 0.759722, "random": 0.766667, "risk": 0.676389}, "ties": 2, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:40:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:37Z", "cpu_seconds": 0.84056862, "created_at": "2026-11-22T03:39:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-388f8625048c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "c7de0af7f0937d02851d76e71fb4f6058e389fb17e1852157e121d7478c63d73", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-388f8625048c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:39:59Z", "seed": 1310499846, "treatment": "gated"}, "priority": 2.004009835263295, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7020b83090dbab7894d1afddda8e8ed5ebaab8e9545245324dabbda089b0548b", "run_seconds": 1.6160595417022705, "seed": 1310499846, "started_at": "2026-11-22T03:41:35Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:41:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-388f8625048c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.439166667, 0.519823789, 0.501754386, 0.612727273], "gated": [0.439166667, 0.519823789, 0.501754386, 0.612727273], "naive": [0.411666667, 0.519823789, 0.372807018, 0.488181818], "scratch": [0.439166667, 0.519823789, 0.501754386, 0.612727273], "ungated": [0.385833333, 0.421145374, 0.398245614, 0.539090909]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c7de0af7f0937d02851d76e71fb4f6058e389fb17e1852157e121d7478c63d73", "primary_control": "scratch", "result_hash": "7020b83090dbab7894d1afddda8e8ed5ebaab8e9545245324dabbda089b0548b", "signature": "dadf0bb08e9ef790f75f37570007a6d566fb942763bcf55a20cd022e2234508dc3d38c87cd27017c19896805d3ffef8f02593266d98501e1364f9d0cc03fed08", "state": "PARTIAL", "summary": {"compressed": 0.518368, "gated": 0.518368, "naive": 0.44812, "scratch": 0.518368, "ungated": 0.436079}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:41:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:11Z", "cpu_seconds": 1.6947136180000002, "created_at": "2026-11-22T03:39:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-7bfbe6862d7d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "prereg_hash": "8c92bfbe85f93f22c5e01d667b53450e980677a98b9d057319c1906ae29edeef", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-7bfbe6862d7d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:39:59Z", "seed": 830756696, "treatment": "gated"}, "priority": 2.004057446925218, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9a2251ad83e1b42de2fa838dc1f3f350268fe3e8c669df71375fc8e090d3113a", "run_seconds": 2.3553504943847656, "seed": 830756696, "started_at": "2026-11-22T03:41:08Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:41:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-7bfbe6862d7d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.410714286, 0.513194444, 0.875, 0.283203125], "gated": [0.410714286, 0.513194444, 0.875, 0.283203125], "naive": [0.410714286, 0.5125, 0.875, 0.26953125], "scratch": [0.410714286, 0.513194444, 0.875, 0.283203125], "ungated": [0.410714286, 0.5125, 0.875, 0.26953125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8c92bfbe85f93f22c5e01d667b53450e980677a98b9d057319c1906ae29edeef", "primary_control": "scratch", "result_hash": "9a2251ad83e1b42de2fa838dc1f3f350268fe3e8c669df71375fc8e090d3113a", "signature": "5eac4ee8c4f5c694122851d962e73d810dbf03c8ceca864509447ed2b350c46d397ce639a2a79a2ae8f824f2d730e9b35fc8879493e5e7ccf0307bdf5ba93b0d", "state": "PARTIAL", "summary": {"compressed": 0.520528, "gated": 0.520528, "naive": 0.516936, "scratch": 0.520528, "ungated": 0.516936}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:41:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:02Z", "cpu_seconds": 1.7671863660000002, "created_at": "2026-11-22T03:39:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-f36e1a2fc165", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "prereg_hash": "e760f2fb457d171cdc09d6c1a8592d07a0efe4dd428345e66c05f0d5cf75e76f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-f36e1a2fc165", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:39:59Z", "seed": 588792017, "treatment": "gated"}, "priority": 2.003263143033603, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "ce9102929dabe98e39bcb2cf9213e4ed4d8b560e0883e53c5dd95621742b5b6f", "run_seconds": 1.8756415843963623, "seed": 588792017, "started_at": "2026-11-22T03:42:00Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:42:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-f36e1a2fc165", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.508317025, 0.547244094, 0.268590998, 0.392857143], "gated": [0.508317025, 0.547244094, 0.268590998, 0.392857143], "naive": [0.257827789, 0.409448819, 0.018101761, 0.337301587], "scratch": [0.508317025, 0.547244094, 0.268590998, 0.392857143], "ungated": [0.38111546, 0.395669291, 0.022015656, 0.305555556]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e760f2fb457d171cdc09d6c1a8592d07a0efe4dd428345e66c05f0d5cf75e76f", "primary_control": "scratch", "result_hash": "ce9102929dabe98e39bcb2cf9213e4ed4d8b560e0883e53c5dd95621742b5b6f", "signature": "15d6f9e3d5af7ed634f7fc38ddc5256797a394d1ab7043519999dfaa5e337f8153eb434b62a5edac7ace5da83bd6fccdb6b294a0f671fe718f32fca41cee0503", "state": "PARTIAL", "summary": {"compressed": 0.429252, "gated": 0.429252, "naive": 0.25567, "scratch": 0.429252, "ungated": 0.276089}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:42:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:40:32Z", "cpu_seconds": 1.524480232, "created_at": "2026-11-22T03:40:16Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b774dd22dea6", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "prereg_hash": "4953f42cfe84feb1b4292322bf792bb3ba83de283d5cf1f58c92af68f0d494e7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b774dd22dea6", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.0, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:40:16Z", "seed": 1411234220, "treatment": "iw_temperature"}, "priority": 2.0095944161715042, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-aa09c6250174", "result_hash": "0021f7d152c8df25030f95832f5f9e63d0b2d037b0aa06e8bfa91bbf4f3831e4", "run_seconds": 2.4217240810394287, "seed": 1411234220, "started_at": "2026-11-22T03:40:29Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020588, -0.003693], "effect": -0.012141, "evaluated_at": "2026-11-22T03:40:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b774dd22dea6", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.272464425, 0.061286645, 0.08937286, 0.035212622, 0.084814238, 0.030863668, 0.06344013, 0.055084147, 0.128526977, 0.021577583, 0.090524623, 0.14602595, 0.180367605, 0.037828175, 0.103980907, 0.102060235], "raw": [0.288282885, 0.07366045, 0.07368209, 0.037954485, 0.062881943, 0.017937562, 0.044467337, 0.035139445, 0.128526977, 0.02780485, 0.084401435, 0.180583885, 0.149972768, 0.028099688, 0.12076578, 0.046134547], "temperature": [0.256862895, 0.051761075, 0.08937286, 0.035212622, 0.049972117, 0.01838111, 0.027864308, 0.047445765, 0.128526977, 0.021577583, 0.084401435, 0.159263255, 0.163838468, 0.024586273, 0.103980907, 0.046134547]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0002, "losses": 10, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "4953f42cfe84feb1b4292322bf792bb3ba83de283d5cf1f58c92af68f0d494e7", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-aa09c6250174", "result_hash": "0021f7d152c8df25030f95832f5f9e63d0b2d037b0aa06e8bfa91bbf4f3831e4", "signature": "acd88454bd4ae2b8ef5079cee4a16e899486e909554b3f5679f7903c9c57555f4b1592a26c0d5d33f9ae4a67f5b91806eed116419eefb0f89185c4daf5b4280f", "state": "VERIFIED", "summary": {"iw_temperature": 0.093964, "raw": 0.087519, "temperature": 0.081824}, "ties": 5, "treatment": "iw_temperature", "wins": 1}, "verified_at": "2026-11-22T03:40:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:15Z", "cpu_seconds": 39.740541313, "created_at": "2026-11-22T03:40:16Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1bb1cd704c76", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 3, "n": 4}, "prereg_hash": "c793458759a03a37a81099f9ed802b322780eee9ca8252095f8991a2953a560f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1bb1cd704c76", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:40:16Z", "seed": 328987934, "treatment": "full"}, "priority": 2.008731635172738, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5d6ad50783a365501177cf63487027c0904dc111cdb8fd58a3a1285f4fda2c94", "run_seconds": 46.21836709976196, "seed": 328987934, "started_at": "2026-11-22T03:40:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.24, 0.74], "effect": 0.25, "evaluated_at": "2026-11-22T03:41:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1bb1cd704c76", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 1.0, 0.0], "fixed_policy": [0.0, 0.75, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 0.0], "no_world_model": [0.5, 1.0, 1.0, 0.0], "random": [0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "c793458759a03a37a81099f9ed802b322780eee9ca8252095f8991a2953a560f", "primary_control": "e141_best", "result_hash": "5d6ad50783a365501177cf63487027c0904dc111cdb8fd58a3a1285f4fda2c94", "signature": "917a7165cb43de848f19bbfb993c97059fafb25c2773e9ae21397c1c50525e5a6cdd8124f9cec752e84cd5ec36505986e7a12851583055103dc0176665987d09", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.1875, "full": 0.75, "no_world_model": 0.625, "random": 0.25}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:41:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:40:30Z", "cpu_seconds": 1.0031196379999998, "created_at": "2026-11-22T03:40:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3816e1a9bfd8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "abfa65626f8d04a7381ed48e8697b101c806a3f7dc4574da8837a77a3a720920", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3816e1a9bfd8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:40:16Z", "seed": 63706491, "treatment": "full"}, "priority": 2.005696103794031, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ef201e9dc65b92e6db7421b03dc4eccffdca1bcb4882c35bb82bd14ae2e0c916", "run_seconds": 1.3253819942474365, "seed": 63706491, "started_at": "2026-11-22T03:40:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.009105, 0.080963], "effect": 0.045034, "evaluated_at": "2026-11-22T03:40:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3816e1a9bfd8", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.560606061, 0.727272727, 0.814814815, 0.916666667], "frozen": [0.348484848, 0.318181818, 0.240740741, 0.3125], "full": [0.636363636, 0.757575758, 0.888888889, 0.916666667], "retrieval": [0.742424242, 0.818181818, 0.888888889, 0.916666667], "rules_nodetect": [0.636363636, 0.818181818, 0.888888889, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "abfa65626f8d04a7381ed48e8697b101c806a3f7dc4574da8837a77a3a720920", "primary_control": "facts", "result_hash": "ef201e9dc65b92e6db7421b03dc4eccffdca1bcb4882c35bb82bd14ae2e0c916", "signature": "fa2df4a806206e406d4b8b1b0d4c79f0f1a6df66b1cf6c57de7355bdae878f7107c15dc86805e1d3bcff8677a7d3e4890d5cd1536ceb898a314f9c7639ee8f04", "state": "VERIFIED", "summary": {"facts": 0.75484, "frozen": 0.304977, "full": 0.799874, "retrieval": 0.84154, "rules_nodetect": 0.815025}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:40:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:43Z", "cpu_seconds": 2.9967141859999984, "created_at": "2026-11-22T03:40:33Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6c4f1974f1a5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "f1ae371098cb338e77afb743037c8ff8203c2fb7d29120842fbf1bc0f8e488de", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6c4f1974f1a5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:40:33Z", "seed": 1819310731, "treatment": "frontier"}, "priority": 2.001450513460327, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b54fa8b6ebb8c8468211d9b5c419c910e8dd3a4145c6479aea63eabcb1e4d6a4", "run_seconds": 3.1482157707214355, "seed": 1819310731, "started_at": "2026-11-22T03:42:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.217515, 0.08432], "effect": -0.066597, "evaluated_at": "2026-11-22T03:42:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6c4f1974f1a5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.344907407, 0.568181818, 0.357857143, 0.684375], "frontier": [0.331790123, 0.439393939, 0.04, 0.753125], "goal": [0.328703704, 0.397727273, 0.063571429, 0.665625], "hypdisc": [0.328703704, 0.409090909, 0.363571429, 0.7], "infogain": [0.339506173, 0.386363636, 0.04, 0.753125], "prederr": [0.339506173, 0.382575758, 0.04, 0.715625], "random": [0.328703704, 0.458333333, 0.334285714, 0.709375], "risk": [0.331790123, 0.439393939, 0.04, 0.753125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "f1ae371098cb338e77afb743037c8ff8203c2fb7d29120842fbf1bc0f8e488de", "primary_control": "random", "result_hash": "b54fa8b6ebb8c8468211d9b5c419c910e8dd3a4145c6479aea63eabcb1e4d6a4", "signature": "795f3b80825617b0142e0cd88d0168c96aef1a022d2096d7650e24d35519d6caa4fc1cca7810a2b18f9e33d839b5f85672992074b9c18c91c2a405df9b5bb901", "state": "VERIFIED", "summary": {"count": 0.48883, "frontier": 0.391077, "goal": 0.363907, "hypdisc": 0.450342, "infogain": 0.379749, "prederr": 0.369427, "random": 0.457674, "risk": 0.391077}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:42:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:11Z", "cpu_seconds": 2.2901223959999997, "created_at": "2026-11-22T03:40:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4fbeeaabf301", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "f38a945d9632247bda276ac373931816113208233361ccfba94938053995888d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4fbeeaabf301", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:40:33Z", "seed": 904128797, "treatment": "full"}, "priority": 2.002791039359098, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1ea13ea03930c8e460b3f41f001861c3a1c9610b7d38af1d606d5163039eb874", "run_seconds": 3.5506653785705566, "seed": 904128797, "started_at": "2026-11-22T03:42:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.150083, 0.650083], "effect": 0.25, "evaluated_at": "2026-11-22T03:42:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4fbeeaabf301", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 0.75, 0.75], "fixed_policy": [1.0, 1.0, 0.25, 0.5], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [0.5, 0.5, 1.0, 0.75], "random": [0.75, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "f38a945d9632247bda276ac373931816113208233361ccfba94938053995888d", "primary_control": "e141_best", "result_hash": "1ea13ea03930c8e460b3f41f001861c3a1c9610b7d38af1d606d5163039eb874", "signature": "62ded70bc65fa15bbd1a3c66c7037d09078d1d330fe8ada2fe916ef220b3b2b959243e496b5498f8127ef68169ca1aa68642a82e0ef1c78a1f93dd66ebb91306", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.6875, "full": 0.9375, "no_world_model": 0.6875, "random": 0.875}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:42:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:37Z", "cpu_seconds": 0.8967214309999999, "created_at": "2026-11-22T03:40:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f9e72f86bc0e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "2f82df2cd2d47fe1b4f1551a2b0869799d70105c52f931346afe061879213b4a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f9e72f86bc0e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:40:33Z", "seed": 1969421382, "treatment": "full"}, "priority": 2.0034996287214537, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "217d837ec2f55c91cb8374b06869170442f98446d38f36568d299463d4749310", "run_seconds": 1.9139904975891113, "seed": 1969421382, "started_at": "2026-11-22T03:41:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.055, 0.68], "effect": 0.3125, "evaluated_at": "2026-11-22T03:41:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f9e72f86bc0e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 0.0, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 0.75, 0.75, 0.0], "no_world_model": [0.0, 0.25, 0.0, 0.0], "random": [0.0, 0.75, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "2f82df2cd2d47fe1b4f1551a2b0869799d70105c52f931346afe061879213b4a", "primary_control": "e141_best", "result_hash": "217d837ec2f55c91cb8374b06869170442f98446d38f36568d299463d4749310", "signature": "d52794346764fe604c7016d40dcc4d8eef8b6ac3b107d13bf0d5209b1789924990f62aef829968a45bb84113214aea5e8fcccca76d794c887a9bbd8dab2e460c", "state": "VERIFIED", "summary": {"e141_best": 0.0625, "fixed_policy": 0.0, "full": 0.375, "no_world_model": 0.0625, "random": 0.1875}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:41:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:20Z", "cpu_seconds": 7.84056161, "created_at": "2026-11-22T03:40:33Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-62059f6e2481", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "81efcb7d868ecdf1e4f68a18ecd54b8dc27b8147ac222e13c39c774e669fddc8", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-62059f6e2481", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:40:33Z", "seed": 1998006510, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "58b23cc8f52f5095d34e5e48868374e7f711cce40b0fe619aee4f1c795f844ab", "run_seconds": 12.140540361404419, "seed": 1998006510, "started_at": "2026-11-22T03:41:08Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [-0.079138, 0.012471], "effect": -0.033333, "evaluated_at": "2026-11-22T03:41:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-62059f6e2481", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.4576635700763785, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0], "root": [0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "81efcb7d868ecdf1e4f68a18ecd54b8dc27b8147ac222e13c39c774e669fddc8", "primary_control": "reset", "result_hash": "58b23cc8f52f5095d34e5e48868374e7f711cce40b0fe619aee4f1c795f844ab", "signature": "430e44007aec6f35788035941ffea89532e17813f16e4c6b37423f7e2985cc7feab7209bf6d31afc391ee464c206cf7fe20eff77f9722687824647e9ecc1b003", "state": "VERIFIED", "summary": {"inherit": 0.5, "reset": 0.5333333333333333, "root": 0.26666666666666666}, "ties": 58, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:41:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:19Z", "cpu_seconds": 7.0718703419999995, "created_at": "2026-11-22T03:40:49Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-3b510e0b23c6", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "893f5bfea0b5b6e46040027b5e1ad25f4d68ec0838b03aa0fa2edd0cf99ad1ff", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3b510e0b23c6", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:40:49Z", "seed": 862937441, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "2a2f5bdf55bff2044c7764bcfa4b8bab66f42eb7c282dcffa55abcdc31858811", "run_seconds": 10.671086072921753, "seed": 862937441, "started_at": "2026-11-22T03:41:08Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:41:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3b510e0b23c6", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.47103432456457295, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0], "reset": [0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "893f5bfea0b5b6e46040027b5e1ad25f4d68ec0838b03aa0fa2edd0cf99ad1ff", "primary_control": "reset", "result_hash": "2a2f5bdf55bff2044c7764bcfa4b8bab66f42eb7c282dcffa55abcdc31858811", "signature": "8a73c252018bf6895789e0a235d51c5ba7559ac45396ecdedb560821e4782cbc5c3499e1c0bdac7d7b33ca5dad9365c59ab2eb495702ced1d94e525bc59d4b0e", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.3333333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:41:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:44Z", "cpu_seconds": 6.750547248, "created_at": "2026-11-22T03:41:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-43e254227bb7", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "ea4e3261c1d45491d3fc5a3042b64d54977601894c4c20cb08fcd5876ac7ab0c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-43e254227bb7", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:41:22Z", "seed": 177046223, "treatment": "frontier"}, "priority": 2.0056473292829775, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-7ffd5c093c52", "result_hash": "96dc02bd76a068575e3b14717089057f17afba441a933cfc1f8055e7ec629891", "run_seconds": 8.45084285736084, "seed": 177046223, "started_at": "2026-11-22T03:41:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023999, -0.001655], "effect": -0.012827, "evaluated_at": "2026-11-22T03:41:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-43e254227bb7", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.538834951, 0.441666667, 0.416666667, 0.249190939], "frontier": [0.469255663, 0.322222222, 0.40625, 0.454692557], "goal": [0.234627832, 0.215277778, 0.336309524, 0.391585761], "hypdisc": [0.511326861, 0.305555556, 0.336309524, 0.496763754], "infogain": [0.381877023, 0.322222222, 0.401785714, 0.252427184], "prederr": [0.511326861, 0.322222222, 0.401785714, 0.514563107], "random": [0.466019417, 0.345833333, 0.422619048, 0.469255663], "risk": [0.194174757, 0.322222222, 0.40625, 0.454692557]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "ea4e3261c1d45491d3fc5a3042b64d54977601894c4c20cb08fcd5876ac7ab0c", "primary_control": "random", "replicated": false, "replication_of": "E141-ow_explore-7ffd5c093c52", "result_hash": "96dc02bd76a068575e3b14717089057f17afba441a933cfc1f8055e7ec629891", "signature": "aa9e10c30aacf2654cc36170c5a29150f32a1ef0d7fa0a3903547a4f5dbb6b2952d39a593a1416aec9486fac1c20404e2c0974ce9532fb3365c3a60fc9eea102", "state": "VERIFIED", "summary": {"count": 0.41159, "frontier": 0.413105, "goal": 0.29445, "hypdisc": 0.412489, "infogain": 0.339578, "prederr": 0.437474, "random": 0.425932, "risk": 0.344335}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:41:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:49Z", "cpu_seconds": 0.20343619299999993, "created_at": "2026-11-22T03:41:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b87f4c824c71", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "026a338d7d56306f7c9107d3b251c7929fe70dbae9fdabbfc0a380460b103ca0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b87f4c824c71", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:41:39Z", "seed": 832548520, "treatment": "full"}, "priority": 2.003960698665826, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "20296ce4900265660505f7b99c99677766f01a54664d867b520de378e5c697cb", "run_seconds": 0.2564871311187744, "seed": 832548520, "started_at": "2026-11-22T03:41:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019083, 0.855917], "effect": 0.4375, "evaluated_at": "2026-11-22T03:41:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b87f4c824c71", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.5, 0.0, 0.75], "fixed_policy": [0.75, 0.75, 0.5, 0.5], "full": [1.0, 1.0, 1.0, 0.75], "no_world_model": [1.0, 1.0, 0.75, 0.75], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "026a338d7d56306f7c9107d3b251c7929fe70dbae9fdabbfc0a380460b103ca0", "primary_control": "e141_best", "result_hash": "20296ce4900265660505f7b99c99677766f01a54664d867b520de378e5c697cb", "signature": "8f400cb4a3730d74e0a474e75ae1e8712692da08faa9cd2fdd5b99f09a42d6f6827b1228661a248b85df2bf74cb96214b66b7a669ebd041351cd66ee81e7db0a", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.625, "full": 0.9375, "no_world_model": 0.875, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:41:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:41:45Z", "cpu_seconds": 2.065115535, "created_at": "2026-11-22T03:41:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-dbc66e32e556", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "468909c501cb9b2b6d55e362c61675af6bb0d90d3172395a2d944ba506d7015a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-dbc66e32e556", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:41:39Z", "seed": 1075764825, "treatment": "full"}, "priority": 2.0084911875237657, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-720329efb621", "result_hash": "1eddc5403c83f112598cac618b678e30797bec60bc5542758888aa479f0a8b88", "run_seconds": 2.6168649196624756, "seed": 1075764825, "started_at": "2026-11-22T03:41:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.082778, 0.198472], "effect": 0.140625, "evaluated_at": "2026-11-22T03:41:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-dbc66e32e556", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.854166667, 0.819444444, 0.944444444, 0.819444444], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.972222222, 0.979166667, 0.972222222, 0.944444444], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "468909c501cb9b2b6d55e362c61675af6bb0d90d3172395a2d944ba506d7015a", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-720329efb621", "result_hash": "1eddc5403c83f112598cac618b678e30797bec60bc5542758888aa479f0a8b88", "signature": "82852d2b5273d4c37659e07e46f2dfa8b9ce28e8b6a23fb8f308b7a752f2b73d907792df0c1dd452ed1693f792199021271856039aa8b800182d7e0f0286b108", "state": "VERIFIED", "summary": {"facts": 0.859375, "frozen": 0.0, "full": 1.0, "retrieval": 0.967014, "rules_nodetect": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:41:55Z"}
{"attempts": 3, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T03:41:39Z", "error": "lease expired; lease expired; lease expired; ", "family": "ow_rsi", "fence": 3, "id": "E141-ow_rsi-7777e45987b3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "b9bda549c1101d44ab3bf3a030d16a85a3d93c28cf37d4ee9d1a7bdc49053485", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-7777e45987b3", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:41:39Z", "seed": 259066812, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 259066812, "started_at": "2026-11-22T03:41:39Z", "state": "FAILED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T03:41:45Z", "cpu_seconds": 2.5606048109999997, "created_at": "2026-11-22T03:41:39Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-64736630ece8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "prereg_hash": "60a320b0785bdeedafb5d5557920e651f4d2cb69329654bb6bce5b189a06bb2c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-64736630ece8", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:41:39Z", "seed": 843021123, "treatment": "gated"}, "priority": 2.0097665679843124, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "977df3ae21818ea1b1bd894297b3d2467c59954e49593ec93ac487640a5649c6", "run_seconds": 3.2783915996551514, "seed": 843021123, "started_at": "2026-11-22T03:41:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.032, 0.098667], "effect": 0.033333, "evaluated_at": "2026-11-22T03:41:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-64736630ece8", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.75, 0.5, 0.516666667, 0.616666667], "gated": [0.75, 0.5, 0.65, 0.616666667], "naive": [0.1, 0.1, 0.266666667, 0.233333333], "scratch": [0.75, 0.5, 0.516666667, 0.616666667], "ungated": [0.4, 0.483333333, 0.616666667, 0.416666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "60a320b0785bdeedafb5d5557920e651f4d2cb69329654bb6bce5b189a06bb2c", "primary_control": "scratch", "result_hash": "977df3ae21818ea1b1bd894297b3d2467c59954e49593ec93ac487640a5649c6", "signature": "09be9a694265a57536caba35c238cb18376c1d86a1af36694042b3e636eac1d6d8218758dd918635beb9fb9c4beeb1192e3bdc0d0a6eae51034756746c00210e", "state": "VERIFIED", "summary": {"compressed": 0.595833, "gated": 0.629167, "naive": 0.175, "scratch": 0.595833, "ungated": 0.479167}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:41:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:02Z", "cpu_seconds": 1.576598831, "created_at": "2026-11-22T03:41:55Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-b5d4d4d3bfbb", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "prereg_hash": "380755807668c1b5af6d646c90f805a6c052bc4a3cbceb081b372c2026b990f7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-b5d4d4d3bfbb", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:41:55Z", "seed": 586311980, "treatment": "iw_temperature"}, "priority": 2.0031560443224055, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-190890b683d7", "result_hash": "141ff604bca80f0d1f87564ba95c07c3052820ea12bd0ee7016d60e97423cf6f", "run_seconds": 1.679607629776001, "seed": 586311980, "started_at": "2026-11-22T03:42:00Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.020339, 0.032687], "effect": 0.006174, "evaluated_at": "2026-11-22T03:42:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-b5d4d4d3bfbb", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.304247698, 0.34544653, 0.299507903, 0.24241237, 0.25204161, 0.01709927, 0.337570153, 0.180492425, 0.178469595, 0.241380275, 0.173928815, 0.294481158, 0.265399445, 0.24426297, 0.25625938, 0.274371102], "raw": [0.24848324, 0.245756938, 0.287619187, 0.195985775, 0.303071952, 0.168342365, 0.306770208, 0.12976293, 0.178469595, 0.219183058, 0.28100303, 0.29065502, 0.191184095, 0.177867062, 0.216210203, 0.247681392], "temperature": [0.27524992, 0.270927495, 0.32401999, 0.205296168, 0.312865942, 0.168342365, 0.31612869, 0.170408435, 0.204176955, 0.253776495, 0.245047195, 0.300402092, 0.216041213, 0.21906393, 0.25625938, 0.268154042]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1572, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.696", "prereg_hash": "380755807668c1b5af6d646c90f805a6c052bc4a3cbceb081b372c2026b990f7", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-190890b683d7", "result_hash": "141ff604bca80f0d1f87564ba95c07c3052820ea12bd0ee7016d60e97423cf6f", "signature": "778bd550e3242f2a8c6a55de9a581199b2481201decbffd613a5b76dc634d679df493dc2adc5c4ea8b035be0874a032fc83575a63ed9f212c011b3c0e4e3280d", "state": "VERIFIED", "summary": {"iw_temperature": 0.244211, "raw": 0.230503, "temperature": 0.250385}, "ties": 1, "treatment": "iw_temperature", "wins": 7}, "verified_at": "2026-11-22T03:42:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:23Z", "cpu_seconds": 0.3809858860000004, "created_at": "2026-11-22T03:41:55Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-41f34a940150", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "311d8155d069e402a2e9b70e5b6c4fd75ad41b5d03fe2c5a9ce235ad4118c91a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-41f34a940150", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:41:55Z", "seed": 155513149, "treatment": "full"}, "priority": 2.0022539809958366, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "8a9763479e522db31b150e59f28f996105057869c83b4ccb0b55622192fc82f5", "run_seconds": 0.9378156661987305, "seed": 155513149, "started_at": "2026-11-22T03:42:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004, 0.012333], "effect": 0.004167, "evaluated_at": "2026-11-22T03:42:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-41f34a940150", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 1.0, 0.983333333, 1.0], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 1.0, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "311d8155d069e402a2e9b70e5b6c4fd75ad41b5d03fe2c5a9ce235ad4118c91a", "primary_control": "facts", "result_hash": "8a9763479e522db31b150e59f28f996105057869c83b4ccb0b55622192fc82f5", "signature": "c33eb7003cc36ebaf76c0727a4ec75a416a29794d823ab13c61436345ea83494965f375dfe52484ee0b51659938a9cf6844ab81ffdb841e7e6394ecf4f347f0f", "state": "VERIFIED", "summary": {"facts": 0.995833, "frozen": 0.0, "full": 1.0, "retrieval": 1.0, "rules_nodetect": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:42:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:53Z", "cpu_seconds": 1.250507860999999, "created_at": "2026-11-22T03:41:55Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a251833ff307", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "ad600a44b9ff379aa41edb1c7a70a0b01cfdafbe35c7d6e27d78630138129c50", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a251833ff307", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:41:55Z", "seed": 898459526, "treatment": "full"}, "priority": 1.9966941289379783, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7341f8f0ed11089fd7efd7cb5fc44cf9f87c6622e8c2cbc9732a415b64777cd1", "run_seconds": 2.058228015899658, "seed": 898459526, "started_at": "2026-11-22T03:43:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001558, 0.208522], "effect": 0.10504, "evaluated_at": "2026-11-22T03:44:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a251833ff307", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.875, 0.095982143, 0.384615385, 0.169753086], "frozen": [0.125, 0.042410714, 0.0, 0.053240741], "full": [0.875, 0.301339286, 0.413461538, 0.355709877], "retrieval": [0.833333333, 0.322544643, 0.403846154, 0.426697531], "rules_nodetect": [0.541666667, 0.272321429, 0.153846154, 0.090277778]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "ad600a44b9ff379aa41edb1c7a70a0b01cfdafbe35c7d6e27d78630138129c50", "primary_control": "facts", "result_hash": "7341f8f0ed11089fd7efd7cb5fc44cf9f87c6622e8c2cbc9732a415b64777cd1", "signature": "6b040f632d393484ac4cfc66cbb8e29efa8d394f68fe2e4aa754b6d6e3fdd8938bfc1ed2a834281fc6469e186226a7e4b1426aec53aadf81a96bf8a3ca02d40e", "state": "VERIFIED", "summary": {"facts": 0.381338, "frozen": 0.055163, "full": 0.486378, "retrieval": 0.496605, "rules_nodetect": 0.264528}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:44:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:07Z", "cpu_seconds": 6.3183672799999995, "created_at": "2026-11-22T03:41:55Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a8d95cf71334", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6b345a04eaeb2c9e820c0738293be935b42caacdbed111ee62f78ae743b532f5", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a8d95cf71334", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:41:55Z", "seed": 2060318044, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "818c92b0960458ca6b9a11a55d52d47b69ff42a7b59d95bb508aecdc1e5a1bda", "run_seconds": 6.57002067565918, "seed": 2060318044, "started_at": "2026-11-22T03:42:00Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [-0.299802, -0.033532], "effect": -0.166667, "evaluated_at": "2026-11-22T03:42:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a8d95cf71334", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.3038564031864097, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0016, "losses": 14, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.996", "prereg_hash": "6b345a04eaeb2c9e820c0738293be935b42caacdbed111ee62f78ae743b532f5", "primary_control": "reset", "result_hash": "818c92b0960458ca6b9a11a55d52d47b69ff42a7b59d95bb508aecdc1e5a1bda", "signature": "879d5d521be96c608136d2d43fe4cbeaf3fd23f3ccf9020d837fdd03ed6349cc30e9a74aef55ef2e8ae07caa7c1edc6ce6979667ab2a32708d220d00392a6e01", "state": "VERIFIED", "summary": {"inherit": 0.6666666666666666, "reset": 0.8333333333333334, "root": 0.2833333333333333}, "ties": 42, "treatment": "inherit", "wins": 4}, "verified_at": "2026-11-22T03:42:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:17Z", "cpu_seconds": 4.576718505999999, "created_at": "2026-11-22T03:41:55Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-cbc0a6900818", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "385acb2bbddfb71bbe606dec837e861d5dc1e30869f0615579994fabbf30039a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-cbc0a6900818", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:41:55Z", "seed": 1665579674, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f98061e35d343589b20d1c2f6541017145e8d75c8d2b5c4627c267a330026d33", "run_seconds": 5.704143762588501, "seed": 1665579674, "started_at": "2026-11-22T03:42:12Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:42:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-cbc0a6900818", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0], "reset": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "385acb2bbddfb71bbe606dec837e861d5dc1e30869f0615579994fabbf30039a", "primary_control": "reset", "result_hash": "f98061e35d343589b20d1c2f6541017145e8d75c8d2b5c4627c267a330026d33", "signature": "176479d8ed36f2fc84416fa6db5d431ebfa88ea5762e9163452a143672737a4fb9a76ac935eb34195a263b45948a05db84354f29d04213e5096e0bdcbdbecb0b", "state": "PARTIAL", "summary": {"inherit": 0.35, "reset": 0.35, "root": 0.3333333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:42:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:17Z", "cpu_seconds": 1.5443501560000001, "created_at": "2026-11-22T03:42:12Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-958474b15e2c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "815fe8aacd8a04de446e920071328c41eb2db7e7860b7ed6e64804249eb051c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-958474b15e2c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:42:12Z", "seed": 552462934, "treatment": "frontier"}, "priority": 2.0044485570998867, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6e89bbecba5824c6e8ad2ad423090571992f5325f95c7ff7a2030cde5c7e0009", "run_seconds": 2.509571075439453, "seed": 552462934, "started_at": "2026-11-22T03:42:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.166113, 0.30186], "effect": 0.067874, "evaluated_at": "2026-11-22T03:42:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-958474b15e2c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.847222222, 0.857142857, 0.87037037, 0.919642857], "frontier": [0.541666667, 0.976190476, 0.981481481, 0.8125], "goal": [0.541666667, 0.964285714, 0.962962963, 0.508928571], "hypdisc": [0.930555556, 0.916666667, 0.981481481, 0.5], "infogain": [0.541666667, 0.976190476, 0.981481481, 0.8125], "prederr": [0.541666667, 0.976190476, 0.981481481, 0.8125], "random": [0.805555556, 0.678571429, 0.824074074, 0.732142857], "risk": [0.541666667, 0.976190476, 0.981481481, 0.8125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "815fe8aacd8a04de446e920071328c41eb2db7e7860b7ed6e64804249eb051c3", "primary_control": "random", "result_hash": "6e89bbecba5824c6e8ad2ad423090571992f5325f95c7ff7a2030cde5c7e0009", "signature": "5b703d8358b89223bc36cd4ccf866c10d6a44d556621bd92264da4d0779176192cb9f3f3b641140d550eb82745c2349a9f4e6c51efa944d7443ff0a04ddd1d03", "state": "VERIFIED", "summary": {"count": 0.873595, "frontier": 0.82796, "goal": 0.744461, "hypdisc": 0.832176, "infogain": 0.82796, "prederr": 0.82796, "random": 0.760086, "risk": 0.82796}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:42:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:25Z", "cpu_seconds": 1.5020544069999993, "created_at": "2026-11-22T03:42:12Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a1efff0af354", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "34665f478ff995e16b80f332f4873b95edc727de5e601a23e5729170b52eb0c7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a1efff0af354", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:42:12Z", "seed": 1491499124, "treatment": "frontier"}, "priority": 2.00221707014383, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a85381a5345542ef3ecb1f8471c2cd2272621fd0b0da2fa405d7cb0310690dcd", "run_seconds": 2.3055365085601807, "seed": 1491499124, "started_at": "2026-11-22T03:42:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.019821, 0.422305], "effect": 0.201242, "evaluated_at": "2026-11-22T03:42:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a1efff0af354", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.8515625, 0.542857143, 0.644736842], "frontier": [0.840909091, 0.9765625, 0.871428571, 0.815789474], "goal": [0.818181818, 0.9453125, 0.828571429, 0.815789474], "hypdisc": [0.954545455, 0.8203125, 0.7, 0.684210526], "infogain": [0.840909091, 0.9765625, 0.871428571, 0.815789474], "prederr": [0.840909091, 0.9765625, 0.871428571, 0.815789474], "random": [0.931818182, 0.78125, 0.414285714, 0.572368421], "risk": [0.840909091, 0.9765625, 0.871428571, 0.815789474]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "34665f478ff995e16b80f332f4873b95edc727de5e601a23e5729170b52eb0c7", "primary_control": "random", "result_hash": "a85381a5345542ef3ecb1f8471c2cd2272621fd0b0da2fa405d7cb0310690dcd", "signature": "9f75c758c3f9f211843b610b4a33d1813ca7e5dc4f9163f7c315f1648d40dc682fda9a18185809c61b62ba4f93b778d0e4b18325c2a210a8e31f62aa00cfb60d", "state": "VERIFIED", "summary": {"count": 0.759789, "frontier": 0.876172, "goal": 0.851964, "hypdisc": 0.789767, "infogain": 0.876172, "prederr": 0.876172, "random": 0.674931, "risk": 0.876172}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:42:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:35Z", "cpu_seconds": 12.136302664999999, "created_at": "2026-11-22T03:42:12Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e9b29eb2c308", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 7, "n": 4}, "prereg_hash": "97ee6dfa410e2bc3a8b0cf7ccc7ce9f1370307dbc8e1d7713ebf8e6d157d8a48", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e9b29eb2c308", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:42:12Z", "seed": 851121844, "treatment": "frontier"}, "priority": 2.0027109355829507, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9a684984796600dc9ef628d9576479514530a689bf0d4957d71c2140a8e4a5a8", "run_seconds": 13.511546611785889, "seed": 851121844, "started_at": "2026-11-22T03:42:21Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:42:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e9b29eb2c308", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "adversarial", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "97ee6dfa410e2bc3a8b0cf7ccc7ce9f1370307dbc8e1d7713ebf8e6d157d8a48", "primary_control": "random", "result_hash": "9a684984796600dc9ef628d9576479514530a689bf0d4957d71c2140a8e4a5a8", "signature": "f8a0a8aa9cfe599ed8c3fea552138f90f4039cb6b3d9b080a3ec4455e24b7671d7f1143b17396ae5aea5bdc5d93f6ba46c2546c06770f5e2f85ef0e489c5cd06", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:42:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:16Z", "cpu_seconds": 0.12639614700000035, "created_at": "2026-11-22T03:42:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4614996b408c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "784c3a38caa66f88c69c8b5a90c20b237e4013463d3474565f4dd1e4d0d2b7b7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4614996b408c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:42:12Z", "seed": 589082184, "treatment": "full"}, "priority": 2.003665349087453, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cf7bde8a7f1b8d75fbd457997f499347ba6fed2a5b22eddb7a81af2c85d4446e", "run_seconds": 0.31478309631347656, "seed": 589082184, "started_at": "2026-11-22T03:42:16Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:42:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4614996b408c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 1.0, 1.0, 1.0], "frozen": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 1.0, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "784c3a38caa66f88c69c8b5a90c20b237e4013463d3474565f4dd1e4d0d2b7b7", "primary_control": "facts", "result_hash": "cf7bde8a7f1b8d75fbd457997f499347ba6fed2a5b22eddb7a81af2c85d4446e", "signature": "9be7980e8f072fd01052be1b0e098defc95c3d73dbadcc5712341df87b144be41a4df1cf23653e429ff3bedda99f437b823936b270f426bb1e7a704d25d7630e", "state": "PARTIAL", "summary": {"facts": 1.0, "frozen": 1.0, "full": 1.0, "retrieval": 1.0, "rules_nodetect": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:42:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:12Z", "cpu_seconds": 4.912716803999999, "created_at": "2026-11-22T03:42:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-830e90c37250", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "dcd24abfc8598a6a6063e7f4f73ed278d76e87209a070bf4a8731663d7e9f20b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-830e90c37250", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:42:28Z", "seed": 365825597, "treatment": "frontier"}, "priority": 1.9941417941197985, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ae286e09953f6d94aec7c4dbd1a020ec34dc26b432ce32745616a9b66aff7983", "run_seconds": 5.032175779342651, "seed": 365825597, "started_at": "2026-11-22T03:44:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.056903, 0.014537], "effect": -0.021183, "evaluated_at": "2026-11-22T03:44:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-830e90c37250", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.50212766, 0.472340426, 0.506382979, 0.593406593], "frontier": [0.463829787, 0.438297872, 0.423404255, 0.331868132], "goal": [0.368085106, 0.387234043, 0.468085106, 0.342857143], "hypdisc": [0.444680851, 0.410638298, 0.480851064, 0.468131868], "infogain": [0.476595745, 0.438297872, 0.423404255, 0.49010989], "prederr": [0.476595745, 0.438297872, 0.423404255, 0.505494505], "random": [0.444680851, 0.438297872, 0.472340426, 0.386813187], "risk": [0.463829787, 0.438297872, 0.423404255, 0.193406593]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "dcd24abfc8598a6a6063e7f4f73ed278d76e87209a070bf4a8731663d7e9f20b", "primary_control": "random", "result_hash": "ae286e09953f6d94aec7c4dbd1a020ec34dc26b432ce32745616a9b66aff7983", "signature": "d7f87fc99ccde2dc3c9f446301f3d799f1b8d4d03d96276008ffd8823a21fda1c0a9be171a3c402d72e46b6eb6fd3c5b0e00df51aae0a8be2d59f6aedec44b00", "state": "VERIFIED", "summary": {"count": 0.518564, "frontier": 0.41435, "goal": 0.391565, "hypdisc": 0.451076, "infogain": 0.457102, "prederr": 0.460948, "random": 0.435533, "risk": 0.379735}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:44:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:54Z", "cpu_seconds": 12.474703082, "created_at": "2026-11-22T03:42:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e93e3ed79bb3", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 7, "n": 4}, "prereg_hash": "b86fd22ef8638e853839fbc786ab2b43c4f4af435497cdb1ec43ff8fbb1451da", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e93e3ed79bb3", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:42:28Z", "seed": 1905923638, "treatment": "frontier"}, "priority": 2.0096372491688963, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "22b236c6bbacc4fdd994168d898de43ee6c63f2bc4ba029aee09cc74277375a3", "run_seconds": 13.413541316986084, "seed": 1905923638, "started_at": "2026-11-22T03:42:40Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:42:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e93e3ed79bb3", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "adversarial", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b86fd22ef8638e853839fbc786ab2b43c4f4af435497cdb1ec43ff8fbb1451da", "primary_control": "random", "result_hash": "22b236c6bbacc4fdd994168d898de43ee6c63f2bc4ba029aee09cc74277375a3", "signature": "b307916aa9a1cdeedb68de1392b73cd685ee9d0c12ad55508537facda907dd3867e1a1650dd586c43b4d53c615efeda9dff9ce284478e2c94e4b6b357baa5509", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:42:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:23Z", "cpu_seconds": 0.23390618799999885, "created_at": "2026-11-22T03:42:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-0f9fb351d14e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "prereg_hash": "3f4cb1baaa2caf349134eb59c44c95f2e70232ae6f12946b8287e50fd8bdb341", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-0f9fb351d14e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:42:28Z", "seed": 1252402387, "treatment": "gated"}, "priority": 1.9898757352842464, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d3c128ca05e5274441bcb5b5ee38bea0bb3d96c52c33f3a181ba46abf1c1f117", "run_seconds": 0.24340510368347168, "seed": 1252402387, "started_at": "2026-11-22T03:44:23Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-0f9fb351d14e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.121212121, 0.106060606, 0.090909091, 0.106060606], "gated": [0.121212121, 0.106060606, 0.090909091, 0.106060606], "naive": [0.121212121, 0.106060606, 0.090909091, 0.106060606], "scratch": [0.121212121, 0.106060606, 0.090909091, 0.106060606], "ungated": [0.121212121, 0.106060606, 0.090909091, 0.106060606]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3f4cb1baaa2caf349134eb59c44c95f2e70232ae6f12946b8287e50fd8bdb341", "primary_control": "scratch", "result_hash": "d3c128ca05e5274441bcb5b5ee38bea0bb3d96c52c33f3a181ba46abf1c1f117", "signature": "1ac4a9bf971f134444350dcfd7bc814523cd572cfef15abbbcb8ce80e5196e07fd138b4e32d76255199b835b3032d7593b47b59d89008c31e20bd56b310b580d", "state": "PARTIAL", "summary": {"compressed": 0.106061, "gated": 0.106061, "naive": 0.106061, "scratch": 0.106061, "ungated": 0.106061}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:44:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:13Z", "cpu_seconds": 0.9917736909999988, "created_at": "2026-11-22T03:42:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2a628472909b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "43621bbfd8a9d20e3034b3238e6d2c9647b425332afb663e17d7da5245942d0e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2a628472909b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:42:28Z", "seed": 1296127189, "treatment": "gated"}, "priority": 1.9928613488329632, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "e374fccbfc72b409ef6e3ca02e802477bc577636a561bdb16be7ae2d8e9aad7f", "run_seconds": 1.0479731559753418, "seed": 1296127189, "started_at": "2026-11-22T03:44:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.065351, 0.023848], "effect": -0.020752, "evaluated_at": "2026-11-22T03:44:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2a628472909b", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.341176471, 0.115789474, 0.488888889, 0.178082192], "gated": [0.341176471, 0.115789474, 0.488888889, 0.178082192], "naive": [0.288235294, 0.123684211, 0.488888889, 0.178082192], "scratch": [0.335294118, 0.115789474, 0.577777778, 0.178082192], "ungated": [0.294117647, 0.123684211, 0.488888889, 0.175342466]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "43621bbfd8a9d20e3034b3238e6d2c9647b425332afb663e17d7da5245942d0e", "primary_control": "scratch", "result_hash": "e374fccbfc72b409ef6e3ca02e802477bc577636a561bdb16be7ae2d8e9aad7f", "signature": "98880e7f7c89ef1db0c570b43cf28b50a377e2dd142763a398ceb940892f6b6a1d630e2e71c4921327ee9437d5479838047f8db041c9ffe384778da0f98cd501", "state": "VERIFIED", "summary": {"compressed": 0.280984, "gated": 0.280984, "naive": 0.269723, "scratch": 0.301736, "ungated": 0.270508}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:44:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:13Z", "cpu_seconds": 2.3315906670000004, "created_at": "2026-11-22T03:42:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2e4b7833989d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lights|lights|1|9"}, "prereg_hash": "e105dd2e587af56a9f6baf7877d10ba515ca8d1cc835e2178cf0edc884d0a9c4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2e4b7833989d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lights|lights|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:42:28Z", "seed": 1271993868, "treatment": "gated"}, "priority": 1.9939587282861337, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "ebd56d3923c56abf4a9d90008b1c590ae89e56b023b2a38031385f1eebbceecd", "run_seconds": 2.4295642375946045, "seed": 1271993868, "started_at": "2026-11-22T03:44:11Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2e4b7833989d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [1.0, 1.0, 0.861111111, 0.854166667], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [1.0, 1.0, 0.861111111, 0.854166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e105dd2e587af56a9f6baf7877d10ba515ca8d1cc835e2178cf0edc884d0a9c4", "primary_control": "scratch", "result_hash": "ebd56d3923c56abf4a9d90008b1c590ae89e56b023b2a38031385f1eebbceecd", "signature": "3700e7af8ac443164a5b26d288c8661a3709ab1ef73ebe3752aaec3e6a81753daad1e57d01efe853f50cb1bb1aea0fc2a01a90c65f4727cf5561ae575389a503", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.928819, "scratch": 1.0, "ungated": 0.928819}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:44:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:02Z", "cpu_seconds": 6.961592733, "created_at": "2026-11-22T03:42:28Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0e5a549b9300", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "a7c3920ea5178fcb319e4eab9cca74a227ccc6e02c60b0078626c3a5e235923d", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0e5a549b9300", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:42:28Z", "seed": 907770757, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "a5225c972316094a58567f046ca3aeec531a58f985b92c417c731f603c9f6d68", "run_seconds": 7.9978392124176025, "seed": 907770757, "started_at": "2026-11-22T03:42:54Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 4}, "better": "higher", "ci95": [-0.136218, 0.036218], "effect": -0.05, "evaluated_at": "2026-11-22T03:43:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0e5a549b9300", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.95, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.028, "losses": 5, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.938", "prereg_hash": "a7c3920ea5178fcb319e4eab9cca74a227ccc6e02c60b0078626c3a5e235923d", "primary_control": "reset", "result_hash": "a5225c972316094a58567f046ca3aeec531a58f985b92c417c731f603c9f6d68", "signature": "c6d50a061c505f0686f66e583b2889715ecc12312a433229d326f42ba1d1452066283eab117af02c871e9c82b6bf5730072a9503bec9046bcf71b8a067ef5a07", "state": "VERIFIED", "summary": {"inherit": 0.7333333333333333, "reset": 0.7833333333333333, "root": 0.43333333333333335}, "ties": 53, "treatment": "inherit", "wins": 2}, "verified_at": "2026-11-22T03:43:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:47Z", "cpu_seconds": 7.3427459310000005, "created_at": "2026-11-22T03:42:28Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-14d8deb4a038", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d131bc25989cd6937a4be08cf8f9144b31907acea1fad9466ebd53d4b050e2f9", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-14d8deb4a038", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:42:28Z", "seed": 1271406552, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "75d4afa075f2c9be6f971f09cde4ad46386fcfa087640fff34315f32754380f6", "run_seconds": 7.691810131072998, "seed": 1271406552, "started_at": "2026-11-22T03:42:39Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.156298, 0.089632], "effect": -0.033333, "evaluated_at": "2026-11-22T03:42:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-14d8deb4a038", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6414113323179722, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "d131bc25989cd6937a4be08cf8f9144b31907acea1fad9466ebd53d4b050e2f9", "primary_control": "reset", "result_hash": "75d4afa075f2c9be6f971f09cde4ad46386fcfa087640fff34315f32754380f6", "signature": "dea435799a25f702afb26503352ddaf1a9a52b1b04e55abc7cf2e2a7cf6a8eb005ce5b543bce7febbd64f24768f63cfceb1c2843e11b39173bdd9dd164f32f0d", "state": "VERIFIED", "summary": {"inherit": 0.6333333333333333, "reset": 0.6666666666666666, "root": 0.36666666666666664}, "ties": 46, "treatment": "inherit", "wins": 6}, "verified_at": "2026-11-22T03:42:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:42:59Z", "cpu_seconds": 3.662015856, "created_at": "2026-11-22T03:42:45Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4a3de7e34988", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "2b95217209bfcf6783a2271590ce78bd5ad422c88bf842df354d83c59b4b6f24", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4a3de7e34988", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:42:45Z", "seed": 165997384, "treatment": "frontier"}, "priority": 2.008189941533071, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e96806dfcd4920ba3fabc8f29edc1300fc408c8afc52d6b82c1b5221e2ea2b9d", "run_seconds": 4.79778528213501, "seed": 165997384, "started_at": "2026-11-22T03:42:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.273008, 0.091593], "effect": -0.090707, "evaluated_at": "2026-11-22T03:43:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4a3de7e34988", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.348189415, 0.971153846, 0.412202381], "frontier": [0.982142857, 0.299442897, 0.528846154, 0.429315476], "goal": [0.928571429, 0.300139276, 0.605769231, 0.366071429], "hypdisc": [1.0, 0.336350975, 0.865384615, 0.370535714], "infogain": [0.982142857, 0.299442897, 0.528846154, 0.429315476], "prederr": [0.982142857, 0.299442897, 0.528846154, 0.429315476], "random": [1.0, 0.325905292, 0.894230769, 0.382440476], "risk": [0.982142857, 0.299442897, 0.528846154, 0.429315476]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "2b95217209bfcf6783a2271590ce78bd5ad422c88bf842df354d83c59b4b6f24", "primary_control": "random", "result_hash": "e96806dfcd4920ba3fabc8f29edc1300fc408c8afc52d6b82c1b5221e2ea2b9d", "signature": "1e4f43891cb4f237fd0849f2f40a244ebe9114b866d3a620ea6ea885c1066ee8ec06d8c535c588630de4f9ddfa8772999383267e29003915983d426eaead480f", "state": "VERIFIED", "summary": {"count": 0.682886, "frontier": 0.559937, "goal": 0.550138, "hypdisc": 0.643068, "infogain": 0.559937, "prederr": 0.559937, "random": 0.650644, "risk": 0.559937}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:43:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:11Z", "cpu_seconds": 0.8145913249999994, "created_at": "2026-11-22T03:43:02Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6c43d0e9849c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "f5b66f611959667e886926b3065b8590dd05ad358d2ed7736d455948629ab574", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6c43d0e9849c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:02Z", "seed": 559492643, "treatment": "frontier"}, "priority": 2.0028060521864273, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6fede75972e39c1500f7e1fe6ac09f80342a884a18c93a461adc74e966332bbc", "run_seconds": 1.0452079772949219, "seed": 559492643, "started_at": "2026-11-22T03:43:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003341, 0.057285], "effect": 0.026972, "evaluated_at": "2026-11-22T03:43:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6c43d0e9849c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.895833333, 1.0, 1.0, 0.979166667], "goal": [0.875, 1.0, 0.964285714, 0.979166667], "hypdisc": [0.9375, 1.0, 1.0, 0.979166667], "infogain": [0.895833333, 1.0, 1.0, 0.979166667], "prederr": [0.895833333, 1.0, 1.0, 0.979166667], "random": [0.895833333, 0.984375, 0.928571429, 0.958333333], "risk": [0.895833333, 1.0, 1.0, 0.979166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "f5b66f611959667e886926b3065b8590dd05ad358d2ed7736d455948629ab574", "primary_control": "random", "result_hash": "6fede75972e39c1500f7e1fe6ac09f80342a884a18c93a461adc74e966332bbc", "signature": "ec9eef32a7ce389fad69f93773f86e2d742ace82cf8830b66d48d53bfc012fa3d834264cc96d7e1d49096f390b212e1fb6b9690ceedfaf481e334ade8f498301", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.96875, "goal": 0.954613, "hypdisc": 0.979167, "infogain": 0.96875, "prederr": 0.96875, "random": 0.941778, "risk": 0.96875}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:43:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:33Z", "cpu_seconds": 22.26695354, "created_at": "2026-11-22T03:43:02Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-22fc75aa7ed3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 7, "n": 4}, "prereg_hash": "17a3c17a3bed1c78ba0b14e477ce555f58c71441ae86ac49e9b8fb901a47f0b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-22fc75aa7ed3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:43:02Z", "seed": 1871530208, "treatment": "full"}, "priority": 2.0045050428855586, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b744e8fa20254b5ce6b691037155097c973996633e2d3d770e5c57a1189bdf75", "run_seconds": 28.10171675682068, "seed": 1871530208, "started_at": "2026-11-22T03:43:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.483549, 0.766451], "effect": 0.625, "evaluated_at": "2026-11-22T03:43:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-22fc75aa7ed3", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.25, 0.0, 0.5], "fixed_policy": [0.25, 0.25, 0.0, 0.0], "full": [1.0, 0.75, 0.75, 1.0], "no_world_model": [0.25, 0.25, 0.75, 0.5], "random": [1.0, 0.25, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "17a3c17a3bed1c78ba0b14e477ce555f58c71441ae86ac49e9b8fb901a47f0b6", "primary_control": "e141_best", "result_hash": "b744e8fa20254b5ce6b691037155097c973996633e2d3d770e5c57a1189bdf75", "signature": "3e570b6a734da6b303a3c68593bf8dd86da3c76de595f5305543029b2553750ead2b3175e79cb19ce90467e374159393214fbd25771b53ed8f089f96703f4e00", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.125, "full": 0.875, "no_world_model": 0.4375, "random": 0.5625}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:44:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:05Z", "cpu_seconds": 0.5169775830000001, "created_at": "2026-11-22T03:43:02Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1586e9a1ec3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "486cce257090946b1d5356c19b1f60efd284a3f6b32ae4474d9617950e7c2cb6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1586e9a1ec3b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:43:02Z", "seed": 1138462845, "treatment": "full"}, "priority": 2.0080976422368506, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f372fd825d1e51b24e6dd3457212215d01930a4392be93e7939d43c8e3e39dd0", "run_seconds": 0.5412039756774902, "seed": 1138462845, "started_at": "2026-11-22T03:43:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.017373, 0.102725], "effect": 0.060049, "evaluated_at": "2026-11-22T03:43:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1586e9a1ec3b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.745098039, 0.715686275, 0.789215686, 0.754901961], "frozen": [0.068627451, 0.117647059, 0.068627451, 0.068627451], "full": [0.823529412, 0.828431373, 0.808823529, 0.784313725], "retrieval": [0.848039216, 0.87254902, 0.848039216, 0.833333333], "rules_nodetect": [0.823529412, 0.828431373, 0.808823529, 0.784313725]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "486cce257090946b1d5356c19b1f60efd284a3f6b32ae4474d9617950e7c2cb6", "primary_control": "facts", "result_hash": "f372fd825d1e51b24e6dd3457212215d01930a4392be93e7939d43c8e3e39dd0", "signature": "d3ff48a78c4a0ea6aa673eef83c659c8521729d2ab50135f60e4cabd656a0fc7919e6ab90df84f4d3bc5db34cb83f8d7c68b88e79c814eed4cd71b74f0e2430f", "state": "VERIFIED", "summary": {"facts": 0.751225, "frozen": 0.080882, "full": 0.811275, "retrieval": 0.85049, "rules_nodetect": 0.811275}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:43:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:18Z", "cpu_seconds": 9.652194025, "created_at": "2026-11-22T03:43:02Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-244855503a7f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "745ab0f2740a22cfd3e0524e03f7e1ae8e061e017a9ad84b10e74d5f632ec184", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-244855503a7f", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:43:02Z", "seed": 1564866444, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "236a57aed0189448e07fecdac98f342b8faafb122484eda4324ff9e499a8f8d0", "run_seconds": 10.192275047302246, "seed": 1564866444, "started_at": "2026-11-22T03:43:07Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:43:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-244855503a7f", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "745ab0f2740a22cfd3e0524e03f7e1ae8e061e017a9ad84b10e74d5f632ec184", "primary_control": "reset", "result_hash": "236a57aed0189448e07fecdac98f342b8faafb122484eda4324ff9e499a8f8d0", "signature": "e82f922b279604a5b935a12b7423aeb7fe58b5ba084dd5ef8a13a35a819eb5019570b6538b38d61ac086ce4b2b401089e5b63aa17ae7457a939a4f03b1e7b80b", "state": "PARTIAL", "summary": {"inherit": 0.6333333333333333, "reset": 0.6333333333333333, "root": 0.3}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:43:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:01Z", "cpu_seconds": 1.7334206499999993, "created_at": "2026-11-22T03:43:19Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-104a31beabfa", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "c58cf72d388c941bafd804b0bca2117f319f90129c7007522c4d10cab8340f3b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-104a31beabfa", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 400, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:43:19Z", "seed": 793063894, "treatment": "iw_temperature"}, "priority": 1.9960241793861369, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-c7cd6d2c2c2c", "result_hash": "4569e755b6edc720d205a4e11066e7dcd3662847b6541179700d97026e6f65a6", "run_seconds": 3.19530987739563, "seed": 793063894, "started_at": "2026-11-22T03:43:58Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.021139, -0.003999], "effect": -0.012569, "evaluated_at": "2026-11-22T03:44:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-104a31beabfa", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.010767918, 0.003266615, 0.030547423, 0.008794248, 0.061206425, 0.057707993, 0.052108598, 0.066623585, 0.026674695, 0.03196516, 0.06415065, 0.040706072, 0.082584167, 0.033690715, 0.07351197, 0.01187039], "raw": [0.013716323, 0.009211195, 0.056493518, 0.012809425, 0.031326892, 0.057013897, 0.063374063, 0.04161361, 0.03764883, 0.046447928, 0.028037098, 0.035650233, 0.037663515, 0.03326724, 0.051479565, 0.020350407], "temperature": [0.009296692, 0.007564833, 0.030547423, 0.013137055, 0.0150192, 0.057013897, 0.03542348, 0.04161361, 0.028939383, 0.026765303, 0.026561025, 0.031769332, 0.037663515, 0.030405203, 0.051479565, 0.01187039]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0028, "losses": 11, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.994", "prereg_hash": "c58cf72d388c941bafd804b0bca2117f319f90129c7007522c4d10cab8340f3b", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-c7cd6d2c2c2c", "result_hash": "4569e755b6edc720d205a4e11066e7dcd3662847b6541179700d97026e6f65a6", "signature": "a0d3e9c395eeb3bbe4688e433b22186f665b70574615770e63d88f81e7d2e88167587d7339f1a4018d9fe1657386fa83dfb191caaa3a5f8d86a45bafa67ff20d", "state": "VERIFIED", "summary": {"iw_temperature": 0.041011, "raw": 0.036006, "temperature": 0.028442}, "ties": 2, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:44:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:37Z", "cpu_seconds": 6.873150156000001, "created_at": "2026-11-22T03:43:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6b6b8506d51a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 3, "n": 4}, "prereg_hash": "a1af3b90c24ab625970c829e9f8b5fe19f397b8a1d299885e4a1f960b1163dec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6b6b8506d51a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:19Z", "seed": 957061840, "treatment": "frontier"}, "priority": 2.0068893912405916, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1fe5a2157ac1fe6f5175a4ff9b05cb0314179105c6fc641fdd43460f9cfb4938", "run_seconds": 7.771984577178955, "seed": 957061840, "started_at": "2026-11-22T03:43:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037882, 0.162522], "effect": 0.06232, "evaluated_at": "2026-11-22T03:44:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6b6b8506d51a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.632183908, 0.986111111, 0.819444444, 0.365900383], "frontier": [0.798850575, 0.979166667, 0.722222222, 0.415708812], "goal": [0.687739464, 0.895833333, 0.708333333, 0.415708812], "hypdisc": [0.689655172, 1.0, 0.722222222, 0.404214559], "infogain": [0.798850575, 0.979166667, 0.722222222, 0.415708812], "prederr": [0.798850575, 0.979166667, 0.722222222, 0.415708812], "random": [0.603448276, 0.951388889, 0.770833333, 0.340996169], "risk": [0.798850575, 0.979166667, 0.722222222, 0.415708812]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "a1af3b90c24ab625970c829e9f8b5fe19f397b8a1d299885e4a1f960b1163dec", "primary_control": "random", "result_hash": "1fe5a2157ac1fe6f5175a4ff9b05cb0314179105c6fc641fdd43460f9cfb4938", "signature": "f97346f15492ff59e5ce1e661be7005457bcdad24529227757f8738ec480ea66fde1789ebfb3bc2f31fa603190bc1492cbad0eb2ad95226784aa3804d4f66101", "state": "VERIFIED", "summary": {"count": 0.70091, "frontier": 0.728987, "goal": 0.676904, "hypdisc": 0.704023, "infogain": 0.728987, "prederr": 0.728987, "random": 0.666667, "risk": 0.728987}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:44:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:46Z", "cpu_seconds": 2.1865896640000013, "created_at": "2026-11-22T03:43:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c4e8ca5c308d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 10, "n": 4}, "prereg_hash": "fbd06ffb6e6b23d46d5a7e220a0c65c83e6f3900e3fd79fd5dbfd70b9eb8cca7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c4e8ca5c308d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:19Z", "seed": 1744226122, "treatment": "frontier"}, "priority": 2.0003570081489723, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4fe9e2ab141a15256670a77e833ea57e723ca268441f708be75e5a84e41ec996", "run_seconds": 3.875638246536255, "seed": 1744226122, "started_at": "2026-11-22T03:43:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.060957, 0.116148], "effect": 0.027596, "evaluated_at": "2026-11-22T03:44:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c4e8ca5c308d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.459027778, 0.417162698, 0.598148148, 0.4859375], "frontier": [0.523611111, 0.459325397, 0.52962963, 0.394791667], "goal": [0.523611111, 0.459325397, 0.52962963, 0.394791667], "hypdisc": [0.599305556, 0.493055556, 0.591666667, 0.394791667], "infogain": [0.552777778, 0.459325397, 0.616666667, 0.394791667], "prederr": [0.5375, 0.459325397, 0.383333333, 0.394791667], "random": [0.392361111, 0.411210317, 0.511111111, 0.482291667], "risk": [0.523611111, 0.459325397, 0.52962963, 0.394791667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "fbd06ffb6e6b23d46d5a7e220a0c65c83e6f3900e3fd79fd5dbfd70b9eb8cca7", "primary_control": "random", "result_hash": "4fe9e2ab141a15256670a77e833ea57e723ca268441f708be75e5a84e41ec996", "signature": "6413cd955932bed33cb3ad3f1b398fc1a9dc28e08904d6007c708c6d42add52a85ee1cbd47916b2b7d540cb08255db84f20498a9cf31f498dcff5cf0eee29305", "state": "VERIFIED", "summary": {"count": 0.490069, "frontier": 0.476839, "goal": 0.476839, "hypdisc": 0.519705, "infogain": 0.50589, "prederr": 0.443738, "random": 0.449244, "risk": 0.476839}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:44:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:24Z", "cpu_seconds": 1.2457829829999998, "created_at": "2026-11-22T03:43:19Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3072b608d87d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "prereg_hash": "49649264d500e04ce9ce204b4ba7e4277b9a8a979fb9d1a1d9d7006744dfbf84", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3072b608d87d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:43:19Z", "seed": 1536570198, "treatment": "gated"}, "priority": 2.008076586077058, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "74c8cba6a594b932378baa27c24c20a802912b51cb00ee3e4aa49afb0a91d1c6", "run_seconds": 1.5733387470245361, "seed": 1536570198, "started_at": "2026-11-22T03:43:22Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:43:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-3072b608d87d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.783333333, 0.388505747, 0.875, 0.141414141], "gated": [0.783333333, 0.388505747, 0.875, 0.141414141], "naive": [0.783333333, 0.388505747, 0.775, 0.163636364], "scratch": [0.783333333, 0.388505747, 0.875, 0.141414141], "ungated": [0.783333333, 0.333333333, 0.675, 0.163636364]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "49649264d500e04ce9ce204b4ba7e4277b9a8a979fb9d1a1d9d7006744dfbf84", "primary_control": "scratch", "result_hash": "74c8cba6a594b932378baa27c24c20a802912b51cb00ee3e4aa49afb0a91d1c6", "signature": "f92d2376298a9dfd669399317a41aeae534d0a9c3e30277bee91eb056374d4bbbf50a4107a3f0887b283f642d4e17894ee33a40302a3c54761ad2e6557ce9800", "state": "PARTIAL", "summary": {"compressed": 0.547063, "gated": 0.547063, "naive": 0.527619, "scratch": 0.547063, "ungated": 0.488826}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:43:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:32Z", "cpu_seconds": 6.983575092999999, "created_at": "2026-11-22T03:43:19Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-631cf230c09e", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6451945eee9bd81faf3fef5f446aabb46a0808d7b324292a551e2f2636237cbc", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-631cf230c09e", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:43:19Z", "seed": 572397977, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9e9a5099cc352d6ba842b06c88238b93c9b381123b69fd681d4bcd4d3113d329", "run_seconds": 9.939548969268799, "seed": 572397977, "started_at": "2026-11-22T03:43:23Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 1, "reset": 1}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:43:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-631cf230c09e", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6451945eee9bd81faf3fef5f446aabb46a0808d7b324292a551e2f2636237cbc", "primary_control": "reset", "result_hash": "9e9a5099cc352d6ba842b06c88238b93c9b381123b69fd681d4bcd4d3113d329", "signature": "06d30a8fc59d230e68ba34b08312fb71d3996edbfbc628905b982c6f0f9886600d1dd9d6fd6238b220c262ceb1c5dfde3d8f8196f4c3653cbf2c760bd930c907", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.43333333333333335}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:43:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:02Z", "cpu_seconds": 8.948677421, "created_at": "2026-11-22T03:43:36Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9f87c4d490e9", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 7, "n": 4}, "prereg_hash": "d3e7d7983391c57550627d60b2ebdd6db546ed89f947b3d0d8963e925dfe4ab2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9f87c4d490e9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:36Z", "seed": 1220405852, "treatment": "frontier"}, "priority": 1.999504008952875, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-27c7d8e7d46e", "result_hash": "d32180b8e7b6773156d74538e516e3c589cc53ea34d17258fb215d0b9f1189b6", "run_seconds": 10.896010398864746, "seed": 1220405852, "started_at": "2026-11-22T03:43:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.090419, 0.005677], "effect": -0.042371, "evaluated_at": "2026-11-22T03:44:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9f87c4d490e9", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.463473054, 0.381355932, 0.330729167, 0.492028986], "frontier": [0.401796407, 0.328813559, 0.274479167, 0.273913043], "goal": [0.341916168, 0.340677966, 0.3, 0.285507246], "hypdisc": [0.40239521, 0.362711864, 0.297916667, 0.383333333], "infogain": [0.401796407, 0.328813559, 0.274479167, 0.273913043], "prederr": [0.401796407, 0.328813559, 0.274479167, 0.273913043], "random": [0.411377246, 0.347457627, 0.300520833, 0.389130435], "risk": [0.401796407, 0.328813559, 0.274479167, 0.273913043]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d3e7d7983391c57550627d60b2ebdd6db546ed89f947b3d0d8963e925dfe4ab2", "primary_control": "random", "replicated": false, "replication_of": "E141-ow_explore-27c7d8e7d46e", "result_hash": "d32180b8e7b6773156d74538e516e3c589cc53ea34d17258fb215d0b9f1189b6", "signature": "f15e0204677923f91bd8d176542e331a05961f0c129ee0f9c1165cd71ac611323fbdc1eb26ddc28115b0ba8a4bb3486dcfa95b30b38e0d08252798eac40c8202", "state": "VERIFIED", "summary": {"count": 0.416897, "frontier": 0.319751, "goal": 0.317025, "hypdisc": 0.361589, "infogain": 0.319751, "prederr": 0.319751, "random": 0.362122, "risk": 0.319751}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:44:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:53Z", "cpu_seconds": 11.583871367, "created_at": "2026-11-22T03:43:36Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-cd45e13524cb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 7, "n": 4}, "prereg_hash": "91a5c34c4225660ef00ee7edb77daa6feef992d986080c135082403fa79df3a8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-cd45e13524cb", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:36Z", "seed": 630890955, "treatment": "frontier"}, "priority": 2.002397086808018, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "801a7e561e3f3b0d1a9fc1542871c66768666f6a529dfc50fa0948dd471a3d84", "run_seconds": 14.524779319763184, "seed": 630890955, "started_at": "2026-11-22T03:43:38Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-cd45e13524cb", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "91a5c34c4225660ef00ee7edb77daa6feef992d986080c135082403fa79df3a8", "primary_control": "random", "result_hash": "801a7e561e3f3b0d1a9fc1542871c66768666f6a529dfc50fa0948dd471a3d84", "signature": "afca4b03fae9b303b106a1211e31cb16eb77e6d8d0971351a8a44a18384cce5012a98db549f0f8480b9e69033a58e8b585b76b75faff6cad8bc631d28282210d", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:44:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:43:46Z", "cpu_seconds": 6.573951870000002, "created_at": "2026-11-22T03:43:36Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-819744ee184e", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "84253d7400caa85da0ba509e37adb0ff849aebc5c166f9b9c13577854fc6f07c", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-819744ee184e", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:43:36Z", "seed": 1593141921, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "be6c259334d8662d95958c406273053eef99d050d9f130ef03ade777f2695b2a", "run_seconds": 8.501245498657227, "seed": 1593141921, "started_at": "2026-11-22T03:43:37Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-819744ee184e", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "84253d7400caa85da0ba509e37adb0ff849aebc5c166f9b9c13577854fc6f07c", "primary_control": "reset", "result_hash": "be6c259334d8662d95958c406273053eef99d050d9f130ef03ade777f2695b2a", "signature": "f7d05c0ecf1acb9a92b6bb6b205a6a5aec271adefa0aa35e9f50f20a4cbbb6667cbc71a5d67ffe025a12a2ff9da29f6433e6821533f16eeebe0763ccecccc803", "state": "PARTIAL", "summary": {"inherit": 0.55, "reset": 0.55, "root": 0.31666666666666665}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:44:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:06Z", "cpu_seconds": 5.819508117999998, "created_at": "2026-11-22T03:43:52Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-9a746fb76e61", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "150d4efccd57cb373b4ab6a18325a13eff79a5691b56e492a318246177298a2a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-9a746fb76e61", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:43:52Z", "seed": 1089273448, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "ab87a08235ac85d1e922dc8d6561b0fa21b6972abdb59c5107ee02571cba8b79", "run_seconds": 7.5860865116119385, "seed": 1089273448, "started_at": "2026-11-22T03:43:58Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-9a746fb76e61", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "150d4efccd57cb373b4ab6a18325a13eff79a5691b56e492a318246177298a2a", "primary_control": "reset", "result_hash": "ab87a08235ac85d1e922dc8d6561b0fa21b6972abdb59c5107ee02571cba8b79", "signature": "c70a93ba0520b5c8a2e37b4cb3c7e0c69ae9b600c7df7749777868ce392d3a86a37b2e3c34787ff8882465fa720ffcbb711e06bad75b6750c3ba4636b8d59903", "state": "PARTIAL", "summary": {"inherit": 0.6833333333333333, "reset": 0.6833333333333333, "root": 0.4166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:44:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:07Z", "cpu_seconds": 0.7994918260000006, "created_at": "2026-11-22T03:43:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4c30b0094249", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "fd5db41ba508124e8ef6cfd20620d6ba50ff13c310d3f7b24d79e4c4d68a913b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4c30b0094249", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:52Z", "seed": 1810657401, "treatment": "frontier"}, "priority": 1.9955054531954373, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "feb21636e0c392bcfd89601b734923af4a2533232e4abb0af67c0c6bd005cc8b", "run_seconds": 0.8057496547698975, "seed": 1810657401, "started_at": "2026-11-22T03:44:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.030982, 0.109018], "effect": 0.07, "evaluated_at": "2026-11-22T03:44:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4c30b0094249", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 0.986666667, 1.0, 1.0], "goal": [0.96, 0.893333333, 0.893333333, 0.973333333], "hypdisc": [0.986666667, 1.0, 0.96, 1.0], "infogain": [1.0, 0.986666667, 1.0, 1.0], "prederr": [1.0, 0.986666667, 1.0, 1.0], "random": [0.893333333, 0.973333333, 0.92, 0.92], "risk": [1.0, 0.986666667, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "fd5db41ba508124e8ef6cfd20620d6ba50ff13c310d3f7b24d79e4c4d68a913b", "primary_control": "random", "result_hash": "feb21636e0c392bcfd89601b734923af4a2533232e4abb0af67c0c6bd005cc8b", "signature": "a6b7ed547170e52ac1b03754b52e42b72a6b6e5f07d68b8c02d26a6bf5f355b1bc99b1d9d4a806a5ce5267af2e7602171c21e684c0f45f8eb184fb7da4268d05", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.996667, "goal": 0.93, "hypdisc": 0.986667, "infogain": 0.996667, "prederr": 0.996667, "random": 0.926667, "risk": 0.996667}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:44:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:24Z", "cpu_seconds": 0.9692857700000062, "created_at": "2026-11-22T03:43:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9c12be0153c6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "fb726b5158b691c3de0a09724fccb197c94ef5fab6b6f5127a307978fdfd3941", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9c12be0153c6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:53Z", "seed": 304248315, "treatment": "frontier"}, "priority": 1.9824770703097963, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ec1d6e3074f1c7238a7918ca0fe8b0f34eea48404c2c9cd2bd9140ebd787e387", "run_seconds": 1.0085957050323486, "seed": 304248315, "started_at": "2026-11-22T03:44:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.186443, 0.278231], "effect": 0.232337, "evaluated_at": "2026-11-22T03:44:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9c12be0153c6", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.989130435, 0.994565217, 1.0], "frontier": [1.0, 1.0, 0.994565217, 0.97826087], "goal": [0.668478261, 0.581521739, 0.72826087, 0.663043478], "hypdisc": [0.782608696, 0.630434783, 0.777173913, 0.820652174], "infogain": [1.0, 1.0, 0.994565217, 0.97826087], "prederr": [1.0, 1.0, 0.994565217, 0.97826087], "random": [0.717391304, 0.739130435, 0.793478261, 0.793478261], "risk": [1.0, 1.0, 0.994565217, 0.97826087]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "fb726b5158b691c3de0a09724fccb197c94ef5fab6b6f5127a307978fdfd3941", "primary_control": "random", "result_hash": "ec1d6e3074f1c7238a7918ca0fe8b0f34eea48404c2c9cd2bd9140ebd787e387", "signature": "42e9cc0c7c605859c5e2a9ac9f67f3e0fae45fed381eec7407795e1cd8cc18fe47f88b7b9f09342fc1fe7ef2db8cdd92f5a05fed133dffba9e2aae5424c17508", "state": "VERIFIED", "summary": {"count": 0.995924, "frontier": 0.993207, "goal": 0.660326, "hypdisc": 0.752717, "infogain": 0.993207, "prederr": 0.993207, "random": 0.76087, "risk": 0.993207}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:44:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:18Z", "cpu_seconds": 0.4891258260000022, "created_at": "2026-11-22T03:43:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c4c08bd7aaf0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "1c17365658e01d28a7489c39deddfab328f1c9cd49ea2791a1f0360665dcdab7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c4c08bd7aaf0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:43:53Z", "seed": 1016430086, "treatment": "frontier"}, "priority": 1.992820379469088, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "58c42a49974688f7c118e949b1d0bf7181b2b8c2596c96daedbcb78a7755a679", "run_seconds": 0.4909977912902832, "seed": 1016430086, "started_at": "2026-11-22T03:44:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.373447, 0.238031], "effect": -0.067708, "evaluated_at": "2026-11-22T03:44:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c4c08bd7aaf0", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.9375, 0.916666667, 0.895833333, 0.916666667], "frontier": [0.958333333, 0.708333333, 0.291666667, 0.958333333], "goal": [0.8125, 0.8125, 0.875, 0.833333333], "hypdisc": [0.916666667, 0.791666667, 0.958333333, 0.979166667], "infogain": [0.895833333, 0.8125, 0.75, 0.916666667], "prederr": [0.729166667, 0.0625, 0.75, 0.875], "random": [0.791666667, 0.729166667, 0.8125, 0.854166667], "risk": [0.958333333, 0.708333333, 0.291666667, 0.958333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "1c17365658e01d28a7489c39deddfab328f1c9cd49ea2791a1f0360665dcdab7", "primary_control": "random", "result_hash": "58c42a49974688f7c118e949b1d0bf7181b2b8c2596c96daedbcb78a7755a679", "signature": "e9cb873c81bd8759eb36528a38dff47383c6c515c83d9945280d7afb383bb7a073bf7a9d2c555e3f7f79d05a8b49f9407b629046511bb4af6e83c8b7d2a83b09", "state": "VERIFIED", "summary": {"count": 0.916667, "frontier": 0.729167, "goal": 0.833333, "hypdisc": 0.911458, "infogain": 0.84375, "prederr": 0.604167, "random": 0.796875, "risk": 0.729167}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:44:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:21Z", "cpu_seconds": 2.459646847000002, "created_at": "2026-11-22T03:43:53Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c20cb0045770", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 8, "n": 4}, "prereg_hash": "727462a36d6f70c5a2d0e55c707d0a335eeaa54781ec37c2467665b06dfae903", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c20cb0045770", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:43:53Z", "seed": 585678982, "treatment": "full"}, "priority": 1.9902039181888023, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5cde8807b99de5fb8b10369a49f1ae9b219475d288789a8c1b16bf507b1908ce", "run_seconds": 2.646644115447998, "seed": 585678982, "started_at": "2026-11-22T03:44:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00834, 0.133772], "effect": 0.062716, "evaluated_at": "2026-11-22T03:44:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c20cb0045770", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.965517241, 1.0, 0.435714286, 0.697777778], "frozen": [0.379310345, 0.358974359, 0.110714286, 0.126666667], "full": [0.972413793, 1.0, 0.528571429, 0.848888889], "retrieval": [0.986206897, 1.0, 0.692857143, 0.84], "rules_nodetect": [0.972413793, 1.0, 0.528571429, 0.848888889]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "727462a36d6f70c5a2d0e55c707d0a335eeaa54781ec37c2467665b06dfae903", "primary_control": "facts", "result_hash": "5cde8807b99de5fb8b10369a49f1ae9b219475d288789a8c1b16bf507b1908ce", "signature": "ec55c4129158db260eb30d7f3ae7fcb0a3b5a7f246d5854a632bc2caa3f51b60013a54581cb10509c782b2c770379334707165361be27df5e89a1d68dfa92706", "state": "VERIFIED", "summary": {"facts": 0.774752, "frozen": 0.243916, "full": 0.837469, "retrieval": 0.879766, "rules_nodetect": 0.837469}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:44:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:29Z", "cpu_seconds": 0.739389709000001, "created_at": "2026-11-22T03:44:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-049e313ba9c5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "67ec985190f27b70db491d026f471a675efdc00a1c5af58c257c121745ee9c76", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-049e313ba9c5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:44:09Z", "seed": 972568201, "treatment": "frontier"}, "priority": 0.5515529550858504, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "05acca7e36cde073d9d9b3a3161266fa42dee0c45ffaaf78de29f93627b70e53", "run_seconds": 0.7716493606567383, "seed": 972568201, "started_at": "2026-11-22T03:44:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.103916, 0.030999], "effect": -0.036458, "evaluated_at": "2026-11-22T03:44:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-049e313ba9c5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.9375, 1.0, 0.916666667], "frontier": [1.0, 0.854166667, 0.958333333, 0.8125], "goal": [1.0, 0.854166667, 0.958333333, 0.8125], "hypdisc": [1.0, 0.9375, 0.958333333, 0.875], "infogain": [1.0, 0.895833333, 0.958333333, 0.833333333], "prederr": [1.0, 0.895833333, 0.958333333, 0.895833333], "random": [0.958333333, 0.9375, 0.958333333, 0.916666667], "risk": [1.0, 0.854166667, 0.958333333, 0.8125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "67ec985190f27b70db491d026f471a675efdc00a1c5af58c257c121745ee9c76", "primary_control": "random", "result_hash": "05acca7e36cde073d9d9b3a3161266fa42dee0c45ffaaf78de29f93627b70e53", "signature": "d6887ffa5ed56d85d70720e2834fd235f1e9f2b1e77140e74dcb0d63fb377fc0a53b5a93ae9e5ebeed443ccaf6d75158e95f3d54839291d107723ddb7eb88f02", "state": "VERIFIED", "summary": {"count": 0.963542, "frontier": 0.90625, "goal": 0.90625, "hypdisc": 0.942708, "infogain": 0.921875, "prederr": 0.9375, "random": 0.942708, "risk": 0.90625}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:44:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:18Z", "cpu_seconds": 0.1735522439999997, "created_at": "2026-11-22T03:44:09Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4753193c754e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "d4b9bf9556d34b4268de3c0c874f70d6203c30dce968368855988ada826a9c4c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4753193c754e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:09Z", "seed": 1123271333, "treatment": "full"}, "priority": 1.9917394829973012, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0b99ebbf8d459fcbc868af08adeaa155ea58d400ba5584ce1639cb6fce5ff92b", "run_seconds": 0.18031835556030273, "seed": 1123271333, "started_at": "2026-11-22T03:44:18Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4753193c754e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.75, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 0.75], "random": [1.0, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d4b9bf9556d34b4268de3c0c874f70d6203c30dce968368855988ada826a9c4c", "primary_control": "e141_best", "result_hash": "0b99ebbf8d459fcbc868af08adeaa155ea58d400ba5584ce1639cb6fce5ff92b", "signature": "5eedc61055268a80dc11dfaf005b76f61e1829364644234263c9809428208f978e153ae4b15b129a48c14b5e219c52af6c3c712ac7aa737c4958bc8ff3f1d207", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 0.9375, "random": 0.9375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:44:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:30Z", "cpu_seconds": 0.4680245510000063, "created_at": "2026-11-22T03:44:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4f6eccb98fd3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "ba4ce158b2b2576b01bfd3e2af4e88441c10ffe0c9981a80dd82976868cf3ad0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4f6eccb98fd3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:44:09Z", "seed": 1369752140, "treatment": "full"}, "priority": 0.5508315654491601, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "67212a6c7f348f113b4a2f2cf9a053f1ede831afa33c1d6c36a67337143a7a5f", "run_seconds": 0.503565788269043, "seed": 1369752140, "started_at": "2026-11-22T03:44:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005492, 0.084075], "effect": 0.039292, "evaluated_at": "2026-11-22T03:44:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4f6eccb98fd3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.666666667, 0.791666667, 0.692810458, 0.441860465], "frozen": [0.372093023, 0.395833333, 0.183006536, 0.426356589], "full": [0.751937984, 0.791666667, 0.764705882, 0.441860465], "retrieval": [0.751937984, 0.791666667, 0.692810458, 0.441860465], "rules_nodetect": [0.821705426, 0.791666667, 0.908496732, 0.441860465]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "ba4ce158b2b2576b01bfd3e2af4e88441c10ffe0c9981a80dd82976868cf3ad0", "primary_control": "facts", "result_hash": "67212a6c7f348f113b4a2f2cf9a053f1ede831afa33c1d6c36a67337143a7a5f", "signature": "2c46f7e24a0aa7f1c2d6230d003cd5b135f6b6757ebdbb6253e0ae9bbc977dc5eadaa50ae907a91ba6dee4c6e4d369b4ef6c10409890a0cd8cd841044ab5b908", "state": "VERIFIED", "summary": {"facts": 0.648251, "frozen": 0.344322, "full": 0.687543, "retrieval": 0.669569, "rules_nodetect": 0.740932}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:44:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:32Z", "cpu_seconds": 0.615760212000005, "created_at": "2026-11-22T03:44:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1be24f7f836d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "5c1c6c41f0cbfa3551b78cbd4670cd08099f7348288a8517490e36a1002e72ea", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1be24f7f836d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:25Z", "seed": 280523234, "treatment": "full"}, "priority": 0.5503786402278734, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "345330d0d781b792e2fd9bbcb231a15b2bd2e21692f936c866981633e9c6ca2c", "run_seconds": 0.6376636028289795, "seed": 280523234, "started_at": "2026-11-22T03:44:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:44:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1be24f7f836d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5c1c6c41f0cbfa3551b78cbd4670cd08099f7348288a8517490e36a1002e72ea", "primary_control": "e141_best", "result_hash": "345330d0d781b792e2fd9bbcb231a15b2bd2e21692f936c866981633e9c6ca2c", "signature": "6f7a340310e1d16e4893547b6ae10c5cf2042c197ec0fdcd91faedf2595d8a0e581f533c54a612501fa76458c45d9ba639c6ac33beeeaf498eb25cb7d82a5805", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.9375, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:44:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:41Z", "cpu_seconds": 0.40997330700000134, "created_at": "2026-11-22T03:44:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4a982e9fb73e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "d79807e4bdaba6a6040ecfe9e0450a953ee7157fd12819d5f86e5b6bec168cb0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4a982e9fb73e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:25Z", "seed": 1387392569, "treatment": "full"}, "priority": 0.5442274774527376, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "3992b059ee272a0a04590e853ba66425d4da2a4446171e07e75fb2a996771e2e", "run_seconds": 0.5700767040252686, "seed": 1387392569, "started_at": "2026-11-22T03:44:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.74, 0.24], "effect": -0.25, "evaluated_at": "2026-11-22T03:44:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4a982e9fb73e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.75, 1.0, 0.75], "fixed_policy": [0.25, 0.25, 0.25, 0.0], "full": [0.0, 0.75, 0.0, 0.75], "no_world_model": [0.75, 0.75, 0.0, 0.75], "random": [0.75, 0.75, 0.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d79807e4bdaba6a6040ecfe9e0450a953ee7157fd12819d5f86e5b6bec168cb0", "primary_control": "e141_best", "result_hash": "3992b059ee272a0a04590e853ba66425d4da2a4446171e07e75fb2a996771e2e", "signature": "8c132ca2ba0e2b980ea4ab6645f7b00f5b62788890f2ea7b64666f3d3b93a8ca9ed52d9da9439fa4afb8b94215baeac473ec76cd4e423a5186e36ba456248d08", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.1875, "full": 0.375, "no_world_model": 0.5625, "random": 0.5625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:44:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:39Z", "cpu_seconds": 1.693146382000002, "created_at": "2026-11-22T03:44:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5f97a7faf803", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "711c9a0264430c8fc8d5a3a10beb63dd88cdec29f3ecac7a24c8a6116b9bfbff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5f97a7faf803", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:25Z", "seed": 1483545643, "treatment": "full"}, "priority": 0.545093919164603, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "3ab13753bf430fc2c592f9b5afbafc297546060e9545bb68671205949503f4f1", "run_seconds": 1.955394983291626, "seed": 1483545643, "started_at": "2026-11-22T03:44:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.032902, 0.532902], "effect": 0.25, "evaluated_at": "2026-11-22T03:44:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5f97a7faf803", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.0, 0.5], "fixed_policy": [0.0, 0.25, 0.0, 0.25], "full": [1.0, 0.0, 0.5, 1.0], "no_world_model": [1.0, 0.0, 0.75, 0.5], "random": [1.0, 0.0, 0.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "711c9a0264430c8fc8d5a3a10beb63dd88cdec29f3ecac7a24c8a6116b9bfbff", "primary_control": "e141_best", "result_hash": "3ab13753bf430fc2c592f9b5afbafc297546060e9545bb68671205949503f4f1", "signature": "e1a3a43106acf008a2bbdc8452c0c61febafa39aae082ed201fd0e931f0eaf8d4cd17ac29ef06ae558510403fe38fd83293322d67513d6b3c53132badc11b900", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.125, "full": 0.625, "no_world_model": 0.5625, "random": 0.3125}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:44:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:27Z", "cpu_seconds": 0.489512843, "created_at": "2026-11-22T03:44:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-624713fd0702", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "b829e82aecf51b2bdc14751b9dcd21418d5af3c933be86e19a6e0c445c4da6e1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-624713fd0702", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:25Z", "seed": 1330832076, "treatment": "full"}, "priority": 0.5649274108105318, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "42f0b45c7ce18573aea80ce28617b8e42f08b024bf022e023bb8b1ac76dc6a93", "run_seconds": 0.6599462032318115, "seed": 1330832076, "started_at": "2026-11-22T03:44:26Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-624713fd0702", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b829e82aecf51b2bdc14751b9dcd21418d5af3c933be86e19a6e0c445c4da6e1", "primary_control": "e141_best", "result_hash": "42f0b45c7ce18573aea80ce28617b8e42f08b024bf022e023bb8b1ac76dc6a93", "signature": "8b985de1789dc4c4f04b14cf9cfc59a7c17a44b1a8eed26bc74e5681ce83c842c82d50198b6e994616cffb3ba413880b995db85654ae2f1e876dfe663c1db805", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:44:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:34Z", "cpu_seconds": 0.5399951590000001, "created_at": "2026-11-22T03:44:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c19d9f049f06", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "ab27a4a80f2c47334b1e2eb6850c8a13cca4c82ddb62d00addfb6c51c837b24f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c19d9f049f06", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:25Z", "seed": 1624449604, "treatment": "full"}, "priority": 0.549407279782457, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "815e5c8373face3eaeff8ebf6f27852a18a0a0ea0e9d7368eb1ad9e6da33d820", "run_seconds": 0.5638539791107178, "seed": 1624449604, "started_at": "2026-11-22T03:44:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T03:44:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c19d9f049f06", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 0.75, 0.75], "random": [0.5, 0.5, 0.25, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "ab27a4a80f2c47334b1e2eb6850c8a13cca4c82ddb62d00addfb6c51c837b24f", "primary_control": "e141_best", "result_hash": "815e5c8373face3eaeff8ebf6f27852a18a0a0ea0e9d7368eb1ad9e6da33d820", "signature": "30ddbb1f4801d6e2751d0332e352eac8c3a25feb3e37cbcd3c13d6686c812e77218fee4521c88aff5417e08b60dd341e57e195d6b961841bb8c3f983dcb0e80d", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0625, "full": 0.75, "no_world_model": 0.625, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:44:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:35Z", "cpu_seconds": 0.4930556449999983, "created_at": "2026-11-22T03:44:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f2b0a00caf3f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "03a31f7a313f54221cb3c2ecf0d2f48cd41d088e49a094de9ba611722110908b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f2b0a00caf3f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:25Z", "seed": 413876224, "treatment": "full"}, "priority": 0.5457356685737107, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b72ac4c7bc3c9a6b7225c608d200e4f707263446f968fc70b12cdf8ca8919b75", "run_seconds": 0.5519952774047852, "seed": 413876224, "started_at": "2026-11-22T03:44:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:44:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f2b0a00caf3f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [0.75, 0.25, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "03a31f7a313f54221cb3c2ecf0d2f48cd41d088e49a094de9ba611722110908b", "primary_control": "e141_best", "result_hash": "b72ac4c7bc3c9a6b7225c608d200e4f707263446f968fc70b12cdf8ca8919b75", "signature": "879b9f985fd4a3f2c8a14a4436fca997a02416862bd3248474ba75eecef27a5086e27da63e3247deea33e54aa2fa18adbe1a8b66e7f84f35484cebc2ae801807", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0625, "full": 0.9375, "no_world_model": 0.75, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:44:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:41Z", "cpu_seconds": 1.3088199090000003, "created_at": "2026-11-22T03:44:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-06fc9bc99ca6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "aed5fc9f6d0bd0842a3f136f5f21d94c75472c56baa92ae10832d662b3840b67", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-06fc9bc99ca6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:44:25Z", "seed": 1840781949, "treatment": "full"}, "priority": 0.5449782602113818, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "75bbffd09a27e0d94e8b2dcb8e254b45abeb1ac0c7f73722446681511fa13fb0", "run_seconds": 1.5514931678771973, "seed": 1840781949, "started_at": "2026-11-22T03:44:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.07108, 0.028588], "effect": -0.021246, "evaluated_at": "2026-11-22T03:44:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-06fc9bc99ca6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.275482094, 0.606575964, 0.345864662, 0.507440476], "frozen": [0.239669421, 0.301587302, 0.142857143, 0.157738095], "full": [0.191460055, 0.64399093, 0.340225564, 0.474702381], "retrieval": [0.110192837, 0.646258503, 0.360902256, 0.511904762], "rules_nodetect": [0.192837466, 0.732426304, 0.554511278, 0.398809524]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "aed5fc9f6d0bd0842a3f136f5f21d94c75472c56baa92ae10832d662b3840b67", "primary_control": "facts", "result_hash": "75bbffd09a27e0d94e8b2dcb8e254b45abeb1ac0c7f73722446681511fa13fb0", "signature": "458c41fd2e248b82be051020f2247970dde46ef88d3f9189adc439f378246dfb092b95e90fc5e0ea7d61af25ec795488153d7b76dcda8a525e5c4717dda4ff03", "state": "VERIFIED", "summary": {"facts": 0.433841, "frozen": 0.210463, "full": 0.412595, "retrieval": 0.407315, "rules_nodetect": 0.469646}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:44:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:30Z", "cpu_seconds": 0.8907030489999954, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1dea239d1b63", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "5674849f62f8d8052285e773c57e5851fff8f5798e68003f8a87d93aec8e61d6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1dea239d1b63", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:44:42Z", "seed": 977567378, "treatment": "frontier"}, "priority": 0.5288338556068553, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "46f5a471b96793e3ab0c26a05dcc62d4a3cfec2e91644e9c522bab492941fd89", "run_seconds": 0.9457075595855713, "seed": 977567378, "started_at": "2026-11-22T03:45:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037483, 0.155551], "effect": 0.059034, "evaluated_at": "2026-11-22T03:45:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1dea239d1b63", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.724137931, 0.987654321, 0.974358974, 0.96124031], "frontier": [0.689655172, 0.987654321, 0.897435897, 0.868217054], "goal": [0.666666667, 0.901234568, 0.897435897, 0.821705426], "hypdisc": [0.632183908, 0.950617284, 0.897435897, 0.906976744], "infogain": [0.574712644, 0.950617284, 0.897435897, 0.945736434], "prederr": [0.689655172, 0.901234568, 0.692307692, 0.891472868], "random": [0.689655172, 0.987654321, 0.692307692, 0.837209302], "risk": [0.689655172, 0.987654321, 0.897435897, 0.868217054]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "5674849f62f8d8052285e773c57e5851fff8f5798e68003f8a87d93aec8e61d6", "primary_control": "random", "result_hash": "46f5a471b96793e3ab0c26a05dcc62d4a3cfec2e91644e9c522bab492941fd89", "signature": "9063960c30cad682e060dc1034707d982e4e645304c220312c4d25a7ee9a35725c50074aa141e7e55b28af793d9a944f0f11e2198cf99cc88d3595cda585cd06", "state": "VERIFIED", "summary": {"count": 0.911848, "frontier": 0.860741, "goal": 0.821761, "hypdisc": 0.846803, "infogain": 0.842126, "prederr": 0.793668, "random": 0.801707, "risk": 0.860741}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:45:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:59Z", "cpu_seconds": 1.1157119869999974, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1e2d7bd8ade1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "9225d5aeb9b3bca4f4a8132324044b359bd0c49289fe3a4ea64f634784151a95", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1e2d7bd8ade1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:44:42Z", "seed": 666465720, "treatment": "frontier"}, "priority": 0.5422043135379216, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a29e850e30028fd37f9d2b82702f28e301584086cecd496f1f2fadddd803df21", "run_seconds": 1.1537635326385498, "seed": 666465720, "started_at": "2026-11-22T03:44:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.163913, 0.071606], "effect": -0.046153, "evaluated_at": "2026-11-22T03:45:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1e2d7bd8ade1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.796296296, 0.65625, 0.67251462, 0.710144928], "frontier": [0.611111111, 0.625, 0.795321637, 0.666666667], "goal": [0.537037037, 0.708333333, 0.660818713, 0.666666667], "hypdisc": [0.555555556, 0.625, 0.649122807, 0.420289855], "infogain": [0.666666667, 0.604166667, 0.719298246, 0.782608696], "prederr": [0.462962963, 0.697916667, 0.602339181, 0.304347826], "random": [0.796296296, 0.708333333, 0.812865497, 0.565217391], "risk": [0.611111111, 0.625, 0.795321637, 0.666666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "9225d5aeb9b3bca4f4a8132324044b359bd0c49289fe3a4ea64f634784151a95", "primary_control": "random", "result_hash": "a29e850e30028fd37f9d2b82702f28e301584086cecd496f1f2fadddd803df21", "signature": "d80ca2fa847cf1f92961965d1bb8f844fded140930d8fb9970e447dbaf1e8ae087307475736b70170873b69e18d99db51a47d62d1b192b8670d1dfec7e3c3308", "state": "VERIFIED", "summary": {"count": 0.708801, "frontier": 0.674525, "goal": 0.643214, "hypdisc": 0.562492, "infogain": 0.693185, "prederr": 0.516892, "random": 0.720678, "risk": 0.674525}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:45:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:57Z", "cpu_seconds": 0.6764138030000026, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b5ac633b6227", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "421e5e7cffdfba1aa6242930a2ffe0629066f4e6b7cb503a1d2e514468a7795f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b5ac633b6227", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:44:42Z", "seed": 1124318000, "treatment": "frontier"}, "priority": 0.5422494345840144, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "8244f4cdf5abff4e8dc977ddd122968a4acd86cd3ce72ba8554a94677ba037e4", "run_seconds": 0.7608356475830078, "seed": 1124318000, "started_at": "2026-11-22T03:44:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004113, 0.066613], "effect": 0.03125, "evaluated_at": "2026-11-22T03:45:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b5ac633b6227", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.958333333, 1.0, 1.0, 1.0], "frontier": [0.958333333, 0.975, 0.958333333, 0.958333333], "goal": [0.895833333, 0.925, 0.895833333, 0.9375], "hypdisc": [0.916666667, 1.0, 0.9375, 0.9375], "infogain": [0.958333333, 0.975, 0.958333333, 0.958333333], "prederr": [0.958333333, 0.975, 0.958333333, 0.958333333], "random": [0.875, 0.975, 0.9375, 0.9375], "risk": [0.958333333, 0.975, 0.958333333, 0.958333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "421e5e7cffdfba1aa6242930a2ffe0629066f4e6b7cb503a1d2e514468a7795f", "primary_control": "random", "result_hash": "8244f4cdf5abff4e8dc977ddd122968a4acd86cd3ce72ba8554a94677ba037e4", "signature": "abcb9ac2162b386f4c97d21f761d0058e5753a1c0849e922e0793ff6d9c5f430dd9d081beb2efbfef1fa196553f7bb48f48051b833919e57b12ccc408fbd9c0c", "state": "VERIFIED", "summary": {"count": 0.989583, "frontier": 0.9625, "goal": 0.913542, "hypdisc": 0.947917, "infogain": 0.9625, "prederr": 0.9625, "random": 0.93125, "risk": 0.9625}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:45:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:11Z", "cpu_seconds": 0.8139206439999995, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-277e59853183", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "4d294c9b0f955fee9c5550b41c6bfbfec09f8bf936d9a32b91aff377e788fbc9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-277e59853183", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:42Z", "seed": 1426976917, "treatment": "full"}, "priority": 0.5351925578885484, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5b48f88a6a9aec139744b307fb835ad4c4960b959a483be29f46a1f5093feba3", "run_seconds": 0.8221893310546875, "seed": 1426976917, "started_at": "2026-11-22T03:45:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.42207, 0.04707], "effect": -0.1875, "evaluated_at": "2026-11-22T03:45:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-277e59853183", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.25, 1.0, 0.0], "full": [0.5, 1.0, 1.0, 0.75], "no_world_model": [1.0, 1.0, 1.0, 0.75], "random": [0.5, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "4d294c9b0f955fee9c5550b41c6bfbfec09f8bf936d9a32b91aff377e788fbc9", "primary_control": "e141_best", "result_hash": "5b48f88a6a9aec139744b307fb835ad4c4960b959a483be29f46a1f5093feba3", "signature": "f7c534765fce05d3e6514fa4d82f67259ee7de1754df96a9c5fc5dba2a2ff2245246ebf00106926f8748bd3b9bd1831b1822cadbe9b2192187b9d47484c6fa09", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 0.8125, "no_world_model": 0.9375, "random": 0.875}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:45:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:05Z", "cpu_seconds": 0.6735524879999986, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-39c3b3eb890d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "50132279010dd829fb132f4647eb0e50fa27f25e2b4c32bd2048c359a8659d27", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-39c3b3eb890d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:42Z", "seed": 431401238, "treatment": "full"}, "priority": 0.5421915970348831, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "65c74a1b0d8e2192d6609e2dd0bf20ea34d635cf09cf2a3c089dcb309da2584e", "run_seconds": 1.065434217453003, "seed": 431401238, "started_at": "2026-11-22T03:45:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T03:45:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-39c3b3eb890d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.25, 0.25], "full": [0.75, 1.0, 1.0, 0.75], "no_world_model": [0.5, 1.0, 0.75, 0.75], "random": [0.75, 0.5, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "50132279010dd829fb132f4647eb0e50fa27f25e2b4c32bd2048c359a8659d27", "primary_control": "e141_best", "result_hash": "65c74a1b0d8e2192d6609e2dd0bf20ea34d635cf09cf2a3c089dcb309da2584e", "signature": "67c6776b0e41bec55754e798b7b17a0bc05f57f4a1d06b2856c51f5545aaef5201a7a8fd11d22dd28466f068c1f7319d07083c7e6dd8d3c1133a1b2d2d274d01", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.125, "full": 0.875, "no_world_model": 0.75, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:45:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:47Z", "cpu_seconds": 0.7198790509999995, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-65fd2d802147", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "9a0526af0057b415ca47e3e63eeb2c4511e0fc1ccb692dd97ce2af87cb6d461f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-65fd2d802147", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:42Z", "seed": 1778709133, "treatment": "full"}, "priority": 0.5454881557327841, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "389f8ea58e38fed80c596436e37bcffc683278278a046a735c040813d1a0af24", "run_seconds": 0.9406816959381104, "seed": 1778709133, "started_at": "2026-11-22T03:44:46Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-65fd2d802147", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.0, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9a0526af0057b415ca47e3e63eeb2c4511e0fc1ccb692dd97ce2af87cb6d461f", "primary_control": "e141_best", "result_hash": "389f8ea58e38fed80c596436e37bcffc683278278a046a735c040813d1a0af24", "signature": "054a22f951094bb78abc1bff2510690708e897ce0b65a34a8fa7a95e8a3aedcd6a0ce2ae5bbe0a1dbbdd2f33c7a320760a80fe37e55d35eadb6500f9a0624d0a", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 1.0, "random": 0.875}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:44:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:53Z", "cpu_seconds": 0.8199970370000003, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f5b61a140b16", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "043380b3f8d9846b993a4ef49867a9675fda0d25c708c3c527bf8db37f3f6129", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f5b61a140b16", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:42Z", "seed": 217675390, "treatment": "full"}, "priority": 0.5440398086741224, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "74953177560145ee456c5d86609f7c3703a9939c5a65c3b949dcef3c7bd0e7ca", "run_seconds": 0.8775222301483154, "seed": 217675390, "started_at": "2026-11-22T03:44:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.549958, 0.950042], "effect": 0.75, "evaluated_at": "2026-11-22T03:44:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f5b61a140b16", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.25, 0.0, 0.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [0.75, 1.0, 0.75, 1.0], "no_world_model": [0.25, 1.0, 0.75, 1.0], "random": [0.75, 0.25, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "043380b3f8d9846b993a4ef49867a9675fda0d25c708c3c527bf8db37f3f6129", "primary_control": "e141_best", "result_hash": "74953177560145ee456c5d86609f7c3703a9939c5a65c3b949dcef3c7bd0e7ca", "signature": "6b0bb84a36f22c96bdb6ab0ba76dd235de2fff94d4dff947e89826600eddafd362064d38c0f205e680efea73204ce1865bda7000aeb8f5caeba45af9958e7d0f", "state": "VERIFIED", "summary": {"e141_best": 0.125, "fixed_policy": 0.0625, "full": 0.875, "no_world_model": 0.75, "random": 0.25}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:44:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:51Z", "cpu_seconds": 1.2109330600000021, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e7d9485ed8fe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "999e0a22b9fa36927fe3767713d83a039e918d7d33e9d32b1f7c44f66336c5f0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e7d9485ed8fe", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:44:42Z", "seed": 1170231526, "treatment": "full"}, "priority": 0.5450444115744045, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "136fb32afda80296b01ed02727f9d55c7668047cb819f7082ff97ab12a77eea2", "run_seconds": 1.2672863006591797, "seed": 1170231526, "started_at": "2026-11-22T03:44:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017619, 0.005714], "effect": -0.005952, "evaluated_at": "2026-11-22T03:44:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e7d9485ed8fe", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.925, 0.743421053, 0.80952381, 0.9375], "frozen": [0.625, 0.388157895, 0.369047619, 0.322916667], "full": [0.925, 0.743421053, 0.785714286, 0.9375], "retrieval": [0.925, 0.743421053, 0.785714286, 0.9375], "rules_nodetect": [0.675, 0.743421053, 0.547619048, 0.9375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "999e0a22b9fa36927fe3767713d83a039e918d7d33e9d32b1f7c44f66336c5f0", "primary_control": "facts", "result_hash": "136fb32afda80296b01ed02727f9d55c7668047cb819f7082ff97ab12a77eea2", "signature": "a06c9908b61df811f37a74c8572b92c4c15c36d4253cfae6b39a6188ef8eb999e6e2aeb1c156f2e84cd6691452cc832da7352094993b90ec6f809039f461f702", "state": "VERIFIED", "summary": {"facts": 0.853861, "frozen": 0.426281, "full": 0.847909, "retrieval": 0.847909, "rules_nodetect": 0.725885}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:44:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:45Z", "cpu_seconds": 0.7010448320000009, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-6dae28820d79", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "d0658263452d7249915efdc0c2dbe339268f71479d48c5f291d93f458340405b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-6dae28820d79", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:44:42Z", "seed": 1038602535, "treatment": "gated"}, "priority": 0.5537734628360732, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6654f1e9766bf39d8c20fd3ebf039ea8d55beda9b581751bcac8304de8028fc3", "run_seconds": 0.777860164642334, "seed": 1038602535, "started_at": "2026-11-22T03:44:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024362, 0.007901], "effect": -0.00823, "evaluated_at": "2026-11-22T03:44:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-6dae28820d79", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.637860082, 0.666666667, 0.632850242, 0.631578947], "gated": [0.637860082, 0.666666667, 0.632850242, 0.631578947], "naive": [0.736625514, 0.689655172, 0.719806763, 0.640350877], "scratch": [0.670781893, 0.666666667, 0.632850242, 0.631578947], "ungated": [0.736625514, 0.666666667, 0.323671498, 0.640350877]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d0658263452d7249915efdc0c2dbe339268f71479d48c5f291d93f458340405b", "primary_control": "scratch", "result_hash": "6654f1e9766bf39d8c20fd3ebf039ea8d55beda9b581751bcac8304de8028fc3", "signature": "49ed9765b6d301cb30fc3eaadabf3bb154f087ecd95a2f47e0d5f7dff94e1eb5462ce746f19c65c64c2d4d58be40c126777a2a672dec4cdf0cd09c2954bf800d", "state": "VERIFIED", "summary": {"compressed": 0.642239, "gated": 0.642239, "naive": 0.69661, "scratch": 0.650469, "ungated": 0.591829}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:44:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:10Z", "cpu_seconds": 8.621316319999998, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-2224ef6ad9f0", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7e729c7886e3c81113b828baabba6b7428cd71411cb0463338e8121782ea289a", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-2224ef6ad9f0", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:44:42Z", "seed": 1843479715, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "cee052a08b040a7524edd7475286be2e3b4a84380f8c851bbc10501a94e2408b", "run_seconds": 10.50613284111023, "seed": 1843479715, "started_at": "2026-11-22T03:44:59Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:45:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-2224ef6ad9f0", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "7e729c7886e3c81113b828baabba6b7428cd71411cb0463338e8121782ea289a", "primary_control": "reset", "result_hash": "cee052a08b040a7524edd7475286be2e3b4a84380f8c851bbc10501a94e2408b", "signature": "42990d8c173dfdd2fa99a8dfe5976589123506ef3985943fc043ee89490d75439dee1c25157fa5f8b94ac9cb64313e877a6c22dc323c0e77ffcedefbd1234108", "state": "PARTIAL", "summary": {"inherit": 0.5333333333333333, "reset": 0.5333333333333333, "root": 0.21666666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:45:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:44:54Z", "cpu_seconds": 7.960912422999996, "created_at": "2026-11-22T03:44:42Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-72f8afa4d00f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "1bc97279343e6bf2a95660ee9edac2664efcdd11963d111448ad28d7505145e0", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-72f8afa4d00f", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:44:42Z", "seed": 1683689078, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "5e895e4bd6a443aed5413c8ab8ee3f28ee385dcbfc42f83470b4cf0f44665d48", "run_seconds": 8.320196151733398, "seed": 1683689078, "started_at": "2026-11-22T03:44:46Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:44:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-72f8afa4d00f", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0], "reset": [0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0], "root": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1bc97279343e6bf2a95660ee9edac2664efcdd11963d111448ad28d7505145e0", "primary_control": "reset", "result_hash": "5e895e4bd6a443aed5413c8ab8ee3f28ee385dcbfc42f83470b4cf0f44665d48", "signature": "af7d4fea01cbfa06a3af1e26d40a703c4dc96d791d88e5f3410bddcd9145898b06e2c1b2a8e4bb7f75f7a862dcee07a960b7ecaec48f85d048e87a7fe0565005", "state": "PARTIAL", "summary": {"inherit": 0.5833333333333334, "reset": 0.5833333333333334, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:45:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:32Z", "cpu_seconds": 0.47743796899999325, "created_at": "2026-11-22T03:44:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-71b4e786db38", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "44e6c2a95c0f682bb94c10782a6864bab05fe22c2c70fe968c6966072354bf2a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-71b4e786db38", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:58Z", "seed": 1049856538, "treatment": "full"}, "priority": 0.526785160984487, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2e722f8d828035b8f4e86dd16bca415e8c1c82a926414a94be30cdc29517d0f7", "run_seconds": 0.4792306423187256, "seed": 1049856538, "started_at": "2026-11-22T03:45:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T03:45:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-71b4e786db38", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 1.0, 1.0], "random": [0.75, 0.75, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "44e6c2a95c0f682bb94c10782a6864bab05fe22c2c70fe968c6966072354bf2a", "primary_control": "e141_best", "result_hash": "2e722f8d828035b8f4e86dd16bca415e8c1c82a926414a94be30cdc29517d0f7", "signature": "97283847fd4f7336b5f66770b880e9f3a528f7936a79e8bbd8bdb42cf40852427dc0e3796d7b35379c9a0a4bb2cca3c554a6fda6f12f9c9564f33a41af50a10f", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 0.75, "no_world_model": 0.75, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:45:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:10Z", "cpu_seconds": 0.4046121340000042, "created_at": "2026-11-22T03:44:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8365d0b0226b", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "62016a8a875741e1cd4f0581cb4465568b2e03277798660abeeb5cddb69da711", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8365d0b0226b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:44:58Z", "seed": 1766626515, "treatment": "full"}, "priority": 0.5408272440176606, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "286af9ca3463a9e5db876fe3fdc441a2fbbe9b0a32dffeaf271cd27b0fd22586", "run_seconds": 0.40747976303100586, "seed": 1766626515, "started_at": "2026-11-22T03:45:10Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:45:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8365d0b0226b", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.375], "fixed_policy": [1.0, 0.0], "full": [1.0, 0.375], "no_world_model": [1.0, 0.25], "random": [1.0, 0.375]}, "prereg_hash": "62016a8a875741e1cd4f0581cb4465568b2e03277798660abeeb5cddb69da711", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "286af9ca3463a9e5db876fe3fdc441a2fbbe9b0a32dffeaf271cd27b0fd22586", "signature": "1a6f3d5446c91f96b6cc66ec3f565bd051c21fd84e9c084c398dd1644e36d95e4d5f1e26d3a1c7fc4c247133d703121822b9ddde7186d10d8f8d75a49b3a1707", "state": "REJECTED", "summary": {"e141_best": 0.375, "fixed_policy": 0.0, "full": 0.375, "no_world_model": 0.25, "random": 0.375}}, "verified_at": "2026-11-22T03:45:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:17Z", "cpu_seconds": 1.2695610660000014, "created_at": "2026-11-22T03:44:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4031fa0cc008", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "2b81a1270bead55748fc403392bff90eaeb4cc136f4016076e09167c25226024", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4031fa0cc008", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:44:58Z", "seed": 597552850, "treatment": "full"}, "priority": 0.5332612264433114, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4ca7d50cd3e58b5d1192bd905675fbdc03d829657fb28d50ab219faf55ba1504", "run_seconds": 1.8356554508209229, "seed": 597552850, "started_at": "2026-11-22T03:45:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037099, 0.052563], "effect": 0.007732, "evaluated_at": "2026-11-22T03:45:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4031fa0cc008", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.442105263, 0.690909091, 0.67761194, 0.649411765], "frozen": [0.110526316, 0.109090909, 0.125373134, 0.12], "full": [0.413157895, 0.763636364, 0.68358209, 0.630588235], "retrieval": [0.4, 0.709090909, 0.68358209, 0.64], "rules_nodetect": [0.413157895, 0.763636364, 0.764179104, 0.630588235]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "2b81a1270bead55748fc403392bff90eaeb4cc136f4016076e09167c25226024", "primary_control": "facts", "result_hash": "4ca7d50cd3e58b5d1192bd905675fbdc03d829657fb28d50ab219faf55ba1504", "signature": "1c6cdb212b3d7019948f532fc4b09e5870afac81ad2d77d233c7a190cef3d64cd77b71ab4eaf0aff858e0f993d64f640262a52a8c0d2cc69f1fe7c49c90bc902", "state": "VERIFIED", "summary": {"facts": 0.61501, "frozen": 0.116248, "full": 0.622741, "retrieval": 0.608168, "rules_nodetect": 0.64289}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:45:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:16Z", "cpu_seconds": 0.7460573600000018, "created_at": "2026-11-22T03:44:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-74618c030e23", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "eb94a3176db57598161c43c8d0356d096fc0c3e4080817733b6b82a49eb750a3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-74618c030e23", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:44:58Z", "seed": 748497272, "treatment": "full"}, "priority": 0.5337433871442286, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2a08a1ba412f1981a3b3ab255ae6c4a1a06f8c5708e91c2ad3e817dd5d25a9c6", "run_seconds": 0.9002292156219482, "seed": 748497272, "started_at": "2026-11-22T03:45:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016302, 0.148783], "effect": 0.06624, "evaluated_at": "2026-11-22T03:45:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-74618c030e23", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.493243243, 1.0, 0.767857143, 1.0], "frozen": [0.297297297, 0.346153846, 0.339285714, 0.410714286], "full": [0.668918919, 1.0, 0.857142857, 1.0], "retrieval": [0.72972973, 1.0, 0.892857143, 1.0], "rules_nodetect": [0.668918919, 1.0, 0.857142857, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "eb94a3176db57598161c43c8d0356d096fc0c3e4080817733b6b82a49eb750a3", "primary_control": "facts", "result_hash": "2a08a1ba412f1981a3b3ab255ae6c4a1a06f8c5708e91c2ad3e817dd5d25a9c6", "signature": "e2249d0b9b78a6f1a580f63c3a84deb08ea04edc68d02917cf8293697cae5e36f5462cb319453c856ccddc54585cff24a950af395fc77aebc9e346adcc5b010d", "state": "VERIFIED", "summary": {"facts": 0.815275, "frozen": 0.348363, "full": 0.881515, "retrieval": 0.905647, "rules_nodetect": 0.881515}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:45:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:05Z", "cpu_seconds": 1.4035393380000016, "created_at": "2026-11-22T03:44:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7e8082ca7a9e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "8e4c90949c5cecd20e7673c9560563cc6c118c9bfa1f01cf011fc60d3a628c04", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7e8082ca7a9e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:44:58Z", "seed": 395146820, "treatment": "full"}, "priority": 0.5434806062558073, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "90ed52f4984caa198005e84183359372d234802a3df11bc3fc8ca0af26bf162c", "run_seconds": 2.519587516784668, "seed": 395146820, "started_at": "2026-11-22T03:45:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.05106, 0.020799], "effect": -0.015131, "evaluated_at": "2026-11-22T03:45:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7e8082ca7a9e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.613333333, 0.755555556, 0.450485437, 0.742105263], "frozen": [0.128, 0.155555556, 0.133980583, 0.110526316], "full": [0.594666667, 0.777777778, 0.386407767, 0.742105263], "retrieval": [0.608, 0.822222222, 0.436893204, 0.742105263], "rules_nodetect": [0.594666667, 0.777777778, 0.74368932, 0.742105263]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "8e4c90949c5cecd20e7673c9560563cc6c118c9bfa1f01cf011fc60d3a628c04", "primary_control": "facts", "result_hash": "90ed52f4984caa198005e84183359372d234802a3df11bc3fc8ca0af26bf162c", "signature": "d6b0cb01118098691036776f0a3f6d4c8a058c59490a646c0529f96b56ea8f0d2baf679c167b109f22156579cb9899c7e4f20d95b8bfc9c2c15afbe965a56a00", "state": "VERIFIED", "summary": {"facts": 0.64037, "frozen": 0.132016, "full": 0.625239, "retrieval": 0.652305, "rules_nodetect": 0.71456}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:45:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:24Z", "cpu_seconds": 1.3494220079999977, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6780eb324e8f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "40a4ddd7355fec6ef6c6f8afe9dd2b37b264d38cab26959545f09590bd19f48c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6780eb324e8f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:45:15Z", "seed": 1357464275, "treatment": "frontier"}, "priority": 0.5381075989296897, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "03af3fbce0039552ba3ac2cc53c6336127b11e36ac7710537eb245b7cfc33fbe", "run_seconds": 1.7920856475830078, "seed": 1357464275, "started_at": "2026-11-22T03:45:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.145277, 0.128975], "effect": -0.008151, "evaluated_at": "2026-11-22T03:45:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6780eb324e8f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.75, 0.690909091, 0.833333333, 0.907894737], "frontier": [0.617647059, 0.768181818, 0.75, 0.868421053], "goal": [0.573529412, 0.736363636, 0.75, 0.868421053], "hypdisc": [0.632352941, 0.754545455, 0.666666667, 0.868421053], "infogain": [0.617647059, 0.768181818, 0.75, 0.868421053], "prederr": [0.617647059, 0.768181818, 0.75, 0.868421053], "random": [0.808823529, 0.686363636, 0.791666667, 0.75], "risk": [0.617647059, 0.768181818, 0.75, 0.868421053]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "40a4ddd7355fec6ef6c6f8afe9dd2b37b264d38cab26959545f09590bd19f48c", "primary_control": "random", "result_hash": "03af3fbce0039552ba3ac2cc53c6336127b11e36ac7710537eb245b7cfc33fbe", "signature": "a8c1df15b065dbd3fba50b65db24e3178d09a2063d751d9c60f973c7862fd8e7a193191fe631c21a2fe6ba1eb9436857d84d3141345109ceb7991c4fc1233804", "state": "VERIFIED", "summary": {"count": 0.795534, "frontier": 0.751062, "goal": 0.732079, "hypdisc": 0.730497, "infogain": 0.751062, "prederr": 0.751062, "random": 0.759213, "risk": 0.751062}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:45:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:42Z", "cpu_seconds": 0.7624921900000032, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ddf871015a60", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "79621c3e3f0b9c9c72664c345b18349dd07418f9fdafb8826ea7df5f6b40e673", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ddf871015a60", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:45:15Z", "seed": 253395321, "treatment": "frontier"}, "priority": 0.521015433648833, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f8b16e4fcae2cc8bb009c29f257989c170609f6454f58dc6cf521d353c10a954", "run_seconds": 0.8124701976776123, "seed": 253395321, "started_at": "2026-11-22T03:45:41Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:45:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ddf871015a60", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.856667], "frontier": [1.0, 0.9075], "goal": [1.0, 0.835833], "hypdisc": [1.0, 0.873333], "infogain": [1.0, 0.886667], "prederr": [1.0, 0.861667], "random": [1.0, 0.850833], "risk": [1.0, 0.9075]}, "prereg_hash": "79621c3e3f0b9c9c72664c345b18349dd07418f9fdafb8826ea7df5f6b40e673", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "f8b16e4fcae2cc8bb009c29f257989c170609f6454f58dc6cf521d353c10a954", "signature": "e18ecabca63fa69ed3d4db2157345ed55f5fb82842d67b56a58097383b9b91ca34891838f32cb110bbadf7ded9fc99326416f616714c20c1fda79ea8079d5b0c", "state": "REJECTED", "summary": {"count": 0.856667, "frontier": 0.9075, "goal": 0.835833, "hypdisc": 0.873333, "infogain": 0.886667, "prederr": 0.861667, "random": 0.850833, "risk": 0.9075}}, "verified_at": "2026-11-22T03:45:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:23Z", "cpu_seconds": 1.3641291469999999, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4f8e0166cf3c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "77d839a7f619415b737851d88d258dc91f72fe90ba1ac5c12c4cd9f631e05270", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4f8e0166cf3c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:45:15Z", "seed": 701863005, "treatment": "full"}, "priority": 0.5406394946331836, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7d4699287c5484a55c878d951a11348863666a71b60b2979585b89c87d1472bb", "run_seconds": 1.6137197017669678, "seed": 701863005, "started_at": "2026-11-22T03:45:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.29707, 0.17207], "effect": -0.0625, "evaluated_at": "2026-11-22T03:45:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4f8e0166cf3c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 0.75, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 0.75], "full": [0.75, 0.75, 0.75, 0.75], "no_world_model": [0.25, 0.5, 0.25, 0.75], "random": [0.25, 1.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "77d839a7f619415b737851d88d258dc91f72fe90ba1ac5c12c4cd9f631e05270", "primary_control": "e141_best", "result_hash": "7d4699287c5484a55c878d951a11348863666a71b60b2979585b89c87d1472bb", "signature": "64a74f91d512aba8e7645e0b3358b125b0b6a05697ff7924f367245c50729c5818682e23a45400feb75bbea81b7574e6f8c65385f684e42c032d0068b8dcb20d", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.9375, "full": 0.75, "no_world_model": 0.4375, "random": 0.6875}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:45:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:17Z", "cpu_seconds": 0.42174458000000214, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-533c29f8a90c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "046a5ac19d094a354ca0463f4ecee67d3931db9bf595ca29715686f7c4ea83dd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-533c29f8a90c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:45:15Z", "seed": 1774040548, "treatment": "full"}, "priority": 0.5426118134394077, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c38d5b36b32b56791849c7d903265569c7979e5f3195f2b503553b69cce17b3b", "run_seconds": 0.6869912147521973, "seed": 1774040548, "started_at": "2026-11-22T03:45:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.620785, -0.004215], "effect": -0.3125, "evaluated_at": "2026-11-22T03:45:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-533c29f8a90c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.0, 0.25, 0.25], "fixed_policy": [0.25, 0.25, 0.0, 0.0], "full": [0.0, 0.0, 0.0, 0.0], "no_world_model": [0.0, 0.0, 0.0, 0.0], "random": [0.0, 0.0, 0.25, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "046a5ac19d094a354ca0463f4ecee67d3931db9bf595ca29715686f7c4ea83dd", "primary_control": "e141_best", "result_hash": "c38d5b36b32b56791849c7d903265569c7979e5f3195f2b503553b69cce17b3b", "signature": "5bf8385f57218196531317cb57832954c6d8f36fd291b4b75614babde595078d77adc37368c2acf843418f939c0190ff11ce69fa580c56c3802e92e88748fa04", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.125, "full": 0.0, "no_world_model": 0.0, "random": 0.0625}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:45:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:48Z", "cpu_seconds": 1.1855458819999996, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-649fbdc7b5c4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "fe9e198e42c3b7cd6cbf0056b8b36a49c9538d1cc21ccc1a5cc0e33bcbfb2dea", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-649fbdc7b5c4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:45:15Z", "seed": 1947823320, "treatment": "full"}, "priority": 0.5149090931625927, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f086091b44be723134e9d651cae3c4f2622066a3ba662ea59d4a9f61e16a4ac6", "run_seconds": 1.507772445678711, "seed": 1947823320, "started_at": "2026-11-22T03:45:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016823, -0.005341], "effect": -0.011082, "evaluated_at": "2026-11-22T03:45:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-649fbdc7b5c4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.485815603, 0.446808511, 0.54964539, 0.455673759], "frozen": [0.106382979, 0.106382979, 0.106382979, 0.106382979], "full": [0.473404255, 0.436170213, 0.546099291, 0.437943262], "retrieval": [0.482269504, 0.457446809, 0.55141844, 0.44858156], "rules_nodetect": [0.473404255, 0.436170213, 0.546099291, 0.737588652]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "fe9e198e42c3b7cd6cbf0056b8b36a49c9538d1cc21ccc1a5cc0e33bcbfb2dea", "primary_control": "facts", "result_hash": "f086091b44be723134e9d651cae3c4f2622066a3ba662ea59d4a9f61e16a4ac6", "signature": "0729025aed2c37598709359224dd64016b3d66bac9c9dbcca3d18fd77b36982d9858db66354df44b6a5a92f3ebc295eb475a08b95b6e07b95040be3067314e04", "state": "VERIFIED", "summary": {"facts": 0.484486, "frozen": 0.106383, "full": 0.473404, "retrieval": 0.484929, "rules_nodetect": 0.548316}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:45:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:29Z", "cpu_seconds": 0.696415309999999, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9e60def65ab5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "00df0414e62166e0910b34848bd4396677dd315d83d67d6ea58b7ac6a085ab9d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9e60def65ab5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:45:15Z", "seed": 1439444165, "treatment": "full"}, "priority": 0.5312060637767363, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "388c5a8a01e7fe9c45299ac66f2fc82c992781baa99b00e9613c6f97de925073", "run_seconds": 0.7271709442138672, "seed": 1439444165, "started_at": "2026-11-22T03:45:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0064, 0.019733], "effect": 0.006667, "evaluated_at": "2026-11-22T03:45:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9e60def65ab5", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.960526316, 0.36, 0.627777778, 0.911111111], "frozen": [0.407894737, 0.266666667, 0.627777778, 0.733333333], "full": [0.960526316, 0.386666667, 0.627777778, 0.911111111], "retrieval": [0.960526316, 0.573333333, 0.627777778, 0.911111111], "rules_nodetect": [0.565789474, 0.306666667, 0.8, 0.911111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "00df0414e62166e0910b34848bd4396677dd315d83d67d6ea58b7ac6a085ab9d", "primary_control": "facts", "result_hash": "388c5a8a01e7fe9c45299ac66f2fc82c992781baa99b00e9613c6f97de925073", "signature": "9cb75a12c29315bb5e422bf2238e4c8cf98ad45753198986d4d51d0eab6866b9a513cdd4c9880d0d259d6990e17c5ac25d0ed1d1d0f5fdf2f49ca1236fbd2e0d", "state": "VERIFIED", "summary": {"facts": 0.714854, "frozen": 0.508918, "full": 0.72152, "retrieval": 0.768187, "rules_nodetect": 0.645892}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:45:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:47Z", "cpu_seconds": 0.6166504580000023, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a574fa9240e5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "86dab5a7bdc792bfbb45b904086b2df515a277b64ce4542a5fb7cf5c26725344", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a574fa9240e5", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:45:15Z", "seed": 528452937, "treatment": "gated"}, "priority": 0.5189661701598006, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "4ed60508b50ee9571e693ae7d22f84c963c2e2d1228d71c07ca8b77b7d3d8c63", "run_seconds": 0.8899121284484863, "seed": 528452937, "started_at": "2026-11-22T03:45:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0304, 0.093733], "effect": 0.031667, "evaluated_at": "2026-11-22T03:45:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a574fa9240e5", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.39869281, 0.248366013, 0.244897959, 0.413333333], "gated": [0.39869281, 0.248366013, 0.244897959, 0.413333333], "naive": [0.254901961, 0.22875817, 0.197278912, 0.4], "scratch": [0.39869281, 0.248366013, 0.244897959, 0.286666667], "ungated": [0.346405229, 0.202614379, 0.176870748, 0.213333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "86dab5a7bdc792bfbb45b904086b2df515a277b64ce4542a5fb7cf5c26725344", "primary_control": "scratch", "result_hash": "4ed60508b50ee9571e693ae7d22f84c963c2e2d1228d71c07ca8b77b7d3d8c63", "signature": "ce4db4743a882b88177bd691e056e6a565b02b5582ce796c0c8bd0d469845a04ab12c1d5617ff7242ad7c5bf1cdb16951acb52ce603514a6403b1c2dac3a7b0c", "state": "VERIFIED", "summary": {"compressed": 0.326323, "gated": 0.326323, "naive": 0.270235, "scratch": 0.294656, "ungated": 0.234806}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:45:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:26Z", "cpu_seconds": 4.823958715999993, "created_at": "2026-11-22T03:45:15Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-a662b0310700", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "961c76c3fc423ea4651e17136718b44f327ce93b00659934c13ff690f0bd41a7", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-a662b0310700", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:45:15Z", "seed": 928252830, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "75c1a41f99f7fb5f420e4d8c07df1f7e34a769ae521cfa77e0ef4669842becda", "run_seconds": 5.685998201370239, "seed": 928252830, "started_at": "2026-11-22T03:45:21Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:45:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-a662b0310700", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "961c76c3fc423ea4651e17136718b44f327ce93b00659934c13ff690f0bd41a7", "primary_control": "reset", "result_hash": "75c1a41f99f7fb5f420e4d8c07df1f7e34a769ae521cfa77e0ef4669842becda", "signature": "1a79985737feb5f5b74127efb4c6206d54459e03b3953f5c6f1114ed48af2f303065dd8c73bd9303cd01d11afd68cea52cb0e93c789906e0d432aaec89a61f06", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.3333333333333333}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:45:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:35Z", "cpu_seconds": 0.8849365579999997, "created_at": "2026-11-22T03:45:32Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1d0fcda32c23", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "4609f816ad65e059f4823203dd07ae5f73b1bcd9e8514a195dc5662c3b059978", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1d0fcda32c23", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:45:32Z", "seed": 1243401297, "treatment": "frontier"}, "priority": 0.5367711734158587, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "52a206ee7e6a1153153ed2ac143ea39d79245e5c9c5f3ee8e942a57891b80b80", "run_seconds": 0.8929958343505859, "seed": 1243401297, "started_at": "2026-11-22T03:45:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.026656, 0.128344], "effect": 0.0775, "evaluated_at": "2026-11-22T03:45:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1d0fcda32c23", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.98, 1.0, 0.933333333, 1.0], "frontier": [0.97, 0.958333333, 0.966666667, 1.0], "goal": [0.92, 0.958333333, 0.866666667, 1.0], "hypdisc": [0.95, 1.0, 0.966666667, 1.0], "infogain": [0.97, 1.0, 0.966666667, 1.0], "prederr": [0.97, 0.875, 0.766666667, 1.0], "random": [0.86, 0.958333333, 0.866666667, 0.9], "risk": [0.97, 0.958333333, 0.966666667, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "4609f816ad65e059f4823203dd07ae5f73b1bcd9e8514a195dc5662c3b059978", "primary_control": "random", "result_hash": "52a206ee7e6a1153153ed2ac143ea39d79245e5c9c5f3ee8e942a57891b80b80", "signature": "ba9cc8e05dbb474ff5fee1cbc7ec3da5fdddb8f1b842d1da67e614d30ce18fd49cb40134b6d0782b812b17d2016fbf739847cf2d95e7568a989a3061b7c35407", "state": "VERIFIED", "summary": {"count": 0.978333, "frontier": 0.97375, "goal": 0.93625, "hypdisc": 0.979167, "infogain": 0.984167, "prederr": 0.902917, "random": 0.89625, "risk": 0.97375}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:45:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:20Z", "cpu_seconds": 3.122733712000006, "created_at": "2026-11-22T03:45:32Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c886eff291b6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "821f3966061226e0b5e45dcdc921ae30c83bff5571f644c1b2b669d36981025e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c886eff291b6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:45:32Z", "seed": 1026843568, "treatment": "frontier"}, "priority": 0.49929859652746406, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "91646c26eb6b9109c9e11bb04dbcf9371ff441844845058755d717b959f6d239", "run_seconds": 3.5042757987976074, "seed": 1026843568, "started_at": "2026-11-22T03:46:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037262, -0.023148], "effect": -0.030205, "evaluated_at": "2026-11-22T03:46:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c886eff291b6", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.409482759, 0.351728723, 0.338888889, 0.56884058], "frontier": [0.313218391, 0.310505319, 0.280555556, 0.483695652], "goal": [0.324712644, 0.327792553, 0.286111111, 0.39673913], "hypdisc": [0.354885057, 0.36037234, 0.327777778, 0.52173913], "infogain": [0.313218391, 0.310505319, 0.280555556, 0.483695652], "prederr": [0.313218391, 0.310505319, 0.280555556, 0.483695652], "random": [0.344827586, 0.331117021, 0.311111111, 0.52173913], "risk": [0.313218391, 0.310505319, 0.280555556, 0.483695652]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "821f3966061226e0b5e45dcdc921ae30c83bff5571f644c1b2b669d36981025e", "primary_control": "random", "result_hash": "91646c26eb6b9109c9e11bb04dbcf9371ff441844845058755d717b959f6d239", "signature": "81a32c002afa763dfdd75fc7e08cb5dde93de88544fc28e326823a355a04827fb72cc4081227b7d46ad34225599e17283b985c8b02320a7427645fa922bafe06", "state": "VERIFIED", "summary": {"count": 0.417235, "frontier": 0.346994, "goal": 0.333839, "hypdisc": 0.391194, "infogain": 0.346994, "prederr": 0.346994, "random": 0.377199, "risk": 0.346994}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:46:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:05Z", "cpu_seconds": 6.410047460999998, "created_at": "2026-11-22T03:45:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-131d732c8b95", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "a1fb88ede7fe375cd5cd658a9696f7e9b6dd0273ee5f3f79eca5940b6a4511a0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-131d732c8b95", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:45:32Z", "seed": 54974738, "treatment": "full"}, "priority": 0.509002939143029, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "fda5f0ed4177693cf58474e8cc0f3c27dd128f16b59fca3e5739231c4c9465bc", "run_seconds": 7.294137477874756, "seed": 54974738, "started_at": "2026-11-22T03:45:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058706, 0.691294], "effect": 0.375, "evaluated_at": "2026-11-22T03:46:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-131d732c8b95", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.5, 0.25], "fixed_policy": [0.25, 0.5, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "a1fb88ede7fe375cd5cd658a9696f7e9b6dd0273ee5f3f79eca5940b6a4511a0", "primary_control": "e141_best", "result_hash": "fda5f0ed4177693cf58474e8cc0f3c27dd128f16b59fca3e5739231c4c9465bc", "signature": "4fd69cfa6364cc0e36b5f54deb98dcabbe7882a9437606bcc163d76259b90acbd959ac25f95edf6e66084d66e28477d16607c8c0ef1ad4079d011f4f5d9b220b", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:46:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:02Z", "cpu_seconds": 0.20451511299999936, "created_at": "2026-11-22T03:45:32Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b163c8385661", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "2fa024169141e23f7e2c640a85d7026f97e0505959d746703eb1bd824e548515", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b163c8385661", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:45:32Z", "seed": 532652751, "treatment": "full"}, "priority": 0.5045169824367879, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-3e401a927b40", "result_hash": "378f86622065577c3b74da0276c4e719aa1996db5a9df434dbcaa6d5d036724e", "run_seconds": 0.22849059104919434, "seed": 532652751, "started_at": "2026-11-22T03:46:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.172898, 0.360435], "effect": 0.266667, "evaluated_at": "2026-11-22T03:46:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b163c8385661", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.791666667, 0.791666667, 0.708333333, 0.375], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.979166667, 1.0, 0.979166667, 0.775], "retrieval": [0.875, 0.979166667, 0.833333333, 0.65], "rules_nodetect": [0.666666667, 0.3125, 0.270833333, 0.5]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2fa024169141e23f7e2c640a85d7026f97e0505959d746703eb1bd824e548515", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-3e401a927b40", "result_hash": "378f86622065577c3b74da0276c4e719aa1996db5a9df434dbcaa6d5d036724e", "signature": "e6a3da3cac216c3fbb69fe46dbd2ea3a2b564898925ba24db7bb677e08b92afd7c7eaada361a2f106fcd8caacf63b177652a3a6c0fa62984b0fdf8aca874dc0f", "state": "VERIFIED", "summary": {"facts": 0.666667, "frozen": 0.0, "full": 0.933333, "retrieval": 0.834375, "rules_nodetect": 0.4375}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:46:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:41Z", "cpu_seconds": 1.3856782339999967, "created_at": "2026-11-22T03:45:32Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c508786eabb8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "e7fc8bdee9ea5791615071e6aa42d150790af16ad1813fa008fe3c8a72279d96", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c508786eabb8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:45:32Z", "seed": 1464764376, "treatment": "full"}, "priority": 0.5234083438658119, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "895d5073ed6811fd01460f8f16743ab75ce861796a4c8fae30d9827668f5b6ec", "run_seconds": 1.4297807216644287, "seed": 1464764376, "started_at": "2026-11-22T03:45:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018244, 0.018949], "effect": 0.000353, "evaluated_at": "2026-11-22T03:45:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c508786eabb8", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.585714286, 0.543478261, 0.562857143, 0.70212766], "frozen": [0.12, 0.134782609, 0.12, 0.127659574], "full": [0.608571429, 0.532608696, 0.571428571, 0.682978723], "retrieval": [0.591428571, 0.552173913, 0.557142857, 0.682978723], "rules_nodetect": [0.608571429, 0.773913043, 0.571428571, 0.682978723]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "e7fc8bdee9ea5791615071e6aa42d150790af16ad1813fa008fe3c8a72279d96", "primary_control": "facts", "result_hash": "895d5073ed6811fd01460f8f16743ab75ce861796a4c8fae30d9827668f5b6ec", "signature": "d2b2fe4cd047600220b6ae58a03e2f7ccfb92f37c6831ff341f2084733faec5d7e64c328806b2366314b0e740c6f72fa0840afdada377c8670dd6f82b75aad0f", "state": "VERIFIED", "summary": {"facts": 0.598544, "frozen": 0.125611, "full": 0.598897, "retrieval": 0.595931, "rules_nodetect": 0.659223}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:45:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:36Z", "cpu_seconds": 1.2532033119999966, "created_at": "2026-11-22T03:45:32Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f17de1b9f519", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "17745d56a1e971a6fc371ae70ca9d214e8a81fde32b960a5d153746daa96f5ca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f17de1b9f519", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:45:32Z", "seed": 598039117, "treatment": "full"}, "priority": 0.5334655452701179, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "317db0bb705cb83b34e577227d569926f0fa080f65f667c850344001288a69bd", "run_seconds": 1.2927727699279785, "seed": 598039117, "started_at": "2026-11-22T03:45:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.033733, -0.027931], "effect": -0.030832, "evaluated_at": "2026-11-22T03:45:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f17de1b9f519", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.703883495, 0.688405797, 0.417391304, 0.485507246], "frozen": [0.181229773, 0.179710145, 0.179710145, 0.179710145], "full": [0.674757282, 0.655072464, 0.384057971, 0.457971014], "retrieval": [0.694174757, 0.689855072, 0.420289855, 0.472463768], "rules_nodetect": [0.674757282, 0.655072464, 0.384057971, 0.630434783]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "17745d56a1e971a6fc371ae70ca9d214e8a81fde32b960a5d153746daa96f5ca", "primary_control": "facts", "result_hash": "317db0bb705cb83b34e577227d569926f0fa080f65f667c850344001288a69bd", "signature": "be2eaacee7827a1b5c2209d185ad966d3396d617479b70911d750a57b86c38bcaa2d875b5674c1acc39c5dffcb30beed215060dbdb7818d6ba978d295f068706", "state": "VERIFIED", "summary": {"facts": 0.573797, "frozen": 0.18009, "full": 0.542965, "retrieval": 0.569196, "rules_nodetect": 0.586081}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:45:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:44Z", "cpu_seconds": 7.147152407999997, "created_at": "2026-11-22T03:45:32Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-09c76f7560da", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "f68a128b62fc7a479086d763da2a967b352b1d8cde7d4e07446b4373c1e6e0a4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-09c76f7560da", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:45:32Z", "seed": 192466767, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "0594a542af0f12478fa70bee0b52c5e1d13a79e3507afef10c1d8b956134e1cf", "run_seconds": 7.302015781402588, "seed": 192466767, "started_at": "2026-11-22T03:45:37Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 4}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:45:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-09c76f7560da", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5806407493679443, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0], "root": [1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f68a128b62fc7a479086d763da2a967b352b1d8cde7d4e07446b4373c1e6e0a4", "primary_control": "reset", "result_hash": "0594a542af0f12478fa70bee0b52c5e1d13a79e3507afef10c1d8b956134e1cf", "signature": "2fd0a16159ebcdff17a891af4362258f337f4a49d0c34cdc131189112aa0a1d4df60a38e20851f275b1e03e9e655a3c0c3789e9ecd4d758d19ced3b3088fc70c", "state": "PARTIAL", "summary": {"inherit": 0.6833333333333333, "reset": 0.6833333333333333, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:45:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:57Z", "cpu_seconds": 7.260297238000007, "created_at": "2026-11-22T03:45:48Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-bcf0d0b8a845", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "8186582ed86ad668909d994f9ca7edc0eba401f1fcbc4f04343694e19cb2c2ad", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-bcf0d0b8a845", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:45:48Z", "seed": 1535141782, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "995f1b040d5df7908cde44385471a530511de9a073650a884e5a77f632cbe8b5", "run_seconds": 7.978156089782715, "seed": 1535141782, "started_at": "2026-11-22T03:45:49Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:46:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-bcf0d0b8a845", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "reset": [1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8186582ed86ad668909d994f9ca7edc0eba401f1fcbc4f04343694e19cb2c2ad", "primary_control": "reset", "result_hash": "995f1b040d5df7908cde44385471a530511de9a073650a884e5a77f632cbe8b5", "signature": "ce3f860e8220822e2be6d7701917a59843001e9a7a38c03bfb1d6c12b8a7608464c15355dde36a0881574a002c66099915597728d12dc4c33c08d80d79f21a04", "state": "PARTIAL", "summary": {"inherit": 0.7166666666666667, "reset": 0.7166666666666667, "root": 0.26666666666666666}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:46:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:27Z", "cpu_seconds": 2.7037505469999985, "created_at": "2026-11-22T03:45:49Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ff61c737b999", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "2d5529ff3d41add53d5c652914f44d72ccc002795e86cbcecd54b2157d6022a8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ff61c737b999", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:45:49Z", "seed": 1167784306, "treatment": "frontier"}, "priority": 0.49596544473318344, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ded4500e782e6fc0af24ba686267194f6b4076c6a6623189d0145def2bf64cb5", "run_seconds": 3.192379951477051, "seed": 1167784306, "started_at": "2026-11-22T03:46:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005557, 0.005557], "effect": 0.0, "evaluated_at": "2026-11-22T03:46:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ff61c737b999", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 0.993055556, 1.0], "goal": [1.0, 1.0, 0.993055556, 0.993055556], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 0.993055556, 1.0], "prederr": [1.0, 1.0, 0.993055556, 1.0], "random": [1.0, 0.993055556, 1.0, 1.0], "risk": [1.0, 1.0, 0.993055556, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "2d5529ff3d41add53d5c652914f44d72ccc002795e86cbcecd54b2157d6022a8", "primary_control": "random", "result_hash": "ded4500e782e6fc0af24ba686267194f6b4076c6a6623189d0145def2bf64cb5", "signature": "35bb7f3a2860d88847a541c4410d8677bc0640abd7f631140243e2ae47eeeb995f7c03efa9f2133a286b9440315505a59a4096cc7cfef0b7a0e3d2ac43f5950d", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.998264, "goal": 0.996528, "hypdisc": 1.0, "infogain": 0.998264, "prederr": 0.998264, "random": 0.998264, "risk": 0.998264}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:46:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:57Z", "cpu_seconds": 3.016287441000003, "created_at": "2026-11-22T03:45:49Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-35924941a43b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "21d51e9e919e70ecb85457d277dc0f3439c753eabc34fba5d77afcd0c2470e39", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-35924941a43b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:45:49Z", "seed": 66713878, "treatment": "full"}, "priority": 0.5110855777753726, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bb1dc80650b159d0fe2edf7f4acea77d4a2545de99eced859abcdf1a76631ce0", "run_seconds": 3.1937551498413086, "seed": 66713878, "started_at": "2026-11-22T03:45:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.230917, 0.605917], "effect": 0.1875, "evaluated_at": "2026-11-22T03:45:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-35924941a43b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.25, 0.25, 0.0], "fixed_policy": [0.75, 0.0, 0.0, 0.0], "full": [1.0, 0.5, 0.0, 0.0], "no_world_model": [0.25, 0.25, 0.25, 0.0], "random": [0.5, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "21d51e9e919e70ecb85457d277dc0f3439c753eabc34fba5d77afcd0c2470e39", "primary_control": "e141_best", "result_hash": "bb1dc80650b159d0fe2edf7f4acea77d4a2545de99eced859abcdf1a76631ce0", "signature": "cdf1818ed7a9a705ec94f37de9a52caa126da0a644f4dc097a5f2b88ffc1617837e3aac75cab811eb1acf2fe54f5fc7445b74f97fb821cd093ab34b6ca4c2702", "state": "VERIFIED", "summary": {"e141_best": 0.1875, "fixed_policy": 0.1875, "full": 0.375, "no_world_model": 0.1875, "random": 0.125}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:46:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:45:53Z", "cpu_seconds": 0.5525412240000023, "created_at": "2026-11-22T03:45:49Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-bf7c8b6315be", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "4be41872a408851f6d0e501efeb85ca37035728022d17bf6d5fb28ad2eb2fccc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bf7c8b6315be", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:45:49Z", "seed": 1678029968, "treatment": "full"}, "priority": 0.5212830809042313, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "bd1fce4069904409947d04808f12b60d90aba2f9100b05dd3cdf463abbe3fd58", "run_seconds": 0.814122200012207, "seed": 1678029968, "started_at": "2026-11-22T03:45:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01098, 0.101525], "effect": 0.045272, "evaluated_at": "2026-11-22T03:45:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-bf7c8b6315be", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.466145833, 0.432692308, 0.371794872, 0.756410256], "frozen": [0.075520833, 0.072115385, 0.275641026, 0.096153846], "full": [0.5703125, 0.509615385, 0.346153846, 0.782051282], "retrieval": [0.703125, 0.644230769, 0.346153846, 0.846153846], "rules_nodetect": [0.5703125, 0.326923077, 0.346153846, 0.782051282]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "4be41872a408851f6d0e501efeb85ca37035728022d17bf6d5fb28ad2eb2fccc", "primary_control": "facts", "result_hash": "bd1fce4069904409947d04808f12b60d90aba2f9100b05dd3cdf463abbe3fd58", "signature": "d24faa8c14f56ff45ca36fbdfc9773ba3d43a4f41055a5990ad4f8bb45daef844f978c828ace1e449303e60d0dad2882a38c9ffada7c660bd2eddc3f0cd26207", "state": "VERIFIED", "summary": {"facts": 0.506761, "frozen": 0.129858, "full": 0.552033, "retrieval": 0.634916, "rules_nodetect": 0.50636}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:45:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:30Z", "cpu_seconds": 3.680548084999998, "created_at": "2026-11-22T03:45:49Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-043a68818f5f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "2b933cd58ee6ba2264a56b2a9ee2f7b3a48a7b9afd442dd955ed998e5418b402", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-043a68818f5f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:45:49Z", "seed": 1307913890, "treatment": "gated"}, "priority": 0.48494184659685746, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "8b9f46c01409705a6ad1dc1585262bcfb96a5355e5cb4e46974d74ebd6b9a256", "run_seconds": 4.259759187698364, "seed": 1307913890, "started_at": "2026-11-22T03:46:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.076723, 0.025885], "effect": -0.025419, "evaluated_at": "2026-11-22T03:46:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-043a68818f5f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.534482759, 0.5, 0.568965517, 0.543478261], "gated": [0.534482759, 0.5, 0.568965517, 0.543478261], "naive": [0.732758621, 0.634615385, 0.75862069, 0.554347826], "scratch": [0.620689655, 0.548076923, 0.568965517, 0.510869565], "ungated": [0.655172414, 0.528846154, 0.655172414, 0.489130435]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "2b933cd58ee6ba2264a56b2a9ee2f7b3a48a7b9afd442dd955ed998e5418b402", "primary_control": "scratch", "result_hash": "8b9f46c01409705a6ad1dc1585262bcfb96a5355e5cb4e46974d74ebd6b9a256", "signature": "333ef5cc96d86c40f66e10c60b57aca3a7c5e18577e62c4b2597c23ac63955d26abfcb55a68ea63fb4aa47cdacfd6e3279d88b9d662e7dc0b8faed4892cf7700", "state": "VERIFIED", "summary": {"compressed": 0.536732, "gated": 0.536732, "naive": 0.670086, "scratch": 0.56215, "ungated": 0.58208}, "ties": 1, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:46:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:19Z", "cpu_seconds": 0.7265443600000054, "created_at": "2026-11-22T03:45:49Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-154b1cf8fc23", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "2a3da0df9d83904e0882716939995a3922a3488306c32245c2f05c80105aff3c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-154b1cf8fc23", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:45:49Z", "seed": 954270439, "treatment": "gated"}, "priority": 0.49678827665748476, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "0623050d4cf38bfca146369b3794580dd746b5ec779cec5a91d10ca72f2a3984", "run_seconds": 0.8896687030792236, "seed": 954270439, "started_at": "2026-11-22T03:46:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.033856, 0.01098], "effect": -0.011438, "evaluated_at": "2026-11-22T03:46:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-154b1cf8fc23", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.411764706, 0.424836601, 0.437908497, 0.411764706], "gated": [0.411764706, 0.424836601, 0.437908497, 0.411764706], "naive": [0.411764706, 0.267973856, 0.437908497, 0.411764706], "scratch": [0.411764706, 0.424836601, 0.437908497, 0.45751634], "ungated": [0.437908497, 0.424836601, 0.31372549, 0.418300654]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2a3da0df9d83904e0882716939995a3922a3488306c32245c2f05c80105aff3c", "primary_control": "scratch", "result_hash": "0623050d4cf38bfca146369b3794580dd746b5ec779cec5a91d10ca72f2a3984", "signature": "dce41e94c3e3d7d9161565f233cf9293ee26da4b68a46f7e331fc9d1bd4b9bae5241c2e1d89565d7f6eddb8b215b8dff5e77128e0c3461d7ffda733522154907", "state": "VERIFIED", "summary": {"compressed": 0.421569, "gated": 0.421569, "naive": 0.382353, "scratch": 0.433007, "ungated": 0.398693}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:46:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:07Z", "cpu_seconds": 4.477210067999991, "created_at": "2026-11-22T03:45:49Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-704cde8b8168", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "prereg_hash": "1503235e81e3941a2583ac3c6f18bf0d75dbf3a8bff6ada0c29c8336924e4237", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-704cde8b8168", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.15}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:45:49Z", "seed": 1124753108, "treatment": "frontier"}, "priority": 0.5088678967940805, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-b392e83d7ef2", "result_hash": "af359216b5622cd7985138cb5e790cb9cadaff45bc3c9491ca859550d42f4638", "run_seconds": 5.061286687850952, "seed": 1124753108, "started_at": "2026-11-22T03:46:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.04392, 0.05877], "effect": 0.051345, "evaluated_at": "2026-11-22T03:46:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-704cde8b8168", "family": "world_model_frontier", "hypothesis_supported": true, "instance_scores": {"curiosity": [0.914634146, 0.911016949, 0.922916667, 0.891393443, 0.899159664, 0.931451613, 0.91733871, 0.905172414, 0.909836066, 0.93699187, 0.911637931, 0.955426357], "frontier": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "random_walk": [0.953252033, 0.947033898, 0.952083333, 0.950819672, 0.966386555, 0.959677419, 0.929435484, 0.935344828, 0.94057377, 0.959349593, 0.926724138, 0.963178295]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -3.6124, "losses": 0, "n": 12, "outcome": "SUPPORTED", "p_sign_display": "0.000244", "prereg_hash": "1503235e81e3941a2583ac3c6f18bf0d75dbf3a8bff6ada0c29c8336924e4237", "primary_control": "random_walk", "replicated": true, "replication_of": "E141-world_model_frontier-b392e83d7ef2", "result_hash": "af359216b5622cd7985138cb5e790cb9cadaff45bc3c9491ca859550d42f4638", "signature": "62944c56e5398ebc71e412d9ef02186336ef5e4bbacaa0b21cab61aaef89cb94b0c79c3beee08df60bacde7ea098a126e6d708d6ce63f7dbee14eb4a505d9105", "state": "VERIFIED", "summary": {"curiosity": 0.917248, "frontier": 1.0, "random_walk": 0.948655}, "ties": 0, "treatment": "frontier", "wins": 12}, "verified_at": "2026-11-22T03:46:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:13Z", "cpu_seconds": 1.5090111070000063, "created_at": "2026-11-22T03:46:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c1e29a090e47", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "dc967c7ce056f1e313ea815f46982d58074c762f69dc033584e33f9b5312073f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c1e29a090e47", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:05Z", "seed": 67425748, "treatment": "frontier"}, "priority": 0.5114789879851791, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "109916fac8300640f92c3bb9b795a0aba513ee1efc56c0a1d62997c2655ada60", "run_seconds": 1.6093099117279053, "seed": 67425748, "started_at": "2026-11-22T03:46:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024667, 0.008], "effect": -0.008333, "evaluated_at": "2026-11-22T03:46:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c1e29a090e47", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.766666667, 0.775, 0.814814815, 0.933333333], "frontier": [0.866666667, 0.916666667, 0.814814815, 0.933333333], "goal": [0.833333333, 0.85, 0.814814815, 0.933333333], "hypdisc": [0.766666667, 0.85, 0.814814815, 0.866666667], "infogain": [0.866666667, 0.916666667, 0.814814815, 0.933333333], "prederr": [0.866666667, 0.916666667, 0.814814815, 0.933333333], "random": [0.9, 0.916666667, 0.814814815, 0.933333333], "risk": [0.866666667, 0.916666667, 0.814814815, 0.933333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "dc967c7ce056f1e313ea815f46982d58074c762f69dc033584e33f9b5312073f", "primary_control": "random", "result_hash": "109916fac8300640f92c3bb9b795a0aba513ee1efc56c0a1d62997c2655ada60", "signature": "54539a126d44bd81aa74bbdb0cad9149824db76cee16aa833c1e66b708600907cba77b967726563a49a33410614a38d1307e6ebe5a3b18d741c459c9addfd60d", "state": "VERIFIED", "summary": {"count": 0.822454, "frontier": 0.88287, "goal": 0.85787, "hypdisc": 0.824537, "infogain": 0.88287, "prederr": 0.88287, "random": 0.891204, "risk": 0.88287}, "ties": 3, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:46:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:12Z", "cpu_seconds": 1.245026487000004, "created_at": "2026-11-22T03:46:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f5dd7315ae71", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "80d0ce5d3aa995142fb620c92ee5d12946dedebb49ffea8d7a20cace6c1dcb7a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f5dd7315ae71", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:05Z", "seed": 2019804528, "treatment": "frontier"}, "priority": 0.5255331835655205, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "247bfe6050ca063ca324be0badb75cbd7968f8508de7d84f5542d3cd897250dd", "run_seconds": 1.3241360187530518, "seed": 2019804528, "started_at": "2026-11-22T03:46:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.128152, 0.190402], "effect": 0.031125, "evaluated_at": "2026-11-22T03:46:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f5dd7315ae71", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.67, 0.875, 0.592, 0.825], "frontier": [0.94, 0.825, 0.592, 0.775], "goal": [0.82, 0.825, 0.592, 0.8625], "hypdisc": [0.87, 0.775, 0.584, 0.825], "infogain": [0.94, 0.825, 0.584, 0.925], "prederr": [0.94, 0.675, 0.656, 0.725], "random": [0.67, 0.875, 0.6, 0.8625], "risk": [0.94, 0.825, 0.664, 0.775]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "80d0ce5d3aa995142fb620c92ee5d12946dedebb49ffea8d7a20cace6c1dcb7a", "primary_control": "random", "result_hash": "247bfe6050ca063ca324be0badb75cbd7968f8508de7d84f5542d3cd897250dd", "signature": "f23118ea921fc98131a8a1406191e3be0a577011f4ca7abe8454c90a67be0e8897808593b3cc92060cfcbf8f1e74977704e18d0019d8c9b198c52ef602dabd0d", "state": "VERIFIED", "summary": {"count": 0.7405, "frontier": 0.783, "goal": 0.774875, "hypdisc": 0.7635, "infogain": 0.8185, "prederr": 0.749, "random": 0.751875, "risk": 0.801}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:46:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:35Z", "cpu_seconds": 1.9653612710000061, "created_at": "2026-11-22T03:46:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-eeb8e96496f2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "e4de88e21ed7699c0ad222c62f0daf7019542e3b27d83addf748ca46210d9abf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-eeb8e96496f2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:46:05Z", "seed": 623075126, "treatment": "full"}, "priority": 0.4813061735349451, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c23698a708aad6a738d12e44fc1eb5ec75eab1c3cdb3e1d9257a19bc0d4d827d", "run_seconds": 2.089383602142334, "seed": 623075126, "started_at": "2026-11-22T03:46:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.035448, 0.248789], "effect": 0.142119, "evaluated_at": "2026-11-22T03:46:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-eeb8e96496f2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.604444444, 0.11625, 0.817683881, 0.579288026], "frozen": [0.302222222, 0.0, 0.319248826, 0.310679612], "full": [0.817777778, 0.15, 0.883411581, 0.834951456], "retrieval": [0.911851852, 0.7425, 0.966353678, 0.902265372], "rules_nodetect": [0.422962963, 0.235, 0.883411581, 0.401941748]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "e4de88e21ed7699c0ad222c62f0daf7019542e3b27d83addf748ca46210d9abf", "primary_control": "facts", "result_hash": "c23698a708aad6a738d12e44fc1eb5ec75eab1c3cdb3e1d9257a19bc0d4d827d", "signature": "7701f8fe4d6a4f891bc9acb9e3259e469a5c4355a6ef5eabedcbc053560b6eb53c18f51b6fefb70327055b046eca74708937d87222cc8e52f422360f7e436202", "state": "VERIFIED", "summary": {"facts": 0.529417, "frozen": 0.233038, "full": 0.671535, "retrieval": 0.880743, "rules_nodetect": 0.485829}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:46:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:33Z", "cpu_seconds": 2.459861105999991, "created_at": "2026-11-22T03:46:05Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2ca1bfc3fe42", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "e6d58d9eb6f05204ba435aea9197235a80fec7301549a3779452c437816c50e4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2ca1bfc3fe42", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:46:05Z", "seed": 173064993, "treatment": "gated"}, "priority": 0.4831547431162687, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "a762b7cd55fb8f13682ce1c6136e5dd7da06464eee41e7e0e829032f45c0f68d", "run_seconds": 2.576368570327759, "seed": 173064993, "started_at": "2026-11-22T03:46:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007982, 0.107341], "effect": 0.049679, "evaluated_at": "2026-11-22T03:46:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2ca1bfc3fe42", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.558333333, 0.45, 0.691666667, 0.490384615], "gated": [0.558333333, 0.45, 0.691666667, 0.490384615], "naive": [0.708333333, 0.45, 0.691666667, 0.701923077], "scratch": [0.558333333, 0.45, 0.608333333, 0.375], "ungated": [0.633333333, 0.266666667, 0.691666667, 0.615384615]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "e6d58d9eb6f05204ba435aea9197235a80fec7301549a3779452c437816c50e4", "primary_control": "scratch", "result_hash": "a762b7cd55fb8f13682ce1c6136e5dd7da06464eee41e7e0e829032f45c0f68d", "signature": "8f8a86e298eb6eeeed88e52781baba39ad4bd7763bedb4d0d480671e613d708fded5a5f901d4db15e2c70c925ed48d5248ee31f3e01edd958c19f0dd188f6205", "state": "VERIFIED", "summary": {"compressed": 0.547596, "gated": 0.547596, "naive": 0.637981, "scratch": 0.497917, "ungated": 0.551763}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:46:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:25Z", "cpu_seconds": 3.068542738000005, "created_at": "2026-11-22T03:46:05Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-af111947b2fa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "8e63912bbd6da54a849c119e39fe054001a35305788ab9193178af5d81486286", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-af111947b2fa", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:46:05Z", "seed": 1693461284, "treatment": "gated"}, "priority": 0.49618825991497917, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "629a7c9bd6d711763a36597a0123c722f0281f1ecbd690d327fc140f3acbe1f0", "run_seconds": 3.233581066131592, "seed": 1693461284, "started_at": "2026-11-22T03:46:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013782, 0.030449], "effect": 0.008333, "evaluated_at": "2026-11-22T03:46:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-af111947b2fa", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.808333333, 0.866666667, 0.875, 0.808333333], "gated": [0.808333333, 0.866666667, 0.875, 0.808333333], "naive": [0.9, 0.866666667, 0.875, 0.783333333], "scratch": [0.816666667, 0.866666667, 0.875, 0.766666667], "ungated": [0.9, 0.866666667, 0.875, 0.783333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "8e63912bbd6da54a849c119e39fe054001a35305788ab9193178af5d81486286", "primary_control": "scratch", "result_hash": "629a7c9bd6d711763a36597a0123c722f0281f1ecbd690d327fc140f3acbe1f0", "signature": "03da0bd9418509ed5259eb09830c2f72e4995bc3f26646e4569a7e177479d4cfcae896f15919e260cd8375d2567d59d49dffd0af39e7b82f8687bef6b50e1404", "state": "VERIFIED", "summary": {"compressed": 0.839583, "gated": 0.839583, "naive": 0.85625, "scratch": 0.83125, "ungated": 0.85625}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:46:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:17Z", "cpu_seconds": 8.949093848999993, "created_at": "2026-11-22T03:46:05Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-d52f35705559", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "fce15d04df0f55e7a62644893ae03539107120a3b32054890e93325ae10388a1", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-d52f35705559", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:46:05Z", "seed": 1286501638, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "f6eedb525ab748a6c54065c0b448e585c3412e69774dfff5355426e451005911", "run_seconds": 9.392147779464722, "seed": 1286501638, "started_at": "2026-11-22T03:46:07Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.194016, 0.160682], "effect": -0.016667, "evaluated_at": "2026-11-22T03:46:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-d52f35705559", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.7705387610974597, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0], "reset": [1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1908, "losses": 15, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.644", "prereg_hash": "fce15d04df0f55e7a62644893ae03539107120a3b32054890e93325ae10388a1", "primary_control": "reset", "result_hash": "f6eedb525ab748a6c54065c0b448e585c3412e69774dfff5355426e451005911", "signature": "395a091b33dc7fe6474e0e8974a7f68c7f4e7dbd70aa1c7aabfda49ba582f26866fe6cd49bafe7f28f632a1bcd51e018e3fb8ef343ae96a8ca5fc88470c3500e", "state": "VERIFIED", "summary": {"inherit": 0.6166666666666667, "reset": 0.6333333333333333, "root": 0.3}, "ties": 31, "treatment": "inherit", "wins": 14}, "verified_at": "2026-11-22T03:46:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:43Z", "cpu_seconds": 7.969689959, "created_at": "2026-11-22T03:46:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-2f439a4a8487", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "929f23bf97eaa767d688e6dc2dad66b8c17b59db65008bc2c09d00f5566511fa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2f439a4a8487", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:22Z", "seed": 474184296, "treatment": "frontier"}, "priority": 0.4550521508721952, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "339d471a95852dba97ff44c154772f73150d2af3215ec49f4c115193cbe15c27", "run_seconds": 8.754682302474976, "seed": 474184296, "started_at": "2026-11-22T03:46:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.098401, 0.177498], "effect": 0.039548, "evaluated_at": "2026-11-22T03:46:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-2f439a4a8487", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.591428571, 0.512765957, 0.53902439, 0.58021978], "frontier": [0.608571429, 0.536170213, 0.392682927, 0.628571429], "goal": [0.368571429, 0.721276596, 0.331707317, 0.415384615], "hypdisc": [0.608571429, 0.495744681, 0.326829268, 0.41978022], "infogain": [0.614285714, 0.24893617, 0.392682927, 0.650549451], "prederr": [0.614285714, 0.270212766, 0.392682927, 0.556043956], "random": [0.597142857, 0.487234043, 0.514634146, 0.408791209], "risk": [0.608571429, 0.225531915, 0.392682927, 0.628571429]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "929f23bf97eaa767d688e6dc2dad66b8c17b59db65008bc2c09d00f5566511fa", "primary_control": "random", "result_hash": "339d471a95852dba97ff44c154772f73150d2af3215ec49f4c115193cbe15c27", "signature": "d917e3187f62df0d334aa82d113621c9a312cdad28e9f03eb57ba1bbe570a8d965d32904146e10b92bfa44ab25712b66fd1b93e90ef03fba2a747f000be4660e", "state": "VERIFIED", "summary": {"count": 0.55586, "frontier": 0.541499, "goal": 0.459235, "hypdisc": 0.462731, "infogain": 0.476614, "prederr": 0.458306, "random": 0.501951, "risk": 0.463839}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:46:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:41Z", "cpu_seconds": 2.720426102999994, "created_at": "2026-11-22T03:46:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4bbfe6602bf6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "f53460bf62425461201c7fd673d9188bdddda8454579a84c2a9fe7dad580c2cc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4bbfe6602bf6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:22Z", "seed": 1669003176, "treatment": "frontier"}, "priority": 0.42848726267981396, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ca7c8f9013609475152be800f5164411793c34b186087360fe4c124b5c7ce01f", "run_seconds": 3.0493826866149902, "seed": 1669003176, "started_at": "2026-11-22T03:46:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.204308, -0.006161], "effect": -0.105234, "evaluated_at": "2026-11-22T03:46:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4bbfe6602bf6", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.6875, 0.605263158, 0.444444444, 0.833333333], "frontier": [0.588068182, 0.302631579, 0.341269841, 0.760416667], "goal": [0.590909091, 0.276315789, 0.390873016, 0.661458333], "hypdisc": [0.580965909, 0.710526316, 0.438492063, 0.84375], "infogain": [0.588068182, 0.302631579, 0.341269841, 0.760416667], "prederr": [0.588068182, 0.302631579, 0.341269841, 0.760416667], "random": [0.649147727, 0.539473684, 0.464285714, 0.760416667], "risk": [0.588068182, 0.302631579, 0.341269841, 0.760416667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "f53460bf62425461201c7fd673d9188bdddda8454579a84c2a9fe7dad580c2cc", "primary_control": "random", "result_hash": "ca7c8f9013609475152be800f5164411793c34b186087360fe4c124b5c7ce01f", "signature": "e5d313deedf940f7ac84c392313a2404a4c1ed63044703e56abd2de8167f13e1d5bf31964353ccaa5d255ca3e53ff103fb492e7d9928f242e03741e2121a640b", "state": "VERIFIED", "summary": {"count": 0.642635, "frontier": 0.498097, "goal": 0.479889, "hypdisc": 0.643434, "infogain": 0.498097, "prederr": 0.498097, "random": 0.603331, "risk": 0.498097}, "ties": 1, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:46:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:45Z", "cpu_seconds": 4.259792086999994, "created_at": "2026-11-22T03:46:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-fa9d19e7686f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "b272e5d86c2ee017212ed16adfc65c8590ff6e9df6afa9066c3f0fcf5fc66085", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-fa9d19e7686f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:22Z", "seed": 434368727, "treatment": "frontier"}, "priority": 0.4260403485282384, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "70bdb2143fa7bd3f5efdd48720179db5c9df7782cb46ae317cb918c059f9200d", "run_seconds": 4.693549394607544, "seed": 434368727, "started_at": "2026-11-22T03:46:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01788, 0.023684], "effect": 0.002902, "evaluated_at": "2026-11-22T03:46:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-fa9d19e7686f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.647135417, 0.366300366, 0.356900452, 0.324349442], "frontier": [0.640625, 0.357600733, 0.303167421, 0.326208178], "goal": [0.60546875, 0.325091575, 0.312782805, 0.304832714], "hypdisc": [0.639322917, 0.331501832, 0.345022624, 0.312267658], "infogain": [0.640625, 0.357600733, 0.303167421, 0.326208178], "prederr": [0.640625, 0.357600733, 0.303167421, 0.326208178], "random": [0.638020833, 0.328296703, 0.325791855, 0.323884758], "risk": [0.640625, 0.357600733, 0.303167421, 0.326208178]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "b272e5d86c2ee017212ed16adfc65c8590ff6e9df6afa9066c3f0fcf5fc66085", "primary_control": "random", "result_hash": "70bdb2143fa7bd3f5efdd48720179db5c9df7782cb46ae317cb918c059f9200d", "signature": "676797c99e46e3978c3c99023fcd2ba9d4757f5ee396a39609cff1d89ece5d2f5ce3381abf645a65146e18b1d6040a55d4e4700c003f69ae0369a98c4536d405", "state": "VERIFIED", "summary": {"count": 0.423671, "frontier": 0.4069, "goal": 0.387044, "hypdisc": 0.407029, "infogain": 0.4069, "prederr": 0.4069, "random": 0.403999, "risk": 0.4069}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:47:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:55Z", "cpu_seconds": 0.59849093199999, "created_at": "2026-11-22T03:46:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3529c43876e3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "6d6a34a1748520d763ca210f7e06dbcba2e269c039dd0a86a0fa50a635f1929d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3529c43876e3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:46:22Z", "seed": 1880518689, "treatment": "full"}, "priority": 0.4220539978159634, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "00dce24766653818b01134565709a9ebbe79ba50464b17cd69276aa2a1fc04e5", "run_seconds": 0.7805662155151367, "seed": 1880518689, "started_at": "2026-11-22T03:46:54Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:47:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3529c43876e3", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.5, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6d6a34a1748520d763ca210f7e06dbcba2e269c039dd0a86a0fa50a635f1929d", "primary_control": "e141_best", "result_hash": "00dce24766653818b01134565709a9ebbe79ba50464b17cd69276aa2a1fc04e5", "signature": "4962b3760a90bd63a78683a78ed215afbd1627462fd8da01cd5dfcc7371c07e228cd87781c64e943b8b976372bbfc6cfa7df782a46de1daae045b38557311104", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.125, "full": 1.0, "no_world_model": 0.875, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:47:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:12Z", "cpu_seconds": 0.6628116560000024, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-05344540eb1f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "bcc30e2be29fda7a351e5348eacbff492fad6ee375ab14af88810cc9f0b6d577", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-05344540eb1f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:38Z", "seed": 353830569, "treatment": "frontier"}, "priority": 0.4176752165902241, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ee38122aac87c705abd5fd64ab15a5b0c53d07cc83a4995b91420dabc28f38e2", "run_seconds": 0.8269190788269043, "seed": 353830569, "started_at": "2026-11-22T03:47:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.249062, 0.136562], "effect": -0.05625, "evaluated_at": "2026-11-22T03:47:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-05344540eb1f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.883333333, 1.0, 1.0, 0.8], "frontier": [0.9, 1.0, 1.0, 0.4625], "goal": [0.933333333, 0.9, 0.916666667, 0.8625], "hypdisc": [0.9, 1.0, 1.0, 0.85], "infogain": [0.9, 1.0, 1.0, 0.875], "prederr": [0.85, 1.0, 1.0, 0.675], "random": [0.833333333, 0.9625, 0.979166667, 0.8125], "risk": [0.9, 1.0, 1.0, 0.4625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "bcc30e2be29fda7a351e5348eacbff492fad6ee375ab14af88810cc9f0b6d577", "primary_control": "random", "result_hash": "ee38122aac87c705abd5fd64ab15a5b0c53d07cc83a4995b91420dabc28f38e2", "signature": "19a2525efeb8215a987a7b68537d62329d3446243d072a40d493bd75c77d88c9e2fd135d21a85ed832d2f67b230ee322f0ee3232d607104d5acb502ecbf36809", "state": "VERIFIED", "summary": {"count": 0.920833, "frontier": 0.840625, "goal": 0.903125, "hypdisc": 0.9375, "infogain": 0.94375, "prederr": 0.88125, "random": 0.896875, "risk": 0.840625}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:47:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:02Z", "cpu_seconds": 0.5745577589999868, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-253c598dfac1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "0d3ef5063b5dcae3a12817a77c841311cfaa70bbaa960dd2c7b9452375fcf23b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-253c598dfac1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:38Z", "seed": 1458048063, "treatment": "frontier"}, "priority": 0.42149178368664414, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f89d00ab87f3172f54a859ab45408ec401a5faaa5b316025c9d74d35b13e2318", "run_seconds": 0.5910458564758301, "seed": 1458048063, "started_at": "2026-11-22T03:47:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.010305, 0.029651], "effect": 0.009673, "evaluated_at": "2026-11-22T03:47:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-253c598dfac1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 0.958333333, 0.916666667, 1.0], "goal": [0.857142857, 0.9375, 0.791666667, 0.958333333], "hypdisc": [0.928571429, 0.895833333, 0.958333333, 0.979166667], "infogain": [1.0, 0.958333333, 0.916666667, 1.0], "prederr": [1.0, 0.958333333, 0.916666667, 1.0], "random": [0.982142857, 0.9375, 0.9375, 0.979166667], "risk": [1.0, 0.958333333, 0.916666667, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "0d3ef5063b5dcae3a12817a77c841311cfaa70bbaa960dd2c7b9452375fcf23b", "primary_control": "random", "result_hash": "f89d00ab87f3172f54a859ab45408ec401a5faaa5b316025c9d74d35b13e2318", "signature": "6a0ea71dc81a21f9be2dc8fe76e092512be23a97b1b2e712754b32d9d432ca8149d9807071f7366c69f29df4cf9f09ca8056a716f56b123c3e6e40af07d81d0f", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.96875, "goal": 0.886161, "hypdisc": 0.940476, "infogain": 0.96875, "prederr": 0.96875, "random": 0.959077, "risk": 0.96875}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:47:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:38Z", "cpu_seconds": 0.9760125970000075, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-564bf423f6ac", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "d01cf02aa6101e984920b6ca3fa0430d0266d959466bbe95f9b2dc6519476c6b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-564bf423f6ac", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:38Z", "seed": 1309677282, "treatment": "frontier"}, "priority": 0.41506029873338557, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3ec673f52e53ea277335b544afee639538d93ab56e83a9f0294f9366ae97f152", "run_seconds": 1.0067696571350098, "seed": 1309677282, "started_at": "2026-11-22T03:47:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.048374, 0.008738], "effect": -0.019818, "evaluated_at": "2026-11-22T03:47:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-564bf423f6ac", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.797101449, 0.947368421, 0.979166667, 0.995884774], "frontier": [0.811594203, 0.929824561, 0.979166667, 0.934156379], "goal": [0.594202899, 0.929824561, 0.979166667, 0.934156379], "hypdisc": [0.753623188, 0.929824561, 0.958333333, 0.934156379], "infogain": [0.811594203, 0.929824561, 0.979166667, 0.934156379], "prederr": [0.811594203, 0.824561404, 0.979166667, 0.934156379], "random": [0.811594203, 0.947368421, 0.979166667, 0.995884774], "risk": [0.811594203, 0.929824561, 0.979166667, 0.934156379]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d01cf02aa6101e984920b6ca3fa0430d0266d959466bbe95f9b2dc6519476c6b", "primary_control": "random", "result_hash": "3ec673f52e53ea277335b544afee639538d93ab56e83a9f0294f9366ae97f152", "signature": "a0ba3a24e5f9026b0123901a1d142172f6a7e8e076a4b6c20253c3bab4333ae19c8880c6f93a9c8c43437af66a2749ac85757436633cd691a07b66a28d1ea50c", "state": "VERIFIED", "summary": {"count": 0.92988, "frontier": 0.913685, "goal": 0.859338, "hypdisc": 0.893984, "infogain": 0.913685, "prederr": 0.88737, "random": 0.933504, "risk": 0.913685}, "ties": 2, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:47:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:49Z", "cpu_seconds": 0.653964399000003, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-03b01943c552", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "dc5187d9e6c552e3ba2a0af8cd4fb89588e48dd8bc87df2aff8c7c865bd4d129", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-03b01943c552", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:46:38Z", "seed": 552563892, "treatment": "full"}, "priority": 0.4254970338518479, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "48e1211c2f9b707eac093a20f4be68b67687be610e2c8ff86b3e1bef221b36c7", "run_seconds": 0.8834161758422852, "seed": 552563892, "started_at": "2026-11-22T03:46:48Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:47:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-03b01943c552", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.9375], "fixed_policy": [1.0, 0.0], "full": [1.0, 0.9375], "no_world_model": [1.0, 0.9375], "random": [1.0, 0.6875]}, "prereg_hash": "dc5187d9e6c552e3ba2a0af8cd4fb89588e48dd8bc87df2aff8c7c865bd4d129", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "48e1211c2f9b707eac093a20f4be68b67687be610e2c8ff86b3e1bef221b36c7", "signature": "49742f237ae22892e231ecd39f6d302fa086854f5d8d0b6465135bb57d9a5ac8a0b5c86f9cd7e23748aa60fb6f84db492e9f8819c9f9c5781e2bfe2f57e25400", "state": "REJECTED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.9375, "random": 0.6875}}, "verified_at": "2026-11-22T03:47:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:06Z", "cpu_seconds": 0.7867850579999924, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1ff96c7dc73f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "374aba6fe6b78dcffa7152d953a52d10f56e03807f9df9b9acab404c05efb653", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1ff96c7dc73f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:46:38Z", "seed": 984675883, "treatment": "full"}, "priority": 0.42021206290705493, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f21528abb7f3602eb92430ccf066fb110f3c15d9aa30c9f18fd77d96e20fda52", "run_seconds": 0.9248168468475342, "seed": 984675883, "started_at": "2026-11-22T03:47:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:47:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1ff96c7dc73f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.0, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.0, 1.0], "no_world_model": [1.0, 1.0, 0.0, 0.75], "random": [1.0, 1.0, 0.25, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "374aba6fe6b78dcffa7152d953a52d10f56e03807f9df9b9acab404c05efb653", "primary_control": "e141_best", "result_hash": "f21528abb7f3602eb92430ccf066fb110f3c15d9aa30c9f18fd77d96e20fda52", "signature": "ef2f784c6586bff98419ffb58e99ecb622b6a27b490016dce6679bbd52799ce6f29043ed5f2e7399b3508168de5dfffa5bd92543fad78257f87582eabb846801", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.0, "full": 0.75, "no_world_model": 0.6875, "random": 0.8125}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:47:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:44Z", "cpu_seconds": 0.5237779539999963, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-516d53adcbd6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "3010f3d7bcc51b0ff00d6af366f9c6f4cfef4b04f227387674ef2fad1c8b369a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-516d53adcbd6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:46:38Z", "seed": 839972657, "treatment": "full"}, "priority": 0.413556783316091, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "32a800e952611956971e4e35ebb47e9a8d94ab6dab36ab5b4305b1c19efe40ba", "run_seconds": 0.5529851913452148, "seed": 839972657, "started_at": "2026-11-22T03:47:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T03:47:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-516d53adcbd6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 0.75, 1.0, 1.0], "random": [0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "3010f3d7bcc51b0ff00d6af366f9c6f4cfef4b04f227387674ef2fad1c8b369a", "primary_control": "e141_best", "result_hash": "32a800e952611956971e4e35ebb47e9a8d94ab6dab36ab5b4305b1c19efe40ba", "signature": "184edd705d14060dd531034499425e593208baae54fe8a4e98527ef984172810e2141c3fc85c0cf7ba65e98337d8a868d54a569c0a385d6db31523a4a3059108", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0625, "full": 0.75, "no_world_model": 0.6875, "random": 0.25}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:47:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:51Z", "cpu_seconds": 0.8072684790000011, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e482441f2299", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "30f3fd3fc5dc340404896ec8a1d6e29c13771c106492e1fec0cd280a5107ef60", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e482441f2299", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:46:38Z", "seed": 1319425634, "treatment": "full"}, "priority": 0.42223237394316915, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7ead0dec18dfd08d0dd14af5e07dc7ae2b47951d72eb8b288a17a7e6560adb92", "run_seconds": 0.9797852039337158, "seed": 1319425634, "started_at": "2026-11-22T03:46:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.094139, 0.844139], "effect": 0.375, "evaluated_at": "2026-11-22T03:47:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e482441f2299", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.0, 0.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [1.0, 0.0, 0.5, 1.0], "no_world_model": [1.0, 0.0, 0.25, 0.5], "random": [1.0, 0.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "30f3fd3fc5dc340404896ec8a1d6e29c13771c106492e1fec0cd280a5107ef60", "primary_control": "e141_best", "result_hash": "7ead0dec18dfd08d0dd14af5e07dc7ae2b47951d72eb8b288a17a7e6560adb92", "signature": "583747e2aa693b6be55a3167ce6806b0f8775e033733d16b523fa8361aaea6cdf9adabc570beb88fdf4c46dbf21b550f23ab63fe941ee4ec1f96ed3e75890302", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 0.625, "no_world_model": 0.4375, "random": 0.625}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:47:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:56Z", "cpu_seconds": 8.554365544000007, "created_at": "2026-11-22T03:46:38Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-7b2fc6017d28", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "dfe0e3a37453e798a6790b708e74a517c356916aa1e42f124dc6d255682f4ea3", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-7b2fc6017d28", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:46:38Z", "seed": 1530695158, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "4bbefd6c70ad4b7487e857cd4b9d9ee37e0e3b8bfe29bd4c154973570540897f", "run_seconds": 9.88363790512085, "seed": 1530695158, "started_at": "2026-11-22T03:46:46Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:47:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-7b2fc6017d28", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "dfe0e3a37453e798a6790b708e74a517c356916aa1e42f124dc6d255682f4ea3", "primary_control": "reset", "result_hash": "4bbefd6c70ad4b7487e857cd4b9d9ee37e0e3b8bfe29bd4c154973570540897f", "signature": "bcb23ff8384a919652b0dbeca616ca9a9465b296a4b0b90dc1e8c1975493882293ce0741b573341dd17e980c056b6209aa3a96dace20a63aebc3695348454801", "state": "PARTIAL", "summary": {"inherit": 0.5166666666666667, "reset": 0.5166666666666667, "root": 0.26666666666666666}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:47:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:14Z", "cpu_seconds": 2.4751274329999973, "created_at": "2026-11-22T03:46:55Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-640149ca09d6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "82c025747f0597a516a80da9bcacc5c14817546089c66169796fbd1aee5284c4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-640149ca09d6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:46:55Z", "seed": 841629731, "treatment": "frontier"}, "priority": 0.41783684900918355, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "292f360ecf5af3e7f3e239c4cfcd24422f056d8dc8646348a2d3b2449e4dddc5", "run_seconds": 2.7262325286865234, "seed": 841629731, "started_at": "2026-11-22T03:47:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004405, 0.001429], "effect": -0.001488, "evaluated_at": "2026-11-22T03:47:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-640149ca09d6", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.988095238, 0.986111111], "frontier": [0.994047619, 0.997395833, 1.0, 1.0], "goal": [0.994047619, 0.911458333, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [0.994047619, 0.997395833, 1.0, 1.0], "prederr": [0.988095238, 0.997395833, 1.0, 1.0], "random": [1.0, 0.997395833, 1.0, 1.0], "risk": [0.994047619, 0.997395833, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "82c025747f0597a516a80da9bcacc5c14817546089c66169796fbd1aee5284c4", "primary_control": "random", "result_hash": "292f360ecf5af3e7f3e239c4cfcd24422f056d8dc8646348a2d3b2449e4dddc5", "signature": "5384858f47568ce839b1793cdefb0b34893bc39d068e173c7b7fe69d0541deee05ba30764c1bdb212ffa78d934994d9d607162ba28603226c20210d7cee5c90c", "state": "VERIFIED", "summary": {"count": 0.993552, "frontier": 0.997861, "goal": 0.976376, "hypdisc": 1.0, "infogain": 0.997861, "prederr": 0.996373, "random": 0.999349, "risk": 0.997861}, "ties": 3, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:47:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:00Z", "cpu_seconds": 0.3307406880000059, "created_at": "2026-11-22T03:46:55Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-800be0f35b66", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "4a8e3971bbd298de4df14b2a02c0a81050653a15b2b81af191724571a433f307", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-800be0f35b66", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:46:55Z", "seed": 1233012291, "treatment": "full"}, "priority": 0.42799195714038263, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "12cb051676df877d4690ce209c6f0817bcbee23b04ec15aec43c1d261c3808f4", "run_seconds": 0.33188724517822266, "seed": 1233012291, "started_at": "2026-11-22T03:47:00Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:47:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-800be0f35b66", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 0.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4a8e3971bbd298de4df14b2a02c0a81050653a15b2b81af191724571a433f307", "primary_control": "e141_best", "result_hash": "12cb051676df877d4690ce209c6f0817bcbee23b04ec15aec43c1d261c3808f4", "signature": "5ee6b7e1616d96074a003fc2e5e57ea1f96ef5020dab022daf6ae8cb681860deae10074b755aba50b09ac406237a676109186560ebde33cdb2125fe86622f207", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.75, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:47:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:08Z", "cpu_seconds": 0.8341039809999984, "created_at": "2026-11-22T03:46:55Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-848e771e081b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "cb1b9d9c3e10345d8a68188ad422d6328b8513ad819dc2baacf4c38d15096ab8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-848e771e081b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:46:55Z", "seed": 342442011, "treatment": "full"}, "priority": 0.4191404189033521, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "debf7903ce599ec9f9a57dcd29c173eb15c428ed1453923f307e1cf536741765", "run_seconds": 0.9412891864776611, "seed": 342442011, "started_at": "2026-11-22T03:47:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:47:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-848e771e081b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.5, 1.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [1.0, 1.0, 0.5, 1.0], "no_world_model": [1.0, 0.5, 0.25, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "cb1b9d9c3e10345d8a68188ad422d6328b8513ad819dc2baacf4c38d15096ab8", "primary_control": "e141_best", "result_hash": "debf7903ce599ec9f9a57dcd29c173eb15c428ed1453923f307e1cf536741765", "signature": "3127f3990b975d67129a2d73b8e69db1f85c9af39acc16f88e86e15b2e0821f314fd596d29e754379b39ed9409646d819c5bdb39195225e2e7491d3ebdd38b07", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.0625, "full": 0.875, "no_world_model": 0.6875, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:47:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:46:57Z", "cpu_seconds": 0.6606317950000005, "created_at": "2026-11-22T03:46:55Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c07bb91ffe6e", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "e60ce9c6aa16aa03a2e8c4eaaa7767a24d692919b9508d513f90f6be16148e13", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c07bb91ffe6e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:46:55Z", "seed": 1181944587, "treatment": "full"}, "priority": 0.6389557269475469, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-1ae849c9d6fe", "result_hash": "ccb1c3b73b25fa1de2f8d3f03ff0d18c132e63cf97daff9ca342a31e6c50b48a", "run_seconds": 0.790980339050293, "seed": 1181944587, "started_at": "2026-11-22T03:46:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004805, 0.035475], "effect": 0.015335, "evaluated_at": "2026-11-22T03:47:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c07bb91ffe6e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.484126984, 0.57183908, 0.866666667, 0.545454545], "frozen": [0.317460317, 0.336206897, 0.4, 0.333333333], "full": [0.507936508, 0.568965517, 0.866666667, 0.585858586], "retrieval": [0.555555556, 0.623563218, 0.866666667, 0.666666667], "rules_nodetect": [0.69047619, 0.75862069, 0.866666667, 0.828282828]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "e60ce9c6aa16aa03a2e8c4eaaa7767a24d692919b9508d513f90f6be16148e13", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-1ae849c9d6fe", "result_hash": "ccb1c3b73b25fa1de2f8d3f03ff0d18c132e63cf97daff9ca342a31e6c50b48a", "signature": "3af1eacc268b9cc020bf0c692c960620cb030ca908ecf1f35c139dffc8b351b19a6c4625ada302a7d6f459b8925bf9d2eddf35c2ab3fd9d967d738357200d206", "state": "VERIFIED", "summary": {"facts": 0.617022, "frozen": 0.34675, "full": 0.632357, "retrieval": 0.678113, "rules_nodetect": 0.786012}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:47:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:07Z", "cpu_seconds": 5.479454842999999, "created_at": "2026-11-22T03:46:55Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0395ea7a22af", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "dc315f5b6a03ea573ebb2a233d3b851e67ad5fca8e0458d6010e74a72af31988", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0395ea7a22af", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:46:55Z", "seed": 1771492720, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "0bb30029b530024efe82522d7cc8e46bfbbe63720e358481999b5203326ac88f", "run_seconds": 5.826575040817261, "seed": 1771492720, "started_at": "2026-11-22T03:47:01Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.105613, 0.005613], "effect": -0.05, "evaluated_at": "2026-11-22T03:47:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0395ea7a22af", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.3475412221495311, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0], "reset": [1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0], "root": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "dc315f5b6a03ea573ebb2a233d3b851e67ad5fca8e0458d6010e74a72af31988", "primary_control": "reset", "result_hash": "0bb30029b530024efe82522d7cc8e46bfbbe63720e358481999b5203326ac88f", "signature": "7c559e9d50015fd9af2681664f167f8ef7c5718d02f7d5340c589a1319034244200ed15880859f4cb8a46dbe60b2acca7375c65d91a16c7288cae88a3a664a02", "state": "VERIFIED", "summary": {"inherit": 0.6166666666666667, "reset": 0.6666666666666666, "root": 0.38333333333333336}, "ties": 57, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:47:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:22Z", "cpu_seconds": 0.669858509000008, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0c4c27cb7172", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "1e8c0ae91d04411748a6bf7258e9ad87b557ce7d290825dddf8eb9065afb07d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0c4c27cb7172", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:47:12Z", "seed": 1521190137, "treatment": "full"}, "priority": 0.4212283050138812, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "38259c7141cdf7c458d73349d428fdbac8d7a45384770f0b09ea670108325d14", "run_seconds": 0.6721534729003906, "seed": 1521190137, "started_at": "2026-11-22T03:47:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.099917, 0.900083], "effect": 0.5, "evaluated_at": "2026-11-22T03:47:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0c4c27cb7172", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.5, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 0.5, 0.75, 1.0], "random": [0.75, 1.0, 0.75, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "1e8c0ae91d04411748a6bf7258e9ad87b557ce7d290825dddf8eb9065afb07d2", "primary_control": "e141_best", "result_hash": "38259c7141cdf7c458d73349d428fdbac8d7a45384770f0b09ea670108325d14", "signature": "05379914f4a309ea17851016f0cb248ab52a13c5a988abec02928dfdd76e312f1a07cb76ef4ddf41856a2c9e304cfcfd5e7228b752404fa86b7f4857aed1a70d", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.8125, "random": 0.625}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:47:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:27Z", "cpu_seconds": 0.3705075430000022, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f20e972f58b5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "806b11c1af92bf9dc6c54f5b9821e353b11aa0bc8b7831838e78e4fa520f1e68", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f20e972f58b5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:47:12Z", "seed": 2078743090, "treatment": "full"}, "priority": 0.4202886728853983, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d596282de8753361887085f41924d1406ffd95e571c20e44b37a000ca24cb513", "run_seconds": 0.38059449195861816, "seed": 2078743090, "started_at": "2026-11-22T03:47:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.620785, -0.004215], "effect": -0.3125, "evaluated_at": "2026-11-22T03:47:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f20e972f58b5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 0.25, 0.5], "fixed_policy": [0.25, 0.25, 0.0, 0.0], "full": [0.0, 0.75, 0.0, 0.25], "no_world_model": [0.0, 0.75, 0.0, 0.25], "random": [0.75, 0.75, 0.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "806b11c1af92bf9dc6c54f5b9821e353b11aa0bc8b7831838e78e4fa520f1e68", "primary_control": "e141_best", "result_hash": "d596282de8753361887085f41924d1406ffd95e571c20e44b37a000ca24cb513", "signature": "6d79083e3dc625e943c3abe0617b25ff730bd3f03601b4b1608fa7527953b6af7e91913dcea3a8169c11f8acec5596698a56836df0dfd8bfb8774931130c680b", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.125, "full": 0.25, "no_world_model": 0.25, "random": 0.5}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:47:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:28Z", "cpu_seconds": 0.6692933230000051, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-67a5d99ed6e4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "ac15221fe4bb1180f2c7a9ec05722e1a242fcd786286f2073ea38bc1dece8eb4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-67a5d99ed6e4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:47:12Z", "seed": 1220052319, "treatment": "full"}, "priority": 0.41589778258740406, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0e684213f992a42ba6627d9a78988d2d7c5268f8ffcddf0efe94ec10dec7ce07", "run_seconds": 0.7212882041931152, "seed": 1220052319, "started_at": "2026-11-22T03:47:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.060898, 0.037466], "effect": -0.011716, "evaluated_at": "2026-11-22T03:47:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-67a5d99ed6e4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.611111111, 0.233695652, 0.427884615, 0.775641026], "frozen": [0.092592593, 0.081521739, 0.264423077, 0.397435897], "full": [0.604938272, 0.29076087, 0.375, 0.730769231], "retrieval": [0.62962963, 0.448369565, 0.399038462, 0.775641026], "rules_nodetect": [0.604938272, 0.279891304, 0.442307692, 0.512820513]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "ac15221fe4bb1180f2c7a9ec05722e1a242fcd786286f2073ea38bc1dece8eb4", "primary_control": "facts", "result_hash": "0e684213f992a42ba6627d9a78988d2d7c5268f8ffcddf0efe94ec10dec7ce07", "signature": "29fad2c25092153a07586049adae4287bf5f5db5cc8d4ddeebf7fbccd8e1d0017b13d2f1bcdb5e81ce8e35ec7431fffce04c0e03d4647006e230c2b9e4272706", "state": "VERIFIED", "summary": {"facts": 0.512083, "frozen": 0.208993, "full": 0.500367, "retrieval": 0.56317, "rules_nodetect": 0.459989}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:47:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:25Z", "cpu_seconds": 0.5092704019999985, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c5458eb7aeda", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "ea2f1b1cc09480a52a3f0244dbd773b660baa9ac4f9b28a0f0ff2160a9c65b58", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c5458eb7aeda", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:47:12Z", "seed": 1964759880, "treatment": "full"}, "priority": 0.4204143179398443, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "50d9b6800c4b155bd828a97d9c15f5600eab636495ccdfb069369572b37b8c72", "run_seconds": 0.5274298191070557, "seed": 1964759880, "started_at": "2026-11-22T03:47:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.011827, 0.010787], "effect": -0.00052, "evaluated_at": "2026-11-22T03:47:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c5458eb7aeda", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.758169935, 0.310077519, 0.295454545, 0.426356589], "frozen": [0.176470588, 0.294573643, 0.28030303, 0.426356589], "full": [0.77124183, 0.310077519, 0.28030303, 0.426356589], "retrieval": [0.764705882, 0.302325581, 0.295454545, 0.426356589], "rules_nodetect": [0.875816993, 0.302325581, 0.28030303, 0.457364341]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "ea2f1b1cc09480a52a3f0244dbd773b660baa9ac4f9b28a0f0ff2160a9c65b58", "primary_control": "facts", "result_hash": "50d9b6800c4b155bd828a97d9c15f5600eab636495ccdfb069369572b37b8c72", "signature": "796d333572a9d79eab0640861d1927d7ac5f68a994e01de12c1b775ae0796e0746356aa024fedd317facd9cd528ff203959a0e9a781993e549e92217009c130d", "state": "VERIFIED", "summary": {"facts": 0.447515, "frozen": 0.294426, "full": 0.446995, "retrieval": 0.447211, "rules_nodetect": 0.478952}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:47:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:52Z", "cpu_seconds": 2.957519769000001, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e2d337512643", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "5e1dbd99ae3128fc1a5aca397956e6698c5f64d76fa09294b6fa4f4795a0f9b0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e2d337512643", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:47:12Z", "seed": 568996189, "treatment": "full"}, "priority": 0.4129393544687234, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "72d4d2a3713cbfbbab5690ef87eab401b0db5aa1a2dc70f7c6d8efe8236902c0", "run_seconds": 3.6137490272521973, "seed": 568996189, "started_at": "2026-11-22T03:47:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.050397, 0.284876], "effect": 0.167637, "evaluated_at": "2026-11-22T03:47:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e2d337512643", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.713580247, 0.721495327, 0.966666667, 0.515555556], "frozen": [0.325925926, 0.300934579, 0.3125, 0.306666667], "full": [0.859259259, 0.897196262, 0.995833333, 0.835555556], "retrieval": [0.908641975, 0.946728972, 1.0, 0.871111111], "rules_nodetect": [0.859259259, 0.392523364, 0.995833333, 0.269333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "5e1dbd99ae3128fc1a5aca397956e6698c5f64d76fa09294b6fa4f4795a0f9b0", "primary_control": "facts", "result_hash": "72d4d2a3713cbfbbab5690ef87eab401b0db5aa1a2dc70f7c6d8efe8236902c0", "signature": "2e795f093a607579de2ead4a7260cbd56a9b95e8d0a1d68789ef8a647c999e0a41d5e0dfc7179cd74d451b40ec5f1089ee6ecb31db1c47dc2682796396009f0d", "state": "VERIFIED", "summary": {"facts": 0.729324, "frozen": 0.311507, "full": 0.896961, "retrieval": 0.931621, "rules_nodetect": 0.629237}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:47:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:20Z", "cpu_seconds": 0.669297254, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1072b46c1611", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "74875448092db6a8cd2375d10714b75354f3e880c4af2488b511ece10139ce59", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1072b46c1611", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:47:12Z", "seed": 1176824766, "treatment": "gated"}, "priority": 0.42209089524439897, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "0f122777980c975d537b56928636ffd4c329a2f72397de640727feac795df409", "run_seconds": 0.7655317783355713, "seed": 1176824766, "started_at": "2026-11-22T03:47:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.04, 0.012973], "effect": -0.013514, "evaluated_at": "2026-11-22T03:47:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1072b46c1611", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.630630631, 0.622222222, 0.583333333, 0.857142857], "gated": [0.630630631, 0.622222222, 0.583333333, 0.857142857], "naive": [0.72972973, 0.644444444, 0.676666667, 0.904761905], "scratch": [0.684684685, 0.622222222, 0.583333333, 0.857142857], "ungated": [0.72972973, 0.644444444, 0.503333333, 0.904761905]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "74875448092db6a8cd2375d10714b75354f3e880c4af2488b511ece10139ce59", "primary_control": "scratch", "result_hash": "0f122777980c975d537b56928636ffd4c329a2f72397de640727feac795df409", "signature": "03109a9c85ea3908f47e70df790fc29621d8f51b5ff926aa14307ebe96c91576e3b5c1f1108d7feb9667a2f64820b3472fa6910b72cf225b3faeba73336e5e08", "state": "VERIFIED", "summary": {"compressed": 0.673332, "gated": 0.673332, "naive": 0.738901, "scratch": 0.686846, "ungated": 0.695567}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:47:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:16Z", "cpu_seconds": 3.0164487610000066, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-dad3d088bb2d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "c1980929059b2a705f75d17942e5ba8e6f5d4b4141a736e8aa1fea7252d780dd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-dad3d088bb2d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:47:12Z", "seed": 435126751, "treatment": "gated"}, "priority": 0.6307929667854494, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "129640d26afb96e717c2b6474a0dcce11f9f3d8808186401cfeade8df0227a46", "run_seconds": 3.1646101474761963, "seed": 435126751, "started_at": "2026-11-22T03:47:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.074, 0.024], "effect": -0.025, "evaluated_at": "2026-11-22T03:47:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-dad3d088bb2d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.6953125, 0.65, 0.54, 0.541666667], "gated": [0.6953125, 0.65, 0.54, 0.541666667], "naive": [0.78125, 0.65, 0.52, 0.75], "scratch": [0.6953125, 0.65, 0.54, 0.641666667], "ungated": [0.6953125, 0.65, 0.59, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c1980929059b2a705f75d17942e5ba8e6f5d4b4141a736e8aa1fea7252d780dd", "primary_control": "scratch", "result_hash": "129640d26afb96e717c2b6474a0dcce11f9f3d8808186401cfeade8df0227a46", "signature": "0f6f22eeb723e2be0d1799a7ef5300884d61af843781ae6ae8074c4d45b6cf4101c068dd6c3ca8b9b2716c1faff05e2fb0493b1da241d9c9db19ae7b2bb86d08", "state": "VERIFIED", "summary": {"compressed": 0.606745, "gated": 0.606745, "naive": 0.675312, "scratch": 0.631745, "ungated": 0.671328}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:47:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:22Z", "cpu_seconds": 4.3737831489999905, "created_at": "2026-11-22T03:47:12Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-c1b81fcb73ea", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "b7c47d514ead10f5d929cc9d0ec153072cd78d334f84b0727c734f51d6a2dda7", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-c1b81fcb73ea", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:47:12Z", "seed": 180868641, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "8825b9bc236679d23e6473284ed921513a51ef283512dd251993ce21a02c73d8", "run_seconds": 4.589384078979492, "seed": 180868641, "started_at": "2026-11-22T03:47:17Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 2, "reset": 3}, "better": "higher", "ci95": [-0.079138, 0.012471], "effect": -0.033333, "evaluated_at": "2026-11-22T03:47:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-c1b81fcb73ea", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.3378560851856869, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "reset": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0], "root": [1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "b7c47d514ead10f5d929cc9d0ec153072cd78d334f84b0727c734f51d6a2dda7", "primary_control": "reset", "result_hash": "8825b9bc236679d23e6473284ed921513a51ef283512dd251993ce21a02c73d8", "signature": "8aca7db70e223c8cf203fae9b39fbc782953e3d15f176cf3a57a9c241ebc9f39b9b5a6ba808aaa2ee201b619df3928c80884749679f91785e2548dae0f9a7d0f", "state": "VERIFIED", "summary": {"inherit": 0.7333333333333333, "reset": 0.7666666666666667, "root": 0.38333333333333336}, "ties": 58, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:47:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:41Z", "cpu_seconds": 8.32789224599999, "created_at": "2026-11-22T03:47:28Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-ef80b58d40f5", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "7d9571025b58b0d2020ee0ced22393ea3f2a96d7f433bdf2ac101c0e2d6245c3", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-ef80b58d40f5", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:47:28Z", "seed": 1379848691, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "9896254590b6f3265d6452ea749e34acf34846164fdd1518c92cce745fa51a13", "run_seconds": 8.71245265007019, "seed": 1379848691, "started_at": "2026-11-22T03:47:32Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.190501, 0.023835], "effect": -0.083333, "evaluated_at": "2026-11-22T03:47:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-ef80b58d40f5", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.8647315821990931, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "reset": [0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 60, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "7d9571025b58b0d2020ee0ced22393ea3f2a96d7f433bdf2ac101c0e2d6245c3", "primary_control": "reset", "result_hash": "9896254590b6f3265d6452ea749e34acf34846164fdd1518c92cce745fa51a13", "signature": "90490dc3b9622ee0ddac5a38e2ff7f0d166951a9d9ee89abf7772731e8286d95d2f3179dcf2285c423c06ca16399b75111313c20eb3e39203db138c2e9e7d40f", "state": "VERIFIED", "summary": {"inherit": 0.5833333333333334, "reset": 0.6666666666666666, "root": 0.4166666666666667}, "ties": 49, "treatment": "inherit", "wins": 3}, "verified_at": "2026-11-22T03:47:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:59Z", "cpu_seconds": 0.7294990880000114, "created_at": "2026-11-22T03:47:29Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c988209c1104", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "31eaa9970a334b1af44f7ddff237fa6936ef6ffb229ebaa94de8292c964b90bb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c988209c1104", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:47:29Z", "seed": 243661039, "treatment": "frontier"}, "priority": 0.40603690115605773, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3a3bfc7d62cf114c92a0b5c4f929760f7a3b0ac8a3022668ef585967685c601f", "run_seconds": 0.8439300060272217, "seed": 243661039, "started_at": "2026-11-22T03:47:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014551, 0.025563], "effect": 0.005506, "evaluated_at": "2026-11-22T03:48:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c988209c1104", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.958333333, 0.9, 0.928571429, 1.0], "frontier": [0.958333333, 0.9, 0.928571429, 0.958333333], "goal": [0.875, 0.775, 0.928571429, 0.9375], "hypdisc": [0.958333333, 0.9, 0.928571429, 0.9375], "infogain": [0.958333333, 0.9, 0.928571429, 0.958333333], "prederr": [0.958333333, 0.9, 0.928571429, 0.958333333], "random": [0.958333333, 0.875, 0.910714286, 0.979166667], "risk": [0.958333333, 0.9, 0.928571429, 0.958333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "31eaa9970a334b1af44f7ddff237fa6936ef6ffb229ebaa94de8292c964b90bb", "primary_control": "random", "result_hash": "3a3bfc7d62cf114c92a0b5c4f929760f7a3b0ac8a3022668ef585967685c601f", "signature": "c44c33b9c8ed2e7b381c9d18300fc3b2711f2c1a28ff140ef516ef2080edd59c6e45f736fa077c3468414d0dadd3a300f6066f6f3efa48094e7ac246b1a0350a", "state": "VERIFIED", "summary": {"count": 0.946726, "frontier": 0.93631, "goal": 0.879018, "hypdisc": 0.931101, "infogain": 0.93631, "prederr": 0.93631, "random": 0.930804, "risk": 0.93631}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:48:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:41Z", "cpu_seconds": 0.385775369000001, "created_at": "2026-11-22T03:47:29Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4340ef3eb815", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "495cdf7413eafb040b217ae10685748fc881c3baadd312136dbe560a6a0ffa50", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4340ef3eb815", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:47:29Z", "seed": 1646668599, "treatment": "full"}, "priority": 0.4142018130271928, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "df7d74edf3ef385a93ea81b8b331680b293b67b89434c69f185d5be4d8f4a5bd", "run_seconds": 0.4049568176269531, "seed": 1646668599, "started_at": "2026-11-22T03:47:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.799352, 0.049352], "effect": -0.375, "evaluated_at": "2026-11-22T03:47:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4340ef3eb815", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.75, 0.25, 0.75], "fixed_policy": [0.0, 0.25, 0.25, 0.25], "full": [0.5, 0.0, 0.25, 0.0], "no_world_model": [0.5, 0.0, 0.25, 0.0], "random": [0.5, 0.0, 0.25, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "495cdf7413eafb040b217ae10685748fc881c3baadd312136dbe560a6a0ffa50", "primary_control": "e141_best", "result_hash": "df7d74edf3ef385a93ea81b8b331680b293b67b89434c69f185d5be4d8f4a5bd", "signature": "ee621537d61323c620c364a70d589014860aa6f0fe28486ce99e367a2c4e79ddcbb58a05627c6e7bcff30edefc7afda240145503e11aaf03cb56c5c69f8a0e0e", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.1875, "full": 0.1875, "no_world_model": 0.1875, "random": 0.1875}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:47:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:23Z", "cpu_seconds": 1.5535027439999993, "created_at": "2026-11-22T03:47:29Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-49dd44e8f96c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "9f6ea7a5f739d9f0954713dcfb0d4e3b4f8625637ad8a07e221536168fd30146", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-49dd44e8f96c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:47:29Z", "seed": 552164435, "treatment": "full"}, "priority": 0.40411720184767785, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5173d69095abeac4d3c678e09f0f3e04563ce180bf2450512d5d68be8b45fbf4", "run_seconds": 1.6793537139892578, "seed": 552164435, "started_at": "2026-11-22T03:48:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.29707, 0.17207], "effect": -0.0625, "evaluated_at": "2026-11-22T03:48:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-49dd44e8f96c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 0.5, 1.0], "fixed_policy": [1.0, 1.0, 0.75, 0.75], "full": [0.75, 0.75, 0.5, 0.75], "no_world_model": [1.0, 0.5, 0.5, 0.5], "random": [1.0, 0.5, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "9f6ea7a5f739d9f0954713dcfb0d4e3b4f8625637ad8a07e221536168fd30146", "primary_control": "e141_best", "result_hash": "5173d69095abeac4d3c678e09f0f3e04563ce180bf2450512d5d68be8b45fbf4", "signature": "aa1765d93a8147905eea1a5e3107a7c2b07c375f73f9f75a1db244fb8b2d98822219f14342db798220d1c32ebf8adcd47487ce4f02b42e87cc73325c1b675a01", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.875, "full": 0.6875, "no_world_model": 0.625, "random": 0.8125}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:48:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:59Z", "cpu_seconds": 1.3510961340000023, "created_at": "2026-11-22T03:47:29Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6c22899d2520", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "194cb2265f04373463066a6f09a3ba67824b60c487004a7e41acbce229445925", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6c22899d2520", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:47:29Z", "seed": 797719911, "treatment": "full"}, "priority": 0.4079405416887535, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9455e24d8d6cc50ecda67c6ce3ff9f28814c2d27b880142b1b58ebf33c9f094e", "run_seconds": 1.6423466205596924, "seed": 797719911, "started_at": "2026-11-22T03:47:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.044139, 0.016002], "effect": -0.014069, "evaluated_at": "2026-11-22T03:48:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6c22899d2520", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.787234043, 0.519503546, 0.712765957, 0.55615942], "frozen": [0.106382979, 0.177304965, 0.131205674, 0.114130435], "full": [0.787234043, 0.473404255, 0.680851064, 0.577898551], "retrieval": [0.787234043, 0.491134752, 0.712765957, 0.56884058], "rules_nodetect": [0.787234043, 0.473404255, 0.680851064, 0.577898551]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "194cb2265f04373463066a6f09a3ba67824b60c487004a7e41acbce229445925", "primary_control": "facts", "result_hash": "9455e24d8d6cc50ecda67c6ce3ff9f28814c2d27b880142b1b58ebf33c9f094e", "signature": "01a87c1e8f53a2030604e4c936997bfaadfec329decb84d9c776589584c0a0a3f0072a78a8e83f70ad1f7ab1838366ca7cdfcaebbad94ef0d9397ac13fa9f309", "state": "VERIFIED", "summary": {"facts": 0.643916, "frozen": 0.132256, "full": 0.629847, "retrieval": 0.639994, "rules_nodetect": 0.629847}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:48:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:36Z", "cpu_seconds": 2.3346099820000035, "created_at": "2026-11-22T03:47:29Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-efe59033b3cd", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 3, "n": 4}, "prereg_hash": "1e1387bfccc01250aeb201230c58d6e6db2976d0c62a288882f4881be523ee83", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-efe59033b3cd", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:47:29Z", "seed": 1560146676, "treatment": "full"}, "priority": 0.41767240873231465, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-bcc65026d33f", "result_hash": "d1dc68e640a97ee1682c73ce7c01f52fe3f241c5bfab77fced9298f0b5c57e4d", "run_seconds": 2.4712681770324707, "seed": 1560146676, "started_at": "2026-11-22T03:47:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.010369, 0.163598], "effect": 0.076614, "evaluated_at": "2026-11-22T03:47:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-efe59033b3cd", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.177690802, 0.17964775, 1.0], "frozen": [0.034285714, 0.03444227, 0.03444227, 0.148387097], "full": [1.0, 0.339726027, 0.32407045, 1.0], "retrieval": [1.0, 0.356164384, 0.380039139, 1.0], "rules_nodetect": [1.0, 0.339726027, 0.32407045, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "1e1387bfccc01250aeb201230c58d6e6db2976d0c62a288882f4881be523ee83", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-bcc65026d33f", "result_hash": "d1dc68e640a97ee1682c73ce7c01f52fe3f241c5bfab77fced9298f0b5c57e4d", "signature": "31290e1dc36282df77216bca77d7ea925fcd110e47de0e818ad1421cfc6912eeb32149159acefc0e9a5f8a218532bd2f675f5047aafac85dc3c968d57c863e06", "state": "VERIFIED", "summary": {"facts": 0.589335, "frozen": 0.062889, "full": 0.665949, "retrieval": 0.684051, "rules_nodetect": 0.665949}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:47:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:32Z", "cpu_seconds": 1.6179026989999983, "created_at": "2026-11-22T03:47:29Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1fbe62cc0fef", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "c59fe870f3492abc7b67c10d74f1b12269ee2fbff67f89c4d8d21d0fb1befa11", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1fbe62cc0fef", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:47:29Z", "seed": 390835534, "treatment": "gated"}, "priority": 0.5691589956402602, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6369c09a908d08b9d989f3d0c031b6d1ec0b10b79bddf5ee4850dfe8c28bce47", "run_seconds": 1.6652319431304932, "seed": 390835534, "started_at": "2026-11-22T03:47:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.050274, 0.06694], "effect": 0.008333, "evaluated_at": "2026-11-22T03:47:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1fbe62cc0fef", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.567307692, 0.564814815, 0.525, 0.517857143], "gated": [0.567307692, 0.564814815, 0.5375, 0.517857143], "naive": [0.701923077, 0.685185185, 0.625, 0.633928571], "scratch": [0.567307692, 0.481481481, 0.525, 0.580357143], "ungated": [0.644230769, 0.509259259, 0.625, 0.544642857]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "c59fe870f3492abc7b67c10d74f1b12269ee2fbff67f89c4d8d21d0fb1befa11", "primary_control": "scratch", "result_hash": "6369c09a908d08b9d989f3d0c031b6d1ec0b10b79bddf5ee4850dfe8c28bce47", "signature": "786a53dcc76ee92e76154e73ca8d6c7877611b707eff4d9185973b2ad86c95a948e2a2a90031174e88a615218781cbaf8ff95b5e9dc2f063a7563dae8195a800", "state": "VERIFIED", "summary": {"compressed": 0.543745, "gated": 0.54687, "naive": 0.661509, "scratch": 0.538537, "ungated": 0.580783}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:47:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:19Z", "cpu_seconds": 0.4585600210000109, "created_at": "2026-11-22T03:47:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-00215f973686", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "4fee5c14ba1e1068484d504de95d27edcba547533b695d0ad61996d5fdd7ce5b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-00215f973686", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:47:45Z", "seed": 1643080428, "treatment": "full"}, "priority": 0.4052089105015612, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a7fe6bdcbd9c885ee7cd3d454afcade9bdb8b900423c1fb9a554e8afdafe3bf2", "run_seconds": 0.54736328125, "seed": 1643080428, "started_at": "2026-11-22T03:48:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:48:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-00215f973686", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.75, 0.0, 0.25], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 1.0, 1.0], "random": [0.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "4fee5c14ba1e1068484d504de95d27edcba547533b695d0ad61996d5fdd7ce5b", "primary_control": "e141_best", "result_hash": "a7fe6bdcbd9c885ee7cd3d454afcade9bdb8b900423c1fb9a554e8afdafe3bf2", "signature": "af9479222ad26f9471af5cedb67f41729d6b62ea1c6f3ba9ab757edbf18fd35d2f622c8a7f1244308bba484794160cd60fd9a125748b2e8532aad843acf7060c", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.25, "full": 0.75, "no_world_model": 0.75, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:48:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:57Z", "cpu_seconds": 0.47549827899999286, "created_at": "2026-11-22T03:47:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-21ed52834fb7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "3819d296df0cfa4abd0134e6d1f08f3fc908c6dadefd70c3ceea2877f50455c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-21ed52834fb7", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:47:45Z", "seed": 2023522221, "treatment": "full"}, "priority": 0.41252093272140666, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5bf5e32bd9a5aae93e6ddc5b54037b7124bdca32c9b0df4fc3dbafcf050d6f02", "run_seconds": 0.5323951244354248, "seed": 2023522221, "started_at": "2026-11-22T03:47:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.24, 0.74], "effect": 0.25, "evaluated_at": "2026-11-22T03:48:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-21ed52834fb7", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 1.0, 0.0], "fixed_policy": [0.5, 0.0, 0.0, 0.0], "full": [0.0, 1.0, 1.0, 0.0], "no_world_model": [0.0, 1.0, 1.0, 0.0], "random": [0.0, 0.75, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "3819d296df0cfa4abd0134e6d1f08f3fc908c6dadefd70c3ceea2877f50455c6", "primary_control": "e141_best", "result_hash": "5bf5e32bd9a5aae93e6ddc5b54037b7124bdca32c9b0df4fc3dbafcf050d6f02", "signature": "02f08d4669f6564e43007207b66c37565a6eb1a154f7401cbd30c8bd60df2faf675fa5f0a549366bfefa6256180d5828674e8010d3d46ff77f568dacef8fc301", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.125, "full": 0.5, "no_world_model": 0.5, "random": 0.4375}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:48:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:10Z", "cpu_seconds": 1.2274188119999998, "created_at": "2026-11-22T03:47:45Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ecec5858ba64", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "b8ad420f4faa5bf5275ecc4c142caa8ee2d06f9eaa9374232979824ea8656dd9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ecec5858ba64", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:47:45Z", "seed": 1989841861, "treatment": "full"}, "priority": 0.4037602561284247, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "550673017c047e1fe19f85e97b73607f5e1bfe9eef8edf8b19ef4b1a1d0c8b12", "run_seconds": 1.5089151859283447, "seed": 1989841861, "started_at": "2026-11-22T03:49:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.073926, -0.02984], "effect": -0.051883, "evaluated_at": "2026-11-22T03:49:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ecec5858ba64", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.458333333, 0.46875, 0.363768116, 0.455357143], "frozen": [0.166666667, 0.166666667, 0.175362319, 0.166666667], "full": [0.418154762, 0.430059524, 0.27826087, 0.412202381], "retrieval": [0.467261905, 0.452380952, 0.301449275, 0.436011905], "rules_nodetect": [0.706845238, 0.430059524, 0.27826087, 0.412202381]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "b8ad420f4faa5bf5275ecc4c142caa8ee2d06f9eaa9374232979824ea8656dd9", "primary_control": "facts", "result_hash": "550673017c047e1fe19f85e97b73607f5e1bfe9eef8edf8b19ef4b1a1d0c8b12", "signature": "d1703bc7e87584431efd2c1007fdf5d0c3fbd21cddf08c918652a0780e4a5f892c1b2d3308d36e22edbdeb441ea29c0944ec6716a4c4e82ecb7a93b83d698201", "state": "VERIFIED", "summary": {"facts": 0.436552, "frozen": 0.168841, "full": 0.384669, "retrieval": 0.414276, "rules_nodetect": 0.456842}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:49:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:51Z", "cpu_seconds": 4.382867415999996, "created_at": "2026-11-22T03:47:45Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ad03a114a04f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "687f6cfb1c45718abf8fd27ebfb2510343949895a6dc390d1bd3b010c97b237c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ad03a114a04f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:47:45Z", "seed": 133466773, "treatment": "gated"}, "priority": 0.5425282882257818, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "8d066c9148659cba63c1ed0d672f42a9f12448a180b9b961eac27ca666a52d57", "run_seconds": 5.0870749950408936, "seed": 133466773, "started_at": "2026-11-22T03:47:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.212975, 0.03503], "effect": -0.088973, "evaluated_at": "2026-11-22T03:47:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-ad03a114a04f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.544642857, 0.467741935, 0.554347826, 0.413043478], "gated": [0.544642857, 0.467741935, 0.554347826, 0.413043478], "naive": [0.723214286, 0.717741935, 0.72826087, 0.684782609], "scratch": [0.544642857, 0.475806452, 0.630434783, 0.684782609], "ungated": [0.6875, 0.725806452, 0.641304348, 0.597826087]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "687f6cfb1c45718abf8fd27ebfb2510343949895a6dc390d1bd3b010c97b237c", "primary_control": "scratch", "result_hash": "8d066c9148659cba63c1ed0d672f42a9f12448a180b9b961eac27ca666a52d57", "signature": "355d0f9d81c39d093ef8d1740e39363051bad9dc6dd429d554155e95f7bf99edc90ee95b1f690a3bde038c236f4009992686b5574633ae1a5d739687c4c8d405", "state": "VERIFIED", "summary": {"compressed": 0.494944, "gated": 0.494944, "naive": 0.7135, "scratch": 0.583917, "ungated": 0.663109}, "ties": 1, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:47:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:47:53Z", "cpu_seconds": 6.1061397, "created_at": "2026-11-22T03:47:45Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-8487ec652ef7", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "6e70245cfc2c33bf5c5c0fbcea4e652b264e0053d51827c0d112bc1e1e32f7e5", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-8487ec652ef7", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:47:45Z", "seed": 1647420641, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "b1a26c24c11e965bd5bf1a0cd843f1cd4fd979dc05f1a2342ea7c79eec145799", "run_seconds": 7.0261523723602295, "seed": 1647420641, "started_at": "2026-11-22T03:47:46Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:47:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-8487ec652ef7", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0], "reset": [0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0], "root": [0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6e70245cfc2c33bf5c5c0fbcea4e652b264e0053d51827c0d112bc1e1e32f7e5", "primary_control": "reset", "result_hash": "b1a26c24c11e965bd5bf1a0cd843f1cd4fd979dc05f1a2342ea7c79eec145799", "signature": "7a75bc1cd93c437c806a7ac6667715a41b164b6fdf87931df9831dab6b8a936fe9a688382e492a0cbc9f84e06c56fc7b4061bc145992920c06cf81f05280170d", "state": "PARTIAL", "summary": {"inherit": 0.6833333333333333, "reset": 0.6833333333333333, "root": 0.4166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:48:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:05Z", "cpu_seconds": 2.381550924999999, "created_at": "2026-11-22T03:47:45Z", "error": null, "family": "world_model_frontier", "fence": 1, "id": "E141-world_model_frontier-a03c4ee7cfd7", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "prereg_hash": "92b9e876c82018b2c4360866bbdaa90081b7a732aab65596ce611f23360d8467", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random_walk", "curiosity"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "world modeling / interactive exploration", "evolution": "E142", "experiment_id": "E141-world_model_frontier-a03c4ee7cfd7", "family": "world_model_frontier", "hypothesis": "Next-state accuracy of frontier exploration is higher than random walk (confirmatory: budget_mult in {0.5, 1.0}).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L4", "metric_direction": "higher", "params": {"budget_mult": 2.0, "n": 12, "size": 12, "teleporters": 2, "wall_density": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "random_walk", "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "registered_at": "2026-11-22T03:47:45Z", "seed": 604331829, "treatment": "frontier"}, "priority": 0.4057095996720786, "question": "Does model-based frontier exploration learn a more accurate transition model than a random walk at low step budgets?", "replicates": "E141-world_model_frontier-6b3938d40f05", "result_hash": "ab35a27ca827f0e4c833f5cb11b585241519046e1dba57a730db303e12550b96", "run_seconds": 3.362218141555786, "seed": 604331829, "started_at": "2026-11-22T03:48:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005789, 0.049343], "effect": 0.021777, "evaluated_at": "2026-11-22T03:48:05Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-world_model_frontier-a03c4ee7cfd7", "family": "world_model_frontier", "hypothesis_supported": false, "instance_scores": {"curiosity": [1.0, 1.0, 0.878378378, 0.758333333, 0.87254902, 1.0, 1.0, 0.833333333, 0.866666667, 0.833333333, 0.849489796, 1.0], "frontier": [1.0, 1.0, 1.0, 0.7, 1.0, 1.0, 1.0, 1.0, 1.0, 0.913978495, 0.788265306, 1.0], "random_walk": [1.0, 1.0, 0.936936937, 0.7, 0.93872549, 1.0, 1.0, 0.927083333, 0.9, 0.86827957, 0.869897959, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9611, "losses": 1, "n": 12, "outcome": "INCONCLUSIVE", "p_sign_display": "0.109", "prereg_hash": "92b9e876c82018b2c4360866bbdaa90081b7a732aab65596ce611f23360d8467", "primary_control": "random_walk", "replicated": false, "replication_of": "E141-world_model_frontier-6b3938d40f05", "result_hash": "ab35a27ca827f0e4c833f5cb11b585241519046e1dba57a730db303e12550b96", "signature": "be511adc0c7c35f852b75df53b844085804f28591401521f92ba0ebf96c9755e68e6faa92b0dad8817681dba4e4d26dedca4b403460e9388503623650fa6f701", "state": "VERIFIED", "summary": {"curiosity": 0.907674, "frontier": 0.950187, "random_walk": 0.92841}, "ties": 6, "treatment": "frontier", "wins": 5}, "verified_at": "2026-11-22T03:48:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:11Z", "cpu_seconds": 0.5205647160000098, "created_at": "2026-11-22T03:48:02Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-750f7c032706", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "aaf3edb619ba38efe5c4e31527c4385456e91f5887fb2d09617e7c913e65a369", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-750f7c032706", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:48:02Z", "seed": 952975255, "treatment": "full"}, "priority": 0.41405267579010274, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "910a93488b7986470ce6796f1938a03d89b654bf28cab7aea0632794a2c5b7be", "run_seconds": 0.5487329959869385, "seed": 952975255, "started_at": "2026-11-22T03:48:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.151278, 0.776278], "effect": 0.3125, "evaluated_at": "2026-11-22T03:48:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-750f7c032706", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 0.0], "fixed_policy": [0.5, 0.5, 0.5, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.25, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "aaf3edb619ba38efe5c4e31527c4385456e91f5887fb2d09617e7c913e65a369", "primary_control": "e141_best", "result_hash": "910a93488b7986470ce6796f1938a03d89b654bf28cab7aea0632794a2c5b7be", "signature": "76ba5901f3d2ba2e68531f4836a15659bcda74ab3e84dd991abe8ebec8d589ab94d01152ec6711a7e72e0cca1de509687b603ac59c4f86387fe475ce230ad707", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 0.8125, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:48:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:16Z", "cpu_seconds": 0.5674975530000097, "created_at": "2026-11-22T03:48:02Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d1aaf9a2a807", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "1e1de10783a8ea1d15214badc81d1316da2481b3324817241f0d21e79ca6cb85", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d1aaf9a2a807", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:48:02Z", "seed": 1213120754, "treatment": "full"}, "priority": 0.4076454302746525, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7a40b6eb2393fd629e6d6f3eb467cf0a5313f897b6d2af450a4da8f115eddd91", "run_seconds": 0.5906500816345215, "seed": 1213120754, "started_at": "2026-11-22T03:48:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:48:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d1aaf9a2a807", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.5], "fixed_policy": [0.0, 0.25, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 0.25], "no_world_model": [1.0, 1.0, 1.0, 0.0], "random": [1.0, 1.0, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "1e1de10783a8ea1d15214badc81d1316da2481b3324817241f0d21e79ca6cb85", "primary_control": "e141_best", "result_hash": "7a40b6eb2393fd629e6d6f3eb467cf0a5313f897b6d2af450a4da8f115eddd91", "signature": "1966ffed0510ca6230a8992b3b749452597f34b10a1140d8ff938e1ac309508bf7b02882d1b26b44d4b7bb9d03214d11835c4193a910e02a4839b3d13ca8260c", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.125, "full": 0.8125, "no_world_model": 0.75, "random": 0.875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:48:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:14Z", "cpu_seconds": 0.5341895600000015, "created_at": "2026-11-22T03:48:02Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3afb969105c3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "859012dee3f3c843f749895b61c805229762841ba0ad38170c4058a72561a55f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3afb969105c3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:48:02Z", "seed": 1005313334, "treatment": "full"}, "priority": 0.41294528825287946, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "34d133c051d84ee552a3e4339172f2286745993c9594f3b79cbbf170d7cfbf58", "run_seconds": 0.5457780361175537, "seed": 1005313334, "started_at": "2026-11-22T03:48:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016842, 0.05193], "effect": 0.017544, "evaluated_at": "2026-11-22T03:48:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3afb969105c3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.833333333, 0.946428571, 0.8125, 0.824561404], "frozen": [0.611111111, 0.517857143, 0.5625, 0.245614035], "full": [0.833333333, 0.946428571, 0.8125, 0.894736842], "retrieval": [0.833333333, 0.946428571, 0.8125, 0.824561404], "rules_nodetect": [0.722222222, 0.946428571, 0.8125, 0.438596491]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "859012dee3f3c843f749895b61c805229762841ba0ad38170c4058a72561a55f", "primary_control": "facts", "result_hash": "34d133c051d84ee552a3e4339172f2286745993c9594f3b79cbbf170d7cfbf58", "signature": "6b92b841358ddf4d64a4b63e7e4e78df8d59f28d8ad9a6d53a95a56f67fe94bbdb3996a3459eb66b0872d0b15aaa65d6cadf9acdaae95ab2867383b063084607", "state": "VERIFIED", "summary": {"facts": 0.854206, "frozen": 0.484271, "full": 0.87175, "retrieval": 0.854206, "rules_nodetect": 0.729937}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:48:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:17Z", "cpu_seconds": 1.1372267800000002, "created_at": "2026-11-22T03:48:02Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8d00de31608d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "3cbd784c6d5fc2d43ef7c7f6661332643d3d571f41318bc24e466692629c3b59", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8d00de31608d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:48:02Z", "seed": 547453642, "treatment": "full"}, "priority": 0.3978619332315371, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "94366b92b6e82c13c0f2269cb9199a375399b19f8f1618b2ee2382eb47a6bcf2", "run_seconds": 1.237790822982788, "seed": 547453642, "started_at": "2026-11-22T03:49:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014499, 0.117773], "effect": 0.051637, "evaluated_at": "2026-11-22T03:49:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8d00de31608d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.757142857, 1.0, 0.623809524, 0.9375], "frozen": [0.321428571, 0.4, 0.233333333, 0.25], "full": [0.857142857, 1.0, 0.742857143, 0.925], "retrieval": [0.871428571, 1.0, 0.728571429, 0.9375], "rules_nodetect": [0.6, 1.0, 0.742857143, 0.925]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "3cbd784c6d5fc2d43ef7c7f6661332643d3d571f41318bc24e466692629c3b59", "primary_control": "facts", "result_hash": "94366b92b6e82c13c0f2269cb9199a375399b19f8f1618b2ee2382eb47a6bcf2", "signature": "cb86d86edf66b44ef2383fcc447b4a3d13858022a8e989199e149c210c1a22cb5e9add0e7255db24d0708fb072b1e0e091ba3da5e482386de3a6e3783d8f0d05", "state": "VERIFIED", "summary": {"facts": 0.829613, "frozen": 0.30119, "full": 0.88125, "retrieval": 0.884375, "rules_nodetect": 0.816964}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:49:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:16Z", "cpu_seconds": 0.5749766939999998, "created_at": "2026-11-22T03:48:02Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-b3c98494b060", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "9be28cf1aa958cd822ba4bf3f8f78289400e84ff339fe35edcbadf4c6acae755", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-b3c98494b060", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:48:02Z", "seed": 503538008, "treatment": "gated"}, "priority": 0.3991723750208277, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "54a85c634c2a1bb958c4e92e47feaa39c2bf35f08dac1001494c2d2a73fb7548", "run_seconds": 0.5989863872528076, "seed": 503538008, "started_at": "2026-11-22T03:49:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.009673, 0.003137], "effect": -0.003268, "evaluated_at": "2026-11-22T03:49:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-b3c98494b060", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.39869281, 0.253333333, 0.411764706, 0.202614379], "gated": [0.39869281, 0.253333333, 0.411764706, 0.202614379], "naive": [0.39869281, 0.386666667, 0.411764706, 0.254901961], "scratch": [0.39869281, 0.253333333, 0.411764706, 0.215686275], "ungated": [0.222222222, 0.206666667, 0.222222222, 0.209150327]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "9be28cf1aa958cd822ba4bf3f8f78289400e84ff339fe35edcbadf4c6acae755", "primary_control": "scratch", "result_hash": "54a85c634c2a1bb958c4e92e47feaa39c2bf35f08dac1001494c2d2a73fb7548", "signature": "f37d0fb0f4ec4d652554d52ce0f2a821de16ff3b044d02790c6f45edd8b06bfb63e3e117b552dfcd2ab7f278533c6e1e9484adc5cd77270c2cce7827799d4307", "state": "VERIFIED", "summary": {"compressed": 0.316601, "gated": 0.316601, "naive": 0.363007, "scratch": 0.319869, "ungated": 0.215065}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:49:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:08Z", "cpu_seconds": 4.063992847999998, "created_at": "2026-11-22T03:48:02Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-de91436ac893", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "e5b53e6895ca5e6747a790e38b0421bad778509adaf64df8bac485069421e664", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-de91436ac893", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:48:02Z", "seed": 910265260, "treatment": "gated"}, "priority": 0.44986264031267154, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "5451e41f3ec0d66c6000a5d8df3cb27659a8136d7fd308e7a22da3822bb85202", "run_seconds": 4.4821860790252686, "seed": 910265260, "started_at": "2026-11-22T03:48:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001871, 0.016116], "effect": 0.007123, "evaluated_at": "2026-11-22T03:48:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-de91436ac893", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.527777778, 0.586538462, 0.422413793, 0.5875], "gated": [0.527777778, 0.586538462, 0.422413793, 0.5875], "naive": [0.685185185, 0.711538462, 0.612068966, 0.6125], "scratch": [0.518518519, 0.567307692, 0.422413793, 0.5875], "ungated": [0.601851852, 0.634615385, 0.517241379, 0.625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "e5b53e6895ca5e6747a790e38b0421bad778509adaf64df8bac485069421e664", "primary_control": "scratch", "result_hash": "5451e41f3ec0d66c6000a5d8df3cb27659a8136d7fd308e7a22da3822bb85202", "signature": "7ee2cca0f8d7976b6224c7641c83ba54ba2c18e6d82556d5050b320b171b2963530705f29f69d7241776dd1c581307c5f0db068bc6f37f906faa5a24fac21606", "state": "VERIFIED", "summary": {"compressed": 0.531058, "gated": 0.531058, "naive": 0.655323, "scratch": 0.523935, "ungated": 0.594677}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:48:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:13Z", "cpu_seconds": 8.844078607, "created_at": "2026-11-22T03:48:02Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-8f04684d6591", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "575f35bdb0187cbe02080bd47c7af1930dd98d0691b7e8b8eceb303bd3ff0c0f", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-8f04684d6591", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:48:02Z", "seed": 942180916, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "fdd9e8065071b31f3c2f26c60e349469a2afe884efa220795b6abfed6f43a1b0", "run_seconds": 9.270206928253174, "seed": 942180916, "started_at": "2026-11-22T03:48:04Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:48:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-8f04684d6591", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0], "reset": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "575f35bdb0187cbe02080bd47c7af1930dd98d0691b7e8b8eceb303bd3ff0c0f", "primary_control": "reset", "result_hash": "fdd9e8065071b31f3c2f26c60e349469a2afe884efa220795b6abfed6f43a1b0", "signature": "d6d94c4e94363b387938fe710e1ef63de70d85e04791b22709a8d1cb158d52a63e09c0455a597b8b985c162514424f14fc22ef430f5ba05cdb52162f74600306", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:48:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:10Z", "cpu_seconds": 1.082747975, "created_at": "2026-11-22T03:48:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-709b7ee27467", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "939213d3c7d6d2765f21aaa246a27cbd3973a7ec65102bedea17bc5a634cb9a1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-709b7ee27467", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:48:19Z", "seed": 1284386498, "treatment": "frontier"}, "priority": 0.40326823828648906, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "7f63b02ea1c12a3cf661885c38aca8b3b0991a2a3c65feabd07ef20cd74ea73f", "run_seconds": 1.4509265422821045, "seed": 1284386498, "started_at": "2026-11-22T03:49:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.279967, 0.281615], "effect": 0.000824, "evaluated_at": "2026-11-22T03:49:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-709b7ee27467", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.6, 0.743589744, 0.904761905, 0.568627451], "frontier": [0.455555556, 0.474358974, 0.857142857, 0.764705882], "goal": [0.766666667, 0.666666667, 0.833333333, 0.705882353], "hypdisc": [0.688888889, 0.628205128, 0.523809524, 0.68627451], "infogain": [0.655555556, 0.525641026, 0.833333333, 0.738562092], "prederr": [0.466666667, 0.679487179, 0.571428571, 0.31372549], "random": [0.822222222, 0.461538462, 0.833333333, 0.431372549], "risk": [0.455555556, 0.474358974, 0.857142857, 0.764705882]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "939213d3c7d6d2765f21aaa246a27cbd3973a7ec65102bedea17bc5a634cb9a1", "primary_control": "random", "result_hash": "7f63b02ea1c12a3cf661885c38aca8b3b0991a2a3c65feabd07ef20cd74ea73f", "signature": "070f4792b3ddaa09f81789992a896d6e6789195ba588d0d95bdc7dd809cc9eb33536a114d3cdfd7a701616248006061a27799ce860aee37440f4cd4830f46000", "state": "VERIFIED", "summary": {"count": 0.704245, "frontier": 0.637941, "goal": 0.743137, "hypdisc": 0.631795, "infogain": 0.688273, "prederr": 0.507827, "random": 0.637117, "risk": 0.637941}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:49:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:23Z", "cpu_seconds": 1.3211064120000002, "created_at": "2026-11-22T03:48:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-85844c894800", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "b72cc536e572e47e73a687c42e28f3a87263e8dd4118c65ab7b6664bd15c8be5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-85844c894800", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:48:19Z", "seed": 2054611862, "treatment": "full"}, "priority": 0.3953262053307567, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "67e725771bea3a2dc76448a12254a542a4e827804d6d80c558ec109a9b4c3cec", "run_seconds": 1.328014612197876, "seed": 2054611862, "started_at": "2026-11-22T03:49:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.025444, 0.147605], "effect": 0.06108, "evaluated_at": "2026-11-22T03:49:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-85844c894800", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.611683849, 0.211904762, 0.415686275, 0.164772727], "frozen": [0.0, 0.0, 0.160784314, 0.0], "full": [0.611683849, 0.23452381, 0.607843137, 0.194318182], "retrieval": [0.864261168, 0.777380952, 0.654901961, 0.764772727], "rules_nodetect": [0.611683849, 0.286904762, 0.623529412, 0.296590909]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b72cc536e572e47e73a687c42e28f3a87263e8dd4118c65ab7b6664bd15c8be5", "primary_control": "facts", "result_hash": "67e725771bea3a2dc76448a12254a542a4e827804d6d80c558ec109a9b4c3cec", "signature": "7bde6de664ebdab679e53cac19bdaa476cd9c4657d72f7a868d756ae69ef2dab2c4f032cfab9622e80cf948912d28d0ca01a4948bfba84e8a551b899a8a97105", "state": "VERIFIED", "summary": {"facts": 0.351012, "frozen": 0.040196, "full": 0.412092, "retrieval": 0.765329, "rules_nodetect": 0.454677}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:49:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:25Z", "cpu_seconds": 1.0413324409999998, "created_at": "2026-11-22T03:48:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9399c3d73ec7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "b37d15558572878e98cb2013a38878473bebe47fbd76e45eaed6f1eeaf0d8f17", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9399c3d73ec7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:48:19Z", "seed": 1916343944, "treatment": "full"}, "priority": 0.3942661137557089, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "75d5bacefbdc0dd90b7e18cb372131a22b8d402b2b2caeafeb5711e006b73796", "run_seconds": 1.0427124500274658, "seed": 1916343944, "started_at": "2026-11-22T03:49:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.092293, 0.21478], "effect": 0.061243, "evaluated_at": "2026-11-22T03:49:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9399c3d73ec7", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.357954545, 1.0, 0.95, 1.0], "frozen": [0.255681818, 0.265625, 0.4, 0.346153846], "full": [0.653409091, 0.96875, 0.95, 0.980769231], "retrieval": [0.727272727, 0.984375, 0.975, 1.0], "rules_nodetect": [0.653409091, 0.96875, 0.95, 0.980769231]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "b37d15558572878e98cb2013a38878473bebe47fbd76e45eaed6f1eeaf0d8f17", "primary_control": "facts", "result_hash": "75d5bacefbdc0dd90b7e18cb372131a22b8d402b2b2caeafeb5711e006b73796", "signature": "0b5b52102433938e60241d69bd1822680603489952073a0e5c123be63774ed87306c50ceac7fc0a9cad3e02b62370ed7e8a79a7315035c681a9005e148ec8e07", "state": "VERIFIED", "summary": {"facts": 0.826989, "frozen": 0.316865, "full": 0.888232, "retrieval": 0.921662, "rules_nodetect": 0.888232}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:49:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:22Z", "cpu_seconds": 1.2043362610000004, "created_at": "2026-11-22T03:48:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a28335527e5f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "4b5e85e06909fa6b8eb1ab071d010fd324e59d4756f5e783162c119ac04166d4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a28335527e5f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:48:19Z", "seed": 1751315830, "treatment": "full"}, "priority": 0.3955323315235687, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e3f09de53a94f32c4c25ac61c2408f22eceb07033b12cdb21f0db2c68abae6d4", "run_seconds": 1.2054502964019775, "seed": 1751315830, "started_at": "2026-11-22T03:49:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.038255, 0.002405], "effect": -0.017925, "evaluated_at": "2026-11-22T03:49:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a28335527e5f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.615789474, 0.75, 0.726027397, 0.473684211], "frozen": [0.110526316, 0.110526316, 0.131506849, 0.110526316], "full": [0.607894737, 0.75, 0.709589041, 0.426315789], "retrieval": [0.618421053, 0.75, 0.734246575, 0.455263158], "rules_nodetect": [0.765789474, 0.75, 0.709589041, 0.426315789]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "4b5e85e06909fa6b8eb1ab071d010fd324e59d4756f5e783162c119ac04166d4", "primary_control": "facts", "result_hash": "e3f09de53a94f32c4c25ac61c2408f22eceb07033b12cdb21f0db2c68abae6d4", "signature": "61530f0fc98045d050927eb130a061d23210cfe49f02d20b449128b1bf3a229326e3cc341988c8e1b13846dc57bc12088da40f5ef8e63549c73e8d18e5a92905", "state": "VERIFIED", "summary": {"facts": 0.641375, "frozen": 0.115771, "full": 0.62345, "retrieval": 0.639483, "rules_nodetect": 0.662924}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:49:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:48:23Z", "cpu_seconds": 3.6254820759999973, "created_at": "2026-11-22T03:48:19Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-6840ba502289", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "f2aaff260d1edb07c9fd46896eb60a23b3cb1c5884c3203006da858f06934dab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-6840ba502289", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:48:19Z", "seed": 350412204, "treatment": "gated"}, "priority": 0.4407137236358854, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7d10135f210e6e7c8ef06d4ff52b3b9238853bdfa411e1e7c6889a9a4cce5838", "run_seconds": 3.80298113822937, "seed": 350412204, "started_at": "2026-11-22T03:48:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0074, 0.0024], "effect": -0.0025, "evaluated_at": "2026-11-22T03:48:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-6840ba502289", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.625, 0.54, 0.75, 0.645833333], "gated": [0.625, 0.54, 0.75, 0.645833333], "naive": [0.6625, 0.64, 0.75, 0.6875], "scratch": [0.625, 0.55, 0.75, 0.645833333], "ungated": [0.6625, 0.55, 0.641666667, 0.729166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "f2aaff260d1edb07c9fd46896eb60a23b3cb1c5884c3203006da858f06934dab", "primary_control": "scratch", "result_hash": "7d10135f210e6e7c8ef06d4ff52b3b9238853bdfa411e1e7c6889a9a4cce5838", "signature": "c2ab712edfe93172383704bc984604ae752bd03fbc1a569735b36f9a92741bb531e3c6399252cddd83d3d7246e315b92ae2c967f3bdd44b2afebe6f5c20fef0d", "state": "VERIFIED", "summary": {"compressed": 0.640208, "gated": 0.640208, "naive": 0.685, "scratch": 0.642708, "ungated": 0.645833}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:48:26Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:49:34Z", "cpu_seconds": 5.30973373, "created_at": "2026-11-22T03:48:19Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-035784f235d2", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "d4d6430db3dfb16c6b52272ffd30901964a5924652d4944d938231e5c66a1d28", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-035784f235d2", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:48:19Z", "seed": 1369572077, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "1015e9ac6ce120568370d729388854d946e8bd2887ec7ff8c60940f54eaa1abc", "run_seconds": 5.560469627380371, "seed": 1369572077, "started_at": "2026-11-22T03:48:24Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:49:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-035784f235d2", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0], "reset": [1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0], "root": [1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d4d6430db3dfb16c6b52272ffd30901964a5924652d4944d938231e5c66a1d28", "primary_control": "reset", "result_hash": "1015e9ac6ce120568370d729388854d946e8bd2887ec7ff8c60940f54eaa1abc", "signature": "bf87aa691d7814ed310130f28ec0463f2a5b7f35526f147e7c432a7626c0b9a1321ce37b03d7f539a356c348bd6aee80cce1bb72b8a05786228b72ab23b7f00b", "state": "PARTIAL", "summary": {"inherit": 0.6166666666666667, "reset": 0.6166666666666667, "root": 0.36666666666666664}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:49:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:11Z", "cpu_seconds": 1.6391677820000001, "created_at": "2026-11-22T03:49:08Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-0e61d8444a87", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "prereg_hash": "4cc1ef3a197ba456d09ecb9bcdfdb6e97c097e40f662278bd7abb5be38616fba", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-0e61d8444a87", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 1.5, "target_noise": 0.0}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:49:08Z", "seed": 1206376531, "treatment": "iw_temperature"}, "priority": 0.4057734646787093, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-011ed4b03c4b", "result_hash": "ac1ceaa62a8f4ecdd028e029849ce5bb474477301e3cb22cc18cc4cf7135806a", "run_seconds": 2.114161252975464, "seed": 1206376531, "started_at": "2026-11-22T03:49:08Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.011028, 0.00871], "effect": -0.001159, "evaluated_at": "2026-11-22T03:49:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-0e61d8444a87", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.08704291, 0.135047918, 0.13693019, 0.013135027, 0.029544365, 0.030992088, 0.123657538, 0.037501727, 0.08729795, 0.023959155, 0.044255667, 0.09967225, 0.11228406, 0.041752165, 0.039758927, 0.047271288], "raw": [0.037076322, 0.128555665, 0.138147663, 0.016592305, 0.0395121, 0.070918545, 0.114829735, 0.029285312, 0.039663738, 0.02463134, 0.031381157, 0.09967225, 0.105056515, 0.028912823, 0.039758927, 0.05636521], "temperature": [0.046737945, 0.135047918, 0.138147663, 0.010032323, 0.0395121, 0.053993278, 0.131957305, 0.027245593, 0.053687845, 0.019736812, 0.089237458, 0.09581601, 0.12456405, 0.024432882, 0.039758927, 0.041650248]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1035, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.788", "prereg_hash": "4cc1ef3a197ba456d09ecb9bcdfdb6e97c097e40f662278bd7abb5be38616fba", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-011ed4b03c4b", "result_hash": "ac1ceaa62a8f4ecdd028e029849ce5bb474477301e3cb22cc18cc4cf7135806a", "signature": "a0167a9381072344ba555948b96347c0aafa6971000c85735baa1ac0431fb3fd400880764031e238db778afef7bf4565b309f856470e7d8f7d58deeb141b5b0a", "state": "VERIFIED", "summary": {"iw_temperature": 0.068131, "raw": 0.062522, "temperature": 0.066972}, "ties": 2, "treatment": "iw_temperature", "wins": 6}, "verified_at": "2026-11-22T03:49:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:19Z", "cpu_seconds": 4.135795766999999, "created_at": "2026-11-22T03:49:08Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-045c4c271329", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "0b5d52a0da4446d69b478379c6eac83ea66400bb8aaf232270e7e0ce7d513493", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-045c4c271329", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:49:08Z", "seed": 765414720, "treatment": "gated"}, "priority": 0.4015726920645557, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "bf64725df77a0070359afb2d958fef7f16d203af7a5ea368fe6f9958d3448be3", "run_seconds": 4.320124864578247, "seed": 765414720, "started_at": "2026-11-22T03:49:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.034259, 0.011111], "effect": -0.011574, "evaluated_at": "2026-11-22T03:49:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-045c4c271329", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.642857143, 0.588235294, 0.583333333, 0.462962963], "gated": [0.642857143, 0.588235294, 0.583333333, 0.462962963], "naive": [0.732142857, 0.779411765, 0.65, 0.592592593], "scratch": [0.642857143, 0.588235294, 0.583333333, 0.509259259], "ungated": [0.651785714, 0.779411765, 0.5, 0.481481481]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "0b5d52a0da4446d69b478379c6eac83ea66400bb8aaf232270e7e0ce7d513493", "primary_control": "scratch", "result_hash": "bf64725df77a0070359afb2d958fef7f16d203af7a5ea368fe6f9958d3448be3", "signature": "a2df0a6fe4480dd1327b96837d3dc296b571d1def37d9e23a3e53367af6bc75d2c98ec2781f8ce0506fd94a8e995845d7f0a7716ea3a3fc18d4e66e0f8b9100d", "state": "VERIFIED", "summary": {"compressed": 0.569347, "gated": 0.569347, "naive": 0.688537, "scratch": 0.580921, "ungated": 0.60317}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:49:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:12Z", "cpu_seconds": 3.114600558, "created_at": "2026-11-22T03:49:08Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-0715f43d9a42", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "623551fe85cfedf33d364fdf73626853173455a4abb3aac5ecc5ed5b6212486c", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-0715f43d9a42", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:49:08Z", "seed": 537938891, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "e177dd13cbb50ead188464b83a633669e445687d49f34925f57a66d2fa6af974", "run_seconds": 3.3804237842559814, "seed": 537938891, "started_at": "2026-11-22T03:49:08Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 1, "reset": 1}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:49:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-0715f43d9a42", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.46440226167943904, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], "reset": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], "root": [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "623551fe85cfedf33d364fdf73626853173455a4abb3aac5ecc5ed5b6212486c", "primary_control": "reset", "result_hash": "e177dd13cbb50ead188464b83a633669e445687d49f34925f57a66d2fa6af974", "signature": "59ed2fca03b17008908891c4c5bfc0c07aeacd45ce4697c4ee852329ee789ba8ce8828f7270ed1ab06ca2323aee278e9e9d681e657f71468d6a24516124f8605", "state": "PARTIAL", "summary": {"inherit": 0.45, "reset": 0.45, "root": 0.35}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:49:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:42Z", "cpu_seconds": 6.899946518999999, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7c34643fd13e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "85a013920ac4ceed005a7a206a89d949ca25fcc83dfb2a658dc8dbd22fa581cf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7c34643fd13e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:49:25Z", "seed": 931148401, "treatment": "frontier"}, "priority": 0.402173607346491, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "7daf0c83704abb4a2d63fb08008672dde551648466bc7d1524f40d213e7c72d1", "run_seconds": 8.01822018623352, "seed": 931148401, "started_at": "2026-11-22T03:49:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.156693, 0.265611], "effect": 0.211152, "evaluated_at": "2026-11-22T03:49:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7c34643fd13e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.36, 0.578947368, 0.362857143, 0.534328358], "frontier": [0.428235294, 0.752631579, 0.525714286, 0.680597015], "goal": [0.355294118, 0.568421053, 0.262857143, 0.570149254], "hypdisc": [0.268235294, 0.384210526, 0.408571429, 0.43880597], "infogain": [0.44, 0.705263158, 0.517142857, 0.680597015], "prederr": [0.251764706, 0.284210526, 0.317142857, 0.379104478], "random": [0.256470588, 0.463157895, 0.354285714, 0.468656716], "risk": [0.428235294, 0.752631579, 0.525714286, 0.680597015]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "85a013920ac4ceed005a7a206a89d949ca25fcc83dfb2a658dc8dbd22fa581cf", "primary_control": "random", "result_hash": "7daf0c83704abb4a2d63fb08008672dde551648466bc7d1524f40d213e7c72d1", "signature": "92c06121c737f0b06b2f1c72b3d6ca183306bb6690247b0da9a872232b971cbffcede39f1cca07e9be90c73d7d3f5319c6574ef5343e6974c4231af49037b009", "state": "VERIFIED", "summary": {"count": 0.459033, "frontier": 0.596795, "goal": 0.43918, "hypdisc": 0.374956, "infogain": 0.585751, "prederr": 0.308056, "random": 0.385643, "risk": 0.596795}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:49:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:10Z", "cpu_seconds": 1.27857281, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c846f04ef3ee", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "b15c59c5889955356fa3e25a5a0b1ea4573528dbf0d3b6da1a400eacbc0af003", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c846f04ef3ee", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:49:25Z", "seed": 1341404985, "treatment": "frontier"}, "priority": 0.37598644328191316, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c3b144accfeb478f5b2f1c8f6a13496eb8571080a1b137815e112724c9f15f73", "run_seconds": 2.588498115539551, "seed": 1341404985, "started_at": "2026-11-22T03:50:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.129333, 0.0585], "effect": -0.035417, "evaluated_at": "2026-11-22T03:50:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c846f04ef3ee", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.677083333, 0.733333333, 0.726190476, 0.81372549], "frontier": [0.729166667, 0.733333333, 0.595238095, 0.823529412], "goal": [0.8125, 0.733333333, 0.714285714, 0.735294118], "hypdisc": [0.729166667, 0.666666667, 0.547619048, 0.235294118], "infogain": [0.729166667, 0.733333333, 0.619047619, 0.823529412], "prederr": [0.729166667, 0.733333333, 0.619047619, 0.617647059], "random": [0.854166667, 0.833333333, 0.511904762, 0.823529412], "risk": [0.729166667, 0.733333333, 0.595238095, 0.823529412]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "b15c59c5889955356fa3e25a5a0b1ea4573528dbf0d3b6da1a400eacbc0af003", "primary_control": "random", "result_hash": "c3b144accfeb478f5b2f1c8f6a13496eb8571080a1b137815e112724c9f15f73", "signature": "474fc3894bfe3e8e6587afeecce1a3f1566a83b9b2c134eda7ee8289c458fb0896fe56b61de37ca02483f8c1c2182edb500a75127aba67b3ec5c4791585f0f08", "state": "VERIFIED", "summary": {"count": 0.737583, "frontier": 0.720317, "goal": 0.748853, "hypdisc": 0.544687, "infogain": 0.726269, "prederr": 0.674799, "random": 0.755734, "risk": 0.720317}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:50:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:15Z", "cpu_seconds": 5.887476489, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-534da4465e3e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "2bc63783682816b149ebda7a4fa1c9937bb610f30ed1d0ff4a059dd88ada82a5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-534da4465e3e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:49:25Z", "seed": 851892611, "treatment": "full"}, "priority": 0.37707444232539433, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "41a180333089d289ad286ea2422f88a7354532df3ba3c5afe552bcb3b34415fe", "run_seconds": 8.241737127304077, "seed": 851892611, "started_at": "2026-11-22T03:50:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.24, 0.74], "effect": 0.25, "evaluated_at": "2026-11-22T03:50:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-534da4465e3e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 1.0, 1.0], "fixed_policy": [1.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.5, 0.25, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "2bc63783682816b149ebda7a4fa1c9937bb610f30ed1d0ff4a059dd88ada82a5", "primary_control": "e141_best", "result_hash": "41a180333089d289ad286ea2422f88a7354532df3ba3c5afe552bcb3b34415fe", "signature": "2d1dd9bdcdd77e820231e4035bd2e1fcf5ee0c7b19b77493cdd3aea98f263f485c45625a14a4a11f1b2b16b5b2609a36171055f6b3e9bbf2c99ca7441bbcc20d", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 0.625, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:50:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:50Z", "cpu_seconds": 0.46881351800000104, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d16cd990a48f", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "d8dc68b31ab7ea82fefc6beddeb7f0433e63f30c130a9cc5dfefe64b8a81f712", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d16cd990a48f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:49:25Z", "seed": 37441264, "treatment": "full"}, "priority": 0.3839357594399055, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-a8f1939fd3cc", "result_hash": "19df0ce1f1c274a79b7878188f63785f7c22c49bd0afd9185fcfbe0fbf98f187", "run_seconds": 0.49625730514526367, "seed": 37441264, "started_at": "2026-11-22T03:49:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.299352, 0.549352], "effect": 0.125, "evaluated_at": "2026-11-22T03:49:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d16cd990a48f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 1.0, 0.5], "fixed_policy": [0.5, 0.0, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 0.25], "no_world_model": [0.25, 1.0, 1.0, 0.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "d8dc68b31ab7ea82fefc6beddeb7f0433e63f30c130a9cc5dfefe64b8a81f712", "primary_control": "e141_best", "replicated": false, "replication_of": "E141-ow_goal-a8f1939fd3cc", "result_hash": "19df0ce1f1c274a79b7878188f63785f7c22c49bd0afd9185fcfbe0fbf98f187", "signature": "2e811f98d59bdca1f37f14bcf77b7218a931fa261fb5b3084e28489595b8d660d97748f7a07800c89ab251522312ab2ebebe26a1646505caec21fc45cf070d05", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.1875, "full": 0.8125, "no_world_model": 0.5625, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:49:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:31Z", "cpu_seconds": 1.0621943660000008, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d4108865163e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "a10b95b0add8d259ac10e22c41d5486da786974befbc855ee6356d94b81b7799", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d4108865163e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:49:25Z", "seed": 906451815, "treatment": "full"}, "priority": 0.4074889880396768, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "87191a5a14670a5ef1fc84d1e34740ce4daea3cd8ec1b14c5eb7dfffdb0be764", "run_seconds": 1.1003472805023193, "seed": 906451815, "started_at": "2026-11-22T03:49:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.060686, 0.102179], "effect": 0.020746, "evaluated_at": "2026-11-22T03:49:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d4108865163e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.179089027, 0.402810304, 0.648351648, 0.293911007], "frozen": [0.157349896, 0.210772834, 0.285714286, 0.133489461], "full": [0.135610766, 0.395784543, 0.791208791, 0.284543326], "retrieval": [0.14699793, 0.43442623, 0.703296703, 0.346604215], "rules_nodetect": [0.270186335, 0.117096019, 0.791208791, 0.172131148]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "a10b95b0add8d259ac10e22c41d5486da786974befbc855ee6356d94b81b7799", "primary_control": "facts", "result_hash": "87191a5a14670a5ef1fc84d1e34740ce4daea3cd8ec1b14c5eb7dfffdb0be764", "signature": "8da2011c65e7c503373da634773e7d2f9cd83acc6148776b63f8f54fc6dfc91496f54d5d18dab2cd4360d0f980c94cc22739a21464c81657742e14e622b9a403", "state": "VERIFIED", "summary": {"facts": 0.38104, "frozen": 0.196832, "full": 0.401787, "retrieval": 0.407831, "rules_nodetect": 0.337656}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:49:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:38Z", "cpu_seconds": 1.2050771820000001, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f8854cb83ae8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "d7f534ed6c4c6e70bc9c8b6961a3d96ab9713e0417e3a69883f0db754a0e7d10", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f8854cb83ae8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:49:25Z", "seed": 934246229, "treatment": "full"}, "priority": 0.40054884727015017, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ffb2533237905633319e240d414f3dc0fb32a259eb98a97bb831e1fc2169bab9", "run_seconds": 1.318500280380249, "seed": 934246229, "started_at": "2026-11-22T03:49:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.043833, -0.001882], "effect": -0.022857, "evaluated_at": "2026-11-22T03:49:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f8854cb83ae8", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.621276596, 0.440425532, 0.562857143, 0.491489362], "frozen": [0.134042553, 0.127659574, 0.12, 0.127659574], "full": [0.593617021, 0.40212766, 0.571428571, 0.457446809], "retrieval": [0.625531915, 0.440425532, 0.557142857, 0.474468085], "rules_nodetect": [0.712765957, 0.40212766, 0.571428571, 0.457446809]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "d7f534ed6c4c6e70bc9c8b6961a3d96ab9713e0417e3a69883f0db754a0e7d10", "primary_control": "facts", "result_hash": "ffb2533237905633319e240d414f3dc0fb32a259eb98a97bb831e1fc2169bab9", "signature": "08831916dc8b0cf45b8f983c9eb47b6998b06dd260fe90b78ceefb4bba1ee45a130ba519b6fb782ab1763a9879c716e57eb9b04892e4223388c3201526b8fa0a", "state": "VERIFIED", "summary": {"facts": 0.529012, "frozen": 0.12734, "full": 0.506155, "retrieval": 0.524392, "rules_nodetect": 0.535942}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:49:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:56Z", "cpu_seconds": 0.6644296240000003, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-19a8eb8a0872", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "f99ab88aa94030e292b2ed268fd005a1b151a7fd219524590ae982acc7a8a332", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-19a8eb8a0872", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:49:25Z", "seed": 307571791, "treatment": "gated"}, "priority": 0.38249588025248543, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "3b3d120858cea6d75762f661711145508557e32dd78de7def34f94211ede2d99", "run_seconds": 0.6972084045410156, "seed": 307571791, "started_at": "2026-11-22T03:49:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.038693, 0.012549], "effect": -0.013072, "evaluated_at": "2026-11-22T03:50:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-19a8eb8a0872", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.411764706, 0.300653595, 0.431372549, 0.418300654], "gated": [0.411764706, 0.300653595, 0.431372549, 0.418300654], "naive": [0.411764706, 0.281045752, 0.431372549, 0.418300654], "scratch": [0.411764706, 0.300653595, 0.431372549, 0.470588235], "ungated": [0.437908497, 0.300653595, 0.31372549, 0.431372549]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "f99ab88aa94030e292b2ed268fd005a1b151a7fd219524590ae982acc7a8a332", "primary_control": "scratch", "result_hash": "3b3d120858cea6d75762f661711145508557e32dd78de7def34f94211ede2d99", "signature": "bc6ad8c32f1d8c53d863e4cc324f5bd744753a40720d33868ae91738c2f42b8af758642d06e1a892cc3a6419a9a71cbcb56dc2821d2d67df2deae7fdaafbe20d", "state": "VERIFIED", "summary": {"compressed": 0.390523, "gated": 0.390523, "naive": 0.385621, "scratch": 0.403595, "ungated": 0.370915}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:50:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:21Z", "cpu_seconds": 0.39271951700000063, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5f9e7bcdfe6a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "prereg_hash": "8a34c245f88a46d4376fe080435e4de2338b101a3d8b9220b2492ff790dbe0df", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5f9e7bcdfe6a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:49:25Z", "seed": 1473756764, "treatment": "gated"}, "priority": 0.3674082221103996, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-736b57296668", "result_hash": "e55b31958681b3b221a2279e31f2f3ebdd492ba151b4c55689147963209cbede", "run_seconds": 0.48749518394470215, "seed": 1473756764, "started_at": "2026-11-22T03:50:20Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:50:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-5f9e7bcdfe6a", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.607142857, 0.5625, 0.5, 0.428571429], "gated": [0.607142857, 0.5625, 0.5, 0.428571429], "naive": [0.25, 0.145833333, 0.104166667, 0.125], "scratch": [0.607142857, 0.5625, 0.5, 0.428571429], "ungated": [0.160714286, 0.104166667, 0.3125, 0.125]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8a34c245f88a46d4376fe080435e4de2338b101a3d8b9220b2492ff790dbe0df", "primary_control": "scratch", "replicated": true, "replication_of": "E141-ow_transfer-736b57296668", "result_hash": "e55b31958681b3b221a2279e31f2f3ebdd492ba151b4c55689147963209cbede", "signature": "95fb8463467d66ddb87fa49e98f4e325b01bd006ef3539ac61e78a3ef8cb786bcd92c16e20b461a03b2fd6a53c9ca99b6a1a95bceab5d2ad23bb439b713a5501", "state": "PARTIAL", "summary": {"compressed": 0.524554, "gated": 0.524554, "naive": 0.15625, "scratch": 0.524554, "ungated": 0.175595}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:50:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:29Z", "cpu_seconds": 2.2364910709999997, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a6385ba3503e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "c02f5ac6d2ed729f13d6574f06882c28c435e6646c1d4e50c0f622b81eb455cd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a6385ba3503e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:49:25Z", "seed": 857398787, "treatment": "gated"}, "priority": 0.4123784290453184, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "15fa3ddfc579a49cf73e7fa721e832523b5a0b12ab6fe9bbe05e706ea3df4d73", "run_seconds": 2.242236375808716, "seed": 857398787, "started_at": "2026-11-22T03:49:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.096319, 0.000305], "effect": -0.048007, "evaluated_at": "2026-11-22T03:49:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a6385ba3503e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.586206897, 0.564516129, 0.730769231, 0.516666667], "gated": [0.586206897, 0.564516129, 0.730769231, 0.516666667], "naive": [0.655172414, 0.677419355, 0.788461538, 0.633333333], "scratch": [0.629310345, 0.596774194, 0.730769231, 0.633333333], "ungated": [0.655172414, 0.556451613, 0.730769231, 0.566666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c02f5ac6d2ed729f13d6574f06882c28c435e6646c1d4e50c0f622b81eb455cd", "primary_control": "scratch", "result_hash": "15fa3ddfc579a49cf73e7fa721e832523b5a0b12ab6fe9bbe05e706ea3df4d73", "signature": "53f9844dcac9f658f74c796d6ab253f4876f8a916edb6790665644b534635246a85d90af1eae930bb99f092a52e1aec0fc6c9d7c0c14dc445349dec430a7e50d", "state": "VERIFIED", "summary": {"compressed": 0.59954, "gated": 0.59954, "naive": 0.688597, "scratch": 0.647547, "ungated": 0.627265}, "ties": 1, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:49:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:47Z", "cpu_seconds": 7.274333942, "created_at": "2026-11-22T03:49:25Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-6c925e2ed35f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "068e11080cde6d666be195749a50e3356c0ffe29d120e65ad4602cc4059d00b7", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-6c925e2ed35f", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:49:25Z", "seed": 549016664, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "bcb4937cfe63ff144772bce8b183110c996847c0dd7c88a9483c66463c4a3d6c", "run_seconds": 8.275917768478394, "seed": 549016664, "started_at": "2026-11-22T03:49:39Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 2, "reset": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:49:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-6c925e2ed35f", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "068e11080cde6d666be195749a50e3356c0ffe29d120e65ad4602cc4059d00b7", "primary_control": "reset", "result_hash": "bcb4937cfe63ff144772bce8b183110c996847c0dd7c88a9483c66463c4a3d6c", "signature": "a590e01c9f6c31d8ddf7ec5cf78450b23783383bbd500472144d1e8814e1b921ac860c8d8ed6403531f74e3b2f3cebecd57af409fcca33864c20a064d361e20a", "state": "PARTIAL", "summary": {"inherit": 0.65, "reset": 0.65, "root": 0.4166666666666667}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:49:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:45Z", "cpu_seconds": 1.638575813000001, "created_at": "2026-11-22T03:49:41Z", "error": null, "family": "calibration_iw", "fence": 1, "id": "E141-calibration_iw-1f50fdcef4b7", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "prereg_hash": "3cf604f30d8e91d4bef6c1d8428b2aef31e62b93b4bff4f7fca4576dcecb53bb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["temperature", "raw"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "metacognition / calibration under covariate shift", "evolution": "E142", "experiment_id": "E141-calibration_iw-1f50fdcef4b7", "family": "calibration_iw", "hypothesis": "Target-domain ECE of importance-weighted temperature scaling is lower than plain temperature scaling (confirmatory: target_noise = 0).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L3", "metric_direction": "lower", "params": {"n": 16, "n_train": 100, "shift": 0.5, "target_noise": 0.3}, "preregistration_commit": "025499a15", "preregistration_doc": "docs/unified-apex/E142_PREREGISTRATION.md", "primary_control": "temperature", "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "registered_at": "2026-11-22T03:49:41Z", "seed": 350086894, "treatment": "iw_temperature"}, "priority": 0.40445029154153456, "question": "Does importance-weighted temperature scaling reduce target-domain calibration error under covariate shift?", "replicates": "E141-calibration_iw-4e12438970cb", "result_hash": "c9d86b8fa124cb8bf2a3e18b8059e747c611efcfd4c33e5b55e6e60ea4a4fe55", "run_seconds": 1.8429081439971924, "seed": 350086894, "started_at": "2026-11-22T03:49:43Z", "state": "VERIFIED", "verdict": {"better": "lower", "ci95": [-0.017046, 0.007366], "effect": -0.00484, "evaluated_at": "2026-11-22T03:49:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-calibration_iw-1f50fdcef4b7", "family": "calibration_iw", "hypothesis_supported": false, "instance_scores": {"iw_temperature": [0.250586908, 0.248178707, 0.185194575, 0.225123118, 0.091403912, 0.225224882, 0.155832963, 0.238567662, 0.152513165, 0.21341431, 0.265037817, 0.227665995, 0.182852563, 0.117977023, 0.22913419, 0.181469277], "raw": [0.239049035, 0.248178707, 0.185194575, 0.166759847, 0.210331913, 0.160628395, 0.162310123, 0.188127485, 0.21653883, 0.21341431, 0.227121615, 0.21838366, 0.182852563, 0.134311193, 0.186874972, 0.181469277], "temperature": [0.250586908, 0.248178707, 0.185194575, 0.202292397, 0.15698749, 0.225224882, 0.13493897, 0.219323878, 0.159537425, 0.181513252, 0.227121615, 0.227665995, 0.170368295, 0.14675443, 0.213269545, 0.163779475]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0144, "losses": 8, "n": 16, "outcome": "CONTRADICTED", "p_sign_display": "0.967", "prereg_hash": "3cf604f30d8e91d4bef6c1d8428b2aef31e62b93b4bff4f7fca4576dcecb53bb", "primary_control": "temperature", "replicated": true, "replication_of": "E141-calibration_iw-4e12438970cb", "result_hash": "c9d86b8fa124cb8bf2a3e18b8059e747c611efcfd4c33e5b55e6e60ea4a4fe55", "signature": "27a0745e3d5a4ec91276869ca742a67f6b5f9ba9a1e9abd78594bec491847234d30e0493aaa86b679bd78319bb52c4cbcb2c01c1662e43957770dfaee952890d", "state": "VERIFIED", "summary": {"iw_temperature": 0.199386, "raw": 0.195097, "temperature": 0.194546}, "ties": 5, "treatment": "iw_temperature", "wins": 3}, "verified_at": "2026-11-22T03:49:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:18Z", "cpu_seconds": 8.185527071, "created_at": "2026-11-22T03:49:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-acd93ba109b1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "7255c04c164e32911d077d27c494ca4d8ff11e3ad637bc6750eb29b91547d7ab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-acd93ba109b1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:49:41Z", "seed": 1461908629, "treatment": "frontier"}, "priority": 0.37720291433834974, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5364b366632a0cf0d9a3598aa319426b02cff0e4b2228329ead1fcaf42be2257", "run_seconds": 10.563382148742676, "seed": 1461908629, "started_at": "2026-11-22T03:50:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.030818, 0.460859], "effect": 0.215021, "evaluated_at": "2026-11-22T03:50:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-acd93ba109b1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.910714286, 0.8, 0.614583333, 0.451428571], "frontier": [0.785714286, 0.991304348, 0.979166667, 0.788571429], "goal": [0.785714286, 1.0, 0.864583333, 0.668571429], "hypdisc": [0.892857143, 0.904347826, 0.770833333, 0.634285714], "infogain": [0.892857143, 0.991304348, 0.9375, 0.777142857], "prederr": [0.696428571, 0.991304348, 0.90625, 0.777142857], "random": [0.892857143, 0.852173913, 0.5625, 0.377142857], "risk": [0.785714286, 0.991304348, 0.979166667, 0.788571429]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "7255c04c164e32911d077d27c494ca4d8ff11e3ad637bc6750eb29b91547d7ab", "primary_control": "random", "result_hash": "5364b366632a0cf0d9a3598aa319426b02cff0e4b2228329ead1fcaf42be2257", "signature": "9281583bc52f68fc722bd2e81920b33429629dc9f31767d86ec20f9453383afb6173114685f36eb6468d5d675ae686b5c358cd3f694bf9530a374a15a447a30c", "state": "VERIFIED", "summary": {"count": 0.694182, "frontier": 0.886189, "goal": 0.829717, "hypdisc": 0.800581, "infogain": 0.899701, "prederr": 0.842781, "random": 0.671168, "risk": 0.886189}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:50:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:52Z", "cpu_seconds": 4.555610167000001, "created_at": "2026-11-22T03:49:41Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-84dc5de1f97b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "c9e0b0f77557639d29f813a6adde17a9cf084d3af88b6ebe3e112984bc76ff4b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-84dc5de1f97b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:49:41Z", "seed": 782425981, "treatment": "full"}, "priority": 0.389609133440541, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e78bc10117bec528358903f09b98618022f81c3ffa79c8ad7fc89c244ed78f2b", "run_seconds": 4.830458402633667, "seed": 782425981, "started_at": "2026-11-22T03:49:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.254215, 0.870785], "effect": 0.5625, "evaluated_at": "2026-11-22T03:49:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-84dc5de1f97b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.0, 0.0, 0.25], "fixed_policy": [0.0, 0.25, 0.5, 0.5], "full": [1.0, 1.0, 0.5, 0.75], "no_world_model": [0.75, 0.5, 0.5, 0.25], "random": [0.75, 0.75, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "c9e0b0f77557639d29f813a6adde17a9cf084d3af88b6ebe3e112984bc76ff4b", "primary_control": "e141_best", "result_hash": "e78bc10117bec528358903f09b98618022f81c3ffa79c8ad7fc89c244ed78f2b", "signature": "175643d18825dde6412b8a4fe225ed1150723c4191223b95b55f2f89978cc13c78dd4c83ac9b6df780ecb6e517ec4aec5509eade925cf5823c51cf403bd2c60e", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.3125, "full": 0.8125, "no_world_model": 0.5, "random": 0.75}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:50:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:58Z", "cpu_seconds": 1.2843119399999985, "created_at": "2026-11-22T03:49:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ca32dc236fdb", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "f9cd7c38ac68497480d7b1ecffd68bb78d1dc7ad834c2eafe1aff8d3e6bebf54", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ca32dc236fdb", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:49:41Z", "seed": 707472616, "treatment": "full"}, "priority": 0.38057179805865343, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fb2e91a5912c2ac603dca18371d426670748080f43c3f6878c1ca551a9791a8a", "run_seconds": 1.3042113780975342, "seed": 707472616, "started_at": "2026-11-22T03:49:57Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:50:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ca32dc236fdb", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.626542], "frozen": [1.0, 0.134356], "full": [1.0, 0.60213], "retrieval": [1.0, 0.626057], "rules_nodetect": [1.0, 0.753101]}, "prereg_hash": "f9cd7c38ac68497480d7b1ecffd68bb78d1dc7ad834c2eafe1aff8d3e6bebf54", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "fb2e91a5912c2ac603dca18371d426670748080f43c3f6878c1ca551a9791a8a", "signature": "119d65b10f68314e92470d051ccf203a76d4f37f4c9924e64c9325e41bd5da88b9cdfc7368ba78485282c0ca6f80815d733088fb89ead0bf32b18bf7a614fd04", "state": "REJECTED", "summary": {"facts": 0.626542, "frozen": 0.134356, "full": 0.60213, "retrieval": 0.626057, "rules_nodetect": 0.753101}}, "verified_at": "2026-11-22T03:50:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:49:59Z", "cpu_seconds": 6.9530255489999995, "created_at": "2026-11-22T03:49:41Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-611f2c66271a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "ce7a8ba0d83075ddfd690978fbf7a9075a51d01d7a4b17fe0802a27921754da4", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-611f2c66271a", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:49:41Z", "seed": 835133768, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "72e48c72a876b35cf55c35dcca88923f5d5bd9ee045d0af81eb23c2f4cebea06", "run_seconds": 7.1761040687561035, "seed": 835133768, "started_at": "2026-11-22T03:49:52Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"inherit": 3, "reset": 3}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:50:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-611f2c66271a", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.5, "rstep": 2}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0], "root": [1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ce7a8ba0d83075ddfd690978fbf7a9075a51d01d7a4b17fe0802a27921754da4", "primary_control": "reset", "result_hash": "72e48c72a876b35cf55c35dcca88923f5d5bd9ee045d0af81eb23c2f4cebea06", "signature": "cf3fff4af94efbc604089089cab613a2babbbe6019c572bbc30b5b0fbd813b2244e948f13e7538eda669c4e9e5e9daa09df398d464245229c8fdb6fccfa85d0e", "state": "PARTIAL", "summary": {"inherit": 0.6333333333333333, "reset": 0.6333333333333333, "root": 0.38333333333333336}, "ties": 60, "treatment": "inherit", "wins": 0}, "verified_at": "2026-11-22T03:50:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:33Z", "cpu_seconds": 3.8803211100000006, "created_at": "2026-11-22T03:49:58Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-2c235bea3dfb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "a12575af424445ac8b61d4ce9574387713a673fc3eb8210dfb87439126b1a05c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2c235bea3dfb", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:49:58Z", "seed": 2073810281, "treatment": "frontier"}, "priority": 0.3566066786416355, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e9330d06558b23d25025ec5961382056cb01a892b840405857fa467806d62df1", "run_seconds": 4.108219623565674, "seed": 2073810281, "started_at": "2026-11-22T03:50:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.120161, 0.025485], "effect": -0.047338, "evaluated_at": "2026-11-22T03:51:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-2c235bea3dfb", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.65224359, 0.865384615, 0.401692708, 0.471875], "frontier": [0.581730769, 0.451923077, 0.421875, 0.425], "goal": [0.580128205, 0.355769231, 0.35546875, 0.421875], "hypdisc": [0.616987179, 0.692307692, 0.383463542, 0.4125], "infogain": [0.581730769, 0.451923077, 0.421875, 0.425], "prederr": [0.581730769, 0.451923077, 0.421875, 0.425], "random": [0.631410256, 0.596153846, 0.386067708, 0.45625], "risk": [0.581730769, 0.451923077, 0.421875, 0.425]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "a12575af424445ac8b61d4ce9574387713a673fc3eb8210dfb87439126b1a05c", "primary_control": "random", "result_hash": "e9330d06558b23d25025ec5961382056cb01a892b840405857fa467806d62df1", "signature": "2a704d907aeb86fa7d29dd1125fb5fb213a9ef899234b53b1388a079b34b465c13dfe78864b5dc124845518cbb5db234bb2c00c18a1d19df39a1fe6ceefe4004", "state": "VERIFIED", "summary": {"count": 0.597799, "frontier": 0.470132, "goal": 0.42831, "hypdisc": 0.526315, "infogain": 0.470132, "prederr": 0.470132, "random": 0.51747, "risk": 0.470132}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:51:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:03Z", "cpu_seconds": 1.393162220999999, "created_at": "2026-11-22T03:49:58Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-301c1e400a5e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "229add01b9020719c007663bd8e1226ffdd77d2e556714e5b887ddfee87075c7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-301c1e400a5e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:49:58Z", "seed": 1496024619, "treatment": "frontier"}, "priority": 0.4078556393294066, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c37ae9258019263b9b13bbb3243ab934e596d0f3d361df2bbe2cfb24c4bd916e", "run_seconds": 1.7718799114227295, "seed": 1496024619, "started_at": "2026-11-22T03:50:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.010236, 0.072957], "effect": 0.041596, "evaluated_at": "2026-11-22T03:50:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-301c1e400a5e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.9, 0.676190476, 0.532051282, 0.755813953], "frontier": [0.9, 0.676190476, 0.75, 0.790697674], "goal": [0.9, 0.676190476, 0.692307692, 0.755813953], "hypdisc": [0.9, 0.676190476, 0.762820513, 0.726744186], "infogain": [0.9, 0.676190476, 0.75, 0.819767442], "prederr": [0.85, 0.676190476, 0.75, 0.819767442], "random": [0.9, 0.602380952, 0.692307692, 0.755813953], "risk": [0.9, 0.676190476, 0.75, 0.790697674]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "229add01b9020719c007663bd8e1226ffdd77d2e556714e5b887ddfee87075c7", "primary_control": "random", "result_hash": "c37ae9258019263b9b13bbb3243ab934e596d0f3d361df2bbe2cfb24c4bd916e", "signature": "0bb3fb4765c679ae075e1227cb03d7ad23cdeed08bb0eed63e1438fc5de2c0e58c5f05a4d1ac2922d254ddc2ba37810c4bca5fabfbfacb7728d33c8cd2681f0f", "state": "VERIFIED", "summary": {"count": 0.716014, "frontier": 0.779222, "goal": 0.756078, "hypdisc": 0.766439, "infogain": 0.786489, "prederr": 0.773989, "random": 0.737626, "risk": 0.779222}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:50:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:50Z", "cpu_seconds": 4.848945945999999, "created_at": "2026-11-22T03:49:58Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-86dca7c7b995", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "07090b444cdbdc46c513e61e01434375c3fe6098c6ea6d69256405dfe6d819d3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-86dca7c7b995", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:49:58Z", "seed": 724135600, "treatment": "frontier"}, "priority": 0.35371463755544386, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ef91b19636fd8ac2ab61e1532bf27bfc6008c78bd4a96b71c696ee11b95a8a0c", "run_seconds": 5.059571743011475, "seed": 724135600, "started_at": "2026-11-22T03:50:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.101741, 0.034866], "effect": -0.033438, "evaluated_at": "2026-11-22T03:51:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-86dca7c7b995", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.9375, 0.865625, 0.783333333, 0.84375], "frontier": [0.73125, 0.771875, 0.75, 0.8875], "goal": [0.6875, 0.6625, 0.535, 0.925], "hypdisc": [0.7875, 0.846875, 0.621666667, 0.8875], "infogain": [0.825, 0.771875, 0.743333333, 0.925], "prederr": [0.79375, 0.771875, 0.63, 0.8625], "random": [0.86875, 0.778125, 0.74, 0.8875], "risk": [0.73125, 0.771875, 0.75, 0.8875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "07090b444cdbdc46c513e61e01434375c3fe6098c6ea6d69256405dfe6d819d3", "primary_control": "random", "result_hash": "ef91b19636fd8ac2ab61e1532bf27bfc6008c78bd4a96b71c696ee11b95a8a0c", "signature": "81ab007a019920a312d802a0286edd087156673369470c1758c725534dfe370238010e9c43564081a82cc0d49871963b6a5ca341ddca4be75342badcef7e4503", "state": "VERIFIED", "summary": {"count": 0.857552, "frontier": 0.785156, "goal": 0.7025, "hypdisc": 0.785885, "infogain": 0.816302, "prederr": 0.764531, "random": 0.818594, "risk": 0.785156}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:51:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:23Z", "cpu_seconds": 0.7790624870000009, "created_at": "2026-11-22T03:49:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8df3f9bcccd6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "a65a836dc6a6d0c1fdf7a1da694f403dbb457a082ba2aa3948f1a564af542803", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8df3f9bcccd6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:49:58Z", "seed": 545603381, "treatment": "full"}, "priority": 0.3609143532514122, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1406822e21da509ff99a3906679100d54fd0160db2b62658eeb850e5e3dc67f9", "run_seconds": 0.7901384830474854, "seed": 545603381, "started_at": "2026-11-22T03:50:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002382, 0.018255], "effect": 0.007937, "evaluated_at": "2026-11-22T03:50:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8df3f9bcccd6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.688888889, 0.755555556, 0.233333333, 0.741747573], "frozen": [0.155555556, 0.155555556, 0.123809524, 0.133980583], "full": [0.688888889, 0.777777778, 0.242857143, 0.741747573], "retrieval": [0.755555556, 0.822222222, 0.233333333, 0.741747573], "rules_nodetect": [0.688888889, 0.777777778, 0.242857143, 0.741747573]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "a65a836dc6a6d0c1fdf7a1da694f403dbb457a082ba2aa3948f1a564af542803", "primary_control": "facts", "result_hash": "1406822e21da509ff99a3906679100d54fd0160db2b62658eeb850e5e3dc67f9", "signature": "d51918a6ef2bf653e5c7e5023f19cf33cd9ffd4bf8678448e48ddb28556e75fe43ca1ad7c35581176866d7dfc556a6bd1c98c00f991392f66cc9fd6b0e65d407", "state": "VERIFIED", "summary": {"facts": 0.604881, "frozen": 0.142225, "full": 0.612818, "retrieval": 0.638215, "rules_nodetect": 0.612818}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:50:54Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:51:17Z", "cpu_seconds": 2.155911275000001, "created_at": "2026-11-22T03:49:58Z", "error": "lease expired; ", "family": "ow_transfer", "fence": 2, "id": "E141-ow_transfer-28a78787ce41", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "5eb2d4b5891770977f7cf0423b7180b0038be235251114da863cdf520595d116", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-28a78787ce41", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:49:58Z", "seed": 28348710, "treatment": "gated"}, "priority": 0.38741668841223204, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "dd90c9720a2b50ed39e768254541d7a871b365d9d8aad43734b747f8d2bdb874", "run_seconds": 2.491781711578369, "seed": 28348710, "started_at": "2026-11-22T03:50:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002308, 0.007115], "effect": 0.002404, "evaluated_at": "2026-11-22T03:51:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-28a78787ce41", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.55, 0.658333333, 0.58, 0.538461538], "gated": [0.55, 0.658333333, 0.58, 0.538461538], "naive": [0.666666667, 0.658333333, 0.66, 0.625], "scratch": [0.55, 0.658333333, 0.58, 0.528846154], "ungated": [0.691666667, 0.658333333, 0.66, 0.615384615]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "5eb2d4b5891770977f7cf0423b7180b0038be235251114da863cdf520595d116", "primary_control": "scratch", "result_hash": "dd90c9720a2b50ed39e768254541d7a871b365d9d8aad43734b747f8d2bdb874", "signature": "27ae9bbc1551871244cb2492cf6cda2ae5e6eab8331b8e17987f2dccc60676d09f55b11080fcdb5bc89d0247b6b80d3889e3180fa98ee50f75ca767b95fe4c03", "state": "VERIFIED", "summary": {"compressed": 0.581699, "gated": 0.581699, "naive": 0.6525, "scratch": 0.579295, "ungated": 0.656346}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:51:51Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:51:21Z", "cpu_seconds": 5.11858646, "created_at": "2026-11-22T03:49:58Z", "error": "lease expired; ", "family": "rsi_meta", "fence": 2, "id": "E141-rsi_meta-3e772dc30189", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "0238ba9c41892bd9406afdc0b5c66fbdefb58c88397781117a0fc0f303a70372", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-3e772dc30189", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:49:58Z", "seed": 66538731, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "6c23155893ae2756c39df17c02d2692d03e62b5bbb02ae5b1d810b737b0c66df", "run_seconds": 5.836416721343994, "seed": 66538731, "started_at": "2026-11-22T03:50:04Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 3, "reset": 2}, "better": "higher", "ci95": [-0.102032, 0.135366], "effect": 0.016667, "evaluated_at": "2026-11-22T03:51:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-3e772dc30189", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6801049297077442, "rstep": 1}, "hypothesis_supported": false, "instance_scores": {"inherit": [1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0], "root": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.301, "losses": 6, "n": 60, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "0238ba9c41892bd9406afdc0b5c66fbdefb58c88397781117a0fc0f303a70372", "primary_control": "reset", "result_hash": "6c23155893ae2756c39df17c02d2692d03e62b5bbb02ae5b1d810b737b0c66df", "signature": "999fa63fa8c894a47f8204332aefdbef4c889d28613cc3cb8bbd2b3571354ddf9046cb8cd3dce45259e48221209098146ea7da5f98c4b85c1e8671a7f0ba040d", "state": "VERIFIED", "summary": {"inherit": 0.6, "reset": 0.5833333333333334, "root": 0.31666666666666665}, "ties": 47, "treatment": "inherit", "wins": 7}, "verified_at": "2026-11-22T03:51:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:57Z", "cpu_seconds": 2.660481505, "created_at": "2026-11-22T03:50:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9052e07258a9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "7346541b4f25c785c255315a7257e34c7d88d112b4bb59631ce0779d8134066c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9052e07258a9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:50:15Z", "seed": 263516657, "treatment": "frontier"}, "priority": 0.3450187603694571, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d4c93eb764d7175647c9995d882304544b5c650b1c31c36c4be7126a9b04a665", "run_seconds": 2.690354347229004, "seed": 263516657, "started_at": "2026-11-22T03:50:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.010278, 0.003333], "effect": -0.003472, "evaluated_at": "2026-11-22T03:51:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9052e07258a9", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.993055556, 1.0], "frontier": [1.0, 0.993055556, 0.993055556, 0.986111111], "goal": [1.0, 0.993055556, 1.0, 0.986111111], "hypdisc": [1.0, 1.0, 1.0, 0.986111111], "infogain": [1.0, 0.993055556, 1.0, 0.986111111], "prederr": [1.0, 0.993055556, 0.993055556, 0.986111111], "random": [1.0, 0.993055556, 0.993055556, 1.0], "risk": [1.0, 0.993055556, 0.993055556, 0.986111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7346541b4f25c785c255315a7257e34c7d88d112b4bb59631ce0779d8134066c", "primary_control": "random", "result_hash": "d4c93eb764d7175647c9995d882304544b5c650b1c31c36c4be7126a9b04a665", "signature": "beb2f67a21b116937fe6a64811e32f160cc939b9638fcec35f67fa4580f7e821bcbb225badd84e411320eab7acd8617fd1e4710971e8bc70991165131e5b4c0d", "state": "VERIFIED", "summary": {"count": 0.998264, "frontier": 0.993056, "goal": 0.994792, "hypdisc": 0.996528, "infogain": 0.994792, "prederr": 0.993056, "random": 0.996528, "risk": 0.993056}, "ties": 3, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:51:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:49Z", "cpu_seconds": 0.3790546269999986, "created_at": "2026-11-22T03:50:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5c49646123af", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "fcfec7ae26bdea7f78b99e5ade2561fd327fd8dd11a822615500000618cf9f14", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5c49646123af", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:50:15Z", "seed": 55464289, "treatment": "full"}, "priority": 0.34674103687665736, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ba80e9dc90412b13f2ab97d871809688e69297e6476faee57e01a15f5489e0bb", "run_seconds": 0.4821205139160156, "seed": 55464289, "started_at": "2026-11-22T03:50:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.266451, 0.016451], "effect": -0.125, "evaluated_at": "2026-11-22T03:51:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5c49646123af", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 1.0, 0.5], "fixed_policy": [0.0, 0.0, 0.5, 0.25], "full": [0.0, 0.0, 1.0, 0.25], "no_world_model": [0.0, 0.0, 1.0, 0.25], "random": [0.0, 0.25, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "fcfec7ae26bdea7f78b99e5ade2561fd327fd8dd11a822615500000618cf9f14", "primary_control": "e141_best", "result_hash": "ba80e9dc90412b13f2ab97d871809688e69297e6476faee57e01a15f5489e0bb", "signature": "68c292237bad8bec4477b2be3b42d86cb9ee1c84afbcfd261a5bb0119f51519cbd730b4ef03298a4c5950ffbca83d1c344df2268987b22a858fbe1a2a32f7200", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.1875, "full": 0.3125, "no_world_model": 0.3125, "random": 0.4375}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:51:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:22Z", "cpu_seconds": 5.755304966000001, "created_at": "2026-11-22T03:50:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8b538ddaac34", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "31771549dd522775b0ba51dd0b384a931ff0bcc9b2a22f8ea4a678be5510ce1e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8b538ddaac34", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:50:15Z", "seed": 1905591685, "treatment": "full"}, "priority": 0.36960537728437626, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "174a53c3f307e75472a4df117bf7418a229652bf6ea53142067067c63adde745", "run_seconds": 6.7186384201049805, "seed": 1905591685, "started_at": "2026-11-22T03:50:15Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:50:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8b538ddaac34", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.3125], "fixed_policy": [1.0, 0.3125], "full": [1.0, 0.1875], "no_world_model": [1.0, 0.3125], "random": [1.0, 0.1875]}, "prereg_hash": "31771549dd522775b0ba51dd0b384a931ff0bcc9b2a22f8ea4a678be5510ce1e", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "174a53c3f307e75472a4df117bf7418a229652bf6ea53142067067c63adde745", "signature": "0e736c5ff56b2858b776fa73e4a558423dd4e44b38e9edc397a33436cd2e8e1df6790fbec00e0b3a9f4c81fc4cfe5d65c65d32d16e58ef8a0f61230b590f630a", "state": "REJECTED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.3125, "full": 0.1875, "no_world_model": 0.3125, "random": 0.1875}}, "verified_at": "2026-11-22T03:50:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:29Z", "cpu_seconds": 2.591791916, "created_at": "2026-11-22T03:50:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5281cbe47499", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "01288b99e333a71b196ed329c09fb906fc687b41a8a7b855f1527e5ac70c6d2a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5281cbe47499", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:50:15Z", "seed": 1940363060, "treatment": "full"}, "priority": 0.3573821341251595, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "319d6eda0eb41a068c1b23e83ca4ca67b70cd4fdd4884219ca5289345192a0c8", "run_seconds": 2.7763254642486572, "seed": 1940363060, "started_at": "2026-11-22T03:50:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.097913, 0.434595], "effect": 0.266254, "evaluated_at": "2026-11-22T03:50:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5281cbe47499", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.290643275, 0.158333333, 0.540833333, 0.275280899], "frozen": [0.232748538, 0.0, 0.251666667, 0.217228464], "full": [0.74502924, 0.205, 0.775833333, 0.604244694], "retrieval": [0.785380117, 0.64, 0.915833333, 0.766541823], "rules_nodetect": [0.232748538, 0.305, 0.775833333, 0.265293383]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "01288b99e333a71b196ed329c09fb906fc687b41a8a7b855f1527e5ac70c6d2a", "primary_control": "facts", "result_hash": "319d6eda0eb41a068c1b23e83ca4ca67b70cd4fdd4884219ca5289345192a0c8", "signature": "e0cbab718404cb22414c1bb4837b846b785412f2535be43fea274aeb377b3c666934382a1cd32add01aa7b7f7f67365f69c690cedf409d807f4240764b734b08", "state": "VERIFIED", "summary": {"facts": 0.316273, "frozen": 0.175411, "full": 0.582527, "retrieval": 0.776939, "rules_nodetect": 0.394719}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:50:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:30Z", "cpu_seconds": 3.2611988200000006, "created_at": "2026-11-22T03:50:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-12646584a63d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "5c1f3a115848d4f4f7c6f2798f3e5f8c91a5b61429a9fc9a6be411c66d0b45e9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-12646584a63d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:50:15Z", "seed": 1875515282, "treatment": "gated"}, "priority": 0.3598750835706633, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c1390b179f811f3eb09f1002f6efc380f80f687cdaf4b330da48f7a0dc5231ee", "run_seconds": 3.591700792312622, "seed": 1875515282, "started_at": "2026-11-22T03:50:26Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:50:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-12646584a63d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.197979798, 0.808333333, 0.883333333, 0.143434343], "gated": [0.197979798, 0.808333333, 0.883333333, 0.143434343], "naive": [0.197979798, 0.658333333, 0.883333333, 0.165656566], "scratch": [0.197979798, 0.808333333, 0.883333333, 0.143434343], "ungated": [0.171717172, 0.658333333, 0.883333333, 0.165656566]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "5c1f3a115848d4f4f7c6f2798f3e5f8c91a5b61429a9fc9a6be411c66d0b45e9", "primary_control": "scratch", "result_hash": "c1390b179f811f3eb09f1002f6efc380f80f687cdaf4b330da48f7a0dc5231ee", "signature": "323554b5faaa3e1b06f91e2cedc319e9be7f21af01e5e15b8ce5fefbb4e185fd4c3187cf5e82b8ae8e23ed8675ae5e658747164de422750964d42aea21331e0a", "state": "PARTIAL", "summary": {"compressed": 0.50827, "gated": 0.50827, "naive": 0.476326, "scratch": 0.50827, "ungated": 0.46976}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:51:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:51Z", "cpu_seconds": 3.8929987130000008, "created_at": "2026-11-22T03:50:32Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c2a1c0f5a894", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "0aeb9fab531ba00d86f63aaa81b0c264d205dc7e8774c9782af56caf2273a0ce", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c2a1c0f5a894", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:50:32Z", "seed": 1334696805, "treatment": "frontier"}, "priority": 0.35204565162570833, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d407055fef2f8fa685e34a592af5e87235f87d4d6d6b19055aae4033b4c72814", "run_seconds": 4.135716915130615, "seed": 1334696805, "started_at": "2026-11-22T03:50:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.351872, -0.038009], "effect": -0.19494, "evaluated_at": "2026-11-22T03:51:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c2a1c0f5a894", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.927083333, 0.9875, 0.875, 1.0], "frontier": [0.729166667, 0.55, 0.383928571, 1.0], "goal": [0.625, 0.5875, 0.508928571, 1.0], "hypdisc": [0.71875, 0.875, 0.741071429, 1.0], "infogain": [0.729166667, 0.55, 0.383928571, 1.0], "prederr": [0.729166667, 0.55, 0.383928571, 1.0], "random": [0.875, 0.925, 0.642857143, 1.0], "risk": [0.729166667, 0.55, 0.383928571, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "0aeb9fab531ba00d86f63aaa81b0c264d205dc7e8774c9782af56caf2273a0ce", "primary_control": "random", "result_hash": "d407055fef2f8fa685e34a592af5e87235f87d4d6d6b19055aae4033b4c72814", "signature": "c067ad19d0eb09ad5af74b7e742039e44ab45e8dfeceabd2ecb3100ce77f38d8d8b4900489c3c0010252ed54b0aca3dab97dd7654cbfe6a7da131b9520efea02", "state": "VERIFIED", "summary": {"count": 0.947396, "frontier": 0.665774, "goal": 0.680357, "hypdisc": 0.833705, "infogain": 0.665774, "prederr": 0.665774, "random": 0.860714, "risk": 0.665774}, "ties": 1, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:51:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:03Z", "cpu_seconds": 1.1121956380000029, "created_at": "2026-11-22T03:50:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-502119f586c1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "95372ca0cdca3a7f1c351865b10232a4d1b979e9379ef5e688d769af60318188", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-502119f586c1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:50:32Z", "seed": 154555791, "treatment": "full"}, "priority": 0.3421136744747035, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "943e4e35c1857bb19dccea13278a4e174298b810eac03fc0b4f310dadb714eb3", "run_seconds": 1.137207269668579, "seed": 154555791, "started_at": "2026-11-22T03:51:02Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:51:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-502119f586c1", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 1.0, 0.5, 1.0], "random": [0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "95372ca0cdca3a7f1c351865b10232a4d1b979e9379ef5e688d769af60318188", "primary_control": "e141_best", "result_hash": "943e4e35c1857bb19dccea13278a4e174298b810eac03fc0b4f310dadb714eb3", "signature": "309ceceb0fd188ee6c549c301341ff069c7659c1fea1ec1e5b8c8cd60e0e57061c97723a9a13dbd6c98468e6de3db8a2269d9210bb510c5a8bb96adfbe2e3402", "state": "PARTIAL", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.875, "random": 0.5}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:51:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:40Z", "cpu_seconds": 4.627096819, "created_at": "2026-11-22T03:50:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-800bf6fa9dd4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "b5b36b9737711e4f7fc79f5f11419cf3a33613d969a54d71fac19e013ccd6bf2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-800bf6fa9dd4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:50:32Z", "seed": 1125796020, "treatment": "full"}, "priority": 0.3743115760520658, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9ba9e0a52fea2ae18645730f31d858ab51532d690d1cfec77955cdde9aca66a1", "run_seconds": 5.385026216506958, "seed": 1125796020, "started_at": "2026-11-22T03:50:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.299352, 0.549352], "effect": 0.125, "evaluated_at": "2026-11-22T03:51:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-800bf6fa9dd4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.25, 0.0, 0.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [0.5, 1.0, 0.0, 0.0], "no_world_model": [1.0, 0.25, 0.0, 0.0], "random": [0.0, 0.75, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "b5b36b9737711e4f7fc79f5f11419cf3a33613d969a54d71fac19e013ccd6bf2", "primary_control": "e141_best", "result_hash": "9ba9e0a52fea2ae18645730f31d858ab51532d690d1cfec77955cdde9aca66a1", "signature": "7e26c6f04a9af84aff4ebc3124a1d6f4fdf758829b7a34a4e04495bf8bea9b7770c1d0aa6d9197a3961fb4fbc11ae78904f8f17858e94e23e142f45338f84a08", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 0.375, "no_world_model": 0.3125, "random": 0.1875}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:51:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:47Z", "cpu_seconds": 0.5451048069999997, "created_at": "2026-11-22T03:50:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f286d0306f2a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "15415e393ebd3f2f394477730bc9b3865cb94e9f2f1ef6d9898174c5158922d4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f286d0306f2a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:50:32Z", "seed": 2000607643, "treatment": "full"}, "priority": 0.33333172280730905, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a98cb055dd3753ca730afc053ed83b6f5c4bd82fb910cd09a99a9e7ac2586a58", "run_seconds": 0.6471521854400635, "seed": 2000607643, "started_at": "2026-11-22T03:51:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T03:52:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f286d0306f2a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.25, 0.75], "fixed_policy": [0.0, 0.5, 0.0, 0.0], "full": [0.75, 1.0, 0.25, 0.25], "no_world_model": [0.0, 1.0, 0.25, 0.75], "random": [0.0, 1.0, 0.25, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "15415e393ebd3f2f394477730bc9b3865cb94e9f2f1ef6d9898174c5158922d4", "primary_control": "e141_best", "result_hash": "a98cb055dd3753ca730afc053ed83b6f5c4bd82fb910cd09a99a9e7ac2586a58", "signature": "55c1328a68368f5b9407e1179fd0cafbac8ef133f92591506d010caaf480f884c64501ad36b423593fa17368bc1730e0b1be6c50105ee2202683ed791bd4240d", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.125, "full": 0.5625, "no_world_model": 0.5, "random": 0.375}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:52:02Z"}
{"attempts": 2, "completed_at": "2026-11-22T03:52:36Z", "cpu_seconds": 0.580154008, "created_at": "2026-11-22T03:50:32Z", "error": "lease expired; ", "family": "ow_model", "fence": 2, "id": "E141-ow_model-644486274ce0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "9dbf879bc0d2788b77a9a1e26be4c9358df98a38f9967f9c6170a3c47a96dc8c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-644486274ce0", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:50:32Z", "seed": 516813028, "treatment": "full"}, "priority": 0.33406714641940977, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "527269614b97a1cf13ee13da24c8cfa4759ccaa0898a0f47bbb32b2dc05a98dc", "run_seconds": 0.7094595432281494, "seed": 516813028, "started_at": "2026-11-22T03:51:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.050388, 0.09203], "effect": 0.071209, "evaluated_at": "2026-11-22T03:52:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-644486274ce0", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.511627907, 0.310606061, 0.383333333, 0.305220884], "frozen": [0.325581395, 0.090909091, 0.077777778, 0.108433735], "full": [0.610465116, 0.386363636, 0.433333333, 0.365461847], "retrieval": [0.587209302, 0.46969697, 0.705555556, 0.481927711], "rules_nodetect": [0.837209302, 0.462121212, 0.433333333, 0.200803213]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "9dbf879bc0d2788b77a9a1e26be4c9358df98a38f9967f9c6170a3c47a96dc8c", "primary_control": "facts", "result_hash": "527269614b97a1cf13ee13da24c8cfa4759ccaa0898a0f47bbb32b2dc05a98dc", "signature": "cdd6a5b3562d5948cf6c9073f6e23cd9f130ec74872394a7a2f4e72314b1e11ecfe422ed476a3caf62f23454c325802d4b8454741bc3fd976c367a4c8316bd03", "state": "VERIFIED", "summary": {"facts": 0.377697, "frozen": 0.150675, "full": 0.448906, "retrieval": 0.561097, "rules_nodetect": 0.483367}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:52:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:42Z", "cpu_seconds": 4.102882847, "created_at": "2026-11-22T03:50:32Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-151784696354", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "2bcfcc78019bf5c8b5a7b7138569d1da09d4629f1fabb138d3a00b5e787b693d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-151784696354", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:50:32Z", "seed": 712676661, "treatment": "gated"}, "priority": 0.36524378337410945, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "bdf565da260f541948ec0c3581a106a1ecc87fa24f4f9fd0ad994db71b692f93", "run_seconds": 4.591066360473633, "seed": 712676661, "started_at": "2026-11-22T03:50:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.008753, 0.121346], "effect": 0.056297, "evaluated_at": "2026-11-22T03:51:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-151784696354", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.472222222, 0.401785714, 0.782258065, 0.480769231], "gated": [0.472222222, 0.401785714, 0.782258065, 0.480769231], "naive": [0.638888889, 0.75, 0.782258065, 0.557692308], "scratch": [0.472222222, 0.401785714, 0.653225806, 0.384615385], "ungated": [0.638888889, 0.669642857, 0.782258065, 0.615384615]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "2bcfcc78019bf5c8b5a7b7138569d1da09d4629f1fabb138d3a00b5e787b693d", "primary_control": "scratch", "result_hash": "bdf565da260f541948ec0c3581a106a1ecc87fa24f4f9fd0ad994db71b692f93", "signature": "6cb4cd652a699752bae957e776ba26a2d5d6b8d2ca2e0ada31c98db3839a85cf8ddd5308c74bf3d7ef3981298975b85ba1c945e9c10116ec976e7c0dccd1dd03", "state": "VERIFIED", "summary": {"compressed": 0.534259, "gated": 0.534259, "naive": 0.68221, "scratch": 0.477962, "ungated": 0.676544}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:51:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:44Z", "cpu_seconds": 8.101039956000001, "created_at": "2026-11-22T03:50:32Z", "error": null, "family": "rsi_meta", "fence": 1, "id": "E141-rsi_meta-5ea12685f7b0", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "prereg_hash": "3b1f5b983eca2480d8d4a32aa7ef24e420770b86cae454973f69a13b546e3c65", "preregistration": {"analysis": "experiment-level sign test across experiments (confirmatory); per-instance paired test descriptive", "compute": "same seed, 4 generations x 6 children, 30 dev tasks, node budget 2500 for both arms", "controls": ["reset", "root (descriptive)"], "evolution": "E142", "experiment_id": "E141-rsi_meta-5ea12685f7b0", "family": "rsi_meta", "heldout": "60 tasks generated by the evaluator from its secret salt", "hypothesis": "inherit final held-out solve rate > reset final held-out solve rate", "hypothesis_id": "H4", "lab": "RSI_LAB", "params": {"budget": 2500, "generations": 4, "k": 6, "n_dev": 30, "n_heldout": 60, "root_config": {"M": 5, "R": 3, "order": ["poly", "geo", "rec1", "rec2", "alt", "mod"], "proposer": {"mut": 0.5, "rstep": 1}}}, "preregistration_commit": "025499a15", "primary_control": "reset", "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "registered_at": "2026-11-22T03:50:32Z", "seed": 255079430, "treatment": "inherit"}, "priority": 4.0, "question": "Does a lineage that inherits its proposer's mutated meta-parameters end with a better held-out solver than a lineage whose proposer is reset each generation?", "replicates": null, "result_hash": "0b4e935ae9f74f61551434e2f29ee6259c3e7a60324d45f8175dd81528c37b33", "run_seconds": 8.994613885879517, "seed": 255079430, "started_at": "2026-11-22T03:50:35Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"inherit": 4, "reset": 2}, "better": "higher", "ci95": [0.101683, 0.331651], "effect": 0.216667, "evaluated_at": "2026-11-22T03:51:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-rsi_meta-5ea12685f7b0", "family": "rsi_meta", "final_proposer_inherit": {"mut": 0.6302885504314834, "rstep": 1}, "hypothesis_supported": true, "instance_scores": {"inherit": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "reset": [1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0], "root": [0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -3.3113, "losses": 1, "n": 60, "outcome": "SUPPORTED", "p_sign_display": "0.000488", "prereg_hash": "3b1f5b983eca2480d8d4a32aa7ef24e420770b86cae454973f69a13b546e3c65", "primary_control": "reset", "result_hash": "0b4e935ae9f74f61551434e2f29ee6259c3e7a60324d45f8175dd81528c37b33", "signature": "c54d290cd775bf1b2f6fd6ac8b4452382d69656d613e788a6ca9f8322c62bb6ce0eabfba699e07fe13fa9fd6d06e51247bed41d80f01b1a5b9790fd434a99f0e", "state": "VERIFIED", "summary": {"inherit": 0.9666666666666667, "reset": 0.75, "root": 0.45}, "ties": 45, "treatment": "inherit", "wins": 14}, "verified_at": "2026-11-22T03:51:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:58Z", "cpu_seconds": 2.389287938999999, "created_at": "2026-11-22T03:50:48Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-74b2814345e9", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "86a240f25f8f9f1cfec78229d1f6737f01f9510cbc6cdfc1fbb752c3e093c8a0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-74b2814345e9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:50:48Z", "seed": 558107029, "treatment": "frontier"}, "priority": 0.3446700128024768, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-836557f4e661", "result_hash": "5e1529bb365064b557794fdaa715df0164f6fddd89af4ece4a891ed0d2c77875", "run_seconds": 2.4731054306030273, "seed": 558107029, "started_at": "2026-11-22T03:50:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.101205, 0.205223], "effect": 0.052009, "evaluated_at": "2026-11-22T03:51:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-74b2814345e9", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.763636364, 0.8, 0.526027397, 0.528571429], "frontier": [0.781818182, 0.854545455, 0.726027397, 0.295238095], "goal": [0.763636364, 0.872727273, 0.276712329, 0.261904762], "hypdisc": [0.8, 0.854545455, 0.068493151, 0.495238095], "infogain": [0.781818182, 0.854545455, 0.695890411, 0.295238095], "prederr": [0.781818182, 0.854545455, 0.695890411, 0.295238095], "random": [0.781818182, 0.8, 0.463013699, 0.404761905], "risk": [0.781818182, 0.854545455, 0.726027397, 0.295238095]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "86a240f25f8f9f1cfec78229d1f6737f01f9510cbc6cdfc1fbb752c3e093c8a0", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-836557f4e661", "result_hash": "5e1529bb365064b557794fdaa715df0164f6fddd89af4ece4a891ed0d2c77875", "signature": "286b1b5b01a18dc067cc2e5ed65d7558d6897acaed60c98c6e470637de8332b76452cbe661bcff1117979e029e1a6122e232f4e06a956089af328af28ed8940a", "state": "VERIFIED", "summary": {"count": 0.654559, "frontier": 0.664407, "goal": 0.543745, "hypdisc": 0.554569, "infogain": 0.656873, "prederr": 0.656873, "random": 0.612398, "risk": 0.664407}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:51:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:50:57Z", "cpu_seconds": 0.5188022570000008, "created_at": "2026-11-22T03:50:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-da6ec400568a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "be68f11011b44f774fc5a841022539f024cce07ff2d3fa8e02aa7d5f374c2bed", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-da6ec400568a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:50:48Z", "seed": 736213854, "treatment": "full"}, "priority": 0.34291176860599565, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "3e705e221bd8ce560d6fd46a267af7c26dfd7081cbe288352253697ac11e5a6d", "run_seconds": 0.5245459079742432, "seed": 736213854, "started_at": "2026-11-22T03:50:56Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:51:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-da6ec400568a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.5], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 0.5], "no_world_model": [1.0, 1.0, 1.0, 0.0], "random": [1.0, 1.0, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "be68f11011b44f774fc5a841022539f024cce07ff2d3fa8e02aa7d5f374c2bed", "primary_control": "e141_best", "result_hash": "3e705e221bd8ce560d6fd46a267af7c26dfd7081cbe288352253697ac11e5a6d", "signature": "180b3277ead8a9c5924871e5d06172c9544b06cf39d201d5f9a8e83028792e6fd3318773a6624c7e28efedd67ab7ea4a504aad3f79bf8ceeca2c4bff949ea80a", "state": "PARTIAL", "summary": {"e141_best": 0.875, "fixed_policy": 0.0625, "full": 0.875, "no_world_model": 0.75, "random": 0.875}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:51:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:17Z", "cpu_seconds": 0.5078497710000001, "created_at": "2026-11-22T03:50:48Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-cd8f9cbec6f5", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "e7fdd04484a476dddc43d6ca7b3eca7ce8ddb5a2aff48e1b12e0ebf87d428ee8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-cd8f9cbec6f5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:50:48Z", "seed": 1268757786, "treatment": "full"}, "priority": 0.3365751711879087, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e0df6dcc84bc69ea7aef07f4fe6d11360dc72602d46bef7ad8ec68a600342e6a", "run_seconds": 0.605424165725708, "seed": 1268757786, "started_at": "2026-11-22T03:51:16Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:51:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-cd8f9cbec6f5", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.44541], "frozen": [1.0, 0.364711], "full": [1.0, 0.4632], "retrieval": [1.0, 0.447974], "rules_nodetect": [1.0, 0.498021]}, "prereg_hash": "e7fdd04484a476dddc43d6ca7b3eca7ce8ddb5a2aff48e1b12e0ebf87d428ee8", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "e0df6dcc84bc69ea7aef07f4fe6d11360dc72602d46bef7ad8ec68a600342e6a", "signature": "3d2135bbae5b250b2d032e5011c94dfb00ee89d2e2ed4edd4e4bd89efe1e7d597b02c5b8442be86e20952f9057a090ed1b2811d8afc61d77a6eb9293c2b9d50d", "state": "REJECTED", "summary": {"facts": 0.44541, "frozen": 0.364711, "full": 0.4632, "retrieval": 0.447974, "rules_nodetect": 0.498021}}, "verified_at": "2026-11-22T03:51:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:03Z", "cpu_seconds": 0.7380140510000004, "created_at": "2026-11-22T03:50:48Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-864e08f7fec4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "ae987d0826ee3f6875a42d1802f39f224a8703b7e2c13fed2c0f6003f47bffa0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-864e08f7fec4", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:50:48Z", "seed": 65072113, "treatment": "gated"}, "priority": 0.3405021705742847, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "e8d9bb6e41a9aadc260622dfa3973b7086037dc737702054371cde808d477c96", "run_seconds": 0.7891621589660645, "seed": 65072113, "started_at": "2026-11-22T03:51:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037949, 0.012308], "effect": -0.012821, "evaluated_at": "2026-11-22T03:51:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-864e08f7fec4", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.698924731, 0.60952381, 0.660818713, 0.743589744], "gated": [0.698924731, 0.60952381, 0.660818713, 0.743589744], "naive": [0.698924731, 0.714285714, 0.631578947, 0.743589744], "scratch": [0.698924731, 0.60952381, 0.660818713, 0.794871795], "ungated": [0.698924731, 0.60952381, 0.49122807, 0.743589744]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "ae987d0826ee3f6875a42d1802f39f224a8703b7e2c13fed2c0f6003f47bffa0", "primary_control": "scratch", "result_hash": "e8d9bb6e41a9aadc260622dfa3973b7086037dc737702054371cde808d477c96", "signature": "d93090393e92f05861f9d8ec3d0da58bf9580d2969035a521258fedb0ffe396da6fd116222ff33c7be431259183c469a0f5fe2ccf8e1a361ad1670c43a7b140a", "state": "VERIFIED", "summary": {"compressed": 0.678214, "gated": 0.678214, "naive": 0.697095, "scratch": 0.691035, "ungated": 0.635817}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:51:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:10Z", "cpu_seconds": 0.9656908279999996, "created_at": "2026-11-22T03:51:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ad2fa04dbc88", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "0d7c4264c1f9b94c322dba0fe67daffbb6040561f6894292465d3f5deb742b6d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ad2fa04dbc88", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:51:05Z", "seed": 1606815708, "treatment": "frontier"}, "priority": 0.3395269521710351, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-9c12be0153c6", "result_hash": "50665af497c5ab08a63dce1632321f77bce15b949c2a1c7b68410edfe480855f", "run_seconds": 1.1774845123291016, "seed": 1606815708, "started_at": "2026-11-22T03:51:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.209463, 0.26608], "effect": 0.237772, "evaluated_at": "2026-11-22T03:51:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ad2fa04dbc88", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.983695652, 0.983695652, 0.972826087, 0.983695652], "frontier": [0.961956522, 0.994565217, 1.0, 1.0], "goal": [0.635869565, 0.608695652, 0.592391304, 0.673913043], "hypdisc": [0.711956522, 0.679347826, 0.733695652, 0.711956522], "infogain": [0.961956522, 0.994565217, 1.0, 1.0], "prederr": [0.961956522, 0.994565217, 1.0, 1.0], "random": [0.755434783, 0.744565217, 0.72826087, 0.777173913], "risk": [0.961956522, 0.994565217, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "0d7c4264c1f9b94c322dba0fe67daffbb6040561f6894292465d3f5deb742b6d", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-9c12be0153c6", "result_hash": "50665af497c5ab08a63dce1632321f77bce15b949c2a1c7b68410edfe480855f", "signature": "699c109100e5769402d8f5562b56f1bbde966f534b0851f8c5f504a68cfc1ad9b198b262e21ad8947155060296f60ca0828039cea30cf39d8fe319fcb09fd205", "state": "VERIFIED", "summary": {"count": 0.980978, "frontier": 0.98913, "goal": 0.627717, "hypdisc": 0.709239, "infogain": 0.98913, "prederr": 0.98913, "random": 0.751359, "risk": 0.98913}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T03:51:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:48Z", "cpu_seconds": 0.4099988509999999, "created_at": "2026-11-22T03:51:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-036da62b8723", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "4be48ce99f98a8bd460d1b19e449e46548b29ab019c85de37c6ec066dfeca86e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-036da62b8723", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:51:05Z", "seed": 129870662, "treatment": "full"}, "priority": 0.33322390573246297, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e60f24b065a6a20fff0e04ea4cff829087e3ba20921695e698f725b7e3f2aae8", "run_seconds": 0.4644181728363037, "seed": 129870662, "started_at": "2026-11-22T03:51:47Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:52:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-036da62b8723", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 1.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [0.5, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4be48ce99f98a8bd460d1b19e449e46548b29ab019c85de37c6ec066dfeca86e", "primary_control": "e141_best", "result_hash": "e60f24b065a6a20fff0e04ea4cff829087e3ba20921695e698f725b7e3f2aae8", "signature": "f1712c577c5aa94194f2d1a11bc2353a448c26a257360c96a79963f002ab9714d3c2a0cb11cf0e7fe9fe345f1a7112ccac2259bd6b30fe753a6fe88c6d037b00", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 1.0, "random": 0.8125}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:52:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:07Z", "cpu_seconds": 0.39522053099999965, "created_at": "2026-11-22T03:51:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-79f28b2f17dc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "658fc086c0f4b8cca85e97daca2ca40000fe214f4595d476a27cf8d62b3404ea", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-79f28b2f17dc", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:51:05Z", "seed": 40663227, "treatment": "full"}, "priority": 0.3312423430609884, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "fc3b09f969903c5b6e019b799b0d192496750359926edb81d297b0da22530550", "run_seconds": 0.4104766845703125, "seed": 40663227, "started_at": "2026-11-22T03:52:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.61, 0.86], "effect": 0.125, "evaluated_at": "2026-11-22T03:52:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-79f28b2f17dc", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.0, 0.75, 0.25], "fixed_policy": [0.0, 0.75, 0.25, 0.0], "full": [1.0, 0.75, 0.0, 0.0], "no_world_model": [0.25, 0.25, 0.0, 0.0], "random": [1.0, 0.25, 0.75, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "658fc086c0f4b8cca85e97daca2ca40000fe214f4595d476a27cf8d62b3404ea", "primary_control": "e141_best", "result_hash": "fc3b09f969903c5b6e019b799b0d192496750359926edb81d297b0da22530550", "signature": "8f1fa172b0303a090e14197b958cc19807188faad4c401f9f203b970024732bc27d79e7a9ac3596202a45bb2dd66fb4d89d21f5de5dc2202e20788bee6af9b0c", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.25, "full": 0.4375, "no_world_model": 0.125, "random": 0.5625}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:52:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:53Z", "cpu_seconds": 0.7076259629999999, "created_at": "2026-11-22T03:51:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1ba66ea9555c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "e2bf80e245caa89735724cad769ea6143e5f00c4160f0a29c40f9dab7f6e7fa8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1ba66ea9555c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:51:05Z", "seed": 2017524527, "treatment": "full"}, "priority": 0.33135532107366733, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "8b620f22370521ddb95c76d914010321f79defa7af923f66228efcac08c9a72d", "run_seconds": 0.7992894649505615, "seed": 2017524527, "started_at": "2026-11-22T03:51:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.015635, 0.155887], "effect": 0.085761, "evaluated_at": "2026-11-22T03:52:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1ba66ea9555c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.636842105, 0.705263158, 0.482142857, 0.98], "frozen": [0.242105263, 0.231578947, 0.303571429, 0.36], "full": [0.778947368, 0.852631579, 0.535714286, 0.98], "retrieval": [0.768421053, 0.884210526, 0.625, 0.98], "rules_nodetect": [0.310526316, 0.852631579, 0.625, 0.58]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "e2bf80e245caa89735724cad769ea6143e5f00c4160f0a29c40f9dab7f6e7fa8", "primary_control": "facts", "result_hash": "8b620f22370521ddb95c76d914010321f79defa7af923f66228efcac08c9a72d", "signature": "9d58543cc3e848ce0bc124b5827e6fd0a049e9000bd641864b7537de09a5cf78dca7659b0a4ea6813472dd055d216107045230c95c760579e9d6c8691b62660c", "state": "VERIFIED", "summary": {"facts": 0.701062, "frozen": 0.284314, "full": 0.786823, "retrieval": 0.814408, "rules_nodetect": 0.592039}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:52:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:11Z", "cpu_seconds": 2.788722988, "created_at": "2026-11-22T03:51:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9686113909ff", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "7ce34fcfc5b5b7a900b49cbdbbca80c2fb8d3b641990d872c3b1cff9c29c04e9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9686113909ff", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:51:05Z", "seed": 1909611146, "treatment": "full"}, "priority": 0.35427747204787885, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3717b91d018a8eaef2c7a8baa6f0c6d0f4174c00823a5b05a21625daf96b3aed", "run_seconds": 3.314724922180176, "seed": 1909611146, "started_at": "2026-11-22T03:51:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.086942, 0.275785], "effect": 0.181364, "evaluated_at": "2026-11-22T03:51:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9686113909ff", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.316666667, 0.297348485, 0.733333333, 0.64], "frozen": [0.296666667, 0.09469697, 0.355555556, 0.38], "full": [0.452222222, 0.509469697, 0.811111111, 0.94], "retrieval": [0.67, 0.583333333, 0.894444444, 0.86], "rules_nodetect": [0.323333333, 0.509469697, 0.516666667, 0.64]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "7ce34fcfc5b5b7a900b49cbdbbca80c2fb8d3b641990d872c3b1cff9c29c04e9", "primary_control": "facts", "result_hash": "3717b91d018a8eaef2c7a8baa6f0c6d0f4174c00823a5b05a21625daf96b3aed", "signature": "9c685584c98d1411dcf4dd27d52133a0ce6b91e5923543a74146c9f0d789460bde60e6946bfe111beff768cf513384378a41130792ce4a3eeb734e72e6c90d0f", "state": "VERIFIED", "summary": {"facts": 0.496837, "frozen": 0.28173, "full": 0.678201, "retrieval": 0.751944, "rules_nodetect": 0.497367}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:51:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:53Z", "cpu_seconds": 0.660069182, "created_at": "2026-11-22T03:51:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-af2d09072057", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "21b2e8a25d2d3dc07dd015fdcf92e485246b11ea6eb892f5fd18280359191083", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-af2d09072057", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:51:05Z", "seed": 73224672, "treatment": "full"}, "priority": 0.33164459794916556, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7ff50a7b7085f1c10fd2944c6eba0256be98b98a8a575667f8b8c7272ba3df61", "run_seconds": 0.7584226131439209, "seed": 73224672, "started_at": "2026-11-22T03:51:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017143, 0.052857], "effect": 0.017857, "evaluated_at": "2026-11-22T03:52:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-af2d09072057", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.861702128, 0.888888889, 0.939393939, 0.803571429], "frozen": [0.337765957, 0.555555556, 0.424242424, 0.25], "full": [0.861702128, 0.888888889, 0.939393939, 0.875], "retrieval": [0.861702128, 0.888888889, 0.939393939, 0.839285714], "rules_nodetect": [0.079787234, 0.444444444, 0.606060606, 0.553571429]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "21b2e8a25d2d3dc07dd015fdcf92e485246b11ea6eb892f5fd18280359191083", "primary_control": "facts", "result_hash": "7ff50a7b7085f1c10fd2944c6eba0256be98b98a8a575667f8b8c7272ba3df61", "signature": "46339532085f7986569f62616ad7062ff4888210e27eae6a6889264c40f65e33a7d8e81e19c995296d12ced0d987c5094e36d8a7370755de0ad09a726593cf08", "state": "VERIFIED", "summary": {"facts": 0.873389, "frozen": 0.391891, "full": 0.891246, "retrieval": 0.882318, "rules_nodetect": 0.420966}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:52:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:23Z", "cpu_seconds": 1.0009798960000005, "created_at": "2026-11-22T03:51:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9cceaddbfcf1", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "e7e8397a4084ffad43b81707631445ca3723bd2a78dd75cf86f47637e63f0cd6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9cceaddbfcf1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:51:22Z", "seed": 1766349441, "treatment": "frontier"}, "priority": 0.33741657042528383, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-76e75cc7ed85", "result_hash": "f282b78f99e7fdc15fc400db573a68b298172ad47b30100c82eea3333678fc52", "run_seconds": 1.0380597114562988, "seed": 1766349441, "started_at": "2026-11-22T03:51:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.145003, 0.148159], "effect": 0.001578, "evaluated_at": "2026-11-22T03:51:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9cceaddbfcf1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.9375, 0.765151515, 0.77124183, 0.784313725], "frontier": [0.71875, 0.371212121, 0.803921569, 0.888888889], "goal": [0.645833333, 0.25, 0.764705882, 0.790849673], "hypdisc": [0.8125, 0.651515152, 0.797385621, 0.816993464], "infogain": [0.71875, 0.371212121, 0.803921569, 0.888888889], "prederr": [0.71875, 0.371212121, 0.803921569, 0.888888889], "random": [0.760416667, 0.545454545, 0.764705882, 0.705882353], "risk": [0.71875, 0.371212121, 0.803921569, 0.888888889]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "e7e8397a4084ffad43b81707631445ca3723bd2a78dd75cf86f47637e63f0cd6", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-76e75cc7ed85", "result_hash": "f282b78f99e7fdc15fc400db573a68b298172ad47b30100c82eea3333678fc52", "signature": "7d0560510787859f3ab87a7956abbfce5f99824978a4d257e370c96ad3246c39d5c8f9db5e107e2ff4f194690535ddf957819a17b2afba39a5c12cfa69d09c08", "state": "VERIFIED", "summary": {"count": 0.814552, "frontier": 0.695693, "goal": 0.612847, "hypdisc": 0.769599, "infogain": 0.695693, "prederr": 0.695693, "random": 0.694115, "risk": 0.695693}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:51:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:29Z", "cpu_seconds": 0.6299158309999981, "created_at": "2026-11-22T03:51:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9e546ac12f5f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "bf6130c486163cd0fba6b1126397cb41d27bbce231b4d76b22bf8785b48f7801", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9e546ac12f5f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:51:22Z", "seed": 981382318, "treatment": "frontier"}, "priority": 0.33548290535573605, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "95de6bfc4b96edab9d1ae7d4ca1a24d16b7335046ec68468315e53d06b30eebe", "run_seconds": 0.6497695446014404, "seed": 981382318, "started_at": "2026-11-22T03:51:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007845, 0.070345], "effect": 0.03125, "evaluated_at": "2026-11-22T03:51:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9e546ac12f5f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.958333333, 1.0, 1.0, 1.0], "goal": [0.895833333, 0.895833333, 1.0, 1.0], "hypdisc": [0.979166667, 0.958333333, 1.0, 1.0], "infogain": [0.958333333, 1.0, 1.0, 1.0], "prederr": [0.958333333, 1.0, 1.0, 1.0], "random": [0.875, 0.958333333, 1.0, 1.0], "risk": [0.958333333, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "bf6130c486163cd0fba6b1126397cb41d27bbce231b4d76b22bf8785b48f7801", "primary_control": "random", "result_hash": "95de6bfc4b96edab9d1ae7d4ca1a24d16b7335046ec68468315e53d06b30eebe", "signature": "cedcb214a255e2dcd79f030e4a220852109c339fc6acd85f7f48212065993da8b66436ab5369386dbcaa8a101c3761768f6af4f694ae88719c4355b31995e708", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.989583, "goal": 0.947917, "hypdisc": 0.984375, "infogain": 0.989583, "prederr": 0.989583, "random": 0.958333, "risk": 0.989583}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:51:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:27Z", "cpu_seconds": 4.2439825330000005, "created_at": "2026-11-22T03:51:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-fd7463130e13", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "03173147442d578b78713aaf4d5d74c540b3fe181d129311d1fdbc1140c3e814", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-fd7463130e13", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:51:22Z", "seed": 84411707, "treatment": "frontier"}, "priority": 0.34218607700043263, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "36568482c343be53f4c391a40a1479bc06a7558b05800545d6b194b43cee4f61", "run_seconds": 4.458481311798096, "seed": 84411707, "started_at": "2026-11-22T03:51:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.105144, -0.000902], "effect": -0.053023, "evaluated_at": "2026-11-22T03:51:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-fd7463130e13", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.592436975, 1.0, 0.740625, 0.35], "frontier": [0.589285714, 0.8625, 0.528125, 0.306410256], "goal": [0.567752101, 0.9375, 0.509375, 0.312179487], "hypdisc": [0.583508403, 0.9625, 0.640625, 0.316666667], "infogain": [0.589285714, 0.8625, 0.528125, 0.306410256], "prederr": [0.589285714, 0.8625, 0.528125, 0.306410256], "random": [0.578781513, 0.9625, 0.621875, 0.33525641], "risk": [0.589285714, 0.8625, 0.528125, 0.306410256]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "03173147442d578b78713aaf4d5d74c540b3fe181d129311d1fdbc1140c3e814", "primary_control": "random", "result_hash": "36568482c343be53f4c391a40a1479bc06a7558b05800545d6b194b43cee4f61", "signature": "f46a967e274cf806c4f39c18e0e2e820f3a16ad649c97b319088deb4b6e93c73b498f75c33bb5b443b1429ea6d6d95a4944ffdcdc641a4aff3f853d36eb2c40b", "state": "VERIFIED", "summary": {"count": 0.670765, "frontier": 0.57158, "goal": 0.581702, "hypdisc": 0.625825, "infogain": 0.57158, "prederr": 0.57158, "random": 0.624603, "risk": 0.57158}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:51:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:42Z", "cpu_seconds": 0.8118666429999999, "created_at": "2026-11-22T03:51:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-456b318f1ed2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "1b9a708a123df446dda8c9493aade4d6534526b33dbdd4fd2628a6c4333d58f2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-456b318f1ed2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:51:39Z", "seed": 538055050, "treatment": "full"}, "priority": 0.33509455410965094, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "fbbb80c8fa017f1ef3fbcaffcd0ac7a3ddc805754ebb8d7c75b2d708fd1dc98a", "run_seconds": 2.1456289291381836, "seed": 538055050, "started_at": "2026-11-22T03:51:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.74, 0.24], "effect": -0.25, "evaluated_at": "2026-11-22T03:52:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-456b318f1ed2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 1.0, 1.0], "fixed_policy": [0.0, 0.5, 0.0, 0.0], "full": [1.0, 0.0, 1.0, 0.0], "no_world_model": [1.0, 0.0, 1.0, 0.5], "random": [0.0, 0.0, 0.75, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "1b9a708a123df446dda8c9493aade4d6534526b33dbdd4fd2628a6c4333d58f2", "primary_control": "e141_best", "result_hash": "fbbb80c8fa017f1ef3fbcaffcd0ac7a3ddc805754ebb8d7c75b2d708fd1dc98a", "signature": "60461c3c7372e19ffcf59a5ff601afb57087c7661f96eebb5aa5ef43cbef2ada5c86040383b9687ed55b1c79d9186098fe55abcaa73c145a255c001889704106", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.125, "full": 0.5, "no_world_model": 0.625, "random": 0.1875}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:52:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:13Z", "cpu_seconds": 2.809813261, "created_at": "2026-11-22T03:51:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4805cae02045", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "d20481f86482516209fe2f9859b8e65b228d4a5e2ee242d441c1868302da33d7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4805cae02045", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:51:39Z", "seed": 54462912, "treatment": "full"}, "priority": 0.3285276336627489, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "66db795744020e6269187d2a2884fe7892119e98e986509b3d143a119cc9b40f", "run_seconds": 2.970288038253784, "seed": 54462912, "started_at": "2026-11-22T03:53:10Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:53:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4805cae02045", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.875], "fixed_policy": [1.0, 0.0625], "no_world_model": [1.0, 0.9375]}, "prereg_hash": "d20481f86482516209fe2f9859b8e65b228d4a5e2ee242d441c1868302da33d7", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "66db795744020e6269187d2a2884fe7892119e98e986509b3d143a119cc9b40f", "signature": "d4d016528ced212766fe07e7ec88463c172c71f530a6b6d56c489a02fd4a408aaeb47daed817bd1c6be19d5a46ee1776c8e8f90d91342557ba3683cbec40280e", "state": "REJECTED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}}, "verified_at": "2026-11-22T03:53:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:51:41Z", "cpu_seconds": 1.073854628, "created_at": "2026-11-22T03:51:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7c896b8e6301", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "12d3a74986a05ea88622960ac7c509c30aad848d5802dd67e4bc7a40b99b19dc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7c896b8e6301", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:51:39Z", "seed": 578660807, "treatment": "full"}, "priority": 0.33752463493424206, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8842b384acd7e3a4291563d3fc3413b75a8a8e4be01c3653efc83afba036c3d5", "run_seconds": 1.5480360984802246, "seed": 578660807, "started_at": "2026-11-22T03:51:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T03:51:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7c896b8e6301", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.5], "full": [0.5, 1.0, 1.0, 1.0], "no_world_model": [0.5, 0.75, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "12d3a74986a05ea88622960ac7c509c30aad848d5802dd67e4bc7a40b99b19dc", "primary_control": "e141_best", "result_hash": "8842b384acd7e3a4291563d3fc3413b75a8a8e4be01c3653efc83afba036c3d5", "signature": "8f6b7b9c99e4fe35da431e9075e8af01667a708d0441600f14ead040da2a241c8e419effdb226825332fdf25d397be3298ee412c40e75419dccfe29f1c1fb70d", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.125, "full": 0.875, "no_world_model": 0.75, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:52:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:51Z", "cpu_seconds": 0.50216097, "created_at": "2026-11-22T03:51:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-213c8f80f5c2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "7d24754f21d04a86a157484b08d42d7fa9cb9daebc9b52287392ebcdd2e9e916", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-213c8f80f5c2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:51:39Z", "seed": 554050399, "treatment": "full"}, "priority": 0.3210272994105399, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "96b65edcbd4c88c9c393bbfac238601483ea5fe3e569d0f5fc5519c51df16caa", "run_seconds": 0.8529069423675537, "seed": 554050399, "started_at": "2026-11-22T03:53:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022424, 0.007273], "effect": -0.007576, "evaluated_at": "2026-11-22T03:54:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-213c8f80f5c2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.300653595, 0.294117647, 0.317829457, 0.287878788], "frozen": [0.274509804, 0.274509804, 0.294573643, 0.265151515], "full": [0.300653595, 0.294117647, 0.317829457, 0.257575758], "retrieval": [0.300653595, 0.300653595, 0.317829457, 0.295454545], "rules_nodetect": [0.339869281, 0.307189542, 0.317829457, 0.287878788]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7d24754f21d04a86a157484b08d42d7fa9cb9daebc9b52287392ebcdd2e9e916", "primary_control": "facts", "result_hash": "96b65edcbd4c88c9c393bbfac238601483ea5fe3e569d0f5fc5519c51df16caa", "signature": "114748c333c5dd509f2e4aa8e3f6616f183394455604f02252e36459db7a86bf54aa24dfff8530aa539b4baa32aae431805f46b1f3dfebf209126b0d909f0504", "state": "VERIFIED", "summary": {"facts": 0.30012, "frozen": 0.277186, "full": 0.292544, "retrieval": 0.303648, "rules_nodetect": 0.313192}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:54:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:19Z", "cpu_seconds": 144.743947399, "created_at": "2026-11-22T03:51:39Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-7e678130a4a8", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "c86e81ceec474ec5b7e736304640d4230f9f1198df4c27a84375e08bb8b1a815", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-7e678130a4a8", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:51:39Z", "seed": 597108627, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "17f0e7f4f95982fe143a5f66f647f142a4def95bc9b6afbcede1e0a5474e290c", "run_seconds": 159.09631657600403, "seed": 597108627, "started_at": "2026-11-22T03:51:40Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 1, "inherit_mem": 1, "inherit_nomem": 2, "random_search": 0, "reset_mem": 1, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [-0.043429, 0.065747], "effect": 0.011159, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:54:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-7e678130a4a8", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 0.4}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.815789474, 1.0, 1.0, 0.659313725, 0.252156863, 0.689260563, 0.75, 0.605, 0.833333333, 0.936], "compressed": [0.894736842, 1.0, 1.0, 0.74754902, 0.368529412, 0.55193662, 0.75, 0.467142857, 0.87254902, 1.0], "inherit_mem": [0.894736842, 1.0, 1.0, 0.74754902, 0.368529412, 0.55193662, 0.75, 0.467142857, 0.87254902, 1.0], "inherit_nomem": [0.947368421, 1.0, 1.0, 0.91503268, 0.499803922, 0.443661972, 0.75, 0.605, 0.87254902, 1.0], "random_search": [0.815789474, 1.0, 1.0, 0.659313725, 0.252156863, 0.689260563, 0.75, 0.605, 0.833333333, 0.936], "reset_mem": [0.894736842, 1.0, 1.0, 0.74754902, 0.368529412, 0.55193662, 0.75, 0.467142857, 0.87254902, 1.0], "retrieval": [0.815789474, 1.0, 1.0, 0.659313725, 0.252156863, 0.689260563, 0.75, 0.605, 0.833333333, 0.936], "rootcause": [0.894736842, 1.0, 1.0, 0.74754902, 0.368529412, 0.55193662, 0.75, 0.467142857, 0.87254902, 1.0], "xfer_mem": [0.894736842, 1.0, 1.0, 0.74754902, 0.368529412, 0.55193662, 0.75, 0.467142857, 0.87254902, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.6448, "losses": 2, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.227", "per_generation_sealed": [0.754085, 0.765244, 0.765244, 0.765244], "prereg_hash": "c86e81ceec474ec5b7e736304640d4230f9f1198df4c27a84375e08bb8b1a815", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "random_search", "compressed"], "result_hash": "17f0e7f4f95982fe143a5f66f647f142a4def95bc9b6afbcede1e0a5474e290c", "signature": "9b31e8e66cd61c148d6e4ff7366fdd7f78cae5c249650bd00c53c10614305faafc30a975ebb9f82862ec303c6b6f5c023daede7d328ee1c56b018f0bfe056300", "state": "VERIFIED", "summary": {"G0": 0.754085, "compressed": 0.765244, "inherit_mem": 0.765244, "inherit_nomem": 0.803342, "random_search": 0.754085, "reset_mem": 0.765244, "retrieval": 0.754085, "rootcause": 0.765244, "xfer_mem": 0.765244}, "ties": 3, "treatment": "inherit_mem", "wins": 5}, "verified_at": "2026-11-22T03:56:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:41Z", "cpu_seconds": 111.20213529099999, "created_at": "2026-11-22T03:51:39Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-c6a95fb5a7f6", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "5eec500a0558fd199fe773d7553726f82cee071d02648f9bd6f9bf3997e346f4", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-c6a95fb5a7f6", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:51:39Z", "seed": 568872086, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "e7adecffdda08f9637994532276040f7ed089cc8c13173ab3db62e46f7527c58", "run_seconds": 121.18065524101257, "seed": 568872086, "started_at": "2026-11-22T03:51:40Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 2, "inherit_mem": 3, "inherit_nomem": 2, "random_search": 1, "reset_mem": 3, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [-0.011672, 0.168143], "effect": 0.078236, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:53:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-c6a95fb5a7f6", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 6, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 6, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.7}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 6, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.512, "history2": false, "mix": {"random": 1.0}, "planner_depth": 10, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 6, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.319444444, 1.0, 0.983333333, 0.598039216, 0.251764706, 0.62109375, 0.875, 0.576636905, 0.864583333, 0.759], "compressed": [0.805555556, 1.0, 0.983333333, 0.598039216, 0.251764706, 0.62109375, 0.875, 0.576636905, 0.854166667, 0.759], "inherit_mem": [0.777777778, 1.0, 0.983333333, 0.75, 0.29254902, 0.616210938, 0.875, 0.577380952, 0.875, 0.884], "inherit_nomem": [0.805555556, 1.0, 0.983333333, 0.598039216, 0.251764706, 0.62109375, 0.875, 0.576636905, 0.854166667, 0.759], "random_search": [0.805555556, 1.0, 0.983333333, 0.598039216, 0.251764706, 0.62109375, 0.875, 0.576636905, 0.864583333, 0.759], "reset_mem": [0.777777778, 1.0, 0.983333333, 0.610294118, 0.29254902, 0.616210938, 0.875, 0.576636905, 0.875, 0.884], "retrieval": [0.319444444, 1.0, 0.983333333, 0.598039216, 0.251764706, 0.62109375, 0.875, 0.576636905, 0.864583333, 0.759], "rootcause": [0.361111111, 1.0, 0.991666667, 0.674019608, 0.284313725, 0.65625, 0.920454545, 0.614583333, 0.90625, 0.908], "xfer_mem": [0.361111111, 1.0, 0.991666667, 0.674019608, 0.284313725, 0.65625, 0.920454545, 0.614583333, 0.90625, 0.908]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -1.2041, "losses": 1, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "per_generation_sealed": [0.68489, 0.733501, 0.732459, 0.763125], "prereg_hash": "5eec500a0558fd199fe773d7553726f82cee071d02648f9bd6f9bf3997e346f4", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "rootcause", "random_search"], "result_hash": "e7adecffdda08f9637994532276040f7ed089cc8c13173ab3db62e46f7527c58", "signature": "bfcd88d88baad0adcd6f0fa93d55bc4726246110a0d6378b30a8e709e2277f20b321e626663b763299d548bb2e6d9490f47b5eb347a47c6e3445ddd56901ce05", "state": "VERIFIED", "summary": {"G0": 0.68489, "compressed": 0.732459, "inherit_mem": 0.763125, "inherit_nomem": 0.732459, "random_search": 0.733501, "reset_mem": 0.74908, "retrieval": 0.68489, "rootcause": 0.731665, "xfer_mem": 0.731665}, "ties": 3, "treatment": "inherit_mem", "wins": 6}, "verified_at": "2026-11-22T03:54:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:13Z", "cpu_seconds": 0.501832056, "created_at": "2026-11-22T03:51:54Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5e36d7dbf666", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "48f658e0ecebb108e68f269d2ae7bacc2fee3df7ac6429f8d675d656c0cb058d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5e36d7dbf666", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:51:54Z", "seed": 2042742057, "treatment": "frontier"}, "priority": 0.32818120681033036, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "22eecb8fa84348bd7c9ae8e9a9d629637371a8b4d51b868ff254a1f5a487bb25", "run_seconds": 0.5690977573394775, "seed": 2042742057, "started_at": "2026-11-22T03:53:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.002476, 0.074459], "effect": 0.038467, "evaluated_at": "2026-11-22T03:53:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5e36d7dbf666", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.816666667, 0.757142857, 0.416666667], "frontier": [0.979166667, 0.866666667, 0.9, 0.864583333], "goal": [1.0, 0.766666667, 0.8, 0.864583333], "hypdisc": [0.979166667, 0.866666667, 0.814285714, 0.822916667], "infogain": [0.979166667, 0.883333333, 0.742857143, 0.104166667], "prederr": [0.979166667, 0.733333333, 0.242857143, 0.03125], "random": [0.979166667, 0.8, 0.885714286, 0.791666667], "risk": [0.979166667, 0.866666667, 0.9, 0.864583333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "48f658e0ecebb108e68f269d2ae7bacc2fee3df7ac6429f8d675d656c0cb058d", "primary_control": "random", "result_hash": "22eecb8fa84348bd7c9ae8e9a9d629637371a8b4d51b868ff254a1f5a487bb25", "signature": "76318ccf07a5c510673f920a7f57a4cc8610585c1e210b9bf5ea0b1a0b2f80ff9167fec8f8525e513e2cd02fcbda5d957e179cc9da18e4441149ceab656b1903", "state": "VERIFIED", "summary": {"count": 0.747619, "frontier": 0.902604, "goal": 0.857812, "hypdisc": 0.870759, "infogain": 0.677381, "prederr": 0.496652, "random": 0.864137, "risk": 0.902604}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:53:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:06Z", "cpu_seconds": 0.5371135049999998, "created_at": "2026-11-22T03:51:54Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8ff40873e80b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "1b2b94e97d69ced61d95d57920f4e791267069c63b1694c7567b6e71eede6f74", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8ff40873e80b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:51:54Z", "seed": 1598584617, "treatment": "full"}, "priority": 0.3335254297005753, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2b5093a4a206ad06164f0e1c0daa4dd06a96c85d548d86484231c04f894ca06b", "run_seconds": 0.5729823112487793, "seed": 1598584617, "started_at": "2026-11-22T03:52:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.549352, 0.299352], "effect": -0.125, "evaluated_at": "2026-11-22T03:52:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8ff40873e80b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 1.0], "fixed_policy": [0.5, 0.0, 0.0, 0.0], "full": [1.0, 0.75, 0.25, 1.0], "no_world_model": [1.0, 0.5, 0.25, 0.5], "random": [1.0, 0.5, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "1b2b94e97d69ced61d95d57920f4e791267069c63b1694c7567b6e71eede6f74", "primary_control": "e141_best", "result_hash": "2b5093a4a206ad06164f0e1c0daa4dd06a96c85d548d86484231c04f894ca06b", "signature": "e8e1248e7dc712cf3d053d29cb05cdac9420683df538edcfcb3a3461fc6044e99860808cc9019b6322225986f8fc2295a7d001095d62f4e24dcb68ecd142be04", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.125, "full": 0.75, "no_world_model": 0.5625, "random": 0.875}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:52:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:11Z", "cpu_seconds": 0.6039002570000003, "created_at": "2026-11-22T03:51:54Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bf82673df1d1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "65e5f31644bd558a22f4d76a25cb1d79963c2e80d7cce66b74cf9c860e8910f7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bf82673df1d1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:51:54Z", "seed": 622374729, "treatment": "full"}, "priority": 0.3307938915632509, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d47d492c15a8d4f17aafdc61a1b3480ab934473249bbeccb118e0860cd5dc1fe", "run_seconds": 0.6910898685455322, "seed": 622374729, "started_at": "2026-11-22T03:52:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.18, 0.555], "effect": 0.1875, "evaluated_at": "2026-11-22T03:52:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-bf82673df1d1", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 0.0, 0.25], "fixed_policy": [0.0, 0.0, 0.5, 0.25], "full": [0.0, 1.0, 0.0, 1.0], "no_world_model": [0.0, 1.0, 0.0, 1.0], "random": [0.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "65e5f31644bd558a22f4d76a25cb1d79963c2e80d7cce66b74cf9c860e8910f7", "primary_control": "e141_best", "result_hash": "d47d492c15a8d4f17aafdc61a1b3480ab934473249bbeccb118e0860cd5dc1fe", "signature": "5606fc11dd99b91ff859a9f979b5cedaa7f67e0f646e57d7a09969ee1ebbb7c851431f52d8ef6a17cccf529b86daba9f508315cd707bb9488740aeca7e75250a", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.1875, "full": 0.5, "no_world_model": 0.5, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:52:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:02Z", "cpu_seconds": 2.99785286, "created_at": "2026-11-22T03:51:54Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2e90db8cf64a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "b52b11f2f08575ede93b7a6462c4a427b459704a0d9a6f722e520cc4518a5cef", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2e90db8cf64a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:51:54Z", "seed": 1957426088, "treatment": "full"}, "priority": 0.3379175155128382, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4f74974c607cc071f4619b3b41442909815cca3b836a34b1620386bd33629036", "run_seconds": 3.3124897480010986, "seed": 1957426088, "started_at": "2026-11-22T03:51:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.042981, 0.197684], "effect": 0.120333, "evaluated_at": "2026-11-22T03:52:03Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2e90db8cf64a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.260416667, 0.875, 0.328888889, 0.451523546], "frozen": [0.152777778, 0.3125, 0.184444444, 0.27700831], "full": [0.326388889, 0.970833333, 0.411111111, 0.688827331], "retrieval": [0.427083333, 0.995833333, 0.448888889, 0.870267775], "rules_nodetect": [0.501736111, 0.970833333, 0.308888889, 0.688827331]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "b52b11f2f08575ede93b7a6462c4a427b459704a0d9a6f722e520cc4518a5cef", "primary_control": "facts", "result_hash": "4f74974c607cc071f4619b3b41442909815cca3b836a34b1620386bd33629036", "signature": "cce3767951573da4d2f1cf3a93c2e4aa13696d5b104949b3ff4d4f1418114a3c2a90e1ef18fc50161d5be85463e02061c48df9aa34ae1f459ef2eb14b5721a01", "state": "VERIFIED", "summary": {"facts": 0.478957, "frozen": 0.231683, "full": 0.59929, "retrieval": 0.685518, "rules_nodetect": 0.617571}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:52:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:44Z", "cpu_seconds": 1.0407673409999987, "created_at": "2026-11-22T03:51:54Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9b66f062dd56", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "dae66f0f6a900a8f8bfdca376cd82d951b48a7f7ee271048e334c4c4d439a49c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9b66f062dd56", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:51:54Z", "seed": 1729908322, "treatment": "full"}, "priority": 0.32975818515055133, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "aff91d64f67fdefdf278cfb8de7c52ab35fe8ae802b2cb36cd3af2f5f1648019", "run_seconds": 1.3977694511413574, "seed": 1729908322, "started_at": "2026-11-22T03:52:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013398, 0.102684], "effect": 0.044643, "evaluated_at": "2026-11-22T03:52:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9b66f062dd56", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.755952381, 1.0, 1.0, 0.734375], "frozen": [0.363095238, 0.316666667, 0.294871795, 0.296875], "full": [0.80952381, 1.0, 1.0, 0.859375], "retrieval": [0.898809524, 1.0, 1.0, 0.875], "rules_nodetect": [0.80952381, 1.0, 1.0, 0.859375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "dae66f0f6a900a8f8bfdca376cd82d951b48a7f7ee271048e334c4c4d439a49c", "primary_control": "facts", "result_hash": "aff91d64f67fdefdf278cfb8de7c52ab35fe8ae802b2cb36cd3af2f5f1648019", "signature": "74b71f68c35491d4dedc80d368292a8a3f59cb247b3ba4dcc286ea05bba596363e6bc887bc08be283f4026d411b870410448329d27c1f59b5daba9a878868800", "state": "VERIFIED", "summary": {"facts": 0.872582, "frozen": 0.317877, "full": 0.917225, "retrieval": 0.943452, "rules_nodetect": 0.917225}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:52:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:00Z", "cpu_seconds": 2.1323419300000004, "created_at": "2026-11-22T03:51:54Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-82412d2023f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "adec9932096f33da7a9a83e0cafbf926405a00782e3cec2cb5c1c1b57d0707fa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-82412d2023f0", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:51:54Z", "seed": 1724039217, "treatment": "gated"}, "priority": 0.3980813647280962, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f2669f77aa651919c9a7040b94b59b7519617b90df7147df8e6677a3703b8226", "run_seconds": 2.368633985519409, "seed": 1724039217, "started_at": "2026-11-22T03:51:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.049125, 0.003821], "effect": -0.022652, "evaluated_at": "2026-11-22T03:52:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-82412d2023f0", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.555555556, 0.592105263, 0.696428571, 0.541666667], "gated": [0.555555556, 0.592105263, 0.696428571, 0.541666667], "naive": [0.583333333, 0.776315789, 0.696428571, 0.8], "scratch": [0.592592593, 0.592105263, 0.75, 0.541666667], "ungated": [0.601851852, 0.684210526, 0.696428571, 0.466666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "adec9932096f33da7a9a83e0cafbf926405a00782e3cec2cb5c1c1b57d0707fa", "primary_control": "scratch", "result_hash": "f2669f77aa651919c9a7040b94b59b7519617b90df7147df8e6677a3703b8226", "signature": "f78374da0f7ef7dcc71396bff275413a51be6bd928811b38b2a50f9c5cac734ef4062f6eac58cfc151f3db5b9fa27b24789f59439e96a2fc6b8cf25cc44c6806", "state": "VERIFIED", "summary": {"compressed": 0.596439, "gated": 0.596439, "naive": 0.714019, "scratch": 0.619091, "ungated": 0.612289}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:52:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:22Z", "cpu_seconds": 0.9551461190000001, "created_at": "2026-11-22T03:52:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8bffc904f544", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "ca332a39a291bef949613cffc8107e542342766084b4ab38cd02fd98ab38daf0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8bffc904f544", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:52:11Z", "seed": 681017164, "treatment": "frontier"}, "priority": 0.33202399494906343, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f1b62be66ca50b26954050b94c6156d0ad6d4550a5ce81682bca7e543272e7a1", "run_seconds": 1.0137691497802734, "seed": 681017164, "started_at": "2026-11-22T03:52:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.050555, 0.1269], "effect": 0.038172, "evaluated_at": "2026-11-22T03:52:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-8bffc904f544", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.96124031, 0.444444444, 0.974358974, 0.996563574], "frontier": [0.906976744, 0.725490196, 0.953846154, 0.996563574], "goal": [0.829457364, 0.72875817, 0.953846154, 0.996563574], "hypdisc": [0.96124031, 0.715686275, 0.974358974, 0.996563574], "infogain": [0.589147287, 0.72875817, 0.912820513, 0.996563574], "prederr": [0.534883721, 0.718954248, 0.917948718, 0.945017182], "random": [0.906976744, 0.552287582, 0.974358974, 0.996563574], "risk": [0.906976744, 0.725490196, 0.953846154, 0.996563574]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "ca332a39a291bef949613cffc8107e542342766084b4ab38cd02fd98ab38daf0", "primary_control": "random", "result_hash": "f1b62be66ca50b26954050b94c6156d0ad6d4550a5ce81682bca7e543272e7a1", "signature": "79976d9043f8e17cc08c7b966f5438db76acff505080433f4cebad5108d7c2cd3f4803881f655eb9ed27853173babad8eefea1800b60a6d0dd9600644cecd406", "state": "VERIFIED", "summary": {"count": 0.844152, "frontier": 0.895719, "goal": 0.877156, "hypdisc": 0.911962, "infogain": 0.806822, "prederr": 0.779201, "random": 0.857547, "risk": 0.895719}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:52:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:31Z", "cpu_seconds": 4.856650888, "created_at": "2026-11-22T03:52:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a3984f1bffb7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "abc514100974e0f837bab1512a3009b7d4d94729e24f618b0930a1e63de86ed2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a3984f1bffb7", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:52:11Z", "seed": 968524385, "treatment": "frontier"}, "priority": 0.3307452832480987, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "0a42add5db8be4eeaa7d1be7e61d8997c43826796fac2920f4251006ee386e94", "run_seconds": 5.269759178161621, "seed": 968524385, "started_at": "2026-11-22T03:52:25Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:52:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a3984f1bffb7", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.986111111, 1.0, 0.962962963, 1.0], "frontier": [1.0, 1.0, 0.962962963, 1.0], "goal": [1.0, 1.0, 0.962962963, 1.0], "hypdisc": [1.0, 1.0, 0.962962963, 1.0], "infogain": [1.0, 1.0, 0.962962963, 1.0], "prederr": [1.0, 1.0, 0.962962963, 1.0], "random": [1.0, 1.0, 0.962962963, 1.0], "risk": [1.0, 1.0, 0.962962963, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "abc514100974e0f837bab1512a3009b7d4d94729e24f618b0930a1e63de86ed2", "primary_control": "random", "result_hash": "0a42add5db8be4eeaa7d1be7e61d8997c43826796fac2920f4251006ee386e94", "signature": "a40627ec246bdec62c14ed62788b29c92208cb2e2bc970594d750467420da853f0a7fea3b849eca8711b58a7004f35e6e4037ea6883eaf66fd4e261cba53f308", "state": "PARTIAL", "summary": {"count": 0.987269, "frontier": 0.990741, "goal": 0.990741, "hypdisc": 0.990741, "infogain": 0.990741, "prederr": 0.990741, "random": 0.990741, "risk": 0.990741}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:52:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:30Z", "cpu_seconds": 3.2187740270000003, "created_at": "2026-11-22T03:52:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-21868a418e16", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 7, "n": 4}, "prereg_hash": "e73627aee3a600b9398737678071e6d48b1babe1d85183af4c0e35d95cc95905", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-21868a418e16", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:52:11Z", "seed": 336445857, "treatment": "full"}, "priority": 0.3307318367996683, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-ebe660e948d6", "result_hash": "62e7bab0bdad739b7566e31a5a4390f7779c58f3fa600ae4cabade073d7789ec", "run_seconds": 3.587588310241699, "seed": 336445857, "started_at": "2026-11-22T03:52:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.124173, 0.260827], "effect": 0.1925, "evaluated_at": "2026-11-22T03:52:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-21868a418e16", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.384333333, 0.398, 0.419333333, 0.3928], "frozen": [0.32, 0.322, 0.28, 0.336], "full": [0.526333333, 0.540333333, 0.615, 0.6828], "retrieval": [0.774, 0.776333333, 0.811, 0.81], "rules_nodetect": [0.526333333, 0.540333333, 0.615, 0.6828]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "e73627aee3a600b9398737678071e6d48b1babe1d85183af4c0e35d95cc95905", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-ebe660e948d6", "result_hash": "62e7bab0bdad739b7566e31a5a4390f7779c58f3fa600ae4cabade073d7789ec", "signature": "7b8e2070e90ede85597ff21a8056e45a36e56cbad04bf500308f5d31dee6157620fd79b1438b9758b31ce491d1b383a2fce7e97b1d954a4a2b800c54071a6900", "state": "VERIFIED", "summary": {"facts": 0.398617, "frozen": 0.3145, "full": 0.591117, "retrieval": 0.792833, "rules_nodetect": 0.591117}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:52:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:16Z", "cpu_seconds": 2.9075643940000004, "created_at": "2026-11-22T03:52:11Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-617b8f255d68", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "d3f3ca5da99eeb08eb5b98bb9bb186d23d4d984b7edc027ab93222121897d385", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-617b8f255d68", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:52:11Z", "seed": 1301447064, "treatment": "gated"}, "priority": 0.35084588789295135, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6cf34e8709a8fc50ad63e69e53b647651feafc25eee41b7faf406dfa4c260d48", "run_seconds": 3.440129280090332, "seed": 1301447064, "started_at": "2026-11-22T03:52:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018264, 0.001193], "effect": -0.008536, "evaluated_at": "2026-11-22T03:52:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-617b8f255d68", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.556818182, 0.625, 0.696428571, 0.555555556], "gated": [0.556818182, 0.625, 0.696428571, 0.555555556], "naive": [0.727272727, 0.609375, 0.767857143, 0.685185185], "scratch": [0.556818182, 0.640625, 0.696428571, 0.574074074], "ungated": [0.670454545, 0.59375, 0.767857143, 0.407407407]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d3f3ca5da99eeb08eb5b98bb9bb186d23d4d984b7edc027ab93222121897d385", "primary_control": "scratch", "result_hash": "6cf34e8709a8fc50ad63e69e53b647651feafc25eee41b7faf406dfa4c260d48", "signature": "ae858968718e16a6fe01198128cc90954a0593bdce3aa3f6a06b45658f17dfd0ec60231263371032b7766593cc82f4196abf926cf163f302dbdf54d61ee2ae02", "state": "VERIFIED", "summary": {"compressed": 0.608451, "gated": 0.608451, "naive": 0.697423, "scratch": 0.616986, "ungated": 0.609867}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:52:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:20Z", "cpu_seconds": 0.582519126, "created_at": "2026-11-22T03:52:11Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-c8bceb3c5e92", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "bf85b84beecd437ee1ab1067f22513bfc5e7965c33576326c001846a7ddf2801", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-c8bceb3c5e92", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:52:11Z", "seed": 269234924, "treatment": "gated"}, "priority": 0.3358887997435823, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "596cd037e6cbf7f1deece0220189dc8fb3b135763a653bd4ecc369917fd960b1", "run_seconds": 0.6019105911254883, "seed": 269234924, "started_at": "2026-11-22T03:52:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.028235, 0.087059], "effect": 0.029412, "evaluated_at": "2026-11-22T03:52:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-c8bceb3c5e92", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.267973856, 0.235294118, 0.189542484, 0.39869281], "gated": [0.267973856, 0.235294118, 0.189542484, 0.39869281], "naive": [0.248366013, 0.37254902, 0.267973856, 0.241830065], "scratch": [0.267973856, 0.235294118, 0.189542484, 0.281045752], "ungated": [0.281045752, 0.196078431, 0.189542484, 0.346405229]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "bf85b84beecd437ee1ab1067f22513bfc5e7965c33576326c001846a7ddf2801", "primary_control": "scratch", "result_hash": "596cd037e6cbf7f1deece0220189dc8fb3b135763a653bd4ecc369917fd960b1", "signature": "995dd5cd205f7017038244ee339ca56bb55932b277f60d3556ad3f4f9fe5cabfcb180fd8c187e6a6ea5c86b114db317212a4738f8100a166882ec296b47a470b", "state": "VERIFIED", "summary": {"compressed": 0.272876, "gated": 0.272876, "naive": 0.28268, "scratch": 0.243464, "ungated": 0.253268}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:52:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:42Z", "cpu_seconds": 1.357874842, "created_at": "2026-11-22T03:52:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a76294c23e0b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "7ec99c9acc4a40f1f6da34974d31b64c8c606a07c7081978cdfce7c13a6a05f0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a76294c23e0b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:52:28Z", "seed": 681167378, "treatment": "frontier"}, "priority": 0.3102301032931202, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c52f0436eff57a6b89203dc9af9286f041907a24b628a45793cf69553a857361", "run_seconds": 1.55434250831604, "seed": 681167378, "started_at": "2026-11-22T03:54:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.155189, -0.009602], "effect": -0.082396, "evaluated_at": "2026-11-22T03:56:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a76294c23e0b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.916666667, 0.813333333, 0.854166667, 0.766666667], "frontier": [0.85, 0.706666667, 0.854166667, 0.633333333], "goal": [0.85, 0.706666667, 0.854166667, 0.733333333], "hypdisc": [0.916666667, 0.613333333, 0.854166667, 0.733333333], "infogain": [0.85, 0.706666667, 0.854166667, 0.766666667], "prederr": [0.85, 0.706666667, 0.854166667, 0.7], "random": [0.916666667, 0.813333333, 0.84375, 0.8], "risk": [0.85, 0.706666667, 0.854166667, 0.633333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "7ec99c9acc4a40f1f6da34974d31b64c8c606a07c7081978cdfce7c13a6a05f0", "primary_control": "random", "result_hash": "c52f0436eff57a6b89203dc9af9286f041907a24b628a45793cf69553a857361", "signature": "be332e4ae5e5e8cede5e501093a340399a44b192d0f3191bfc0f6cac7fadc17eb8029000a8dd25e0aa82a7de0273dd47b75094e9242fbb20da4945be83c62608", "state": "VERIFIED", "summary": {"count": 0.837708, "frontier": 0.761042, "goal": 0.786042, "hypdisc": 0.779375, "infogain": 0.794375, "prederr": 0.777708, "random": 0.843438, "risk": 0.761042}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:56:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:34Z", "cpu_seconds": 0.5992763239999999, "created_at": "2026-11-22T03:52:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d6391440041d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "cb5066db40fed7faa41c71e1f2880a5e1b0287f78c59a30bd29407f0128f18ec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d6391440041d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:52:28Z", "seed": 196310635, "treatment": "frontier"}, "priority": 0.31415074267974047, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "79012a99b48b9b32a09a2284032a97f27894d4251942b2bd59fd85927f20f4ab", "run_seconds": 0.7862548828125, "seed": 196310635, "started_at": "2026-11-22T03:54:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.006167, 0.050083], "effect": 0.028125, "evaluated_at": "2026-11-22T03:56:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d6391440041d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.895833333, 1.0, 1.0, 0.925], "frontier": [0.875, 0.979166667, 1.0, 0.85], "goal": [0.5625, 0.958333333, 1.0, 0.825], "hypdisc": [0.895833333, 0.958333333, 1.0, 0.875], "infogain": [0.854166667, 0.979166667, 1.0, 0.925], "prederr": [0.645833333, 0.979166667, 1.0, 0.925], "random": [0.854166667, 0.979166667, 0.958333333, 0.8], "risk": [0.875, 0.979166667, 1.0, 0.85]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "cb5066db40fed7faa41c71e1f2880a5e1b0287f78c59a30bd29407f0128f18ec", "primary_control": "random", "result_hash": "79012a99b48b9b32a09a2284032a97f27894d4251942b2bd59fd85927f20f4ab", "signature": "46daf345e3aa00c322e6b53e2ee13ee7a3c9351490573aa805a0523ac1450ec0c96bc9e9c4795c28ae9784c33809e7842fe19a0ca3081c45c79751122afafc05", "state": "VERIFIED", "summary": {"count": 0.955208, "frontier": 0.926042, "goal": 0.836458, "hypdisc": 0.932292, "infogain": 0.939583, "prederr": 0.8875, "random": 0.897917, "risk": 0.926042}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:56:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:43Z", "cpu_seconds": 0.9012035370000007, "created_at": "2026-11-22T03:52:28Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-10dbca51a22a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "d75db71878df12a1cbcfad90a104e441963cba21af411db5ea718103fc1b7e6f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-10dbca51a22a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:52:28Z", "seed": 1207157336, "treatment": "full"}, "priority": 0.3319820419151001, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e4bf6994a3ef084d8e4a5476ac93cafb09e935d2f981b8b8e0eae7f2f631bd39", "run_seconds": 1.262216329574585, "seed": 1207157336, "started_at": "2026-11-22T03:52:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T03:52:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-10dbca51a22a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.25, 0.25, 0.5], "full": [1.0, 1.0, 0.5, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [1.0, 0.75, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d75db71878df12a1cbcfad90a104e441963cba21af411db5ea718103fc1b7e6f", "primary_control": "e141_best", "result_hash": "e4bf6994a3ef084d8e4a5476ac93cafb09e935d2f981b8b8e0eae7f2f631bd39", "signature": "7d1b2fade403b3cc3aa2b5b557dfac7630be328de1c3afa3c372e09a244bb9e9c1dd3f34e9fd6980c4a851201d770a19e6fa0fa9d2cf86dabfd3b5c3bda19108", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 0.875, "no_world_model": 0.9375, "random": 0.8125}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:52:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:51Z", "cpu_seconds": 0.326697835, "created_at": "2026-11-22T03:52:28Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c32ba0269513", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "0ba0105ca4b01ec394439e3a2454f62b1a62a4e4f2e697c413c4b2d5f493068a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c32ba0269513", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:52:28Z", "seed": 1214216104, "treatment": "full"}, "priority": 0.31703499891407244, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e0e5febb201c617cf54e4d59aa053bc7152843f135e191cfaf93b82844645f3d", "run_seconds": 0.9219624996185303, "seed": 1214216104, "started_at": "2026-11-22T03:53:50Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:54:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c32ba0269513", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0ba0105ca4b01ec394439e3a2454f62b1a62a4e4f2e697c413c4b2d5f493068a", "primary_control": "e141_best", "result_hash": "e0e5febb201c617cf54e4d59aa053bc7152843f135e191cfaf93b82844645f3d", "signature": "892dc5a195d558e6b4f979c9c466158bcecc28a278fa564ada1977ce7887534809156dde14c9b116b3dd3537bb04a68fef949e4323e02d0709b550595cbed402", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:54:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:52:38Z", "cpu_seconds": 2.2383212310000005, "created_at": "2026-11-22T03:52:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a6a0e476f1c4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "bbeda854ea24d7cc5a58283e62e297edf313c8e9d263ba25ae94642c688c2227", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a6a0e476f1c4", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:52:28Z", "seed": 27322184, "treatment": "gated"}, "priority": 0.3667551952055633, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "a17a6c11ea3f33b7db8bdce01ba76b9f61b38596567b612f8c08e094286ddd23", "run_seconds": 2.665787935256958, "seed": 27322184, "started_at": "2026-11-22T03:52:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.061667, 0.02], "effect": -0.020833, "evaluated_at": "2026-11-22T03:52:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a6a0e476f1c4", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.47, 0.452380952, 0.5625, 0.527777778], "gated": [0.47, 0.452380952, 0.5625, 0.527777778], "naive": [0.62, 0.642857143, 0.75, 0.680555556], "scratch": [0.47, 0.452380952, 0.5625, 0.611111111], "ungated": [0.62, 0.642857143, 0.669642857, 0.652777778]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "bbeda854ea24d7cc5a58283e62e297edf313c8e9d263ba25ae94642c688c2227", "primary_control": "scratch", "result_hash": "a17a6c11ea3f33b7db8bdce01ba76b9f61b38596567b612f8c08e094286ddd23", "signature": "01912e9e9edcff551e7f6ae56c52ae924b53a20a8ccfdaf37d56a344e2d9ae28e15e0f9166a02b738f3feecc03a200d241c37b58160ebe332b806b7cac2a5c0c", "state": "VERIFIED", "summary": {"compressed": 0.503165, "gated": 0.503165, "naive": 0.673353, "scratch": 0.523998, "ungated": 0.646319}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:52:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:05Z", "cpu_seconds": 1.013919387, "created_at": "2026-11-22T03:52:45Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f3816029dc02", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "bc4714077a95a5904d6674094a385563056fe6d092f3c7c6999e9920a5a3c879", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f3816029dc02", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:52:45Z", "seed": 1855906516, "treatment": "frontier"}, "priority": 0.32954174189813, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "19fc95d259b1b0dc602953c59b8016b9b5719a0d16504bad7b9ab8ee7d76717b", "run_seconds": 1.2445425987243652, "seed": 1855906516, "started_at": "2026-11-22T03:53:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.235177, 0.189133], "effect": -0.023022, "evaluated_at": "2026-11-22T03:53:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f3816029dc02", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.646666667, 0.564102564, 0.810457516, 0.866666667], "frontier": [0.673333333, 0.794871795, 0.516339869, 0.733333333], "goal": [0.38, 0.794871795, 0.653594771, 0.733333333], "hypdisc": [0.48, 0.641025641, 0.68627451, 0.866666667], "infogain": [0.506666667, 0.692307692, 0.614379085, 0.733333333], "prederr": [0.433333333, 0.641025641, 0.594771242, 0.666666667], "random": [0.646666667, 0.538461538, 0.758169935, 0.866666667], "risk": [0.673333333, 0.794871795, 0.516339869, 0.733333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "bc4714077a95a5904d6674094a385563056fe6d092f3c7c6999e9920a5a3c879", "primary_control": "random", "result_hash": "19fc95d259b1b0dc602953c59b8016b9b5719a0d16504bad7b9ab8ee7d76717b", "signature": "24684c9a998654f41b15cfe252f461dedd6d44ec43866ee4aefd0453cb3029da28746fbfe980e983e479dc81dee96bfa3795c3030c50c02d25e191401898cb00", "state": "VERIFIED", "summary": {"count": 0.721973, "frontier": 0.67947, "goal": 0.64045, "hypdisc": 0.668492, "infogain": 0.636672, "prederr": 0.583949, "random": 0.702491, "risk": 0.67947}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T03:53:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:43Z", "cpu_seconds": 1.189529119, "created_at": "2026-11-22T03:52:45Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c859991e8548", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "3dc9acb5988433cd69b00d6cdd64aa9337f86fb0806d5f85f760c65ce99bedcd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c859991e8548", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:52:45Z", "seed": 1398450357, "treatment": "full"}, "priority": 0.3221498153788749, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7054ca935dca2d04ca46c4c276d905d904338e21f61b15ccb404ef08a1e6bd4e", "run_seconds": 1.5950074195861816, "seed": 1398450357, "started_at": "2026-11-22T03:53:42Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:54:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c859991e8548", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.444827], "frozen": [1.0, 0.205924], "full": [1.0, 0.435813], "retrieval": [1.0, 0.446188], "rules_nodetect": [1.0, 0.422066]}, "prereg_hash": "3dc9acb5988433cd69b00d6cdd64aa9337f86fb0806d5f85f760c65ce99bedcd", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "7054ca935dca2d04ca46c4c276d905d904338e21f61b15ccb404ef08a1e6bd4e", "signature": "db54c7f8ddc58ae0a8646ac149320e603e8dcfa3cd71c245414149f19e17790925c44cd32bf21b31f6b67a09e677c6d7eeaeb1ffccc97943dedf0c4b0f117702", "state": "REJECTED", "summary": {"facts": 0.444827, "frozen": 0.205924, "full": 0.435813, "retrieval": 0.446188, "rules_nodetect": 0.422066}}, "verified_at": "2026-11-22T03:54:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:09Z", "cpu_seconds": 177.649699779, "created_at": "2026-11-22T03:52:45Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-f9293fa8edd0", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "14fce088feb62f8b8a94e0fe25e65f4952ccd884f12307794c922d1faf438385", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-f9293fa8edd0", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:52:45Z", "seed": 1853578035, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "97ad4d784c0735812afae9385ad3b96b321ccc0c37b97a093f82cf90f18c3d5b", "run_seconds": 198.63545393943787, "seed": 1853578035, "started_at": "2026-11-22T03:53:50Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 3, "inherit_mem": 2, "inherit_nomem": 2, "random_search": 1, "reset_mem": 2, "retrieval": 0, "rootcause": 3, "xfer_mem": 3}, "better": "higher", "ci95": [-0.070526, 0.117206], "effect": 0.02334, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:58:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-f9293fa8edd0", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.1, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.1, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "random_search": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 6, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 4, "reset_patience": 0, "rule_threshold": 0.732}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0, "risk": 0.6}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.7}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.61637931, 0.486111111, 0.958333333, 0.661821705, 0.254509804, 0.546195652, 0.897727273, 0.686111111, 0.48015873, 0.617424242], "compressed": [0.974137931, 1.0, 0.989583333, 0.486434109, 0.280392157, 0.548233696, 0.914772727, 0.596527778, 0.48015873, 0.792929293], "inherit_mem": [0.974137931, 0.375, 0.989583333, 0.486434109, 0.280392157, 0.548233696, 0.914772727, 0.596527778, 0.48015873, 0.792929293], "inherit_nomem": [0.952586207, 0.236111111, 0.979166667, 0.626937984, 0.270196078, 0.658967391, 0.943181818, 0.618055556, 0.48015873, 0.787878788], "random_search": [0.362068966, 1.0, 0.96875, 0.80620155, 0.263137255, 0.543817935, 0.914772727, 0.686111111, 0.452380952, 0.742424242], "reset_mem": [0.965517241, 1.0, 0.979166667, 0.626937984, 0.280392157, 0.667798913, 0.914772727, 0.484027778, 0.48015873, 0.761616162], "retrieval": [0.61637931, 0.486111111, 0.958333333, 0.661821705, 0.254509804, 0.546195652, 0.897727273, 0.686111111, 0.48015873, 0.617424242], "rootcause": [0.956896552, 0.993055556, 0.96875, 0.615310078, 0.280392157, 0.42017663, 0.914772727, 0.477777778, 0.48015873, 0.761616162], "xfer_mem": [0.961206897, 0.75, 0.96875, 0.48255814, 0.270588235, 0.671535326, 0.9375, 0.740277778, 0.48015873, 0.826262626]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.5953, "losses": 3, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.254", "per_generation_sealed": [0.620477, 0.691733, 0.643817, 0.643817], "prereg_hash": "14fce088feb62f8b8a94e0fe25e65f4952ccd884f12307794c922d1faf438385", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "rootcause", "reset_mem"], "result_hash": "97ad4d784c0735812afae9385ad3b96b321ccc0c37b97a093f82cf90f18c3d5b", "signature": "d8b0482f7470bcf5fb1b601374b4c48450b3f1829e552deea984bb7cdf7904df9628c4b981c4c6f59f2fea635e82c8479c07664114f7d306fe79dfb593b56100", "state": "VERIFIED", "summary": {"G0": 0.620477, "compressed": 0.706317, "inherit_mem": 0.643817, "inherit_nomem": 0.655324, "random_search": 0.673966, "reset_mem": 0.716039, "retrieval": 0.620477, "rootcause": 0.686891, "xfer_mem": 0.708884}, "ties": 1, "treatment": "inherit_mem", "wins": 6}, "verified_at": "2026-11-22T04:00:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:07Z", "cpu_seconds": 2.957212055, "created_at": "2026-11-22T03:52:45Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-341016fe9381", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "350a0319edd61bb49bb5090c9c62de5636a126c522e4f8b715b756f15a57a6f3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-341016fe9381", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:52:45Z", "seed": 272910487, "treatment": "gated"}, "priority": 0.3472461879517455, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "1e44c966d8c750d4c72690fd2cb38254752f4f32c2d5355b9879c3186eea747f", "run_seconds": 3.6251065731048584, "seed": 272910487, "started_at": "2026-11-22T03:53:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.027724, 0.005403], "effect": -0.011161, "evaluated_at": "2026-11-22T03:53:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-341016fe9381", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.677419355, 0.57, 0.857142857, 0.526785714], "gated": [0.677419355, 0.57, 0.857142857, 0.526785714], "naive": [0.806451613, 0.57, 0.857142857, 0.6875], "scratch": [0.677419355, 0.57, 0.892857143, 0.535714286], "ungated": [0.459677419, 0.57, 0.892857143, 0.6875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "350a0319edd61bb49bb5090c9c62de5636a126c522e4f8b715b756f15a57a6f3", "primary_control": "scratch", "result_hash": "1e44c966d8c750d4c72690fd2cb38254752f4f32c2d5355b9879c3186eea747f", "signature": "b3000bc8553b387ca1ac584398aff88f3ef398fd2b59c60094ad7a91151fb9f78e494a946edc0deef842865fd4d40767f27565d2aa771cee01ef9e250f35f70d", "state": "VERIFIED", "summary": {"compressed": 0.657837, "gated": 0.657837, "naive": 0.730274, "scratch": 0.668998, "ungated": 0.652509}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:53:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:44Z", "cpu_seconds": 2.170677432, "created_at": "2026-11-22T03:53:18Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-bd55def9c659", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "ec0ab3d00d0ff8e28c83bbf54f4ce484c00efa9c9b54840b9bfb430068ffd687", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bd55def9c659", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:53:18Z", "seed": 670878709, "treatment": "full"}, "priority": 0.323763790862418, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b43452adddc9460effcbf9dc0ed2d27677cc4ea34e0f071dc044d2b1b6de974a", "run_seconds": 2.525912284851074, "seed": 670878709, "started_at": "2026-11-22T03:53:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01458, 0.177511], "effect": 0.081465, "evaluated_at": "2026-11-22T03:54:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-bd55def9c659", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.340677966, 1.0, 0.502777778, 0.5], "frozen": [0.118644068, 0.36, 0.0, 0.112244898], "full": [0.537288136, 1.0, 0.502777778, 0.629251701], "retrieval": [0.66779661, 1.0, 0.825, 0.736394558], "rules_nodetect": [0.337288136, 0.344, 0.538888889, 0.629251701]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "ec0ab3d00d0ff8e28c83bbf54f4ce484c00efa9c9b54840b9bfb430068ffd687", "primary_control": "facts", "result_hash": "b43452adddc9460effcbf9dc0ed2d27677cc4ea34e0f071dc044d2b1b6de974a", "signature": "23f83ba2a13a2a7ebe3433fb8f8af87bfd341ce87cf5c4c23745bf3abd520f8e0aa9189778bfcacc3cc5eb75d5ed991dea92978ad5bd953ec65058983e98ac08", "state": "VERIFIED", "summary": {"facts": 0.585864, "frozen": 0.147722, "full": 0.667329, "retrieval": 0.807298, "rules_nodetect": 0.462357}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:54:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:36Z", "cpu_seconds": 1.006872991, "created_at": "2026-11-22T03:53:18Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-dfbb96835167", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "0789a09b51c7b234971b17fef74a89780980cbd305559ac0407be3f483b13c12", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-dfbb96835167", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:53:18Z", "seed": 1258696272, "treatment": "full"}, "priority": 0.32835653362082207, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5ef731eaf861d532577a3cafc6859af2589c5ef25226c371490d21ae19d39008", "run_seconds": 1.0654747486114502, "seed": 1258696272, "started_at": "2026-11-22T03:53:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.075856, -0.030511], "effect": -0.053184, "evaluated_at": "2026-11-22T03:53:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-dfbb96835167", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.416666667, 0.454692557, 0.346376812, 0.504166667], "frozen": [0.166666667, 0.181229773, 0.179710145, 0.180555556], "full": [0.355654762, 0.427184466, 0.265217391, 0.461111111], "retrieval": [0.421130952, 0.438511327, 0.297101449, 0.520833333], "rules_nodetect": [0.355654762, 0.697411003, 0.265217391, 0.461111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "0789a09b51c7b234971b17fef74a89780980cbd305559ac0407be3f483b13c12", "primary_control": "facts", "result_hash": "5ef731eaf861d532577a3cafc6859af2589c5ef25226c371490d21ae19d39008", "signature": "e19df678d3fe8b8a8f46cd6279021ad940a192cbda05d8a3c6e19d25d4ebedf7d5fde75376528d4ed8563f3c6397e82d659a1bc75e80643601a1e4c903f1ff01", "state": "VERIFIED", "summary": {"facts": 0.430476, "frozen": 0.177041, "full": 0.377292, "retrieval": 0.419394, "rules_nodetect": 0.444849}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:53:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:53:37Z", "cpu_seconds": 1.3929916070000001, "created_at": "2026-11-22T03:53:18Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-b532e7bc5e1f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "06f01b19c12a77e7234c98caa9f8475dfe4b8e97f146b2fc208a27e506758da8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-b532e7bc5e1f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:53:18Z", "seed": 983295597, "treatment": "gated"}, "priority": 0.3318295855343543, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "159c631e7d02770abb0a7784026c72578753b85e3dabf38bf0b72191d0376cfe", "run_seconds": 1.5023550987243652, "seed": 983295597, "started_at": "2026-11-22T03:53:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.053163, 0.041854], "effect": -0.005655, "evaluated_at": "2026-11-22T03:53:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-b532e7bc5e1f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.857142857, 0.564516129, 0.691666667, 0.55], "gated": [0.857142857, 0.564516129, 0.691666667, 0.55], "naive": [0.857142857, 0.790322581, 0.791666667, 0.7], "scratch": [0.821428571, 0.564516129, 0.675, 0.625], "ungated": [0.857142857, 0.669354839, 0.691666667, 0.7]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "06f01b19c12a77e7234c98caa9f8475dfe4b8e97f146b2fc208a27e506758da8", "primary_control": "scratch", "result_hash": "159c631e7d02770abb0a7784026c72578753b85e3dabf38bf0b72191d0376cfe", "signature": "0b709297a7369ff2dbbc36204c8541f41407a40b4417b1a308890e5f4d61247b79e4aa86c3aa168b5af2c1048784a66309931a70a3801b0078ac895686764c03", "state": "VERIFIED", "summary": {"compressed": 0.665831, "gated": 0.665831, "naive": 0.784783, "scratch": 0.671486, "ungated": 0.729541}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:53:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:46Z", "cpu_seconds": 5.021406862999999, "created_at": "2026-11-22T03:53:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1f0ef9a35e62", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "4767162c730c5b086975488e99578c9d7d76a53eb5fde4397e4519f89786c33b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1f0ef9a35e62", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:53:52Z", "seed": 2098588587, "treatment": "full"}, "priority": 0.3058855470572433, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "33716d0918eb13c90f9b8a8f40e040d761b24762cc46a5f9f292457ba76879c4", "run_seconds": 5.296701908111572, "seed": 2098588587, "started_at": "2026-11-22T03:54:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.04707, 0.42207], "effect": 0.1875, "evaluated_at": "2026-11-22T03:56:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1f0ef9a35e62", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.75, 0.5, 0.75], "fixed_policy": [0.0, 0.75, 0.5, 0.5], "full": [0.0, 1.0, 1.0, 0.75], "no_world_model": [0.0, 0.75, 0.5, 0.75], "random": [0.0, 1.0, 0.25, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "4767162c730c5b086975488e99578c9d7d76a53eb5fde4397e4519f89786c33b", "primary_control": "e141_best", "result_hash": "33716d0918eb13c90f9b8a8f40e040d761b24762cc46a5f9f292457ba76879c4", "signature": "c1c8afe4bc4904ef26356c56aec5432f72e0bec1e22322d2fd7661a5debfac3c3c62e36f034ab1f132a7a498c9614eb4b57c1e4f6abecf75e5f922f4b2b1cb0c", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.4375, "full": 0.6875, "no_world_model": 0.5, "random": 0.5625}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:56:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:35Z", "cpu_seconds": 1.2777753029999999, "created_at": "2026-11-22T03:53:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5e1d77c5e69e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "9574e4ce83048a63b36e23d65c8a5022480a777d5b84f4c855a38857d52054a4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5e1d77c5e69e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:53:52Z", "seed": 468169660, "treatment": "full"}, "priority": 0.31368802738695317, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b807a3993b6313550abf137a0832a2f773e8b1207f0c20f26cda40fa831e8a2b", "run_seconds": 1.5586633682250977, "seed": 468169660, "started_at": "2026-11-22T03:54:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.43, 0.305], "effect": -0.0625, "evaluated_at": "2026-11-22T03:56:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5e1d77c5e69e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 0.5], "fixed_policy": [1.0, 1.0, 0.25, 1.0], "full": [1.0, 0.75, 0.5, 0.75], "no_world_model": [0.5, 0.75, 0.75, 0.25], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "9574e4ce83048a63b36e23d65c8a5022480a777d5b84f4c855a38857d52054a4", "primary_control": "e141_best", "result_hash": "b807a3993b6313550abf137a0832a2f773e8b1207f0c20f26cda40fa831e8a2b", "signature": "96c969c32d8ca670952189867a1ad9ba265683e648d31a654dbe9b0debdfd1ac9bfc8f09b09327c8975b4e3de881e81cf0cb12c5d2fe9aa050053c125df08c00", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.8125, "full": 0.75, "no_world_model": 0.5625, "random": 1.0}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:56:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:02Z", "cpu_seconds": 3.502205935000001, "created_at": "2026-11-22T03:53:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b2377cc9d223", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "db69f0a0b27f4c90f814d9ed5b4cc1d4e617a697db230d868917cb5a7b953add", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b2377cc9d223", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:53:52Z", "seed": 1169558186, "treatment": "full"}, "priority": 0.2991598768127463, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "3362bfbfe70e7b6d5a6fcb41305102e95d5e15b266db9154f2d4c2e999aa8cfa", "run_seconds": 4.834834575653076, "seed": 1169558186, "started_at": "2026-11-22T03:54:57Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:56:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b2377cc9d223", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.0625], "fixed_policy": [1.0, 0.125], "full": [1.0, 0.125], "no_world_model": [1.0, 0.125], "random": [1.0, 0.25]}, "prereg_hash": "db69f0a0b27f4c90f814d9ed5b4cc1d4e617a697db230d868917cb5a7b953add", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "3362bfbfe70e7b6d5a6fcb41305102e95d5e15b266db9154f2d4c2e999aa8cfa", "signature": "c113acf877e309c65b8b1b562016b044d7050fb55267143f5714e3417395494eeaded89f9e28dd696cc9f498f82063f43fbfea8b4cd22c4d1e9a65c666bb490c", "state": "REJECTED", "summary": {"e141_best": 0.0625, "fixed_policy": 0.125, "full": 0.125, "no_world_model": 0.125, "random": 0.25}}, "verified_at": "2026-11-22T03:56:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:40Z", "cpu_seconds": 0.7154758110000001, "created_at": "2026-11-22T03:53:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-df732dd9e95a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "0d921e52c3f13923e3f87735c1cdad9b0da9ee4c59f3cb3cadbd5b4568394a89", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-df732dd9e95a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:53:52Z", "seed": 1738155857, "treatment": "full"}, "priority": 0.3106663296171684, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a4d4f7914123c652d349cb093cb4c64d210bbe923c448f79e98d98ec3f1692a0", "run_seconds": 0.7288460731506348, "seed": 1738155857, "started_at": "2026-11-22T03:54:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.049345, 0.043989], "effect": -0.002678, "evaluated_at": "2026-11-22T03:56:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-df732dd9e95a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.331521739, 0.45035461, 0.313186813, 0.303191489], "frozen": [0.119565217, 0.175531915, 0.210622711, 0.106382979], "full": [0.309782609, 0.413120567, 0.380952381, 0.283687943], "retrieval": [0.34057971, 0.445035461, 0.336996337, 0.29787234], "rules_nodetect": [0.547101449, 0.741134752, 0.807692308, 0.671985816]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "0d921e52c3f13923e3f87735c1cdad9b0da9ee4c59f3cb3cadbd5b4568394a89", "primary_control": "facts", "result_hash": "a4d4f7914123c652d349cb093cb4c64d210bbe923c448f79e98d98ec3f1692a0", "signature": "ba591137c6cca433c20991a5d4605d9bb261f079c56e0a8aa333f25d5f497c531ded4e702fca1284886b9e1a51c639f087c314ad84e22e790c5db6083465260b", "state": "VERIFIED", "summary": {"facts": 0.349564, "frozen": 0.153026, "full": 0.346886, "retrieval": 0.355121, "rules_nodetect": 0.691979}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:56:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:52Z", "cpu_seconds": 0.6424118729999999, "created_at": "2026-11-22T03:53:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-09f085e9aa1c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "6b9ae1078e3f513f19966e7fa9c8ba658962afa8d047d865a0740320f1fe36b7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-09f085e9aa1c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:53:52Z", "seed": 1632355711, "treatment": "gated"}, "priority": 0.3008781838682819, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9b3e9ac7208ae65d0f522b027bc07b6a24c924cda90f7202a469d33deb5e45d4", "run_seconds": 0.6541967391967773, "seed": 1632355711, "started_at": "2026-11-22T03:54:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01451, 0.004706], "effect": -0.004902, "evaluated_at": "2026-11-22T03:56:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-09f085e9aa1c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.31372549, 0.300653595, 0.274509804, 0.307189542], "gated": [0.31372549, 0.300653595, 0.274509804, 0.307189542], "naive": [0.31372549, 0.281045752, 0.274509804, 0.307189542], "scratch": [0.31372549, 0.300653595, 0.274509804, 0.326797386], "ungated": [0.31372549, 0.300653595, 0.209150327, 0.307189542]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "6b9ae1078e3f513f19966e7fa9c8ba658962afa8d047d865a0740320f1fe36b7", "primary_control": "scratch", "result_hash": "9b3e9ac7208ae65d0f522b027bc07b6a24c924cda90f7202a469d33deb5e45d4", "signature": "f95c359c5c23661329ad4c5a2950d09b2c41e7e17ddd5350f3cb6291150e2e1c77cb66e53527377c053d6e25d53cce48e408d0d32b9240f9941a7e73af47f80d", "state": "VERIFIED", "summary": {"compressed": 0.29902, "gated": 0.29902, "naive": 0.294118, "scratch": 0.303922, "ungated": 0.28268}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:56:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:36Z", "cpu_seconds": 1.404892911, "created_at": "2026-11-22T03:53:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-58969546a769", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "2355ac799c4ce894ad870010efbb712c7a5e9b662f8e3b0f9c87789700004d42", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-58969546a769", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:53:52Z", "seed": 753354996, "treatment": "gated"}, "priority": 0.3532098977868647, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "84eac57f2390938aeb60237467ad9a989b6fe68617cae15660e0188c5fc46606", "run_seconds": 1.9586317539215088, "seed": 753354996, "started_at": "2026-11-22T03:54:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.05518, 0.032392], "effect": -0.011394, "evaluated_at": "2026-11-22T03:56:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-58969546a769", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.646551724, 0.579545455, 0.663461538, 0.461538462], "gated": [0.646551724, 0.579545455, 0.663461538, 0.461538462], "naive": [0.74137931, 0.625, 0.711538462, 0.75], "scratch": [0.637931034, 0.556818182, 0.663461538, 0.538461538], "ungated": [0.646551724, 0.568181818, 0.663461538, 0.490384615]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "2355ac799c4ce894ad870010efbb712c7a5e9b662f8e3b0f9c87789700004d42", "primary_control": "scratch", "result_hash": "84eac57f2390938aeb60237467ad9a989b6fe68617cae15660e0188c5fc46606", "signature": "b521cae7dcb5341144943075931b9de999765437d71a80fee1b605df4e00efc5a886181c3da15b76634a981805714c3ba1f1179da934aab55ee733d8c8716000", "state": "VERIFIED", "summary": {"compressed": 0.587774, "gated": 0.587774, "naive": 0.706979, "scratch": 0.599168, "ungated": 0.592145}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T03:56:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:53Z", "cpu_seconds": 0.8060284009999998, "created_at": "2026-11-22T03:53:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ec40dc61ba82", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "prereg_hash": "a0b11b44b681543965be916a062722d9d44448904344616babdbbdf5bc5d65bf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ec40dc61ba82", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:53:52Z", "seed": 1467364803, "treatment": "gated"}, "priority": 0.2995835972045561, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-6bc75240bf73", "result_hash": "95ce380a0baa2e91bf8e15a140ea29d0bb0ae5b453dd2fd819b9064ef74867d9", "run_seconds": 0.8245131969451904, "seed": 1467364803, "started_at": "2026-11-22T03:54:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.021363, 0.074589], "effect": 0.026613, "evaluated_at": "2026-11-22T03:56:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-ec40dc61ba82", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.153191489, 0.195652174, 0.089361702, 0.189010989], "gated": [0.153191489, 0.195652174, 0.089361702, 0.189010989], "naive": [0.168085106, 0.115217391, 0.117021277, 0.10989011], "scratch": [0.153191489, 0.095652174, 0.085106383, 0.186813187], "ungated": [0.180851064, 0.023913043, 0.117021277, 0.173626374]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "a0b11b44b681543965be916a062722d9d44448904344616babdbbdf5bc5d65bf", "primary_control": "scratch", "replicated": true, "replication_of": "E141-ow_transfer-6bc75240bf73", "result_hash": "95ce380a0baa2e91bf8e15a140ea29d0bb0ae5b453dd2fd819b9064ef74867d9", "signature": "4c6c6142b01a48a4f7176c03308839ed35e8306e16b72b4a328f6a186a5e769b9b088d23f4301f5494a52ccbccf148bbe1d7bc07a27e7148e053b2ca8ddbaf0e", "state": "VERIFIED", "summary": {"compressed": 0.156804, "gated": 0.156804, "naive": 0.127553, "scratch": 0.130191, "ungated": 0.123853}, "ties": 1, "treatment": "gated", "wins": 3}, "verified_at": "2026-11-22T03:56:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:19Z", "cpu_seconds": 2.557009815999999, "created_at": "2026-11-22T03:54:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4857421880ba", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "fe0a7fbff464473f75ad26b47deff307d9928a1a3dcbb5cf2a38db5205a9d7c2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4857421880ba", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:54:41Z", "seed": 408808493, "treatment": "frontier"}, "priority": 0.28965563654068993, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "882ba608424ee20d917e81c1f1b29797fadf48e09d9c4bef26048546e4baa735", "run_seconds": 2.999072790145874, "seed": 408808493, "started_at": "2026-11-22T03:55:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01419, 0.011065], "effect": -0.001562, "evaluated_at": "2026-11-22T03:56:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4857421880ba", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.91875, 0.925, 0.89375, 0.925], "frontier": [0.90625, 0.875, 0.9, 0.925], "goal": [0.83125, 0.8, 0.925, 0.925], "hypdisc": [0.875, 0.85, 0.8875, 0.925], "infogain": [0.90625, 0.85, 0.925, 0.925], "prederr": [0.90625, 0.9, 0.9125, 0.925], "random": [0.925, 0.875, 0.8875, 0.925], "risk": [0.90625, 0.875, 0.9, 0.925]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "fe0a7fbff464473f75ad26b47deff307d9928a1a3dcbb5cf2a38db5205a9d7c2", "primary_control": "random", "result_hash": "882ba608424ee20d917e81c1f1b29797fadf48e09d9c4bef26048546e4baa735", "signature": "024487945b4f6cc924b2c6f381b920e80e005fb792cef520fcabb2d138e09fe7a9241b8c020e137724c449bd6f2c3f68df0c56649c4fb575f3bb0f4ebf843408", "state": "VERIFIED", "summary": {"count": 0.915625, "frontier": 0.901563, "goal": 0.870313, "hypdisc": 0.884375, "infogain": 0.901563, "prederr": 0.910937, "random": 0.903125, "risk": 0.901563}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:56:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:59Z", "cpu_seconds": 1.1752276940000006, "created_at": "2026-11-22T03:54:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b4d8d0cb8ae9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "cbfbd13ecc8c5b9134400a99f172146ce1b4ab94aa0123bed7f8f22216e06a98", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b4d8d0cb8ae9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:54:41Z", "seed": 1833496465, "treatment": "frontier"}, "priority": 0.2939135449988107, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "efd69bf60cead8739958fe612455aa0440d9425c5bf82e78ec340a30e8689e48", "run_seconds": 1.617274522781372, "seed": 1833496465, "started_at": "2026-11-22T03:54:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.30852, 0.035603], "effect": -0.136459, "evaluated_at": "2026-11-22T03:56:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b4d8d0cb8ae9", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.575, 0.830645161, 0.75, 0.881578947], "frontier": [0.39375, 0.733870968, 0.85, 0.407894737], "goal": [0.4125, 0.540322581, 0.86, 0.578947368], "hypdisc": [0.240625, 0.741935484, 0.85, 0.578947368], "infogain": [0.3625, 0.733870968, 0.85, 0.657894737], "prederr": [0.65, 0.733870968, 0.85, 0.578947368], "random": [0.725, 0.701612903, 0.86, 0.644736842], "risk": [0.39375, 0.733870968, 0.85, 0.407894737]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "cbfbd13ecc8c5b9134400a99f172146ce1b4ab94aa0123bed7f8f22216e06a98", "primary_control": "random", "result_hash": "efd69bf60cead8739958fe612455aa0440d9425c5bf82e78ec340a30e8689e48", "signature": "84a1e44862699f250a77004f5e6fedd50130dc4d0cc5e2ad6141afadf19af799dc36c50303e5f1297b1d7ac35eb70517117ce29b8079aadcb43e72e1f1e10a0c", "state": "VERIFIED", "summary": {"count": 0.759306, "frontier": 0.596379, "goal": 0.597942, "hypdisc": 0.602877, "infogain": 0.651066, "prederr": 0.703205, "random": 0.732837, "risk": 0.596379}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:56:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:01Z", "cpu_seconds": 2.8124284829999997, "created_at": "2026-11-22T03:54:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-cd32fbd7becf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "fda5b9d8c6622aadd94ff9dab0925a3c9a4975658a961397555cada44c30669e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-cd32fbd7becf", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:54:41Z", "seed": 1360019946, "treatment": "frontier"}, "priority": 0.2900778230434386, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b2156b0ae70ece0954839fbd48a4868ab53c6753d671fc8dc70dc798f7963f85", "run_seconds": 3.325908899307251, "seed": 1360019946, "started_at": "2026-11-22T03:54:58Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:56:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-cd32fbd7becf", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.986111111, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.986111111, 1.0, 0.986111111, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 0.986111111], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fda5b9d8c6622aadd94ff9dab0925a3c9a4975658a961397555cada44c30669e", "primary_control": "random", "result_hash": "b2156b0ae70ece0954839fbd48a4868ab53c6753d671fc8dc70dc798f7963f85", "signature": "15db7420746dab5557caa152c55a1c575214be7e15ef669a49be6353e1b8ca631929e71b6390da044796451e077bead3a70f75cbaa3655215602677401e5b007", "state": "PARTIAL", "summary": {"count": 0.996528, "frontier": 1.0, "goal": 0.993056, "hypdisc": 1.0, "infogain": 1.0, "prederr": 0.996528, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:56:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:47Z", "cpu_seconds": 1.3959007710000002, "created_at": "2026-11-22T03:54:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0039f6b3353b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "7e376a3c1112c0c2b9dacc9f97c4aaa8adc64eee166d835ef41528dc50c85f0c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0039f6b3353b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:54:41Z", "seed": 570358314, "treatment": "full"}, "priority": 0.3155785664665617, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d015545cad77fd67f0ccda6d330979d581a613c9787f53fda9aabf722d3494d7", "run_seconds": 1.3993728160858154, "seed": 570358314, "started_at": "2026-11-22T03:54:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024286, 0.00574], "effect": -0.009273, "evaluated_at": "2026-11-22T03:56:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0039f6b3353b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.638127854, 0.578947368, 0.446153846, 0.80952381], "frozen": [0.168949772, 0.296992481, 0.252564103, 0.412698413], "full": [0.606164384, 0.578947368, 0.441025641, 0.80952381], "retrieval": [0.652968037, 0.578947368, 0.43974359, 0.80952381], "rules_nodetect": [0.606164384, 0.578947368, 0.51025641, 0.80952381]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7e376a3c1112c0c2b9dacc9f97c4aaa8adc64eee166d835ef41528dc50c85f0c", "primary_control": "facts", "result_hash": "d015545cad77fd67f0ccda6d330979d581a613c9787f53fda9aabf722d3494d7", "signature": "187c06ec0a9a4277440373d7273d0a508a5bf35a9f7f7d235641dcac529cd45228ae97996c6dfa71ee04dbaad3b6d6362003c0d76c871d6bc3c73ab08711750c", "state": "VERIFIED", "summary": {"facts": 0.618188, "frozen": 0.282801, "full": 0.608915, "retrieval": 0.620296, "rules_nodetect": 0.626223}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:56:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:52Z", "cpu_seconds": 0.9634469399999999, "created_at": "2026-11-22T03:54:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6fc2f236287a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "0a4008399d657ee8399a49e484ee9fa979f8f01ad297d699653365ef07dd7fe2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6fc2f236287a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:54:41Z", "seed": 1962703288, "treatment": "full"}, "priority": 0.30353010263059593, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d41d373a7fad94e25d3a19cb9d5ce8215ca9a6d1e6feec7a6283e4d35e9e22f2", "run_seconds": 0.9722659587860107, "seed": 1962703288, "started_at": "2026-11-22T03:54:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.042815, -0.012989], "effect": -0.027902, "evaluated_at": "2026-11-22T03:56:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6fc2f236287a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.226086957, 0.482978723, 0.476595745, 0.491489362], "frozen": [0.134782609, 0.127659574, 0.127659574, 0.127659574], "full": [0.180434783, 0.461702128, 0.465957447, 0.457446809], "retrieval": [0.254347826, 0.491489362, 0.485106383, 0.474468085], "rules_nodetect": [0.180434783, 0.461702128, 0.661702128, 0.457446809]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "0a4008399d657ee8399a49e484ee9fa979f8f01ad297d699653365ef07dd7fe2", "primary_control": "facts", "result_hash": "d41d373a7fad94e25d3a19cb9d5ce8215ca9a6d1e6feec7a6283e4d35e9e22f2", "signature": "84ddbf97c6bb1a676738a595953297519483fab09ff3633db7ee36656826d9ebe8d6ef59baf8339b11e975f0a445da1665b2ce92034dfc8986956d4522385c04", "state": "VERIFIED", "summary": {"facts": 0.419288, "frozen": 0.12944, "full": 0.391385, "retrieval": 0.426353, "rules_nodetect": 0.440321}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:56:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:54:47Z", "cpu_seconds": 0.3451034150000001, "created_at": "2026-11-22T03:54:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f7926f31da9e", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "45cb786137cf4247323acc2dd11806d4cd1df444d0f386427f5632f6c8bc13aa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f7926f31da9e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:54:41Z", "seed": 2123145267, "treatment": "full"}, "priority": 0.3096517724238626, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-f26a07e68523", "result_hash": "8149a9cbec31e497f6382b6f7f1dd2be0043ec2de7ea4b5628f2bfc5edc8848f", "run_seconds": 0.3787393569946289, "seed": 2123145267, "started_at": "2026-11-22T03:54:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.089883, 0.358034], "effect": 0.223958, "evaluated_at": "2026-11-22T03:56:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f7926f31da9e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.953125, 0.680555556, 0.59375, 0.71875], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 0.888888889, 0.96875, 0.984375], "retrieval": [0.984375, 0.861111111, 0.90625, 0.953125], "rules_nodetect": [1.0, 0.888888889, 0.96875, 0.984375]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "45cb786137cf4247323acc2dd11806d4cd1df444d0f386427f5632f6c8bc13aa", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-f26a07e68523", "result_hash": "8149a9cbec31e497f6382b6f7f1dd2be0043ec2de7ea4b5628f2bfc5edc8848f", "signature": "6b5906da2c3f0ba0de4f8dff13975a801d825de85bfed5e4d47bcab35e915b20cca18481f57a3ac82d66cfc8c7e88af1a6b394619421d7c1432a8a51a01bea03", "state": "VERIFIED", "summary": {"facts": 0.736545, "frozen": 0.0, "full": 0.960503, "retrieval": 0.926215, "rules_nodetect": 0.960503}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:56:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:39Z", "cpu_seconds": 3.769864849000001, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0d73d8bca241", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "5908a2f5fd1150ca6367d43dde63a8b9979425d8804dc8baa341f2b528ad91a6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0d73d8bca241", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:54:58Z", "seed": 1761237039, "treatment": "frontier"}, "priority": 0.2824792149684485, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b2427cf14887f795fed2470faaf5370fe44045c1f0baa64e1d7539fa64a41ad0", "run_seconds": 3.9684946537017822, "seed": 1761237039, "started_at": "2026-11-22T03:55:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.050607, 0.14339], "effect": 0.046391, "evaluated_at": "2026-11-22T03:57:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0d73d8bca241", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.50212766, 0.514893617, 0.589130435, 0.608571429], "frontier": [0.463829787, 0.461702128, 0.469565217, 0.591428571], "goal": [0.368085106, 0.380851064, 0.3, 0.571428571], "hypdisc": [0.444680851, 0.461702128, 0.515217391, 0.585714286], "infogain": [0.476595745, 0.461702128, 0.117391304, 0.591428571], "prederr": [0.476595745, 0.461702128, 0.195652174, 0.591428571], "random": [0.444680851, 0.442553191, 0.510869565, 0.402857143], "risk": [0.463829787, 0.461702128, 0.469565217, 0.591428571]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "5908a2f5fd1150ca6367d43dde63a8b9979425d8804dc8baa341f2b528ad91a6", "primary_control": "random", "result_hash": "b2427cf14887f795fed2470faaf5370fe44045c1f0baa64e1d7539fa64a41ad0", "signature": "052396987eb48b46f0c918d636ef7eb76de08062123b7dfafab28b2e2cb45393ee2c27ed3661fbb6b0ae15f68b672538714d346193aee99154126b00883dfe0a", "state": "VERIFIED", "summary": {"count": 0.553681, "frontier": 0.496631, "goal": 0.405091, "hypdisc": 0.501829, "infogain": 0.411779, "prederr": 0.431345, "random": 0.45024, "risk": 0.496631}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:57:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:44Z", "cpu_seconds": 0.6861945400000025, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ae6d7203c50a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "dacfae8b143488eba500d115d4aae6fc47369d0985f364b5f9715e267550f7b0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ae6d7203c50a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:54:58Z", "seed": 2058102178, "treatment": "frontier"}, "priority": 0.27893996614019484, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ae28156bb74c4400c08a367d584c45e5b8348b4648cd392359fa4e0aa4744860", "run_seconds": 0.69215989112854, "seed": 2058102178, "started_at": "2026-11-22T03:55:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.099285, 0.082916], "effect": -0.008185, "evaluated_at": "2026-11-22T03:58:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ae6d7203c50a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 0.982142857, 0.982142857, 0.833333333], "goal": [0.916666667, 0.946428571, 0.928571429, 0.895833333], "hypdisc": [1.0, 0.982142857, 0.982142857, 0.979166667], "infogain": [1.0, 0.982142857, 0.982142857, 0.833333333], "prederr": [1.0, 0.982142857, 0.982142857, 0.833333333], "random": [0.958333333, 0.964285714, 0.928571429, 0.979166667], "risk": [1.0, 0.982142857, 0.982142857, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "dacfae8b143488eba500d115d4aae6fc47369d0985f364b5f9715e267550f7b0", "primary_control": "random", "result_hash": "ae28156bb74c4400c08a367d584c45e5b8348b4648cd392359fa4e0aa4744860", "signature": "8da85e6015ef13b10aeed69c852f5e10439ce696d01de81dee87b16cd34e01ebe8e845fad5807df98a1e46200b276bc940822dbfd33df0bd97dde96f189a1b04", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.949405, "goal": 0.921875, "hypdisc": 0.985863, "infogain": 0.949405, "prederr": 0.949405, "random": 0.957589, "risk": 0.949405}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T03:58:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:33Z", "cpu_seconds": 0.35782584300000053, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3519d127bf79", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "a246873a36d744d67df36e70a083e177caa004166d7a31997e2181f98643b700", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3519d127bf79", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:54:58Z", "seed": 1531679398, "treatment": "full"}, "priority": 0.28344725974163065, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0fb74be0cca74fd220b21bd8e61215ac287aeb389eb7532b4aedaf4901688883", "run_seconds": 0.41495370864868164, "seed": 1531679398, "started_at": "2026-11-22T03:55:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.68, 0.055], "effect": -0.3125, "evaluated_at": "2026-11-22T03:57:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3519d127bf79", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 0.5, 0.75], "fixed_policy": [0.25, 0.5, 0.0, 0.75], "full": [0.75, 0.0, 0.0, 0.75], "no_world_model": [0.75, 0.0, 0.0, 0.75], "random": [0.5, 0.0, 0.5, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "a246873a36d744d67df36e70a083e177caa004166d7a31997e2181f98643b700", "primary_control": "e141_best", "result_hash": "0fb74be0cca74fd220b21bd8e61215ac287aeb389eb7532b4aedaf4901688883", "signature": "7c211a59894966291a5634071a37db57477674afbfce7bce7257aed9aeeab8c09d811e220aaa4c65ad41cf5fe7d16ee927a8934f579e237cc0d17d0d6c4ce70e", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.375, "full": 0.375, "no_world_model": 0.375, "random": 0.4375}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:57:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:40Z", "cpu_seconds": 1.408539632, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-be43e650b48a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "b51e845bd6c7cc7590b7416404f0bf66c7e860bb33f6f6c9944e8fc3c1f40766", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-be43e650b48a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:54:58Z", "seed": 234144935, "treatment": "full"}, "priority": 0.28028014682643887, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9f3936ea1861124cea6c5c911cd96566f5e8d0394df746b2492b29a9561d066f", "run_seconds": 1.8689749240875244, "seed": 234144935, "started_at": "2026-11-22T03:55:38Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:57:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-be43e650b48a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.0, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 0.0, 0.0, 0.0], "no_world_model": [0.0, 0.5, 0.0, 0.0], "random": [0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b51e845bd6c7cc7590b7416404f0bf66c7e860bb33f6f6c9944e8fc3c1f40766", "primary_control": "e141_best", "result_hash": "9f3936ea1861124cea6c5c911cd96566f5e8d0394df746b2492b29a9561d066f", "signature": "7316d3ef3e5b83a3f79ebba73b9a0e7de3277d062ac2148135e3e386e00cec9b0ad681094547387565bd29a615334b4cc5d3906a72328f1cfd96afdcc41ff608", "state": "PARTIAL", "summary": {"e141_best": 0.0, "fixed_policy": 0.0, "full": 0.0, "no_world_model": 0.125, "random": 0.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:58:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:11Z", "cpu_seconds": 4.531964657, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c945a3ecda4c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "9e6ae8f3377c472dec21d8585237fe526789c21a32087678583d2e2e903ae6dc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c945a3ecda4c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:54:58Z", "seed": 1052998108, "treatment": "full"}, "priority": 0.30180193214880735, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d60b8dfc7846a5e86dded4a491d20e33bfec3b5cda7e9aafdb32152f15334153", "run_seconds": 4.876868009567261, "seed": 1052998108, "started_at": "2026-11-22T03:55:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.74157, 0.49157], "effect": -0.125, "evaluated_at": "2026-11-22T03:56:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c945a3ecda4c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.0, 0.25, 1.0], "fixed_policy": [0.0, 0.0, 0.75, 0.0], "full": [1.0, 0.0, 0.25, 0.0], "no_world_model": [0.75, 0.0, 0.25, 0.0], "random": [0.0, 0.0, 0.25, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "9e6ae8f3377c472dec21d8585237fe526789c21a32087678583d2e2e903ae6dc", "primary_control": "e141_best", "result_hash": "d60b8dfc7846a5e86dded4a491d20e33bfec3b5cda7e9aafdb32152f15334153", "signature": "481302a88ff066ce91c8a136797d982bae40acfa0f64cce50cb3eea54f69ad379a7bd8bcaad856c286e5877e10737d2607f6a4fd6b94bfd279ec598d4208c307", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.1875, "full": 0.3125, "no_world_model": 0.25, "random": 0.0625}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:56:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:16Z", "cpu_seconds": 2.3781722320000007, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b9e5ac2658c7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "94c07b17215999b736d36cb759e1c75ea8e838b7062c9e6caf579f7909371e62", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b9e5ac2658c7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:54:58Z", "seed": 1278126695, "treatment": "full"}, "priority": 0.2950196610470993, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "369a718b90508b13fea5d79f8fc06413db675c7ba7cfcf536672a88976195c54", "run_seconds": 2.6686127185821533, "seed": 1278126695, "started_at": "2026-11-22T03:55:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.083588, 0.335069], "effect": 0.125741, "evaluated_at": "2026-11-22T03:56:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b9e5ac2658c7", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.991666667, 0.295555556, 0.325925926, 0.864206642], "frozen": [0.308333333, 0.222222222, 0.0, 0.335055351], "full": [1.0, 0.74, 0.325925926, 0.914391144], "retrieval": [1.0, 0.762962963, 0.746296296, 0.95202952], "rules_nodetect": [1.0, 0.328148148, 0.338888889, 0.914391144]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "94c07b17215999b736d36cb759e1c75ea8e838b7062c9e6caf579f7909371e62", "primary_control": "facts", "result_hash": "369a718b90508b13fea5d79f8fc06413db675c7ba7cfcf536672a88976195c54", "signature": "c5cacfcb7d545aca310f4048cfe5ef0ba280d418aeed045593adb1ddb465ca056a7317af88f8cb34eb894f3aabff5eca77b7c7782d38e427dd1a99857e619501", "state": "VERIFIED", "summary": {"facts": 0.619339, "frozen": 0.216403, "full": 0.745079, "retrieval": 0.865322, "rules_nodetect": 0.645357}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:56:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:08Z", "cpu_seconds": 1.1303117059999988, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c0ebda254617", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "da42119bda30591d8af3254854672c9b60b983d98008c78c5c20b3f247924540", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c0ebda254617", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:54:58Z", "seed": 2079930185, "treatment": "full"}, "priority": 0.2971289356185418, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "365b005325b9008f1e7f679ea389e4251657c40783a8cbeadb5ab5790e5f3e68", "run_seconds": 1.2119231224060059, "seed": 2079930185, "started_at": "2026-11-22T03:55:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.071417, 0.004133], "effect": -0.033642, "evaluated_at": "2026-11-22T03:56:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c0ebda254617", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.58, 0.342857143, 0.642105263, 0.741747573], "frozen": [0.1325, 0.137142857, 0.136842105, 0.133980583], "full": [0.5725, 0.3, 0.557894737, 0.741747573], "retrieval": [0.585, 0.36, 0.610526316, 0.741747573], "rules_nodetect": [0.565, 0.3, 0.557894737, 0.741747573]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "da42119bda30591d8af3254854672c9b60b983d98008c78c5c20b3f247924540", "primary_control": "facts", "result_hash": "365b005325b9008f1e7f679ea389e4251657c40783a8cbeadb5ab5790e5f3e68", "signature": "d0c27d21b7fd9618891f0c49734d5702dfb434999a1c845751e52f34090ec4fe58c027d65e87da943341d66f324175549575db8e2b69f4f91114699078f58600", "state": "VERIFIED", "summary": {"facts": 0.576677, "frozen": 0.135116, "full": 0.543036, "retrieval": 0.574318, "rules_nodetect": 0.541161}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:56:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:38Z", "cpu_seconds": 0.15485687400000003, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c5bf1b597c7e", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "e0f5c6ebb6c9288fcce1ab7a84c543927c70f569db3ed1f15ccf762296403c02", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c5bf1b597c7e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:54:58Z", "seed": 1879481852, "treatment": "full"}, "priority": 0.28146117904260265, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-9b95ca3108a1", "result_hash": "cd5adf2146fe716d5703b2b183cf0b7f8ab4a3610b5776465f4af08e9fbbc118", "run_seconds": 0.16123509407043457, "seed": 1879481852, "started_at": "2026-11-22T03:55:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002727, 0.008409], "effect": 0.002841, "evaluated_at": "2026-11-22T03:57:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c5bf1b597c7e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.931818182, 1.0, 0.977272727, 0.988636364], "frozen": [0.034090909, 0.022727273, 0.034090909, 0.034090909], "full": [0.943181818, 1.0, 0.977272727, 0.988636364], "retrieval": [0.943181818, 1.0, 0.977272727, 0.988636364], "rules_nodetect": [0.943181818, 1.0, 0.977272727, 0.988636364]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "e0f5c6ebb6c9288fcce1ab7a84c543927c70f569db3ed1f15ccf762296403c02", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-9b95ca3108a1", "result_hash": "cd5adf2146fe716d5703b2b183cf0b7f8ab4a3610b5776465f4af08e9fbbc118", "signature": "cdfd9923ac8aa328ddd2837793b4cacd603512405fc5517df89972a5ae82b239527e12bb23329f3cd04664045973044906b31a86fe80f3b2a1b53fbece5d8809", "state": "VERIFIED", "summary": {"facts": 0.974432, "frozen": 0.03125, "full": 0.977273, "retrieval": 0.977273, "rules_nodetect": 0.977273}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:57:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:27Z", "cpu_seconds": 21.317175878, "created_at": "2026-11-22T03:54:58Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-8d490b7152a3", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "b84c390c7f56393874d34dea12037706a64a042f2c1f99a2372b9c718c90a196", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-8d490b7152a3", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:54:58Z", "seed": 282497836, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "c7e93ebb52123e7beb7d81cd7b3e6fcc5f32ff3a25c520b1cab0901b46c7b7ba", "run_seconds": 23.477927446365356, "seed": 282497836, "started_at": "2026-11-22T03:55:04Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 3, "inherit_mem": 3, "inherit_nomem": 3, "random_search": 1, "reset_mem": 2, "retrieval": 0, "rootcause": 3, "xfer_mem": 2}, "better": "higher", "ci95": [-0.032244, 0.282841], "effect": 0.125298, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T03:56:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-8d490b7152a3", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"infogain": 1.0, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.55}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.55}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"hypdisc": 0.8, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.7}, "random_search": {"abstain_threshold": 0.379, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.742}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0, "risk": 0.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"infogain": 1.0, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.55}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"count": 0.6, "infogain": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.39673913, 0.201388889, 0.933333333, 0.647058824, 0.322083333, 0.181475904, 0.609848485, 0.226537217, 0.241134752, 0.543350168], "compressed": [1.0, 0.361111111, 0.986666667, 0.343137255, 0.189583333, 0.192394578, 0.924242424, 0.278317152, 0.283687943, 0.751683502], "inherit_mem": [0.402173913, 1.0, 0.96, 0.551470588, 0.295416667, 0.27936747, 0.833333333, 0.208737864, 0.336879433, 0.688552189], "inherit_nomem": [0.489130435, 1.0, 0.98, 0.558823529, 0.340833333, 0.157756024, 0.882575758, 0.086569579, 0.405141844, 0.665824916], "random_search": [0.39673913, 1.0, 0.953333333, 0.789215686, 0.58, 0.15436747, 0.833333333, 0.226537217, 0.336879433, 0.418350168], "reset_mem": [0.994565217, 0.739583333, 0.96, 0.830882353, 0.319166667, 0.312876506, 0.882575758, 0.66302589, 0.290780142, 0.727272727], "retrieval": [0.39673913, 0.201388889, 0.933333333, 0.647058824, 0.322083333, 0.181475904, 0.609848485, 0.226537217, 0.241134752, 0.543350168], "rootcause": [1.0, 0.361111111, 0.986666667, 0.343137255, 0.189583333, 0.192394578, 0.924242424, 0.278317152, 0.283687943, 0.751683502], "xfer_mem": [1.0, 0.215277778, 0.986666667, 0.681372549, 0.202083333, 0.215361446, 0.890151515, 0.292071197, 0.326241135, 0.71969697]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.7648, "losses": 3, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.172", "per_generation_sealed": [0.430295, 0.468219, 0.464228, 0.555593], "prereg_hash": "b84c390c7f56393874d34dea12037706a64a042f2c1f99a2372b9c718c90a196", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "reset_mem", "inherit_nomem"], "result_hash": "c7e93ebb52123e7beb7d81cd7b3e6fcc5f32ff3a25c520b1cab0901b46c7b7ba", "signature": "b4f715a4604fde62a3d8eab680df4316d2d054bc6b307db78c2a0b420c234e990fcaa5ca670815239cd955a7a8fe2565392d2a19726e4342a2ef2c981b2eee05", "state": "VERIFIED", "summary": {"G0": 0.430295, "compressed": 0.531082, "inherit_mem": 0.555593, "inherit_nomem": 0.556666, "random_search": 0.568876, "reset_mem": 0.672073, "retrieval": 0.430295, "rootcause": 0.531082, "xfer_mem": 0.552892}, "ties": 0, "treatment": "inherit_mem", "wins": 7}, "verified_at": "2026-11-22T03:57:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:28Z", "cpu_seconds": 0.4422154599999981, "created_at": "2026-11-22T03:55:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-10fdca7e6ff5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "2f64d512147e3ee7c4210179ec57d8506c20bb59f0e6a8ee0d25eedf09e75b14", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-10fdca7e6ff5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:55:15Z", "seed": 498961936, "treatment": "full"}, "priority": 0.28635246760675076, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e5ebc4c3e954c9231be0830f831b345cac6035f256ccca721c53b1a9836b7b88", "run_seconds": 0.9470937252044678, "seed": 498961936, "started_at": "2026-11-22T03:55:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T03:57:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-10fdca7e6ff5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 0.75], "fixed_policy": [0.5, 0.25, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 0.25], "no_world_model": [1.0, 0.25, 0.0, 0.0], "random": [1.0, 0.75, 0.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2f64d512147e3ee7c4210179ec57d8506c20bb59f0e6a8ee0d25eedf09e75b14", "primary_control": "e141_best", "result_hash": "e5ebc4c3e954c9231be0830f831b345cac6035f256ccca721c53b1a9836b7b88", "signature": "432046daf28a105affa7675606ff3111fd5fef44352d21a4b87a4da504d7ccb2a5aa57022e1c05c53c98261f781f2b22d8b3d2f18cd5d044b3505a7218655400", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.1875, "full": 0.75, "no_world_model": 0.3125, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:57:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:22Z", "cpu_seconds": 0.9641649769999994, "created_at": "2026-11-22T03:55:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3d3347a063ee", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "6b2c6cfd4da577106b5c88dc86daa62bb22c5b38ed9ded34de8659b42d018afd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3d3347a063ee", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:55:15Z", "seed": 638048694, "treatment": "full"}, "priority": 0.28657353733990976, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "09e7699020c9557180c3bea2a83e2706cc4015d57527952ea51007708a018fa7", "run_seconds": 1.2717316150665283, "seed": 638048694, "started_at": "2026-11-22T03:55:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.011088, 0.077905], "effect": 0.033409, "evaluated_at": "2026-11-22T03:56:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3d3347a063ee", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.442105263, 0.672727273, 0.563934426, 0.690909091], "frozen": [0.110526316, 0.109090909, 0.108196721, 0.109090909], "full": [0.436842105, 0.745454545, 0.557377049, 0.763636364], "retrieval": [0.465789474, 0.690909091, 0.557377049, 0.709090909], "rules_nodetect": [0.436842105, 0.745454545, 0.557377049, 0.763636364]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "6b2c6cfd4da577106b5c88dc86daa62bb22c5b38ed9ded34de8659b42d018afd", "primary_control": "facts", "result_hash": "09e7699020c9557180c3bea2a83e2706cc4015d57527952ea51007708a018fa7", "signature": "dbf98a52dc25a1192c489aa70457d0564714865e33566b79923a5a534c3a0a309ffb7f72b1b3833aed46910a7c1a9baca06618da6fce280ab17b36e4fbac8f0d", "state": "VERIFIED", "summary": {"facts": 0.592419, "frozen": 0.109226, "full": 0.625828, "retrieval": 0.605792, "rules_nodetect": 0.625828}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:56:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:30Z", "cpu_seconds": 4.505076668000001, "created_at": "2026-11-22T03:55:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-36d3dd27861b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "bcf9e09a7d36162f7f442e9783a26178d4ed48920c7964366dcc562886de9a55", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-36d3dd27861b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:55:15Z", "seed": 1572900335, "treatment": "gated"}, "priority": 0.2864218986078561, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f5cf6d035b73abbe906a640b6851124d2e6cdddd643952fecbc3a158f462fe3b", "run_seconds": 5.828556776046753, "seed": 1572900335, "started_at": "2026-11-22T03:55:24Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:57:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-36d3dd27861b", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.133333333, 0.875, 0.195959596, 0.397701149], "gated": [0.133333333, 0.875, 0.195959596, 0.397701149], "naive": [0.082828283, 0.875, 0.119191919, 0.314942529], "scratch": [0.133333333, 0.875, 0.195959596, 0.397701149], "ungated": [0.090909091, 0.875, 0.119191919, 0.314942529]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "bcf9e09a7d36162f7f442e9783a26178d4ed48920c7964366dcc562886de9a55", "primary_control": "scratch", "result_hash": "f5cf6d035b73abbe906a640b6851124d2e6cdddd643952fecbc3a158f462fe3b", "signature": "f4484aa4647f4f4a5a29e5cf0c942413e17112b3b0c6b30ea7b0ba2d089ff3f4a4f3817d5de6363020cd8998e471794520283d19483c95d3dc8ae2da9c8d4209", "state": "PARTIAL", "summary": {"compressed": 0.400499, "gated": 0.400499, "naive": 0.347991, "scratch": 0.400499, "ungated": 0.350011}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:57:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:48Z", "cpu_seconds": 2.451759333000002, "created_at": "2026-11-22T03:55:32Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-05e3154f3bf8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "229dbdda3d64d9f647e0dff8a645c08446ec2cc4a469f951ee7d3ecfe383eb68", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-05e3154f3bf8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:55:32Z", "seed": 246441085, "treatment": "frontier"}, "priority": 0.2764894586025748, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "962290096da67dca26cae3078ec3c80bf315930fdc21929eb86386743ef8ae4a", "run_seconds": 2.5040228366851807, "seed": 246441085, "started_at": "2026-11-22T03:55:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.273738, 0.042726], "effect": -0.115506, "evaluated_at": "2026-11-22T03:58:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-05e3154f3bf8", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.705078125, 0.352539062, 0.6875, 0.690972222], "frontier": [0.650390625, 0.311523438, 0.364583333, 0.565972222], "goal": [0.6015625, 0.288085938, 0.4375, 0.564236111], "hypdisc": [0.685546875, 0.30859375, 0.604166667, 0.585069444], "infogain": [0.650390625, 0.311523438, 0.364583333, 0.565972222], "prederr": [0.650390625, 0.311523438, 0.364583333, 0.565972222], "random": [0.6640625, 0.331054688, 0.71875, 0.640625], "risk": [0.650390625, 0.311523438, 0.364583333, 0.565972222]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "229dbdda3d64d9f647e0dff8a645c08446ec2cc4a469f951ee7d3ecfe383eb68", "primary_control": "random", "result_hash": "962290096da67dca26cae3078ec3c80bf315930fdc21929eb86386743ef8ae4a", "signature": "3252a8a3d7b7e10190bca83d8969b48ac380e579b96026c2325c107a825594d25a139137eb15a3663aa063e9e56219ef13cac1e331073239f8d9248895a64202", "state": "VERIFIED", "summary": {"count": 0.609022, "frontier": 0.473117, "goal": 0.472846, "hypdisc": 0.545844, "infogain": 0.473117, "prederr": 0.473117, "random": 0.588623, "risk": 0.473117}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T03:58:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:56Z", "cpu_seconds": 1.0290681890000002, "created_at": "2026-11-22T03:55:32Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-fb0a8e7a1eab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "7a740ae47ddfd86378273b8a329bbe326be36cda7f27009320ca1d6f61f1b2fd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-fb0a8e7a1eab", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:55:32Z", "seed": 1945484275, "treatment": "frontier"}, "priority": 0.2764255218600812, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c30fdcf27700ae4b7e3c86a92c04d71003737c6d5b8d8bd8cf81c3c141925384", "run_seconds": 1.1316146850585938, "seed": 1945484275, "started_at": "2026-11-22T03:55:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.025263, 0.077895], "effect": 0.026316, "evaluated_at": "2026-11-22T03:58:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-fb0a8e7a1eab", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.888888889, 0.986928105, 0.824561404, 0.944444444], "frontier": [0.888888889, 0.986928105, 0.807017544, 0.944444444], "goal": [0.703703704, 0.836601307, 0.807017544, 0.944444444], "hypdisc": [0.777777778, 0.941176471, 0.807017544, 0.944444444], "infogain": [0.814814815, 0.620915033, 0.771929825, 0.944444444], "prederr": [0.814814815, 0.620915033, 0.807017544, 0.944444444], "random": [0.888888889, 0.986928105, 0.701754386, 0.944444444], "risk": [0.888888889, 0.986928105, 0.807017544, 0.944444444]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7a740ae47ddfd86378273b8a329bbe326be36cda7f27009320ca1d6f61f1b2fd", "primary_control": "random", "result_hash": "c30fdcf27700ae4b7e3c86a92c04d71003737c6d5b8d8bd8cf81c3c141925384", "signature": "8dfbd37836674c35ad6def0d3c6ef3b712b79c8d191322427a6ea55e69bdfdbc66d452d74d08755655da82a51665562db70de990cbe1922ae18796d2e12a9202", "state": "VERIFIED", "summary": {"count": 0.911206, "frontier": 0.90682, "goal": 0.822942, "hypdisc": 0.867604, "infogain": 0.788026, "prederr": 0.796798, "random": 0.880504, "risk": 0.90682}, "ties": 3, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T03:58:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:33Z", "cpu_seconds": 0.7827106949999987, "created_at": "2026-11-22T03:55:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9a4c50b1ae53", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "2b8e160bc2c9057ba3779cb7f391d3918601d6e8c0007058a224cae6945fe933", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9a4c50b1ae53", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:55:32Z", "seed": 218203615, "treatment": "full"}, "priority": 0.28460892305620544, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ab32141af8338223932b1ba73df5f8818baefe6943de2de2120f0e4260d4bd94", "run_seconds": 0.7989456653594971, "seed": 218203615, "started_at": "2026-11-22T03:55:32Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:57:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9a4c50b1ae53", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"fixed_policy": [1.0, 0.0], "random": [1.0, 0.8125]}, "prereg_hash": "2b8e160bc2c9057ba3779cb7f391d3918601d6e8c0007058a224cae6945fe933", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "ab32141af8338223932b1ba73df5f8818baefe6943de2de2120f0e4260d4bd94", "signature": "996573b9fd5c48f0aa1995010875ea2847aad01dcc6ea846520ce57e4c3268ab273b4fafbdf811cfa07424c1afcd4f89521a0aac8c92f2f0cbe7131d13acd503", "state": "REJECTED", "summary": {"e141_best": 1.0, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 1.0, "random": 0.8125}}, "verified_at": "2026-11-22T03:57:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:45Z", "cpu_seconds": 0.5309386790000019, "created_at": "2026-11-22T03:55:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a204d632d772", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "22420d0cea5a43c7e29236c09ae170eaa0a021a1e244152dc4dba7881e43f526", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a204d632d772", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:55:32Z", "seed": 2046935394, "treatment": "full"}, "priority": 0.27887797885277266, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5930d21e45ca5c81d3ce826bd8abe50fe5b878913c319dd5a3b35434fa991062", "run_seconds": 0.5804803371429443, "seed": 2046935394, "started_at": "2026-11-22T03:55:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T03:58:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a204d632d772", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.75, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 0.5], "no_world_model": [0.75, 1.0, 1.0, 0.5], "random": [1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "22420d0cea5a43c7e29236c09ae170eaa0a021a1e244152dc4dba7881e43f526", "primary_control": "e141_best", "result_hash": "5930d21e45ca5c81d3ce826bd8abe50fe5b878913c319dd5a3b35434fa991062", "signature": "778e8b78201fab661fbf6f77c2b012364ce981d3af2a669da16c7b0a392f6fcdcb02d3c360aa3f871c2a657af37c0732bee41c9b178a3265f18a951c8b99830e", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.25, "full": 0.875, "no_world_model": 0.8125, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:58:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:00Z", "cpu_seconds": 0.7053289440000015, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-31ca199a892a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "21aff24707d4fc891e78f8019ef436d60006ae114d5cc171a92b7d953ae82dd4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-31ca199a892a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:55:48Z", "seed": 174836001, "treatment": "full"}, "priority": 0.27535738287027306, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b8ddaf08da8071f6f355205ad615ae01368ba22f565a225d095ac1d93182de53", "run_seconds": 1.0211665630340576, "seed": 174836001, "started_at": "2026-11-22T03:55:59Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:58:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-31ca199a892a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.0, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.0, 0.75], "no_world_model": [1.0, 1.0, 0.0, 0.75], "random": [1.0, 1.0, 0.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "21aff24707d4fc891e78f8019ef436d60006ae114d5cc171a92b7d953ae82dd4", "primary_control": "e141_best", "result_hash": "b8ddaf08da8071f6f355205ad615ae01368ba22f565a225d095ac1d93182de53", "signature": "5e522bf8d538daef812f796e7ee8d952cefd73af74e6a0fb62856d7bd8ac91c1abef4c0aa365e0191f41a8386472a28de561e2a02dfaca1a9bc4d03e6489020c", "state": "PARTIAL", "summary": {"e141_best": 0.6875, "fixed_policy": 0.0, "full": 0.6875, "no_world_model": 0.6875, "random": 0.6875}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:58:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:51Z", "cpu_seconds": 0.7986188100000007, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4d027803365f", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "b675532c674200f1e6b93becd4ae25c1e2273dc70c0383690d8f865f50548385", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4d027803365f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:55:48Z", "seed": 1397850048, "treatment": "full"}, "priority": 0.2819311702225112, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "83d4d083d5d0ba6308ef5eda6baa47f47922331ec898b408ef032b8aa5dba937", "run_seconds": 0.8427963256835938, "seed": 1397850048, "started_at": "2026-11-22T03:55:50Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:58:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4d027803365f", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"fixed_policy": [1.0, 0.0], "full": [1.0, 0.9375], "no_world_model": [1.0, 0.9375], "random": [1.0, 0.5625]}, "prereg_hash": "b675532c674200f1e6b93becd4ae25c1e2273dc70c0383690d8f865f50548385", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "83d4d083d5d0ba6308ef5eda6baa47f47922331ec898b408ef032b8aa5dba937", "signature": "d5b9b68c0ab1c5a00fe995e201ba1d079627209ebe48a372ee58187360601e7269092c37a75858600cb13d044a42fe00c187e84d3066c86eb6239ac101fbfe0f", "state": "REJECTED", "summary": {"e141_best": 1.0, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.9375, "random": 0.5625}}, "verified_at": "2026-11-22T03:58:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:54Z", "cpu_seconds": 0.6704964089999983, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bc29e2dd0c32", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "3275a9ec3d89475c89cf9faf41d3317cf01102851e6cfb1b3ca615ed867694b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bc29e2dd0c32", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:55:48Z", "seed": 1018697962, "treatment": "full"}, "priority": 0.27735504711622927, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "03840d09a3d3111505ca496eeb8d437435624494a60d6576f30e69f4d3da0061", "run_seconds": 0.6738696098327637, "seed": 1018697962, "started_at": "2026-11-22T03:55:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:58:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-bc29e2dd0c32", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "3275a9ec3d89475c89cf9faf41d3317cf01102851e6cfb1b3ca615ed867694b6", "primary_control": "e141_best", "result_hash": "03840d09a3d3111505ca496eeb8d437435624494a60d6576f30e69f4d3da0061", "signature": "592d65e4530d60b44abd4ed76d1139adf2459fe2ac3e0b196ee1e8401b9122f10c5dd09ca0df78a9db85690ddd4d878d2e17ea970f86d1992063265ba8aa9600", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 0.9375, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:58:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:49Z", "cpu_seconds": 0.3213220710000009, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ff34c36b3051", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "629c8fdda369453b3aedc525b8f92bc92fa7dcb058c4d833f88339e086570184", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ff34c36b3051", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:55:48Z", "seed": 286446083, "treatment": "full"}, "priority": 0.28380796631810196, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "95e833608e9fa6dea5045395ae2674f76f0d9df4d2e1a0c69ea46084856765d6", "run_seconds": 0.3346874713897705, "seed": 286446083, "started_at": "2026-11-22T03:55:49Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:58:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ff34c36b3051", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.75, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "629c8fdda369453b3aedc525b8f92bc92fa7dcb058c4d833f88339e086570184", "primary_control": "e141_best", "result_hash": "95e833608e9fa6dea5045395ae2674f76f0d9df4d2e1a0c69ea46084856765d6", "signature": "ee25735dbd28f6cfeafde93c80f7f93e426db26d12b8c48b4a9060da0bcfef4742eaf0d93dffe58f87558ed6765656dc52ace90e324c784b173abc53df5b5a09", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.6875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:58:07Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:05Z", "cpu_seconds": 0.5614343460000022, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5e107c028cd9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "2af59b45c87aab477388ba91be2c0b3914caa5123e916a03349b04869740ab79", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5e107c028cd9", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:55:48Z", "seed": 1704939251, "treatment": "full"}, "priority": 0.26765173853245866, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "770a89a1b141d5944b75996608d9641a6b3b24819091968890d2b5baf3c3a36a", "run_seconds": 0.5896201133728027, "seed": 1704939251, "started_at": "2026-11-22T03:56:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022779, 0.089391], "effect": 0.033306, "evaluated_at": "2026-11-22T03:58:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5e107c028cd9", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.62962963, 0.441176471, 0.743055556, 0.791666667], "frozen": [0.503703704, 0.323529412, 0.590277778, 0.583333333], "full": [0.748148148, 0.455882353, 0.743055556, 0.791666667], "retrieval": [0.748148148, 0.529411765, 0.743055556, 0.791666667], "rules_nodetect": [0.748148148, 0.779411765, 0.756944444, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "2af59b45c87aab477388ba91be2c0b3914caa5123e916a03349b04869740ab79", "primary_control": "facts", "result_hash": "770a89a1b141d5944b75996608d9641a6b3b24819091968890d2b5baf3c3a36a", "signature": "0774c1a45c814009b274a6f102e3e62b72b62f7aece26a51ea874121997650031f2b2d74e629447cd14140912cc1c812000f3ff2c94ea6e72fa628aefcf76604", "state": "VERIFIED", "summary": {"facts": 0.651382, "frozen": 0.500211, "full": 0.684688, "retrieval": 0.703071, "rules_nodetect": 0.758626}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:58:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:02Z", "cpu_seconds": 0.8866578579999995, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6e27e50fa6c0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "868cd93f4f63e0970f92ab0d8b799ef0b12f886f522f4ade79fc65bc617fca24", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6e27e50fa6c0", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:55:48Z", "seed": 1357225437, "treatment": "full"}, "priority": 0.2725115648033902, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f6a2fe9b9b8cef50b9330ebee61934f21c0ddc25fc55e14c4d1fcc251cfe7fd6", "run_seconds": 1.0925495624542236, "seed": 1357225437, "started_at": "2026-11-22T03:56:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.009425, 0.046871], "effect": 0.018723, "evaluated_at": "2026-11-22T03:58:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6e27e50fa6c0", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.964285714, 1.0, 0.734848485, 1.0], "frozen": [0.321428571, 0.378205128, 0.401515152, 0.316666667], "full": [0.978571429, 1.0, 0.795454545, 1.0], "retrieval": [0.978571429, 1.0, 0.863636364, 1.0], "rules_nodetect": [0.978571429, 1.0, 0.795454545, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "868cd93f4f63e0970f92ab0d8b799ef0b12f886f522f4ade79fc65bc617fca24", "primary_control": "facts", "result_hash": "f6a2fe9b9b8cef50b9330ebee61934f21c0ddc25fc55e14c4d1fcc251cfe7fd6", "signature": "a9c0be5b726161d5ac7d5f3786fd9d7b796b25578df652a6dce0120badb548937e07507ed1e8d868c60b5418ba41a1fb41eb69b3fdabcd5d474dcb450575bd02", "state": "VERIFIED", "summary": {"facts": 0.924784, "frozen": 0.354454, "full": 0.943506, "retrieval": 0.960552, "rules_nodetect": 0.943506}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T03:58:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:02Z", "cpu_seconds": 0.1114815440000001, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-72fd1cd49f05", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "5db24adb098a3a6aeaf39e7b31e803b6a5b59b601bddea28d0546c41b54b62e7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-72fd1cd49f05", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:55:48Z", "seed": 1779601037, "treatment": "full"}, "priority": 0.2684641288841664, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2958abb185f9db88d21a54fb372e92a7331f0f6577631702dc29680669ba11c0", "run_seconds": 0.11262893676757812, "seed": 1779601037, "started_at": "2026-11-22T03:56:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.038457, 0.063816], "effect": 0.051136, "evaluated_at": "2026-11-22T03:58:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-72fd1cd49f05", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.818181818, 0.803030303, 0.810606061, 0.795454545], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [0.863636364, 0.871212121, 0.863636364, 0.833333333], "retrieval": [0.863636364, 0.818181818, 0.833333333, 0.803030303], "rules_nodetect": [0.863636364, 0.871212121, 0.863636364, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "5db24adb098a3a6aeaf39e7b31e803b6a5b59b601bddea28d0546c41b54b62e7", "primary_control": "facts", "result_hash": "2958abb185f9db88d21a54fb372e92a7331f0f6577631702dc29680669ba11c0", "signature": "03088858d0fadf5dac5ca3d6897befa04b53c3808ca181ce35de370865e0afab4336a2fa537535acc65a028ef30d6751dabd3186efa8e7f4ae09f5a638b55f09", "state": "VERIFIED", "summary": {"facts": 0.806818, "frozen": 0.030303, "full": 0.857955, "retrieval": 0.829545, "rules_nodetect": 0.857955}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T03:58:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:55:57Z", "cpu_seconds": 1.375237921, "created_at": "2026-11-22T03:55:48Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-f5718a14b643", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "a23ee7be7fcbf108f2d203a179681e0acff8c3fd155342a9c7a0f2d65c8fe7ab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-f5718a14b643", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:55:48Z", "seed": 193542889, "treatment": "gated"}, "priority": 0.2759431579439894, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "2a020d72d84395580b8b9e11ef8ddd0ea1f19a31908ab55edc6e2f718b80b911", "run_seconds": 1.6025300025939941, "seed": 193542889, "started_at": "2026-11-22T03:55:56Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:58:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-f5718a14b643", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.357142857, 0.433333333, 0.661290323, 0.464285714], "gated": [0.357142857, 0.433333333, 0.661290323, 0.464285714], "naive": [0.616071429, 0.625, 0.830645161, 0.55952381], "scratch": [0.357142857, 0.433333333, 0.661290323, 0.464285714], "ungated": [0.553571429, 0.508333333, 0.830645161, 0.55952381]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a23ee7be7fcbf108f2d203a179681e0acff8c3fd155342a9c7a0f2d65c8fe7ab", "primary_control": "scratch", "result_hash": "2a020d72d84395580b8b9e11ef8ddd0ea1f19a31908ab55edc6e2f718b80b911", "signature": "9d39078efba97608bab3510f865a9c0ab27673407c60dfec8b8522a50314d56caafb851544949349faffbbcbd8340cf88bf1b84d7bcd36d2261844f4b5064209", "state": "PARTIAL", "summary": {"compressed": 0.479013, "gated": 0.479013, "naive": 0.65781, "scratch": 0.479013, "ungated": 0.613018}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:58:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:04Z", "cpu_seconds": 0.6727776169999999, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9c64157dcf50", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "9ba33e15ac7bae52cf55e03b3ced608df3f4cb9cd3c07527e5de11eaa64a0244", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9c64157dcf50", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:56:05Z", "seed": 739600787, "treatment": "frontier"}, "priority": 0.26625868549805654, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "79ccb31fe290298a98ecff3abda59a7b141a2c818ee8298a96a40ccea9a5e01a", "run_seconds": 1.1092958450317383, "seed": 739600787, "started_at": "2026-11-22T03:58:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004583, 0.056667], "effect": 0.026042, "evaluated_at": "2026-11-22T04:00:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9c64157dcf50", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.979166667, 1.0, 1.0, 1.0], "frontier": [0.979166667, 1.0, 0.979166667, 0.979166667], "goal": [0.916666667, 0.958333333, 0.9375, 0.895833333], "hypdisc": [0.895833333, 1.0, 1.0, 0.979166667], "infogain": [0.979166667, 1.0, 0.979166667, 0.979166667], "prederr": [0.979166667, 1.0, 0.979166667, 0.979166667], "random": [0.916666667, 0.958333333, 0.979166667, 0.979166667], "risk": [0.979166667, 1.0, 0.979166667, 0.979166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "9ba33e15ac7bae52cf55e03b3ced608df3f4cb9cd3c07527e5de11eaa64a0244", "primary_control": "random", "result_hash": "79ccb31fe290298a98ecff3abda59a7b141a2c818ee8298a96a40ccea9a5e01a", "signature": "ac1ad1004a7186d4ed1ea76b5a8ef46e56fee9653ce34548bb66ac7a76502d60f7b6607be9820a5852b1cf3af670b36895bafcd56d8082d01f6718d16a200602", "state": "VERIFIED", "summary": {"count": 0.994792, "frontier": 0.984375, "goal": 0.927083, "hypdisc": 0.96875, "infogain": 0.984375, "prederr": 0.984375, "random": 0.958333, "risk": 0.984375}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:00:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:12Z", "cpu_seconds": 0.8558642310000018, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-64f84a3bc327", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "d08d3f38d1a67c806d27d75a4a83d73cf32cade39198f72e8791c913f3b734df", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-64f84a3bc327", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:56:05Z", "seed": 84133723, "treatment": "full"}, "priority": 0.277801768076152, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f268f2233e8350a319f437eca37224c6d67abe3d64240ea0011602bd18dd04b2", "run_seconds": 1.3402767181396484, "seed": 84133723, "started_at": "2026-11-22T03:56:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T03:58:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-64f84a3bc327", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 1.0, 1.0], "random": [1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d08d3f38d1a67c806d27d75a4a83d73cf32cade39198f72e8791c913f3b734df", "primary_control": "e141_best", "result_hash": "f268f2233e8350a319f437eca37224c6d67abe3d64240ea0011602bd18dd04b2", "signature": "d521dad6eeee93baf526affc1e9b3cf70cff9c2f3bcce7baa85a107fe5596110bd1e9bf9808520c58ac37f0861ee99eade05fb455f9b1259dfc30cbeddef220e", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 0.875, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T03:58:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:13Z", "cpu_seconds": 0.6720387290000005, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ad3d2186f074", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "c5f3372fbef9c20829a5feda551a9cd8ca35cb7122f0e2ad49907003f83a0ea6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ad3d2186f074", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:56:05Z", "seed": 1325041115, "treatment": "full"}, "priority": 0.2636236449219501, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ccbac2bd74d615ff309ed7fdaf6fb661bba2498ee44eebf067848cae6acd788b", "run_seconds": 0.7628531455993652, "seed": 1325041115, "started_at": "2026-11-22T03:58:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T04:00:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ad3d2186f074", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.5, 1.0], "no_world_model": [1.0, 0.75, 0.25, 1.0], "random": [0.0, 1.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "c5f3372fbef9c20829a5feda551a9cd8ca35cb7122f0e2ad49907003f83a0ea6", "primary_control": "e141_best", "result_hash": "ccbac2bd74d615ff309ed7fdaf6fb661bba2498ee44eebf067848cae6acd788b", "signature": "4c9a2c164a15b252c9757351d60de76cf90b57eee383ec3e01e3154e565ff4974bd903c7c5f04411adfcd695e74a9136bdc0399ca8ca09af949e29f7edd2420b", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0, "full": 0.875, "no_world_model": 0.75, "random": 0.625}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:00:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:08Z", "cpu_seconds": 0.26349476799999927, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f7d15ee93261", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "7eaef4ed4772f537bbece76ec630dabf57fe137f5b7f6fb35434ef3a7c00981c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f7d15ee93261", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:56:05Z", "seed": 2295452, "treatment": "full"}, "priority": 0.2643220745895353, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8c14a96b83530c4b5bb47e950f767a959747eb758d1dc83c6429cf42002b1420", "run_seconds": 0.278057336807251, "seed": 2295452, "started_at": "2026-11-22T03:58:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T04:00:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f7d15ee93261", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [0.75, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "7eaef4ed4772f537bbece76ec630dabf57fe137f5b7f6fb35434ef3a7c00981c", "primary_control": "e141_best", "result_hash": "8c14a96b83530c4b5bb47e950f767a959747eb758d1dc83c6429cf42002b1420", "signature": "9ffe00edf957372244ad567e1d736dcb09c8f0e6bbb5535fc2a5ed3ad6a1106801861c1823bbdbcaa8e9d43e739e3d77ccd294133796709efbe0ade4a01cb90f", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.125, "full": 1.0, "no_world_model": 1.0, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:00:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:58Z", "cpu_seconds": 1.093797511, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3fc4ca89e848", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "7c4ce7eabefb8a822be594c1a625ce1b790129750564e294dddabde1c6ec859c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3fc4ca89e848", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:56:05Z", "seed": 248750505, "treatment": "full"}, "priority": 0.26690612903152794, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "8a373cfa8b763fa1688fb9e3b58dd4cdc6b8f80ff46a922b91c8693c6b0d7804", "run_seconds": 1.1433258056640625, "seed": 248750505, "started_at": "2026-11-22T03:57:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.049896, 0.029742], "effect": -0.010077, "evaluated_at": "2026-11-22T04:00:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3fc4ca89e848", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.461309524, 0.283171521, 0.299352751, 0.512944984], "frozen": [0.180059524, 0.181229773, 0.181229773, 0.181229773], "full": [0.422619048, 0.333333333, 0.275080906, 0.485436893], "retrieval": [0.46875, 0.257281553, 0.328478964, 0.529126214], "rules_nodetect": [0.422619048, 0.694174757, 0.726537217, 0.75566343]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "7c4ce7eabefb8a822be594c1a625ce1b790129750564e294dddabde1c6ec859c", "primary_control": "facts", "result_hash": "8a373cfa8b763fa1688fb9e3b58dd4cdc6b8f80ff46a922b91c8693c6b0d7804", "signature": "c217b76eccddf179257975f01e42e0248a2ce2cb97879c8ab18916e381a38732a620009b972587308eaf3374e8e967749f6cfa9ecda3ab28ee1a3b8fcc087c04", "state": "VERIFIED", "summary": {"facts": 0.389195, "frozen": 0.180937, "full": 0.379118, "retrieval": 0.395909, "rules_nodetect": 0.649749}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:00:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:17Z", "cpu_seconds": 0.8390041109999977, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-728d1bf20d4d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "b4bcb6a4c026d3f6644db256b1ac64faeb46e631d174eb85ed198263ca3d1d24", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-728d1bf20d4d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:56:05Z", "seed": 1237751198, "treatment": "full"}, "priority": 0.27703587228561777, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "471bab63cb68fe7d2a782ae9bad1d3f799cf6178048c9da289c007772ba87112", "run_seconds": 0.8790419101715088, "seed": 1237751198, "started_at": "2026-11-22T03:56:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.065878, 0.075731], "effect": 0.004926, "evaluated_at": "2026-11-22T03:58:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-728d1bf20d4d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.85, 0.8, 0.6875, 0.946078431], "frozen": [0.31, 0.39375, 0.46875, 0.269607843], "full": [0.93, 0.81875, 0.59375, 0.960784314], "retrieval": [0.92, 0.83125, 0.8125, 0.965686275], "rules_nodetect": [0.34, 0.35625, 0.65625, 0.960784314]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "b4bcb6a4c026d3f6644db256b1ac64faeb46e631d174eb85ed198263ca3d1d24", "primary_control": "facts", "result_hash": "471bab63cb68fe7d2a782ae9bad1d3f799cf6178048c9da289c007772ba87112", "signature": "2db84b414206e5a0e1f2d168fabfcbcc6763059e5303610544047aa1ff1526fa039903f5e7cd7e5e0e5767df7047eac84ef672dba8cd960489efa90c7d483008", "state": "VERIFIED", "summary": {"facts": 0.820895, "frozen": 0.360527, "full": 0.825821, "retrieval": 0.882359, "rules_nodetect": 0.578321}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:58:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:02Z", "cpu_seconds": 0.16212502299999976, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-996fdd04d339", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "af2c1802a1104680223cc0a32049a58962f8d3c3066e24cb1f28f0de23eec6c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-996fdd04d339", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:56:05Z", "seed": 1587616575, "treatment": "full"}, "priority": 0.2667380977899182, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b8763f5da0fbb56237eff9f349c12a2271336d8c256b96c94a568a55e0dd22b8", "run_seconds": 0.21781039237976074, "seed": 1587616575, "started_at": "2026-11-22T03:58:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026844, 0.102602], "effect": 0.037879, "evaluated_at": "2026-11-22T04:00:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-996fdd04d339", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.666666667, 0.96969697, 0.696969697], "frozen": [0.03030303, 0.015151515, 0.03030303, 0.03030303], "full": [1.0, 0.803030303, 0.96969697, 0.712121212], "retrieval": [1.0, 0.803030303, 0.954545455, 0.712121212], "rules_nodetect": [1.0, 0.803030303, 0.96969697, 0.712121212]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "af2c1802a1104680223cc0a32049a58962f8d3c3066e24cb1f28f0de23eec6c6", "primary_control": "facts", "result_hash": "b8763f5da0fbb56237eff9f349c12a2271336d8c256b96c94a568a55e0dd22b8", "signature": "d976d2f685cef2025d5f0cd8f835e91fa82d8388903e208851225f8c4c8d5179e473d90e8fad39c4e683b1bd765fc5169d54d338cf31bbafc4147ec053763c00", "state": "VERIFIED", "summary": {"facts": 0.833333, "frozen": 0.026515, "full": 0.871212, "retrieval": 0.867424, "rules_nodetect": 0.871212}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:00:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:52Z", "cpu_seconds": 201.698805757, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-a6b5dceb84f6", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "23714cf42837803b99fee8d4b138da1534e0b04d461610cfcf2f05bae91338cc", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-a6b5dceb84f6", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:56:05Z", "seed": 1200136385, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "86f9b748ace076986165c5db030c1c28c4be10282074f9d1aa7d38c74fa29791", "run_seconds": 225.56608748435974, "seed": 1200136385, "started_at": "2026-11-22T03:56:07Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"compressed": 0, "inherit_mem": 0, "inherit_nomem": 0, "random_search": 0, "reset_mem": 0, "retrieval": 0, "rootcause": 0, "xfer_mem": 0}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:01:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-a6b5dceb84f6", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "compressed": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "inherit_mem": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "inherit_nomem": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "random_search": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "reset_mem": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "retrieval": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "rootcause": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22], "xfer_mem": [0.3125, 0.649305556, 0.989583333, 0.255555556, 1.0, 0.480905512, 0.871212121, 0.857894737, 0.875, 0.22]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "per_generation_sealed": [0.651196, 0.651196, 0.651196, 0.651196], "prereg_hash": "23714cf42837803b99fee8d4b138da1534e0b04d461610cfcf2f05bae91338cc", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "inherit_nomem", "rootcause"], "result_hash": "86f9b748ace076986165c5db030c1c28c4be10282074f9d1aa7d38c74fa29791", "signature": "8426e1da0cdfd6d08ace8e81f15e4cb530a5565ef35676b2dc5d74f0b5faaad27611c791c78c97e62dd1e03ad6fbe9f5010a0e4edba265a691e08d320f7d1209", "state": "PARTIAL", "summary": {"G0": 0.651196, "compressed": 0.651196, "inherit_mem": 0.651196, "inherit_nomem": 0.651196, "random_search": 0.651196, "reset_mem": 0.651196, "retrieval": 0.651196, "rootcause": 0.651196, "xfer_mem": 0.651196}, "ties": 10, "treatment": "inherit_mem", "wins": 0}, "verified_at": "2026-11-22T04:02:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:14Z", "cpu_seconds": 4.907595429000001, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-21e00a741bf8", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "8791c09416ff102862ad20bd492a2a7950d3033b9fa30e58ec1c7626551d94fb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-21e00a741bf8", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:56:05Z", "seed": 1020621642, "treatment": "gated"}, "priority": 0.3235374504174923, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "01a0748fd75172f3d89cdf546f25f9c5892e1a39fad38c64537debe2438486de", "run_seconds": 6.644667148590088, "seed": 1020621642, "started_at": "2026-11-22T03:56:07Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T03:58:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-21e00a741bf8", "family": "ow_transfer", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"compressed": [1.0, 0.568269], "gated": [1.0, 0.568269], "naive": [1.0, 0.6675], "scratch": [1.0, 0.624135], "ungated": [1.0, 0.58]}, "prereg_hash": "8791c09416ff102862ad20bd492a2a7950d3033b9fa30e58ec1c7626551d94fb", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "01a0748fd75172f3d89cdf546f25f9c5892e1a39fad38c64537debe2438486de", "signature": "35b4841a45c61a326cb59a200b8af5905ba30e125cc25e9faf517593f474d908cb8ae51051ecbcac2706f7d6746c0a51c97b1ab250816494047800247a9b9305", "state": "REJECTED", "summary": {"compressed": 0.568269, "gated": 0.568269, "naive": 0.6675, "scratch": 0.624135, "ungated": 0.58}}, "verified_at": "2026-11-22T03:58:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:20Z", "cpu_seconds": 0.7586597459999993, "created_at": "2026-11-22T03:56:05Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-9650a30183d3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "7b16fc63e134289194e8cadaa0e4ffdd8361243b1231def923d90b55e1206220", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-9650a30183d3", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:56:05Z", "seed": 63282553, "treatment": "gated"}, "priority": 0.2739829519533238, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "b35c6abf818dcaf9527b0d0991dc3b277fdb9c6f2eba41db50d5bbeec35aa561", "run_seconds": 0.8963210582733154, "seed": 63282553, "started_at": "2026-11-22T03:56:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.070919, 0.004896], "effect": -0.033011, "evaluated_at": "2026-11-22T03:58:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-9650a30183d3", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.68115942, 0.631578947, 0.566666667, 0.740740741], "gated": [0.68115942, 0.631578947, 0.566666667, 0.740740741], "naive": [0.782608696, 0.707602339, 0.633333333, 0.851851852], "scratch": [0.739130435, 0.631578947, 0.566666667, 0.814814815], "ungated": [0.782608696, 0.631578947, 0.633333333, 0.666666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7b16fc63e134289194e8cadaa0e4ffdd8361243b1231def923d90b55e1206220", "primary_control": "scratch", "result_hash": "b35c6abf818dcaf9527b0d0991dc3b277fdb9c6f2eba41db50d5bbeec35aa561", "signature": "da53ad8e2ae836459b9345982dc875e0a139aeeb4fb6246152b6cb0f5033a511bc969d885f1faffa6c7de13e5dfc69c6fdd6db92dcf15591b55465d1d4ca6b03", "state": "VERIFIED", "summary": {"compressed": 0.655036, "gated": 0.655036, "naive": 0.743849, "scratch": 0.688048, "ungated": 0.678547}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T03:58:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:51Z", "cpu_seconds": 21.019903713999998, "created_at": "2026-11-22T03:56:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5e6b179762cf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "15b1d66254ab847771aed3f1375795244de7094f8cb9098d82f5a1c5e3ee1317", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5e6b179762cf", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:56:22Z", "seed": 326769828, "treatment": "full"}, "priority": 0.30313619158075583, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "177b511ac1a21a36135747d13e7b81962a3439a8eb51709f6f85c95e649e304e", "run_seconds": 25.660141468048096, "seed": 326769828, "started_at": "2026-11-22T03:56:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.052693, 0.947307], "effect": 0.5, "evaluated_at": "2026-11-22T03:58:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5e6b179762cf", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 0.25, 0.0], "fixed_policy": [0.0, 0.0, 1.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 0.75, 0.75, 0.25], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "15b1d66254ab847771aed3f1375795244de7094f8cb9098d82f5a1c5e3ee1317", "primary_control": "e141_best", "result_hash": "177b511ac1a21a36135747d13e7b81962a3439a8eb51709f6f85c95e649e304e", "signature": "567f517db92fcbafe4be7efdc9004fa84e548852a9c6c410f256d1001c0d10794a46be2fe6b41b8c86db762f64d8127c920118fa4368b3e7568f043d7cbac409", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 0.6875, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T03:58:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:52Z", "cpu_seconds": 1.00956393, "created_at": "2026-11-22T03:56:22Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-432f1d23721d", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "4a7e01cd079d47917f7d0cc4c425ca2904830f8def8b38450d7994b1d44a8efc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-432f1d23721d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:56:22Z", "seed": 829196284, "treatment": "full"}, "priority": 0.27156704728647235, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "92723f2611b4e7360da83d8a973d9dfbe0bdcbf3f9c6a7433fe4fe89df90fdd7", "run_seconds": 1.0618493556976318, "seed": 829196284, "started_at": "2026-11-22T03:57:51Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:00:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-432f1d23721d", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.372179], "frozen": [1.0, 0.161804], "full": [1.0, 0.397227], "retrieval": [1.0, 0.411496], "rules_nodetect": [1.0, 0.454246]}, "prereg_hash": "4a7e01cd079d47917f7d0cc4c425ca2904830f8def8b38450d7994b1d44a8efc", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "92723f2611b4e7360da83d8a973d9dfbe0bdcbf3f9c6a7433fe4fe89df90fdd7", "signature": "7529a20f713cad52d3b51480ce93228f844c2415bb1f4a48a8ac0b4bb0bf41f62131503b2bd6e0dac1eb38c67f6a1b237f44d0b1b9866abffb51e4b4d2efe70e", "state": "REJECTED", "summary": {"facts": 0.372179, "frozen": 0.161804, "full": 0.397227, "retrieval": 0.411496, "rules_nodetect": 0.454246}}, "verified_at": "2026-11-22T04:00:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:40Z", "cpu_seconds": 0.610190969, "created_at": "2026-11-22T03:56:22Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4a468a3f11a6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "3e41e4ba7b4602ff87eb93c330cffce530d575ec3c91fc7d5c228428b71b067e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4a468a3f11a6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:56:22Z", "seed": 567113061, "treatment": "full"}, "priority": 0.2735071880608744, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fb86472a85031a802cbde2075ddb3085cf50179693857d21a51739da05bde2db", "run_seconds": 0.6446080207824707, "seed": 567113061, "started_at": "2026-11-22T03:57:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.064261, 0.072109], "effect": 0.003924, "evaluated_at": "2026-11-22T04:00:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4a468a3f11a6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.547368421, 0.427835052, 0.5375, 0.734848485], "frozen": [0.375438596, 0.074742268, 0.28125, 0.590909091], "full": [0.592982456, 0.5, 0.51875, 0.651515152], "retrieval": [0.592982456, 0.605670103, 0.575, 0.590909091], "rules_nodetect": [0.585964912, 0.5, 0.73125, 0.454545455]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "3e41e4ba7b4602ff87eb93c330cffce530d575ec3c91fc7d5c228428b71b067e", "primary_control": "facts", "result_hash": "fb86472a85031a802cbde2075ddb3085cf50179693857d21a51739da05bde2db", "signature": "e5de7dac35f84cef5bc9a29a21385e3c9f0508189f49d34c78492cecf2869c070284b73f1a0758a8fb4f4fd2a40417d5ffa29a0aad1a280fc80e230106157d0f", "state": "VERIFIED", "summary": {"facts": 0.561888, "frozen": 0.330585, "full": 0.565812, "retrieval": 0.59114, "rules_nodetect": 0.56794}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:00:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:25Z", "cpu_seconds": 2.497843883999998, "created_at": "2026-11-22T03:56:22Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5d8b4216a6e2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "a75a89cfa1842bf7d6ea309d563aa8c112312f0b0749e7d5f3e4a205f23ec0b4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5d8b4216a6e2", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:56:22Z", "seed": 341109237, "treatment": "gated"}, "priority": 0.35910715078258165, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "af08f2d8b080df0b2ef7ebfa4ca2abd3d53e0e2e1003d7e24c2dae8c8db0e781", "run_seconds": 2.6379616260528564, "seed": 341109237, "started_at": "2026-11-22T03:56:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.067007, 0.029507], "effect": -0.01875, "evaluated_at": "2026-11-22T03:58:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-5d8b4216a6e2", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.696428571, 0.53, 0.541666667, 0.608333333], "gated": [0.696428571, 0.53, 0.541666667, 0.608333333], "naive": [0.758928571, 0.74, 0.641666667, 0.791666667], "scratch": [0.696428571, 0.53, 0.633333333, 0.591666667], "ungated": [0.696428571, 0.69, 0.641666667, 0.7]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "a75a89cfa1842bf7d6ea309d563aa8c112312f0b0749e7d5f3e4a205f23ec0b4", "primary_control": "scratch", "result_hash": "af08f2d8b080df0b2ef7ebfa4ca2abd3d53e0e2e1003d7e24c2dae8c8db0e781", "signature": "f685ce10e757738d58d45bf6685e94bc7ba0897a11e8e759370c1f2b9f2af512671c43bde730bcd8710ec02fc8c55e4f63a18b21e1d9d29c67c4b7655262820d", "state": "VERIFIED", "summary": {"compressed": 0.594107, "gated": 0.594107, "naive": 0.733065, "scratch": 0.612857, "ungated": 0.682024}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T03:58:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:57Z", "cpu_seconds": 3.6230105270000004, "created_at": "2026-11-22T03:56:37Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-13d559243a9b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "34e1a01fefb5f8e40c6e8bd87a526aa00a1d6e8cf1328b542399c36bcaf13d0d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-13d559243a9b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:56:37Z", "seed": 1464716685, "treatment": "frontier"}, "priority": 0.2687498756142826, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ae8b8dda47402229998bae2f16f657fb6769f2cdd7a326697dc6235528e1af1e", "run_seconds": 4.062561273574829, "seed": 1464716685, "started_at": "2026-11-22T03:57:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.195002, 0.021694], "effect": -0.086654, "evaluated_at": "2026-11-22T04:00:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-13d559243a9b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.415419162, 0.401785714, 0.831521739, 0.741319444], "frontier": [0.368263473, 0.318877551, 0.543478261, 0.692708333], "goal": [0.398952096, 0.30994898, 0.429347826, 0.642361111], "hypdisc": [0.380988024, 0.352040816, 0.690217391, 0.675347222], "infogain": [0.368263473, 0.318877551, 0.543478261, 0.692708333], "prederr": [0.368263473, 0.318877551, 0.543478261, 0.692708333], "random": [0.381736527, 0.400510204, 0.788043478, 0.699652778], "risk": [0.368263473, 0.318877551, 0.543478261, 0.692708333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "34e1a01fefb5f8e40c6e8bd87a526aa00a1d6e8cf1328b542399c36bcaf13d0d", "primary_control": "random", "result_hash": "ae8b8dda47402229998bae2f16f657fb6769f2cdd7a326697dc6235528e1af1e", "signature": "ee8147be106f42d502329de87dfd9a9144bd0fca1e7bf3fc88bbea002c22221a33bd1a2ebd9eb4de47a251c836d867faa22a9e7481803a62934d6cd635226000", "state": "VERIFIED", "summary": {"count": 0.597512, "frontier": 0.480832, "goal": 0.445153, "hypdisc": 0.524648, "infogain": 0.480832, "prederr": 0.480832, "random": 0.567486, "risk": 0.480832}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:00:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:56:40Z", "cpu_seconds": 0.516407488, "created_at": "2026-11-22T03:56:37Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ac5046f08435", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "9f1002acc212c84e79fdec68fbc33851f53017d7fad44b03153d5b83a0c0e70a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ac5046f08435", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:56:37Z", "seed": 583297270, "treatment": "full"}, "priority": 0.27576525507054195, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6011a0a05d0104379fa212906f2ddd4a6daba6ea04b8be51b96778e3013c209b", "run_seconds": 0.5220048427581787, "seed": 583297270, "started_at": "2026-11-22T03:56:39Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T03:58:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ac5046f08435", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.5, 0.5], "full": [0.5, 0.0, 1.0, 1.0], "no_world_model": [0.0, 0.0, 1.0, 1.0], "random": [0.5, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9f1002acc212c84e79fdec68fbc33851f53017d7fad44b03153d5b83a0c0e70a", "primary_control": "e141_best", "result_hash": "6011a0a05d0104379fa212906f2ddd4a6daba6ea04b8be51b96778e3013c209b", "signature": "2ff156e510697513fab05c2f25611776184e25dc0d207278e92c15ae9884d8deb037832db747e85bcb77226025b424ee6335e5fb1dd6744b11a94adf54095f09", "state": "PARTIAL", "summary": {"e141_best": 0.625, "fixed_policy": 0.3125, "full": 0.625, "no_world_model": 0.5, "random": 0.625}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T03:58:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:40Z", "cpu_seconds": 0.472760549, "created_at": "2026-11-22T03:56:53Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-771efb9bc31f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "7dc918cd478db2cc42b2b4ac12425a1c5a355c233957b4ffc5be96fbd1e49346", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-771efb9bc31f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:56:53Z", "seed": 1137268603, "treatment": "full"}, "priority": 0.2842078716801339, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a2a0cbdb2421d952c3cce28d5807a19bf31d5580aabd2a0a0bcc7a4ce4c9a4a8", "run_seconds": 0.5010671615600586, "seed": 1137268603, "started_at": "2026-11-22T03:57:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023318, 0.004898], "effect": -0.00921, "evaluated_at": "2026-11-22T04:00:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-771efb9bc31f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.424836601, 0.424836601, 0.307189542, 0.287878788], "frozen": [0.405228758, 0.418300654, 0.274509804, 0.265151515], "full": [0.424836601, 0.418300654, 0.307189542, 0.257575758], "retrieval": [0.424836601, 0.385620915, 0.300653595, 0.295454545], "rules_nodetect": [0.424836601, 0.405228758, 0.326797386, 0.287878788]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7dc918cd478db2cc42b2b4ac12425a1c5a355c233957b4ffc5be96fbd1e49346", "primary_control": "facts", "result_hash": "a2a0cbdb2421d952c3cce28d5807a19bf31d5580aabd2a0a0bcc7a4ce4c9a4a8", "signature": "c67d95e09b28208457976404879823d2610712bfc56961a45e344896329b1fe0209eaed9aba5a169689df21059bf9bf12a610c9224aaf410081919624e860008", "state": "VERIFIED", "summary": {"facts": 0.361185, "frozen": 0.340798, "full": 0.351976, "retrieval": 0.351641, "rules_nodetect": 0.361185}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:00:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:48Z", "cpu_seconds": 2.247992645, "created_at": "2026-11-22T03:57:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-762ac7f63728", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "82dc6989b91eeb599505ceb9e9c05b8d6b2125317c0d160e4bf1c3ce759a96e5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-762ac7f63728", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:57:43Z", "seed": 382483404, "treatment": "full"}, "priority": 0.2955890434836184, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4d885d0e5ab8c5cbc74f2e30fa953ad5253059fd31bbeb115c14ce98a14e34dc", "run_seconds": 2.6846466064453125, "seed": 382483404, "started_at": "2026-11-22T03:57:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.053448, 0.32494], "effect": 0.189194, "evaluated_at": "2026-11-22T04:00:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-762ac7f63728", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.381443299, 0.92211838, 0.415789474, 0.286592866], "frozen": [0.127835052, 0.267912773, 0.226315789, 0.225092251], "full": [0.604123711, 0.985981308, 0.515789474, 0.656826568], "retrieval": [0.655670103, 0.992990654, 0.645614035, 0.699876999], "rules_nodetect": [0.356701031, 0.985981308, 0.435087719, 0.210332103]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "82dc6989b91eeb599505ceb9e9c05b8d6b2125317c0d160e4bf1c3ce759a96e5", "primary_control": "facts", "result_hash": "4d885d0e5ab8c5cbc74f2e30fa953ad5253059fd31bbeb115c14ce98a14e34dc", "signature": "b1d078e8c18116805662a48c2b49a1642625a4c4cdee0b9e64e574c754f40868b528efbe677222ff67239aa7e4304dc58b90a4da619f2616515a998006d7f80a", "state": "VERIFIED", "summary": {"facts": 0.501486, "frozen": 0.211789, "full": 0.69068, "retrieval": 0.748538, "rules_nodetect": 0.497026}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:00:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:50Z", "cpu_seconds": 113.027616411, "created_at": "2026-11-22T03:57:43Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-629905271c54", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "5a0e6b950da43aacf0da84722985d460001f4e428c88dc9a9b7bea5679d64371", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-629905271c54", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T03:57:43Z", "seed": 1748403966, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "b51186fb7dd21ab911fb15f24a520b116da61f827b65d06e97e530129f77c60c", "run_seconds": 125.61202383041382, "seed": 1748403966, "started_at": "2026-11-22T03:57:45Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 3, "inherit_mem": 3, "inherit_nomem": 3, "random_search": 1, "reset_mem": 3, "retrieval": 2, "rootcause": 2, "xfer_mem": 2}, "better": "higher", "ci95": [-0.038501, 0.24676], "effect": 0.10413, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:00:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-629905271c54", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0, "risk": 0.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.75}, "random_search": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.191, "history2": false, "mix": {"random": 1.0}, "planner_depth": 22, "reset_patience": 0, "rule_threshold": 0.771}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.6}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0, "risk": 0.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.212962963, 1.0, 1.0, 0.560457516, 0.736666667, 0.550257732, 0.942028986, 0.483333333, 0.547222222, 1.0], "compressed": [0.925925926, 1.0, 1.0, 0.908496732, 0.832083333, 0.803801546, 0.963768116, 0.255952381, 0.547222222, 1.0], "inherit_mem": [0.87962963, 1.0, 0.98, 0.911764706, 0.830833333, 0.544136598, 0.97826087, 0.402380952, 0.547222222, 1.0], "inherit_nomem": [0.273148148, 1.0, 0.986666667, 0.852941176, 0.825416667, 0.550257732, 0.942028986, 0.301190476, 0.547222222, 1.0], "random_search": [0.300925926, 1.0, 0.986666667, 0.852941176, 0.785, 0.550257732, 0.942028986, 0.461904762, 0.502777778, 1.0], "reset_mem": [0.87962963, 1.0, 0.98, 0.911764706, 0.830833333, 0.544136598, 0.97826087, 0.402380952, 0.547222222, 1.0], "retrieval": [0.212962963, 1.0, 0.973333333, 0.87254902, 0.825416667, 0.791881443, 0.963768116, 0.483333333, 0.502777778, 1.0], "rootcause": [0.212962963, 1.0, 1.0, 0.908496732, 0.832083333, 0.803801546, 0.963768116, 0.563095238, 0.547222222, 1.0], "xfer_mem": [0.185185185, 1.0, 0.993333333, 0.908496732, 0.750833333, 0.544458763, 1.0, 0.428571429, 0.547222222, 1.0]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.301, "losses": 3, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "per_generation_sealed": [0.703293, 0.735639, 0.745082, 0.807423], "prereg_hash": "5a0e6b950da43aacf0da84722985d460001f4e428c88dc9a9b7bea5679d64371", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "retrieval", "rootcause"], "result_hash": "b51186fb7dd21ab911fb15f24a520b116da61f827b65d06e97e530129f77c60c", "signature": "0219c34dcb0f13dfc00b7ea13d81ebf194e72c85f4fdd431fb51ab3bcc2dd8f35a2732dca4c0bf51fc3f74a2ff14098b2512c4f6db64a5b088ac6ba84f04480e", "state": "VERIFIED", "summary": {"G0": 0.703293, "compressed": 0.823725, "inherit_mem": 0.807423, "inherit_nomem": 0.727887, "random_search": 0.73825, "reset_mem": 0.807423, "retrieval": 0.762602, "rootcause": 0.783143, "xfer_mem": 0.73581}, "ties": 3, "treatment": "inherit_mem", "wins": 4}, "verified_at": "2026-11-22T04:01:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:57:46Z", "cpu_seconds": 0.702443237, "created_at": "2026-11-22T03:57:43Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-c0bd0e8912de", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "878f850e117087210c1b969fdb4b2914bf63eab1c4f315db3f3f6a090143a603", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-c0bd0e8912de", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:57:43Z", "seed": 1507655962, "treatment": "gated"}, "priority": 0.2719679406701825, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d68836ae8fe819d03cd610876f0a11fc6a31a2eff9698d474914f1f882118e15", "run_seconds": 0.8557686805725098, "seed": 1507655962, "started_at": "2026-11-22T03:57:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.009673, 0.003137], "effect": -0.003268, "evaluated_at": "2026-11-22T04:00:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-c0bd0e8912de", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.274509804, 0.248366013, 0.176470588, 0.209150327], "gated": [0.274509804, 0.248366013, 0.176470588, 0.209150327], "naive": [0.274509804, 0.248366013, 0.215686275, 0.22875817], "scratch": [0.274509804, 0.248366013, 0.176470588, 0.222222222], "ungated": [0.215686275, 0.202614379, 0.14379085, 0.104575163]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "878f850e117087210c1b969fdb4b2914bf63eab1c4f315db3f3f6a090143a603", "primary_control": "scratch", "result_hash": "d68836ae8fe819d03cd610876f0a11fc6a31a2eff9698d474914f1f882118e15", "signature": "3cd5f1c9226826dda499ac36df86ec823e15d15e80a9f0d8ac2bd2c136b785a0ad871dc5b0cbdf3e0847fb8460783077da5b55b400871bc16e9499b6db0fd507", "state": "VERIFIED", "summary": {"compressed": 0.227124, "gated": 0.227124, "naive": 0.24183, "scratch": 0.230392, "ungated": 0.166667}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:00:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:39Z", "cpu_seconds": 0.358367404, "created_at": "2026-11-22T03:58:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-47cb67d3ac2f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "852aa78da4b2f9e5666f16363cd8fce8ab0fa677a853dace6fba1dc1774ce6b0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-47cb67d3ac2f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:58:00Z", "seed": 1123625704, "treatment": "frontier"}, "priority": 0.26184545014505817, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "606d15bf7d90e538db13c865920641afd1c99d756134fb29a602186ea609d535", "run_seconds": 0.48227524757385254, "seed": 1123625704, "started_at": "2026-11-22T03:59:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.116032, 0.156696], "effect": 0.136364, "evaluated_at": "2026-11-22T04:00:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-47cb67d3ac2f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.977272727, 0.977272727, 0.977272727, 1.0], "frontier": [0.875, 0.943181818, 0.965909091, 1.0], "goal": [0.738636364, 0.693181818, 0.704545455, 0.840909091], "hypdisc": [0.772727273, 0.784090909, 0.772727273, 0.875], "infogain": [0.875, 0.943181818, 0.965909091, 1.0], "prederr": [0.875, 0.943181818, 0.965909091, 1.0], "random": [0.761363636, 0.795454545, 0.806818182, 0.875], "risk": [0.875, 0.943181818, 0.965909091, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "852aa78da4b2f9e5666f16363cd8fce8ab0fa677a853dace6fba1dc1774ce6b0", "primary_control": "random", "result_hash": "606d15bf7d90e538db13c865920641afd1c99d756134fb29a602186ea609d535", "signature": "5966c75fa9665bdf8e3f48f521a578472e17a6101b59686b2bc68f3ba02f3dc0635ec021a235989fde0ecaac0585d3f910c215cac65cd6bb0dba65ef0ce54f0e", "state": "VERIFIED", "summary": {"count": 0.982955, "frontier": 0.946023, "goal": 0.744318, "hypdisc": 0.801136, "infogain": 0.946023, "prederr": 0.946023, "random": 0.809659, "risk": 0.946023}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:00:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:19Z", "cpu_seconds": 0.258861042999996, "created_at": "2026-11-22T03:58:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-78566b6851cc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "2238c72f3cea4932202034990cef2133564104d8622da6efdd3a8e1e6d02d2fd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-78566b6851cc", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:58:00Z", "seed": 1125033118, "treatment": "frontier"}, "priority": 0.26118123033327284, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a75677d0da56e5d8627002c36c5c409b69d08882efe3319c660fe549c84a2f58", "run_seconds": 0.2887849807739258, "seed": 1125033118, "started_at": "2026-11-22T04:00:18Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-78566b6851cc", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "2238c72f3cea4932202034990cef2133564104d8622da6efdd3a8e1e6d02d2fd", "primary_control": "random", "result_hash": "a75677d0da56e5d8627002c36c5c409b69d08882efe3319c660fe549c84a2f58", "signature": "d90fd944ef2b968a7c5be3d8a53ff45108dc0532988e1d9470b8cc49393531fe1773fe5d4b7acd58bbff7b24532ba1a41c1fa5ecebd131656ddd97688e966b0f", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:02:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:17Z", "cpu_seconds": 1.265440988, "created_at": "2026-11-22T03:58:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4449723c59b4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "1aacf6062b95256e20db4635b8e91ac83fd282e3eb4f0eb40b8bddc3896dc530", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4449723c59b4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:58:00Z", "seed": 135501959, "treatment": "full"}, "priority": 0.2633144260184948, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f3bc2adb7cdf83a0a1b8f1630ec033243783a397c395f44b89491e71a1b83a15", "run_seconds": 1.3560833930969238, "seed": 135501959, "started_at": "2026-11-22T03:58:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.020898, 0.047722], "effect": 0.013412, "evaluated_at": "2026-11-22T04:00:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4449723c59b4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.411347518, 0.462765957, 0.737588652, 0.542857143], "frozen": [0.108156028, 0.106382979, 0.106382979, 0.1], "full": [0.418439716, 0.459219858, 0.723404255, 0.607142857], "retrieval": [0.430851064, 0.468085106, 0.723404255, 0.602380952], "rules_nodetect": [0.418439716, 0.719858156, 0.723404255, 0.607142857]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "1aacf6062b95256e20db4635b8e91ac83fd282e3eb4f0eb40b8bddc3896dc530", "primary_control": "facts", "result_hash": "f3bc2adb7cdf83a0a1b8f1630ec033243783a397c395f44b89491e71a1b83a15", "signature": "b422e889532cf55f2111d73466b35f6a88635a4604a5b926ef09887a2a87354dd580b25f1a778de3926c5c2b91e5fafa20107392df79800d54ecc4a99519b20a", "state": "VERIFIED", "summary": {"facts": 0.53864, "frozen": 0.10523, "full": 0.552052, "retrieval": 0.55618, "rules_nodetect": 0.617211}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:00:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:10Z", "cpu_seconds": 0.7178346849999997, "created_at": "2026-11-22T03:58:00Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3a0a6be87fe4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "0ab8782ebd7bbd76ee98ce13dd6ece45a1e866d11c16d8782fedcf7f1d81dd02", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3a0a6be87fe4", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:58:00Z", "seed": 203949293, "treatment": "gated"}, "priority": 0.2638285016877016, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "89a14a5106c986f19b8df3bfd3ce1ab4ad2d97ed2666d32c3671697977e9e5bd", "run_seconds": 0.73830246925354, "seed": 203949293, "started_at": "2026-11-22T03:58:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01451, 0.004706], "effect": -0.004902, "evaluated_at": "2026-11-22T04:00:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-3a0a6be87fe4", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.31372549, 0.294117647, 0.437908497, 0.31372549], "gated": [0.31372549, 0.294117647, 0.437908497, 0.31372549], "naive": [0.31372549, 0.274509804, 0.307189542, 0.31372549], "scratch": [0.31372549, 0.294117647, 0.437908497, 0.333333333], "ungated": [0.333333333, 0.326797386, 0.267973856, 0.254901961]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "0ab8782ebd7bbd76ee98ce13dd6ece45a1e866d11c16d8782fedcf7f1d81dd02", "primary_control": "scratch", "result_hash": "89a14a5106c986f19b8df3bfd3ce1ab4ad2d97ed2666d32c3671697977e9e5bd", "signature": "285ef0c1e6a62756a0eef0e529ad34c326a19bfb2554be08b964f9ce1b4c017d11ca41ebff7c8f2da5b3908820a7b6ae1e51f51e6d0f232e3f36e01cf96c9104", "state": "VERIFIED", "summary": {"compressed": 0.339869, "gated": 0.339869, "naive": 0.302288, "scratch": 0.344771, "ungated": 0.295752}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:00:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:44Z", "cpu_seconds": 0.4282201430000008, "created_at": "2026-11-22T03:58:00Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e931f5082c1e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "prereg_hash": "301519a30aeedd4da09e5f27e0722bf60c16083552e7b4b3c51f3a7f1bfad3a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e931f5082c1e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:58:00Z", "seed": 2141441819, "treatment": "gated"}, "priority": 0.2631723137181234, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d5526534896f3704359556b05adcd03c4fa629ed869e15b6ae4a4ef4e7db8634", "run_seconds": 0.49033403396606445, "seed": 2141441819, "started_at": "2026-11-22T03:58:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.032, 0.098667], "effect": 0.033333, "evaluated_at": "2026-11-22T04:00:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-e931f5082c1e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.733333333, 1.0, 0.616666667, 0.45], "gated": [0.733333333, 1.0, 0.75, 0.45], "naive": [0.416666667, 0.216666667, 0.75, 0.166666667], "scratch": [0.733333333, 1.0, 0.616666667, 0.45], "ungated": [0.416666667, 0.1, 0.75, 0.166666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "301519a30aeedd4da09e5f27e0722bf60c16083552e7b4b3c51f3a7f1bfad3a2", "primary_control": "scratch", "result_hash": "d5526534896f3704359556b05adcd03c4fa629ed869e15b6ae4a4ef4e7db8634", "signature": "1ab251138135f380f334e40368d1202181d679c04d2688fd52f19f747bff84e15ebc1520bee8f6682a3744c4796f43f7c2d4b2b4393816a7c1f768db7acdd000", "state": "VERIFIED", "summary": {"compressed": 0.7, "gated": 0.733333, "naive": 0.3875, "scratch": 0.7, "ungated": 0.358333}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:00:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:33Z", "cpu_seconds": 1.452280076000001, "created_at": "2026-11-22T03:58:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-14f01f28e41e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "05dceff2890bf57d42c1e3d32d100e6a82d4174fe533ff4d1a202270f8968310", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-14f01f28e41e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:17Z", "seed": 2022792288, "treatment": "full"}, "priority": 0.2680526191649608, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "edd86b4fbab057170b81ee5d19649fb006989e4f9277387a1541f9f8aa588beb", "run_seconds": 1.870220422744751, "seed": 2022792288, "started_at": "2026-11-22T03:58:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T04:00:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-14f01f28e41e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 0.5, 1.0], "fixed_policy": [1.0, 1.0, 0.75, 1.0], "full": [1.0, 0.5, 0.75, 0.75], "no_world_model": [1.0, 0.25, 0.25, 0.75], "random": [1.0, 0.5, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "05dceff2890bf57d42c1e3d32d100e6a82d4174fe533ff4d1a202270f8968310", "primary_control": "e141_best", "result_hash": "edd86b4fbab057170b81ee5d19649fb006989e4f9277387a1541f9f8aa588beb", "signature": "7e89d817d4bb151798cf1af91b5e3fcd221dbdfe7d6a4d70217d62e07ea21f435e466eb5956f745cc35b4f9c00f6d62684de1edf838e00023a8cd1e79e481b0a", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.9375, "full": 0.75, "no_world_model": 0.5625, "random": 0.8125}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:00:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:28Z", "cpu_seconds": 0.7761253689999998, "created_at": "2026-11-22T03:58:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-825053ddfdaa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "94286e2a29aaff71b746397c06554126184f233b09a92243a20376b32d4fee51", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-825053ddfdaa", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:17Z", "seed": 24887025, "treatment": "full"}, "priority": 0.26867055981432175, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "fd4dbf3ecf238364e5e2cb3ddf893a7e85ad19dbb6741526d068284923adefc7", "run_seconds": 1.0739047527313232, "seed": 24887025, "started_at": "2026-11-22T03:58:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:00:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-825053ddfdaa", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 0.0, 0.5], "fixed_policy": [0.0, 0.25, 0.25, 0.0], "full": [1.0, 0.75, 0.0, 0.0], "no_world_model": [1.0, 0.0, 0.0, 0.0], "random": [1.0, 0.75, 0.25, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "94286e2a29aaff71b746397c06554126184f233b09a92243a20376b32d4fee51", "primary_control": "e141_best", "result_hash": "fd4dbf3ecf238364e5e2cb3ddf893a7e85ad19dbb6741526d068284923adefc7", "signature": "323ee38703abbfb7412e2afea82cb9786b72c83d0133b769d2f7d687a4f2bf389b17d6f2cbef21944e0f6db646fa2d891382013529dc2328753851823606000e", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.125, "full": 0.4375, "no_world_model": 0.25, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:00:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:22Z", "cpu_seconds": 0.465070238, "created_at": "2026-11-22T03:58:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9a1bb98f7e7c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "ca15fa1f914fd9acce316439d6a307fc8476f01ba427d09f906b740cf0b209df", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9a1bb98f7e7c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:17Z", "seed": 1683708379, "treatment": "full"}, "priority": 0.2808588978353637, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d73bbf6df71f756fe9cc2a35c519f5c0682ac9d59ce4b8e032e397d7ca310131", "run_seconds": 0.509291410446167, "seed": 1683708379, "started_at": "2026-11-22T03:58:22Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:00:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9a1bb98f7e7c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.0, 0.0, 0.0], "fixed_policy": [0.5, 0.25, 0.5, 0.0], "full": [0.25, 0.0, 0.0, 0.0], "no_world_model": [0.25, 0.0, 0.5, 0.0], "random": [0.25, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ca15fa1f914fd9acce316439d6a307fc8476f01ba427d09f906b740cf0b209df", "primary_control": "e141_best", "result_hash": "d73bbf6df71f756fe9cc2a35c519f5c0682ac9d59ce4b8e032e397d7ca310131", "signature": "ca0829341f6d9c975a00d7babcac01087cc85e545f97d3217a3c3079932056397eeb12b4ad370c06f80b9632360c5964316f09a189b2788dbd358362b6d0bd08", "state": "PARTIAL", "summary": {"e141_best": 0.0625, "fixed_policy": 0.3125, "full": 0.0625, "no_world_model": 0.1875, "random": 0.0625}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:00:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:34Z", "cpu_seconds": 0.7984181450000003, "created_at": "2026-11-22T03:58:17Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ba713f188e9a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "9bddb3df7b0d5421117da770b0e62b84075246eca31eb7d879f5dd4f29aecd1c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ba713f188e9a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:17Z", "seed": 151031768, "treatment": "full"}, "priority": 0.26671100370225254, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-0a8c727b866e", "result_hash": "b294c5e88a15d9cad6350e8e4a1aa441a535b7582fb3733afe6e97de90b776d5", "run_seconds": 1.182680368423462, "seed": 151031768, "started_at": "2026-11-22T03:58:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:00:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ba713f188e9a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "9bddb3df7b0d5421117da770b0e62b84075246eca31eb7d879f5dd4f29aecd1c", "primary_control": "e141_best", "replicated": false, "replication_of": "E141-ow_goal-0a8c727b866e", "result_hash": "b294c5e88a15d9cad6350e8e4a1aa441a535b7582fb3733afe6e97de90b776d5", "signature": "016a621a5083fcf6488a335a719309610aaf1b362f77762c13d5dd475fb8f83531a182858a4848275da221f986e64c74ed2132cc57e482426cac1206b8421500", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:00:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:26Z", "cpu_seconds": 0.9008257989999997, "created_at": "2026-11-22T03:58:17Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3f2957ce0992", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "d8586f63e518e8cd4df397912f831d67f60f52bbafc8a2f2b9501c36607d0d3c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3f2957ce0992", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:58:17Z", "seed": 705382615, "treatment": "full"}, "priority": 0.2713829043513402, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "545efb986f0a405667bb9c0dd188405f648eb96bccfeb21761f817fe1ea4124e", "run_seconds": 1.298396110534668, "seed": 705382615, "started_at": "2026-11-22T03:58:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.036877, -0.024374], "effect": -0.030626, "evaluated_at": "2026-11-22T04:00:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3f2957ce0992", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.487234043, 0.540425532, 0.540659341, 0.46097561], "frozen": [0.127659574, 0.127659574, 0.131868132, 0.131707317], "full": [0.465957447, 0.508510638, 0.505494505, 0.426829268], "retrieval": [0.529787234, 0.49787234, 0.531868132, 0.46097561], "rules_nodetect": [0.465957447, 0.734042553, 0.705494505, 0.426829268]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "d8586f63e518e8cd4df397912f831d67f60f52bbafc8a2f2b9501c36607d0d3c", "primary_control": "facts", "result_hash": "545efb986f0a405667bb9c0dd188405f648eb96bccfeb21761f817fe1ea4124e", "signature": "c7481af6cbeadbb5d3d81404ca0a0103e70d37b60a2702f5a4557fb69efe61df231eb5b88d5f099e4364a2016f9d4a5b4a7361b81968694dfd5c108d87c6c806", "state": "VERIFIED", "summary": {"facts": 0.507324, "frozen": 0.129724, "full": 0.476698, "retrieval": 0.505126, "rules_nodetect": 0.583081}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:00:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:20Z", "cpu_seconds": 1.1832250579999997, "created_at": "2026-11-22T03:58:17Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8ef122cf7669", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "961301613ce3b9f4e734a01b5a8da03ae9417426ae3bbc027cbb6e898fd66fb7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8ef122cf7669", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:58:17Z", "seed": 844336807, "treatment": "gated"}, "priority": 0.3560754093671563, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f15726879c7df5eaf8a42a3d9cb1d422c330cd17449ebbfc0db46365ba5aadb4", "run_seconds": 1.493685245513916, "seed": 844336807, "started_at": "2026-11-22T03:58:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.148, 0.048], "effect": -0.05, "evaluated_at": "2026-11-22T04:00:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-8ef122cf7669", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.714285714, 0.543478261, 0.733333333, 0.44], "gated": [0.714285714, 0.543478261, 0.733333333, 0.44], "naive": [0.776785714, 0.586956522, 0.75, 0.73], "scratch": [0.714285714, 0.543478261, 0.733333333, 0.64], "ungated": [0.714285714, 0.586956522, 0.75, 0.73]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "961301613ce3b9f4e734a01b5a8da03ae9417426ae3bbc027cbb6e898fd66fb7", "primary_control": "scratch", "result_hash": "f15726879c7df5eaf8a42a3d9cb1d422c330cd17449ebbfc0db46365ba5aadb4", "signature": "604ee87e27b755741849ad928ef4e2d19675e5fbe0c98ce74f017789a1a0f6ddb062f0d8ca5fa69fc20030d3334b067813705c74a424ed4f7ddf025ae92a9b09", "state": "VERIFIED", "summary": {"compressed": 0.607774, "gated": 0.607774, "naive": 0.710936, "scratch": 0.657774, "ungated": 0.695311}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:00:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:25Z", "cpu_seconds": 0.6788641260000001, "created_at": "2026-11-22T03:58:34Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-10e82d9ec388", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "82442a4f91f5815819161a2369e1132577a39efbca18c6919299f739822f7864", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-10e82d9ec388", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:34Z", "seed": 1550348811, "treatment": "full"}, "priority": 0.26071081097025367, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-283d71b3a883", "result_hash": "9b51cfd5ffae46ba2e51894ec70c956d33aa6eeeba706d6770c9c124c7e827e6", "run_seconds": 0.7074184417724609, "seed": 1550348811, "started_at": "2026-11-22T04:00:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.12, 0.37], "effect": 0.125, "evaluated_at": "2026-11-22T04:02:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-10e82d9ec388", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.0, 0.0], "fixed_policy": [0.0, 0.25, 0.0, 0.5], "full": [1.0, 1.0, 0.0, 0.5], "no_world_model": [1.0, 1.0, 0.5, 0.25], "random": [1.0, 1.0, 0.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "82442a4f91f5815819161a2369e1132577a39efbca18c6919299f739822f7864", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-283d71b3a883", "result_hash": "9b51cfd5ffae46ba2e51894ec70c956d33aa6eeeba706d6770c9c124c7e827e6", "signature": "572effab2dd5c5f4789be41507e1592c597711284438d88b4738b4a3dffc2cbd066e848f77137e0f5594f36208d0452db296ca96dfdcf4400e216f06faed3f01", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.1875, "full": 0.625, "no_world_model": 0.6875, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:02:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:19Z", "cpu_seconds": 0.6039790389999999, "created_at": "2026-11-22T03:58:34Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-41ea9382afea", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "59ce6f592f1da6b21b17f90763f62d26d28e54cff0da249b64b518e624899306", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-41ea9382afea", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:34Z", "seed": 386345421, "treatment": "full"}, "priority": 0.26166570133040445, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e6c8a05d018d77e2a3b210c0ebd067b0b7b7feedb4847eb36d54316d54894ae5", "run_seconds": 0.6673369407653809, "seed": 386345421, "started_at": "2026-11-22T04:00:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.815, 1.06], "effect": 0.9375, "evaluated_at": "2026-11-22T04:02:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-41ea9382afea", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.0, 0.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [0.5, 0.5, 0.5, 0.25], "random": [1.0, 0.5, 0.75, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "59ce6f592f1da6b21b17f90763f62d26d28e54cff0da249b64b518e624899306", "primary_control": "e141_best", "result_hash": "e6c8a05d018d77e2a3b210c0ebd067b0b7b7feedb4847eb36d54316d54894ae5", "signature": "dd8490865083321d3b4e00ed29ea38284471734ec9389e5a525be7797d57099c1dcced2b3060582ea1c35b4449a556d047d98c5215ae213fad7b17064ff5c200", "state": "VERIFIED", "summary": {"e141_best": 0.0, "fixed_policy": 0.0625, "full": 0.9375, "no_world_model": 0.4375, "random": 0.75}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:02:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:39Z", "cpu_seconds": 0.5240264679999989, "created_at": "2026-11-22T03:58:34Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-97ff8cd0030c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "b191712d1af18c0770ddd3eadfd065e53acea99f25db74d9024e489ada0c2a19", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-97ff8cd0030c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:34Z", "seed": 1103125604, "treatment": "full"}, "priority": 0.28166548907664113, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4fbe018c75fbf43773a17639b169f560146d1e29441b933697dc3f09f30ead06", "run_seconds": 0.8097045421600342, "seed": 1103125604, "started_at": "2026-11-22T03:58:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T04:00:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-97ff8cd0030c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 1.0, 0.25], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 1.0, 1.0, 0.0], "no_world_model": [0.0, 1.0, 1.0, 0.0], "random": [0.0, 1.0, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "b191712d1af18c0770ddd3eadfd065e53acea99f25db74d9024e489ada0c2a19", "primary_control": "e141_best", "result_hash": "4fbe018c75fbf43773a17639b169f560146d1e29441b933697dc3f09f30ead06", "signature": "a87bc17b7f774f3b4235dd3f6599b5541b912e39dc4bd7fad84f5ffb649389e8b668c3c95ae62ecadd7c65387d3cbf57aa34506d0c6b90274c9626f9033ef50b", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.0, "full": 0.5, "no_world_model": 0.5, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:00:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:58:40Z", "cpu_seconds": 0.5701622860000004, "created_at": "2026-11-22T03:58:34Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bd2a647d87f6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "261f1faaa18571f08a1d1c0b5e3c1818d4472e1f3531caf57a863104af394c08", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bd2a647d87f6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:34Z", "seed": 1592118627, "treatment": "full"}, "priority": 0.26914921092957345, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e057fbdc9bf667a0ef4eec603f1e07bb56df2774a8a1fdb88b08372a25492d44", "run_seconds": 0.6348938941955566, "seed": 1592118627, "started_at": "2026-11-22T03:58:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.18, 0.555], "effect": 0.1875, "evaluated_at": "2026-11-22T04:00:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-bd2a647d87f6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.25, 0.0], "fixed_policy": [1.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 0.0], "no_world_model": [0.75, 1.0, 0.25, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "261f1faaa18571f08a1d1c0b5e3c1818d4472e1f3531caf57a863104af394c08", "primary_control": "e141_best", "result_hash": "e057fbdc9bf667a0ef4eec603f1e07bb56df2774a8a1fdb88b08372a25492d44", "signature": "7061cf20cff8b23d21f805e35b81dc29dd1c0ea326ea32cf70baa1e30ef4e164089378bc92909fd936ef286f086a713dda15b72f3945298c85bd4992f4a71c0f", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.25, "full": 0.75, "no_world_model": 0.75, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:00:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:29Z", "cpu_seconds": 0.19364871900000002, "created_at": "2026-11-22T03:58:34Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0fd4be067853", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "fadc214c49e5cb832d8f474fe114a6f0b7a996f5e41209ee23bf0dce31cf9deb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0fd4be067853", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:58:34Z", "seed": 1592866039, "treatment": "full"}, "priority": 0.2601085301584266, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1c3040aa39768ad6ed4963ce4b9eee92e234baa588529ef4fc581510544f0707", "run_seconds": 0.20255517959594727, "seed": 1592866039, "started_at": "2026-11-22T04:00:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.000997, 0.016149], "effect": 0.007576, "evaluated_at": "2026-11-22T04:02:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0fd4be067853", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.96969697, 1.0, 0.984848485, 0.96969697], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [0.984848485, 1.0, 0.984848485, 0.984848485], "retrieval": [0.984848485, 1.0, 0.984848485, 0.984848485], "rules_nodetect": [0.984848485, 1.0, 0.984848485, 0.984848485]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "fadc214c49e5cb832d8f474fe114a6f0b7a996f5e41209ee23bf0dce31cf9deb", "primary_control": "facts", "result_hash": "1c3040aa39768ad6ed4963ce4b9eee92e234baa588529ef4fc581510544f0707", "signature": "2e0fc55da0d4ce2955b09a64ffa6a0af02071dce2fafda3e01e99bb834053908e8d799875012009ea61bb9a338284e32ca2c03a5ab41a9885fa69fc4b9476209", "state": "VERIFIED", "summary": {"facts": 0.981061, "frozen": 0.030303, "full": 0.988636, "retrieval": 0.988636, "rules_nodetect": 0.988636}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:02:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:38Z", "cpu_seconds": 0.1736967430000007, "created_at": "2026-11-22T03:58:34Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e4332e556c20", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "3540c408fde41d98a1b68ea6eca9863308089a1f65e1a78f17cce55e750e4528", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e4332e556c20", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:58:34Z", "seed": 1403702067, "treatment": "full"}, "priority": 0.26279317346437925, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "31b943f9aec7a980ee95942042c4de7a72436c17b21981f026126af19dda5b27", "run_seconds": 0.1904911994934082, "seed": 1403702067, "started_at": "2026-11-22T03:59:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014545, 0.044848], "effect": 0.015152, "evaluated_at": "2026-11-22T04:00:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e4332e556c20", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.984848485, 1.0, 0.863636364, 0.984848485], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [0.984848485, 1.0, 0.924242424, 0.984848485], "retrieval": [0.984848485, 1.0, 0.909090909, 0.984848485], "rules_nodetect": [0.984848485, 1.0, 0.924242424, 0.984848485]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "3540c408fde41d98a1b68ea6eca9863308089a1f65e1a78f17cce55e750e4528", "primary_control": "facts", "result_hash": "31b943f9aec7a980ee95942042c4de7a72436c17b21981f026126af19dda5b27", "signature": "935373374aa49596a334828038ba2270b256c8cb065ea7ae26e6ceb7e200ac169ca54acd6dc32f03fac3b574d734ff7150921e9b28ba31b52fbf9a5bf5431a0e", "state": "VERIFIED", "summary": {"facts": 0.958333, "frozen": 0.030303, "full": 0.973485, "retrieval": 0.969697, "rules_nodetect": 0.973485}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:00:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:27Z", "cpu_seconds": 6.547294623, "created_at": "2026-11-22T03:58:51Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0382559a3d6c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "2b26e54c17443c2dab67832a60a6648efe906b4b9b45d6a7dcb92f1270b48740", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0382559a3d6c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:58:51Z", "seed": 1331283059, "treatment": "frontier"}, "priority": 0.26866654918948046, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a7a6777835a54a6b09379bd92477d892018ccd58154f5ed4f3b0b97dd8c6a13c", "run_seconds": 7.51490330696106, "seed": 1331283059, "started_at": "2026-11-22T03:59:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.169121, 0.14533], "effect": -0.011895, "evaluated_at": "2026-11-22T04:00:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0382559a3d6c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.523529412, 0.722580645, 0.5, 0.396226415], "frontier": [0.611764706, 0.64516129, 0.376470588, 0.226415094], "goal": [0.494117647, 0.664516129, 0.464705882, 0.245283019], "hypdisc": [0.529411765, 0.464516129, 0.541176471, 0.358490566], "infogain": [0.605882353, 0.683870968, 0.376470588, 0.264150943], "prederr": [0.247058824, 0.167741935, 0.376470588, 0.230188679], "random": [0.488235294, 0.580645161, 0.370588235, 0.467924528], "risk": [0.611764706, 0.64516129, 0.376470588, 0.226415094]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "2b26e54c17443c2dab67832a60a6648efe906b4b9b45d6a7dcb92f1270b48740", "primary_control": "random", "result_hash": "a7a6777835a54a6b09379bd92477d892018ccd58154f5ed4f3b0b97dd8c6a13c", "signature": "2b3778951da62626481100574ceaf7a22c531ecfd1fac689c5caeb0dca2321055d77574eea659f7842d0d758e7cf6cee845b1a56421c32ce2500672b4a1a0e0f", "state": "VERIFIED", "summary": {"count": 0.535584, "frontier": 0.464953, "goal": 0.467156, "hypdisc": 0.473399, "infogain": 0.482594, "prederr": 0.255365, "random": 0.476848, "risk": 0.464953}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:00:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:21Z", "cpu_seconds": 1.016418273, "created_at": "2026-11-22T03:58:51Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5ed483235624", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "47ec47b12349daeb9490d5aacc88deddc400b6a4e44b24439ac42365f2420fb5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5ed483235624", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:58:51Z", "seed": 1946847453, "treatment": "frontier"}, "priority": 0.2746699813065931, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "8065d462bbf723c4279d5d897460f3b660e08c48938acfd49db6fc41537d9199", "run_seconds": 1.2853469848632812, "seed": 1946847453, "started_at": "2026-11-22T03:59:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.046134, 0.225051], "effect": 0.089458, "evaluated_at": "2026-11-22T04:00:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5ed483235624", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.777777778, 0.666666667, 0.446808511, 0.585858586], "frontier": [0.777777778, 0.590909091, 0.744680851, 0.797979798], "goal": [0.777777778, 0.651515152, 0.730496454, 0.797979798], "hypdisc": [0.722222222, 0.666666667, 0.368794326, 0.797979798], "infogain": [0.777777778, 0.613636364, 0.29787234, 0.797979798], "prederr": [0.111111111, 0.590909091, 0.170212766, 0.636363636], "random": [0.75, 0.628787879, 0.659574468, 0.515151515], "risk": [0.777777778, 0.590909091, 0.744680851, 0.797979798]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "47ec47b12349daeb9490d5aacc88deddc400b6a4e44b24439ac42365f2420fb5", "primary_control": "random", "result_hash": "8065d462bbf723c4279d5d897460f3b660e08c48938acfd49db6fc41537d9199", "signature": "19569d372ef39358748284e50297afb0e0851e0f66e2464a324c5e016c0e70684c2ac2c5f7e1b1142c40553d5ab543e8ad928daa3a1f17ff1bbfb919ae583704", "state": "VERIFIED", "summary": {"count": 0.619278, "frontier": 0.727837, "goal": 0.739442, "hypdisc": 0.638916, "infogain": 0.621817, "prederr": 0.377149, "random": 0.638378, "risk": 0.727837}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:00:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:33Z", "cpu_seconds": 0.22045623599999953, "created_at": "2026-11-22T03:58:51Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-6f12996b0408", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "5f88ed87cb24fa0b4e8a357ef048f5463293cdcb153ba8270e62a89e95437c92", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-6f12996b0408", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:58:51Z", "seed": 294474916, "treatment": "full"}, "priority": 0.2667678085206485, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0e9d30e57f1bfb83ab42a1a87f36016ca4cd2caffb16e2d78e4523001549bad6", "run_seconds": 0.29116344451904297, "seed": 294474916, "started_at": "2026-11-22T03:59:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T04:00:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-6f12996b0408", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.0, 0.5], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5f88ed87cb24fa0b4e8a357ef048f5463293cdcb153ba8270e62a89e95437c92", "primary_control": "e141_best", "result_hash": "0e9d30e57f1bfb83ab42a1a87f36016ca4cd2caffb16e2d78e4523001549bad6", "signature": "323b3e20a4fcca199b4416b03d498bec7a8cd65697fe6519310dd7065824899ec628d3541eee78faba399ffffd61e3a4de7bef7e9f13aa6861357904f4bfae09", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 0.9375, "no_world_model": 1.0, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:00:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:27Z", "cpu_seconds": 0.697944057, "created_at": "2026-11-22T03:59:24Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6d466d3e6f1a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "e0beb66963dbaffa4c6a01898436d7ab2aec9b32dc6a4f189ff8b278c5ae99fc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6d466d3e6f1a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T03:59:24Z", "seed": 1492506398, "treatment": "frontier"}, "priority": 0.27917211759041166, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ff38ed5947dd5e82f44296375a6733aebd80b7b84eed889cc34f7368f96393ff", "run_seconds": 1.023125410079956, "seed": 1492506398, "started_at": "2026-11-22T03:59:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.031189, 0.081311], "effect": 0.05625, "evaluated_at": "2026-11-22T04:00:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6d466d3e6f1a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.7875, 0.85, 1.0, 0.783333333], "frontier": [0.8125, 0.916666667, 1.0, 0.866666667], "goal": [0.6875, 0.816666667, 1.0, 0.783333333], "hypdisc": [0.8, 0.733333333, 1.0, 0.883333333], "infogain": [0.8875, 0.9, 1.0, 0.883333333], "prederr": [0.8625, 0.766666667, 1.0, 0.733333333], "random": [0.725, 0.85, 0.9625, 0.833333333], "risk": [0.8125, 0.916666667, 1.0, 0.866666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "e0beb66963dbaffa4c6a01898436d7ab2aec9b32dc6a4f189ff8b278c5ae99fc", "primary_control": "random", "result_hash": "ff38ed5947dd5e82f44296375a6733aebd80b7b84eed889cc34f7368f96393ff", "signature": "4d877ebad574a2ffc03d11af695e0edc6db57335f3bff193d32ef2a28269178b5df7cecf90a421395869c69c265fe1336ec3ef1712a27b35e35276f109dfbd01", "state": "VERIFIED", "summary": {"count": 0.855208, "frontier": 0.898958, "goal": 0.821875, "hypdisc": 0.854167, "infogain": 0.917708, "prederr": 0.840625, "random": 0.842708, "risk": 0.898958}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:00:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:33Z", "cpu_seconds": 1.046301632, "created_at": "2026-11-22T03:59:24Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6a9a740495e2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "9dcf92d806ad8e48ffd731d845da440f27d2089f9b02349444cc33135597034f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6a9a740495e2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:59:24Z", "seed": 1409286736, "treatment": "full"}, "priority": 0.2706141053122504, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "53a004ad226154dd1472e5caee83d2022f507e66e96171678e92408ff1f6d11d", "run_seconds": 1.1595451831817627, "seed": 1409286736, "started_at": "2026-11-22T03:59:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022966, 0.007845], "effect": -0.00756, "evaluated_at": "2026-11-22T04:00:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6a9a740495e2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.536009445, 0.273015873, 0.302588997, 0.519927536], "frozen": [0.154663518, 0.144973545, 0.176375405, 0.277173913], "full": [0.52892562, 0.243386243, 0.309061489, 0.519927536], "retrieval": [0.532467532, 0.250793651, 0.28802589, 0.519927536], "rules_nodetect": [0.304604486, 0.205291005, 0.417475728, 0.503623188]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "9dcf92d806ad8e48ffd731d845da440f27d2089f9b02349444cc33135597034f", "primary_control": "facts", "result_hash": "53a004ad226154dd1472e5caee83d2022f507e66e96171678e92408ff1f6d11d", "signature": "0ed07149f9555780964689e492f5504cb6072fb51e0751cf424767cca813ece53509ce3e023916f0d97bac130a20603f617d52e490bdfc20414a5190b89a4507", "state": "VERIFIED", "summary": {"facts": 0.407885, "frozen": 0.188297, "full": 0.400325, "retrieval": 0.397804, "rules_nodetect": 0.357749}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:00:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:57Z", "cpu_seconds": 0.22479176999999595, "created_at": "2026-11-22T03:59:40Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-94408f2fb917", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "f20b3d0747f1e13031bdaa745ed5f6c3fbf6c9547cd4fecce75596f203667846", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-94408f2fb917", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:59:40Z", "seed": 1791489532, "treatment": "full"}, "priority": 0.26585184715729937, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7640cecc32025050c8eb1e82c6e1cf5cc3b281480921871f23d3cecafee72295", "run_seconds": 0.22978758811950684, "seed": 1791489532, "started_at": "2026-11-22T03:59:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.065803, 1.065803], "effect": 0.5, "evaluated_at": "2026-11-22T04:02:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-94408f2fb917", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "f20b3d0747f1e13031bdaa745ed5f6c3fbf6c9547cd4fecce75596f203667846", "primary_control": "e141_best", "result_hash": "7640cecc32025050c8eb1e82c6e1cf5cc3b281480921871f23d3cecafee72295", "signature": "56106075c7527115e6cc2dddd672cdfd8792745789d1cc5e6ff6203602c06dc5b3b525357d125e2d85f0afb5ad6448a11ff207ae47ef1b3b535ab79e07611801", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:02:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:44Z", "cpu_seconds": 1.0234345659999997, "created_at": "2026-11-22T03:59:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-00e606c35a5a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "cc736837d1c60cd18654616a90e65f487b50878dc6535c6886d91692e5c96b3d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-00e606c35a5a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:59:40Z", "seed": 333020091, "treatment": "full"}, "priority": 0.266535737273351, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "164d28cc20e642c9e2e912c59f460ad50936e408d793e33b44fceddf4aa15079", "run_seconds": 1.0771141052246094, "seed": 333020091, "started_at": "2026-11-22T03:59:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.004225, 0.099334], "effect": 0.051779, "evaluated_at": "2026-11-22T04:00:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-00e606c35a5a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.672727273, 0.690909091, 0.548275862, 0.523529412], "frozen": [0.109090909, 0.109090909, 0.103448276, 0.035294118], "full": [0.745454545, 0.763636364, 0.527586207, 0.605882353], "retrieval": [0.690909091, 0.709090909, 0.534482759, 0.547058824], "rules_nodetect": [0.745454545, 0.763636364, 0.527586207, 0.605882353]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "cc736837d1c60cd18654616a90e65f487b50878dc6535c6886d91692e5c96b3d", "primary_control": "facts", "result_hash": "164d28cc20e642c9e2e912c59f460ad50936e408d793e33b44fceddf4aa15079", "signature": "833ff7f26f5ce0db2c098d9b498190e2a240a0a0929fb74e3fcf88513b4a8f1eb132f488841dc0ff3e6bb8b77dbd9431001886d833d8b28a70b27e6c2ea3850a", "state": "VERIFIED", "summary": {"facts": 0.60886, "frozen": 0.089231, "full": 0.66064, "retrieval": 0.620385, "rules_nodetect": 0.66064}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:00:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T03:59:45Z", "cpu_seconds": 1.2924082159999997, "created_at": "2026-11-22T03:59:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-31914539c048", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "1fb73569920bb463695951a8632b05f0f54de046ab625ba7e1c880dd583260f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-31914539c048", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:59:40Z", "seed": 570329681, "treatment": "full"}, "priority": 0.2658888683464864, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-0039f6b3353b", "result_hash": "9d44134e87819c6865654426a553b1ce90354e6837ea0ef6811d1deb53348dad", "run_seconds": 1.351729393005371, "seed": 570329681, "started_at": "2026-11-22T03:59:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.049549, 0.024939], "effect": -0.012305, "evaluated_at": "2026-11-22T04:00:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-31914539c048", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.316770186, 0.542471042, 0.541322314, 0.348370927], "frozen": [0.232919255, 0.201737452, 0.179063361, 0.172932331], "full": [0.346273292, 0.501930502, 0.493112948, 0.35839599], "retrieval": [0.267080745, 0.538610039, 0.527548209, 0.35839599], "rules_nodetect": [0.346273292, 0.501930502, 0.493112948, 0.205513784]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "1fb73569920bb463695951a8632b05f0f54de046ab625ba7e1c880dd583260f4", "primary_control": "facts", "replicated": false, "replication_of": "E141-ow_model-0039f6b3353b", "result_hash": "9d44134e87819c6865654426a553b1ce90354e6837ea0ef6811d1deb53348dad", "signature": "80f88dbe6d4a18399079a383224131c4af4de8323bf3b43e147aff80ef51d29147205a653bf09010acc47e3cb31fb8959c9ffcaa5708951bd595be618287270f", "state": "VERIFIED", "summary": {"facts": 0.437234, "frozen": 0.196663, "full": 0.424928, "retrieval": 0.422909, "rules_nodetect": 0.386708}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:00:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:13Z", "cpu_seconds": 0.5706206679999752, "created_at": "2026-11-22T03:59:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d9f03a91bb46", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "c6ddf7054880b23ef83a3fb7cbc5de082926ccc84e550259a514e718edd39125", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d9f03a91bb46", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T03:59:40Z", "seed": 7958686, "treatment": "full"}, "priority": 0.2626168892071148, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e26616693f81a80ed04761dfb783010af158ffd2446d3c8d82bebdaa90789871", "run_seconds": 0.6028776168823242, "seed": 7958686, "started_at": "2026-11-22T04:00:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.004456, 0.051634], "effect": 0.028045, "evaluated_at": "2026-11-22T04:02:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d9f03a91bb46", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.901960784, 0.87755102, 0.777777778, 0.568627451], "frozen": [0.091503268, 0.081632653, 0.091503268, 0.091503268], "full": [0.901960784, 0.904761905, 0.803921569, 0.62745098], "retrieval": [0.941176471, 0.897959184, 0.875816993, 0.764705882], "rules_nodetect": [0.901960784, 0.904761905, 0.803921569, 0.62745098]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c6ddf7054880b23ef83a3fb7cbc5de082926ccc84e550259a514e718edd39125", "primary_control": "facts", "result_hash": "e26616693f81a80ed04761dfb783010af158ffd2446d3c8d82bebdaa90789871", "signature": "6d0ef009f425493160ae156848415c6af42e166ad0290878288580cd80790431883892545388a101e5150d50e0fac885ace3c9271f324d6b367ad89ed9ba2509", "state": "VERIFIED", "summary": {"facts": 0.781479, "frozen": 0.089036, "full": 0.809524, "retrieval": 0.869915, "rules_nodetect": 0.809524}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:02:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:03Z", "cpu_seconds": 0.1938093030000232, "created_at": "2026-11-22T03:59:40Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-cb046855ed34", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "prereg_hash": "7890c7c6547314034c6d7fb409b61b782387a5a37af81599c3a07ea7c920c9fb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-cb046855ed34", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:59:40Z", "seed": 1489097677, "treatment": "gated"}, "priority": 0.26286939992650116, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "176d528dd1bc05a7b3d6308e7179cde385f82c5ffc4743638bf51d0d7e1e0d2b", "run_seconds": 0.2395946979522705, "seed": 1489097677, "started_at": "2026-11-22T04:00:02Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:02:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-cb046855ed34", "family": "ow_transfer", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"compressed": [1.0, 0.401515], "gated": [1.0, 0.401515], "naive": [1.0, 0.352273], "scratch": [1.0, 0.401515], "ungated": [1.0, 0.352273]}, "prereg_hash": "7890c7c6547314034c6d7fb409b61b782387a5a37af81599c3a07ea7c920c9fb", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "176d528dd1bc05a7b3d6308e7179cde385f82c5ffc4743638bf51d0d7e1e0d2b", "signature": "1aee2778ace55b20e8c9aa5b0377b6518469332dc3555dbb435436ad1b0f62713d39c6fa021bfcec45ee71e2f3f5a50b85ad4a51d3908484570e92a5480c7906", "state": "REJECTED", "summary": {"compressed": 0.401515, "gated": 0.401515, "naive": 0.352273, "scratch": 0.401515, "ungated": 0.352273}}, "verified_at": "2026-11-22T04:02:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:24Z", "cpu_seconds": 0.16864635299999997, "created_at": "2026-11-22T03:59:57Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2475026d0c60", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "dafdd562e58c404e31a74d3fab686c931493237a465b0d76c308e433552877ac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2475026d0c60", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T03:59:57Z", "seed": 1345498824, "treatment": "full"}, "priority": 0.26111962674708733, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f5c7c7f2890db23527a06634d2c3fffb93854e325370cea350b6b79ff68a314c", "run_seconds": 0.2151031494140625, "seed": 1345498824, "started_at": "2026-11-22T04:00:23Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2475026d0c60", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.75, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "dafdd562e58c404e31a74d3fab686c931493237a465b0d76c308e433552877ac", "primary_control": "e141_best", "result_hash": "f5c7c7f2890db23527a06634d2c3fffb93854e325370cea350b6b79ff68a314c", "signature": "07f45a10cc9270a6d9792eda4184fcf6a7cf1d91ed03be47fcdb0605ed77ffac929ec73c51394136967fa97f5548a4e803c040cf80c1d3b73b31414017708b07", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:08Z", "cpu_seconds": 0.20257279000000494, "created_at": "2026-11-22T03:59:57Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-973e863ee250", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "prereg_hash": "46108b3556eba6a634637b5b41848a08fa064e8277010097f5063828991ef24c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-973e863ee250", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T03:59:57Z", "seed": 1846696434, "treatment": "gated"}, "priority": 0.26276515792878186, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "be1cb1c91ce6a576b24569d41a992cbdfea936efa3b3293496b1bfc6c73759a6", "run_seconds": 0.20299768447875977, "seed": 1846696434, "started_at": "2026-11-22T04:00:08Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-973e863ee250", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.560606061, 0.378787879, 0.348484848, 0.196969697], "gated": [0.560606061, 0.378787879, 0.348484848, 0.196969697], "naive": [0.560606061, 0.378787879, 0.212121212, 0.196969697], "scratch": [0.560606061, 0.378787879, 0.348484848, 0.196969697], "ungated": [0.560606061, 0.378787879, 0.212121212, 0.196969697]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "46108b3556eba6a634637b5b41848a08fa064e8277010097f5063828991ef24c", "primary_control": "scratch", "result_hash": "be1cb1c91ce6a576b24569d41a992cbdfea936efa3b3293496b1bfc6c73759a6", "signature": "dd3b531fb556c71cb466ba248652ed2845074017a2702823e2fc17e69b1a7353f686be5c7b3d3220ed305022d2e804f48b1d964889ec5058a9090f1d743a8305", "state": "PARTIAL", "summary": {"compressed": 0.371212, "gated": 0.371212, "naive": 0.337121, "scratch": 0.371212, "ungated": 0.337121}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:02:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:18Z", "cpu_seconds": 0.14101867100000004, "created_at": "2026-11-22T04:00:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-08f52a25ad2f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "544131620464ce361968e80448be3d931e081f10b49882eb151aeccb57f71852", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-08f52a25ad2f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:00:14Z", "seed": 1066381349, "treatment": "full"}, "priority": 0.26389096073027596, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5f0c3b61631f66c5d7a64de3187adaf3e84f0d53dad2136c8070141ca8030952", "run_seconds": 0.14913463592529297, "seed": 1066381349, "started_at": "2026-11-22T04:00:18Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-08f52a25ad2f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.75, 0.25, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "544131620464ce361968e80448be3d931e081f10b49882eb151aeccb57f71852", "primary_control": "e141_best", "result_hash": "5f0c3b61631f66c5d7a64de3187adaf3e84f0d53dad2136c8070141ca8030952", "signature": "f78045fa8da2242832f5a11ca77da4557699031858a2e9afbb76329ac01abe147c14c09ecc26e34b8a9b2f9a7a52692a502f2ac2ea1a733318752b887490840d", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:29Z", "cpu_seconds": 0.18837988800001426, "created_at": "2026-11-22T04:00:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3cd770ed9e96", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "6b75e3716c7786992dc19dcbfa6b71a4502aff79d1db7678f826195af3da78d1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3cd770ed9e96", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:00:14Z", "seed": 504870343, "treatment": "full"}, "priority": 0.25994114418014863, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8af7291e509ab8c8c364fc5ba9996447f91a5062ac87190aaf46b7589c3b70af", "run_seconds": 0.20444512367248535, "seed": 504870343, "started_at": "2026-11-22T04:00:29Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3cd770ed9e96", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.75, 0.25, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6b75e3716c7786992dc19dcbfa6b71a4502aff79d1db7678f826195af3da78d1", "primary_control": "e141_best", "result_hash": "8af7291e509ab8c8c364fc5ba9996447f91a5062ac87190aaf46b7589c3b70af", "signature": "469564bbd27ad9280d04e3b89963938666c33553e1b89d437a9a89972095bcdc9f518f607d3489aed520a1e8d4512960dbcb5eb1e8557c14276572e5d549a807", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:30Z", "cpu_seconds": 0.212678205, "created_at": "2026-11-22T04:00:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a034c2f91a3c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "f1ea1dd177ff9436bc50e4eb501bc0e3364222433418ff567f3812f4dfce3205", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a034c2f91a3c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:00:14Z", "seed": 308991410, "treatment": "full"}, "priority": 0.2589179365262151, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9ee2322bfddef530f7fea3ad0724299255bf3bdd6d8614ab55d61b79ab2d1770", "run_seconds": 0.21476006507873535, "seed": 308991410, "started_at": "2026-11-22T04:00:30Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a034c2f91a3c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.5, 0.5, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f1ea1dd177ff9436bc50e4eb501bc0e3364222433418ff567f3812f4dfce3205", "primary_control": "e141_best", "result_hash": "9ee2322bfddef530f7fea3ad0724299255bf3bdd6d8614ab55d61b79ab2d1770", "signature": "92a460fc73f3e11c45282f322ce6afed5744853506bd3d12502cea8d54eb5529ea6b84dbf537b1e447ca58827a91402218f0a65c29760c5581fe18999068b702", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:24Z", "cpu_seconds": 0.3993638419999854, "created_at": "2026-11-22T04:00:14Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7ebaf86c8dbf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "c9662e575e0babff696e0232fa3c6ff7b98c9158765cbe36a779062f97d04595", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7ebaf86c8dbf", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:14Z", "seed": 2025257068, "treatment": "full"}, "priority": 0.26105686929702787, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e34ad65df5e8e03b39da39c703c7894fd7e98c17a584b429f71e3e9a7a768982", "run_seconds": 0.4189431667327881, "seed": 2025257068, "started_at": "2026-11-22T04:00:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018013, 0.166822], "effect": 0.074405, "evaluated_at": "2026-11-22T04:02:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7ebaf86c8dbf", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.964285714, 0.729166667, 0.928571429, 1.0], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.982142857, 0.9375, 1.0, 1.0], "retrieval": [0.982142857, 0.854166667, 1.0, 1.0], "rules_nodetect": [0.982142857, 0.9375, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c9662e575e0babff696e0232fa3c6ff7b98c9158765cbe36a779062f97d04595", "primary_control": "facts", "result_hash": "e34ad65df5e8e03b39da39c703c7894fd7e98c17a584b429f71e3e9a7a768982", "signature": "fda2eed6ec26a6d12e7b381842172620192f2bb463f0ac3c4d5cc66d0af368d3a132b1439df0e95b8cf5fe5dba964a00c2e0f081be880aa4769d2281d904ff0c", "state": "VERIFIED", "summary": {"facts": 0.905506, "frozen": 0.0, "full": 0.979911, "retrieval": 0.959077, "rules_nodetect": 0.979911}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:02:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:19Z", "cpu_seconds": 0.43612793999999955, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-725f52f61405", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "d5fbf89346ac59600ce9a5d17fb7484b54a5b6963ed762a12679e9eecd3e94f9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-725f52f61405", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:00:30Z", "seed": 1845259847, "treatment": "frontier"}, "priority": 0.2533650005788542, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9ad620a0b02acbb11c48b5c89711c50b1667147db0e97c6a0bb8c67b8f1a61ff", "run_seconds": 0.4808645248413086, "seed": 1845259847, "started_at": "2026-11-22T04:01:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.058881, 0.19979], "effect": 0.070455, "evaluated_at": "2026-11-22T04:02:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-725f52f61405", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.922727273, 0.477272727, 1.0], "frontier": [1.0, 0.95, 0.472727273, 1.0], "goal": [1.0, 0.613636364, 0.472727273, 1.0], "hypdisc": [1.0, 0.604545455, 0.427272727, 1.0], "infogain": [1.0, 0.95, 0.472727273, 1.0], "prederr": [1.0, 0.95, 0.472727273, 1.0], "random": [1.0, 0.681818182, 0.459090909, 1.0], "risk": [1.0, 0.95, 0.472727273, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "d5fbf89346ac59600ce9a5d17fb7484b54a5b6963ed762a12679e9eecd3e94f9", "primary_control": "random", "result_hash": "9ad620a0b02acbb11c48b5c89711c50b1667147db0e97c6a0bb8c67b8f1a61ff", "signature": "bbfc66cc50191aad792a27d011cd71e1c3dadf14dcd62331822c215fd3e8d79ba2b205cd5cc116d872f6f63a23cb723866e4601520633b33d53e06303b58d406", "state": "VERIFIED", "summary": {"count": 0.85, "frontier": 0.855682, "goal": 0.771591, "hypdisc": 0.757955, "infogain": 0.855682, "prederr": 0.855682, "random": 0.785227, "risk": 0.855682}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:02:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:34Z", "cpu_seconds": 0.15990770700000012, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e0174f0af753", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "22a6314ab23601b8bde893e5854e46e859576f7bd63f7dc3ed5badb54c889073", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e0174f0af753", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:00:30Z", "seed": 2037506129, "treatment": "full"}, "priority": 0.42983466620699057, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4660b6857a250f1cabd80592bd51962591498c200c698d00f38a66262049153d", "run_seconds": 0.17308449745178223, "seed": 2037506129, "started_at": "2026-11-22T04:00:34Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e0174f0af753", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.25, 0.0, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "22a6314ab23601b8bde893e5854e46e859576f7bd63f7dc3ed5badb54c889073", "primary_control": "e141_best", "result_hash": "4660b6857a250f1cabd80592bd51962591498c200c698d00f38a66262049153d", "signature": "0f1e3deefa2f04fd33c9bc06a86996c215c4043eb7d34f1565b422e97df8c21d44ea8ee2ec1547af78c1e6705c28b59217e3a1f32c4fc4269e36c9b4448c5b01", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:18Z", "cpu_seconds": 0.8227217360000001, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0e35951e1d54", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "2be6780ed4ab7e9d8cbd7da639f00c9f40fc82f3517ee99780c270624e32b607", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0e35951e1d54", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:30Z", "seed": 1478927010, "treatment": "full"}, "priority": 0.25372874948487056, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5f7cfc72ae9f282b409e9fb316c41acc1b6232286d5e2adcc812ae8e0e2eb7f6", "run_seconds": 0.8304042816162109, "seed": 1478927010, "started_at": "2026-11-22T04:01:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026248, 0.079189], "effect": 0.026471, "evaluated_at": "2026-11-22T04:02:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0e35951e1d54", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.537254902, 0.535686275, 0.534117647, 0.541960784], "frozen": [0.381960784, 0.418823529, 0.418039216, 0.418823529], "full": [0.55372549, 0.490196078, 0.60627451, 0.604705882], "retrieval": [0.814117647, 0.815686275, 0.823529412, 0.844705882], "rules_nodetect": [0.55372549, 0.490196078, 0.60627451, 0.604705882]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "2be6780ed4ab7e9d8cbd7da639f00c9f40fc82f3517ee99780c270624e32b607", "primary_control": "facts", "result_hash": "5f7cfc72ae9f282b409e9fb316c41acc1b6232286d5e2adcc812ae8e0e2eb7f6", "signature": "58489a1c62dfc228077b83084f26eeb353972a1ef8086a2e7924135b65cbe7f4bbd86b0fd850c774fdd9ad461473822173e3e0a1c3da7be551bb4671f7891f0b", "state": "VERIFIED", "summary": {"facts": 0.537255, "frozen": 0.409412, "full": 0.563725, "retrieval": 0.82451, "rules_nodetect": 0.563725}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:02:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:38Z", "cpu_seconds": 2.906836577000007, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-647b60c1a788", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 3, "n": 4}, "prereg_hash": "35d5420e11ca477f80e76d0869a8e87047cde3b5c15d6e93341716c35627b3b5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-647b60c1a788", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:30Z", "seed": 1319152237, "treatment": "full"}, "priority": 0.34478923923827565, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-aabf5555895a", "result_hash": "b01971cd4415da89a483643769bdfcdb2cdbb1731e3cd292b075fa828fa8af9a", "run_seconds": 3.5107197761535645, "seed": 1319152237, "started_at": "2026-11-22T04:00:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.027195, 0.096319], "effect": 0.034562, "evaluated_at": "2026-11-22T04:02:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-647b60c1a788", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.995391705, 1.0, 0.995391705, 0.870967742], "frozen": [0.0, 0.142857143, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 1.0, 0.953917051], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "35d5420e11ca477f80e76d0869a8e87047cde3b5c15d6e93341716c35627b3b5", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-aabf5555895a", "result_hash": "b01971cd4415da89a483643769bdfcdb2cdbb1731e3cd292b075fa828fa8af9a", "signature": "49846f4323ae62d7929da410d9251c85611e586ebce29ee08d818f8541dbb0c7196301d8d7f161eba53709e6fa4e69c8c0124dca315b43134a1f9e0e7990d10b", "state": "VERIFIED", "summary": {"facts": 0.965438, "frozen": 0.035714, "full": 1.0, "retrieval": 0.988479, "rules_nodetect": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:02:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:37Z", "cpu_seconds": 1.603730818, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6e1cab92aa3c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "67c1ec73351d8f05feb673841e533be904bcb6f99c4a467e3edfed4385d1fbce", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6e1cab92aa3c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:30Z", "seed": 1514749873, "treatment": "full"}, "priority": 0.29859197558297756, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d2113cc16a6eaf373d1527a281095097d554c7dcc5cb21c0ab53355a9454a2f1", "run_seconds": 1.8491542339324951, "seed": 1514749873, "started_at": "2026-11-22T04:00:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.07386, 0.277645], "effect": 0.175753, "evaluated_at": "2026-11-22T04:02:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6e1cab92aa3c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.48, 0.426666667, 0.292682927, 0.509259259], "frozen": [0.38, 0.168888889, 0.093495935, 0.236111111], "full": [0.52, 0.668888889, 0.56300813, 0.659722222], "retrieval": [0.52, 0.686666667, 0.56300813, 0.694444444], "rules_nodetect": [0.52, 0.668888889, 0.219512195, 0.62037037]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "67c1ec73351d8f05feb673841e533be904bcb6f99c4a467e3edfed4385d1fbce", "primary_control": "facts", "result_hash": "d2113cc16a6eaf373d1527a281095097d554c7dcc5cb21c0ab53355a9454a2f1", "signature": "a9a308f3dc4d17b054e974c1a62245aa3cb1705614cb788f1b4f68630291df6fcabbe28c9f781d54e1ca3e7d132a69a830fb98059b24a92560ddbd6626863500", "state": "VERIFIED", "summary": {"facts": 0.427152, "frozen": 0.219624, "full": 0.602905, "retrieval": 0.61603, "rules_nodetect": 0.507193}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:02:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:14Z", "cpu_seconds": 0.17128465700000017, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c12948e4dd40", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "53688755416571b1a8f644e251eed669f1837512029e00f1172fe351470985a5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c12948e4dd40", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:30Z", "seed": 1622396855, "treatment": "full"}, "priority": 0.2570195370997863, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cf2d4c6465f806c3fddd49b391164c28cc13761c0d79c8d482140d4ea25f719e", "run_seconds": 0.1717073917388916, "seed": 1622396855, "started_at": "2026-11-22T04:01:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0224, 0.069067], "effect": 0.023333, "evaluated_at": "2026-11-22T04:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c12948e4dd40", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.596590909, 0.272727273, 0.368888889], "frozen": [0.034090909, 0.034090909, 0.022727273, 0.031111111], "full": [1.0, 0.596590909, 0.272727273, 0.462222222], "retrieval": [1.0, 0.596590909, 0.277272727, 0.462222222], "rules_nodetect": [1.0, 0.596590909, 0.272727273, 0.462222222]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "53688755416571b1a8f644e251eed669f1837512029e00f1172fe351470985a5", "primary_control": "facts", "result_hash": "cf2d4c6465f806c3fddd49b391164c28cc13761c0d79c8d482140d4ea25f719e", "signature": "6b7b3f66b9b76a329b0ffa06920cf3332f092b74c96e94b3676e3860c72bd43fe71a12981626431a2be7a94aa7006292b988fcc912428dc09d89d20037bbcb0c", "state": "VERIFIED", "summary": {"facts": 0.559552, "frozen": 0.030505, "full": 0.582885, "retrieval": 0.584021, "rules_nodetect": 0.582885}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:16Z", "cpu_seconds": 0.48573774999999997, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-cafaaf60e172", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "2a18a63ff745788787b5c1cc90553d0c4a3bff0412d1814fdfa5440ba92a736f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-cafaaf60e172", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:30Z", "seed": 1255319232, "treatment": "full"}, "priority": 0.25478871253420365, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "252edff5577f2dbce1c13def53cf27eabe33443bbd431cba59f1dda49e5d0c14", "run_seconds": 0.48963165283203125, "seed": 1255319232, "started_at": "2026-11-22T04:01:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.030182, 0.128525], "effect": 0.049172, "evaluated_at": "2026-11-22T04:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-cafaaf60e172", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.91503268, 0.178294574, 0.888888889, 0.85620915], "frozen": [0.091503268, 0.108527132, 0.091503268, 0.091503268], "full": [0.921568627, 0.348837209, 0.901960784, 0.862745098], "retrieval": [0.934640523, 0.434108527, 0.941176471, 0.91503268], "rules_nodetect": [0.921568627, 0.348837209, 0.901960784, 0.862745098]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2a18a63ff745788787b5c1cc90553d0c4a3bff0412d1814fdfa5440ba92a736f", "primary_control": "facts", "result_hash": "252edff5577f2dbce1c13def53cf27eabe33443bbd431cba59f1dda49e5d0c14", "signature": "a70970a0abbe04661f98eaf70b7132acd1aa51cabc7722e61ea6baec08d186069d8ecd8072cf1299f443c2bdc227e1d8a40b9da839a5efa86697f968ce7a2d09", "state": "VERIFIED", "summary": {"facts": 0.709606, "frozen": 0.095759, "full": 0.758778, "retrieval": 0.80624, "rules_nodetect": 0.758778}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:02:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:43Z", "cpu_seconds": 1.0588491479999997, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e7864aeedb88", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 9, "n": 4}, "prereg_hash": "8392aaf25ae9847e1bd433c149682eace241eeda2343eb4ef2724142698f0a71", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e7864aeedb88", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:30Z", "seed": 908585689, "treatment": "full"}, "priority": 0.2574719608798163, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9027d1e1c3b08cc36765c10e94b71c922e75656a4020d2af199453ad58a23532", "run_seconds": 1.3689491748809814, "seed": 908585689, "started_at": "2026-11-22T04:00:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002295, 0.158469], "effect": 0.078087, "evaluated_at": "2026-11-22T04:02:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e7864aeedb88", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.583333333, 0.513333333, 0.539583333], "frozen": [0.6, 0.430701754, 0.433333333, 0.379166667], "full": [1.0, 0.671929825, 0.701666667, 0.575], "retrieval": [1.0, 0.837719298, 0.819166667, 0.809375], "rules_nodetect": [1.0, 0.671929825, 0.701666667, 0.575]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "8392aaf25ae9847e1bd433c149682eace241eeda2343eb4ef2724142698f0a71", "primary_control": "facts", "result_hash": "9027d1e1c3b08cc36765c10e94b71c922e75656a4020d2af199453ad58a23532", "signature": "3110c383bacbf534b60097fff08d59470783d438aa463d15889f480044d9830e1b71018053cfca3de88242a028a813bfe18f9c0621b2795f00540d04e08d5f03", "state": "VERIFIED", "summary": {"facts": 0.659062, "frozen": 0.4608, "full": 0.737149, "retrieval": 0.866565, "rules_nodetect": 0.737149}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:02:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:35Z", "cpu_seconds": 57.115779163, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-11aca59a451f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "526f00638ef0de321ac23cd61dfc66324f7c3279cc760f8232ceeb3e04e30092", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-11aca59a451f", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:00:30Z", "seed": 1392629020, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "0ac622e22a4185f6b5b04c9c0ef33b631f5751b95c1b96f14c23bd77754b5d55", "run_seconds": 62.250157833099365, "seed": 1392629020, "started_at": "2026-11-22T04:00:32Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 1, "inherit_mem": 2, "inherit_nomem": 2, "random_search": 1, "reset_mem": 2, "retrieval": 0, "rootcause": 1, "xfer_mem": 2}, "better": "higher", "ci95": [-0.108698, 0.040812], "effect": -0.033943, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:02:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-11aca59a451f", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0, "risk": 1.061}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.6, "random": 1.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.3, 1.0, 0.96875, 0.821705426, 0.848958333, 0.684151786, 0.875, 0.599137931, 0.767307692, 0.89], "compressed": [0.367857143, 1.0, 0.979166667, 0.579457364, 0.852083333, 0.871651786, 0.880681818, 0.592241379, 0.757051282, 0.8675], "inherit_mem": [0.335714286, 1.0, 0.958333333, 0.677325581, 0.861458333, 0.868303571, 0.886363636, 0.448275862, 0.517307692, 0.8625], "inherit_nomem": [0.382142857, 1.0, 0.96875, 0.848837209, 0.878125, 0.600446429, 0.880681818, 0.573275862, 0.757051282, 0.855], "random_search": [0.946428571, 1.0, 1.0, 0.841085271, 0.726041667, 0.751116071, 0.965909091, 0.45, 0.517307692, 0.905], "reset_mem": [0.335714286, 1.0, 0.958333333, 0.677325581, 0.861458333, 0.868303571, 0.886363636, 0.448275862, 0.517307692, 0.8625], "retrieval": [0.3, 1.0, 0.96875, 0.821705426, 0.848958333, 0.684151786, 0.875, 0.599137931, 0.767307692, 0.89], "rootcause": [0.367857143, 1.0, 0.979166667, 0.579457364, 0.852083333, 0.871651786, 0.880681818, 0.592241379, 0.757051282, 0.8675], "xfer_mem": [0.335714286, 1.0, 0.958333333, 0.677325581, 0.861458333, 0.868303571, 0.886363636, 0.448275862, 0.517307692, 0.8625]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.1272, "losses": 5, "n": 10, "n_sealed": 10, "outcome": "CONTRADICTED", "p_sign_display": "0.746", "per_generation_sealed": [0.775501, 0.774769, 0.741558, 0.741558], "prereg_hash": "526f00638ef0de321ac23cd61dfc66324f7c3279cc760f8232ceeb3e04e30092", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "random_search", "reset_mem"], "result_hash": "0ac622e22a4185f6b5b04c9c0ef33b631f5751b95c1b96f14c23bd77754b5d55", "signature": "8ba504f54d8ddce87eb0aee5bd6b974cea7c38d57c7a6b29c56e97d53b6fff02a34d164200b34d951198f1baaba6d8be0babf81f43adc1b4ff20dc2be231ab02", "state": "VERIFIED", "summary": {"G0": 0.775501, "compressed": 0.774769, "inherit_mem": 0.741558, "inherit_nomem": 0.774431, "random_search": 0.810289, "reset_mem": 0.741558, "retrieval": 0.775501, "rootcause": 0.774769, "xfer_mem": 0.741558}, "ties": 1, "treatment": "inherit_mem", "wins": 4}, "verified_at": "2026-11-22T04:03:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:00:39Z", "cpu_seconds": 0.15760575799999987, "created_at": "2026-11-22T04:00:30Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1408127cc5a5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "prereg_hash": "644e5ea288cd5446fcda69cce7517e4a2c6e246e17e4db912d86c16ab0744838", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1408127cc5a5", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:00:30Z", "seed": 142090643, "treatment": "gated"}, "priority": 0.26608816233444127, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "66d80a228df87f8343cb4ecdeb48578c953cf1b4f753f1eeb59a394aca5ebf3a", "run_seconds": 0.18163585662841797, "seed": 142090643, "started_at": "2026-11-22T04:00:39Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1408127cc5a5", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.287878788, 0.333333333, 0.242424242, 0.106060606], "gated": [0.287878788, 0.333333333, 0.242424242, 0.106060606], "naive": [0.287878788, 0.318181818, 0.257575758, 0.106060606], "scratch": [0.287878788, 0.333333333, 0.242424242, 0.106060606], "ungated": [0.287878788, 0.318181818, 0.257575758, 0.106060606]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "644e5ea288cd5446fcda69cce7517e4a2c6e246e17e4db912d86c16ab0744838", "primary_control": "scratch", "result_hash": "66d80a228df87f8343cb4ecdeb48578c953cf1b4f753f1eeb59a394aca5ebf3a", "signature": "5b1d55097bcdd610b2364a5e53e273b1f4b3d040fd5640be9f9d15fe46fa74711f4f604401e242a74d5b2eb01fb72415465f40abe6aed747a7b422477f55550c", "state": "PARTIAL", "summary": {"compressed": 0.242424, "gated": 0.242424, "naive": 0.242424, "scratch": 0.242424, "ungated": 0.242424}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:02:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:12Z", "cpu_seconds": 0.28460178400000014, "created_at": "2026-11-22T04:00:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d7de2b0d599e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "4074c94475932f0052eb8c5d63a143fd762387b2827c979a4ca4a7b49e5dd417", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d7de2b0d599e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:00:47Z", "seed": 982453235, "treatment": "full"}, "priority": 0.25746326773541695, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5565e2d16493c9762ce0aa6a80b494fad8392f23c186c13909bccd614884eeb1", "run_seconds": 0.3215675354003906, "seed": 982453235, "started_at": "2026-11-22T04:01:12Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d7de2b0d599e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4074c94475932f0052eb8c5d63a143fd762387b2827c979a4ca4a7b49e5dd417", "primary_control": "e141_best", "result_hash": "5565e2d16493c9762ce0aa6a80b494fad8392f23c186c13909bccd614884eeb1", "signature": "ccc6ba1a05a7d5775730bb28541fdad522c0e7bd9c2a8431f69c3da0ecf98fd4830932d981a0bab4be73b45226186f8f4a8930e743e0d0833558791a49528a0d", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 1.0, "random": 0.8125}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:05Z", "cpu_seconds": 0.5197995310000001, "created_at": "2026-11-22T04:00:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ebd210812a15", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "ef15f4a09ba1ac93e414089ac19fa9600f4f4538877e82ed3bd7b836b937cfc4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ebd210812a15", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:00:47Z", "seed": 1204071452, "treatment": "full"}, "priority": 0.2590501155584054, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1341150d6309b139520b1940e9af8864b0b8fa6e1b538d0743e9ceaa98b775b4", "run_seconds": 0.8809103965759277, "seed": 1204071452, "started_at": "2026-11-22T04:01:05Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ebd210812a15", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.25, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ef15f4a09ba1ac93e414089ac19fa9600f4f4538877e82ed3bd7b836b937cfc4", "primary_control": "e141_best", "result_hash": "1341150d6309b139520b1940e9af8864b0b8fa6e1b538d0743e9ceaa98b775b4", "signature": "e96993ffefab2b166e899170ad61d4b767712f032eaf17ed261113ea3ce712c95684686af432181a0b028189c6d5a5c9b739c9bf50a21c458625e768e144750a", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:07Z", "cpu_seconds": 1.3913091640000002, "created_at": "2026-11-22T04:00:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b32bc7148acc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "893f1e006f22c928c9cb0b9b55bb58412e9ab4c0843ef801bd4a70a3058792d5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b32bc7148acc", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:00:47Z", "seed": 102997814, "treatment": "full"}, "priority": 0.2605380834509329, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b3a4ec8d89abf4f63c3aa2e6af35e2b8d9b049de557d59868c99935756e9f898", "run_seconds": 2.042192220687866, "seed": 102997814, "started_at": "2026-11-22T04:01:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.015562, 0.060532], "effect": 0.022485, "evaluated_at": "2026-11-22T04:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b32bc7148acc", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.524137931, 0.547058824, 0.621333333, 0.742105263], "frozen": [0.103448276, 0.035294118, 0.128, 0.110526316], "full": [0.548275862, 0.623529412, 0.610666667, 0.742105263], "retrieval": [0.551724138, 0.594117647, 0.608, 0.742105263], "rules_nodetect": [0.548275862, 0.623529412, 0.610666667, 0.742105263]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "893f1e006f22c928c9cb0b9b55bb58412e9ab4c0843ef801bd4a70a3058792d5", "primary_control": "facts", "result_hash": "b3a4ec8d89abf4f63c3aa2e6af35e2b8d9b049de557d59868c99935756e9f898", "signature": "5d72153a64fc00cde06b7423ed8e7da8c80bf93adad6bd8c63bdd5874f8e0ea644550059cad0d746510509f77df7a2f9d04e6f928f4b94177c4de586d335b80f", "state": "VERIFIED", "summary": {"facts": 0.608659, "frozen": 0.094317, "full": 0.631144, "retrieval": 0.623987, "rules_nodetect": 0.631144}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:11Z", "cpu_seconds": 0.42998440300000007, "created_at": "2026-11-22T04:00:47Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-478b966eeb3e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "prereg_hash": "0deac9705abab417b3b0b5b110ac42c03b6a9962dc7e3909dd708863aee2965c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-478b966eeb3e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:00:47Z", "seed": 1101179312, "treatment": "gated"}, "priority": 0.25785913497751084, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9da56a223b88bf5b699ab888e0d14d5399b9a5c2f05a93f7e070a438b1eb7fc0", "run_seconds": 0.4746122360229492, "seed": 1101179312, "started_at": "2026-11-22T04:01:10Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-478b966eeb3e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.543010753, 0.644444444, 0.666666667, 0.61827957], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.543010753, 0.766666667, 0.666666667, 0.629032258]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0deac9705abab417b3b0b5b110ac42c03b6a9962dc7e3909dd708863aee2965c", "primary_control": "scratch", "result_hash": "9da56a223b88bf5b699ab888e0d14d5399b9a5c2f05a93f7e070a438b1eb7fc0", "signature": "8b4f492dd40e4cc97d45015fbe3d034138bc81e1266a938a215cbcf9f48e07cb897abc50d6840ea767d6e05aebef398e5b2356aae0d2dcae9f5b1ff02968be08", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.6181, "scratch": 1.0, "ungated": 0.651344}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:08Z", "cpu_seconds": 3.156309053, "created_at": "2026-11-22T04:00:47Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d97d533ae901", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "5227b91b2dbe8af8310ed69b0d3cbbe96a8bb8fe6f2e55a917714fa9f559435d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d97d533ae901", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:00:47Z", "seed": 1049137238, "treatment": "gated"}, "priority": 0.31427492123522105, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "57349c532abf3e7a787b21a078f91ee9854b1e3c9a023a88c50bd4704853d3cc", "run_seconds": 3.9072048664093018, "seed": 1049137238, "started_at": "2026-11-22T04:01:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0925, 0.03], "effect": -0.03125, "evaluated_at": "2026-11-22T04:02:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d97d533ae901", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.553571429, 0.583333333, 0.56, 0.541666667], "gated": [0.553571429, 0.583333333, 0.56, 0.541666667], "naive": [0.669642857, 0.714285714, 0.56, 0.633333333], "scratch": [0.553571429, 0.583333333, 0.56, 0.666666667], "ungated": [0.553571429, 0.702380952, 0.63, 0.541666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5227b91b2dbe8af8310ed69b0d3cbbe96a8bb8fe6f2e55a917714fa9f559435d", "primary_control": "scratch", "result_hash": "57349c532abf3e7a787b21a078f91ee9854b1e3c9a023a88c50bd4704853d3cc", "signature": "fc4657d213e4460b395f9502a39e12a01839c06d04782ada98d6a0d9f0968c134bcd8c5549e3b7c143bbf1d8dbaf0d83bba105cb9ef55d5ec340971a0966fc0e", "state": "VERIFIED", "summary": {"compressed": 0.559643, "gated": 0.559643, "naive": 0.644315, "scratch": 0.590893, "ungated": 0.606905}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:02:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:33Z", "cpu_seconds": 0.26164454899999967, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7f8240a21895", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "71ef7dfd5432bdc7b08b0896c6f5d689fb9a6b6b517def8c5a6f56863ea06614", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7f8240a21895", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:01:19Z", "seed": 1798508732, "treatment": "frontier"}, "priority": 0.2557197458365689, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4112c1b2071bee824e6bf08837f3abbb5d36a3ab16a08da36c870532e1041fae", "run_seconds": 0.3783106803894043, "seed": 1798508732, "started_at": "2026-11-22T04:01:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.147864, 0.221454], "effect": 0.184659, "evaluated_at": "2026-11-22T04:02:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7f8240a21895", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.988636364, 0.943181818, 1.0, 0.988636364], "frontier": [1.0, 0.909090909, 1.0, 1.0], "goal": [0.681818182, 0.75, 0.772727273, 0.852272727], "hypdisc": [0.75, 0.761363636, 0.784090909, 0.886363636], "infogain": [1.0, 0.909090909, 1.0, 1.0], "prederr": [1.0, 0.909090909, 1.0, 1.0], "random": [0.772727273, 0.761363636, 0.795454545, 0.840909091], "risk": [1.0, 0.909090909, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "71ef7dfd5432bdc7b08b0896c6f5d689fb9a6b6b517def8c5a6f56863ea06614", "primary_control": "random", "result_hash": "4112c1b2071bee824e6bf08837f3abbb5d36a3ab16a08da36c870532e1041fae", "signature": "60c93a480d989ae8ac6aa29f6b9d4bab5eccdde25faaacfeef04036bef6348c40755cbf2a1cd3dafd3da69072c473a4b3385770b1d00bb865e8e2f2a4d38cd0b", "state": "VERIFIED", "summary": {"count": 0.980114, "frontier": 0.977273, "goal": 0.764205, "hypdisc": 0.795455, "infogain": 0.977273, "prederr": 0.977273, "random": 0.792614, "risk": 0.977273}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:02:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:23Z", "cpu_seconds": 0.3241501239999991, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b51b5de8ac17", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "a73a243cd046607d105ba3cbbd0629dfe64e2f1d8579f459f92a5739957ee708", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b51b5de8ac17", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:01:19Z", "seed": 291721690, "treatment": "frontier"}, "priority": 0.25112906651581884, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "74bb835b80442d67bfa956b57132a0494085c5d19bb5dc8db833480a25852d43", "run_seconds": 0.4160151481628418, "seed": 291721690, "started_at": "2026-11-22T04:03:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.049129, 0.216023], "effect": 0.132576, "evaluated_at": "2026-11-22T04:03:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b51b5de8ac17", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.984848485, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.787878788, 0.833333333, 0.772727273, 0.878787879], "hypdisc": [0.893939394, 0.909090909, 0.893939394, 0.863636364], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.909090909, 0.848484848, 0.757575758, 0.954545455], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a73a243cd046607d105ba3cbbd0629dfe64e2f1d8579f459f92a5739957ee708", "primary_control": "random", "result_hash": "74bb835b80442d67bfa956b57132a0494085c5d19bb5dc8db833480a25852d43", "signature": "1386e263291aecdb9d18fa902ad11acfee8b456a9aac0034e8dfb51a079b8885563844bebb93b87f75ef256e28dfe4aee2462ce0b47b3c55c2b1b71cc2713205", "state": "VERIFIED", "summary": {"count": 0.996212, "frontier": 1.0, "goal": 0.818182, "hypdisc": 0.890152, "infogain": 1.0, "prederr": 1.0, "random": 0.867424, "risk": 1.0}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:03:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:27Z", "cpu_seconds": 0.6406863769999998, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-155e7a44a4a4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "86517ad462685173687a7d76962cac9c9fb8f399dc0a9ea5172d78c0597e3a6b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-155e7a44a4a4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:01:19Z", "seed": 1903721786, "treatment": "full"}, "priority": 0.257921749211814, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1677bbedf27d4ff268bbe809416a59b5b43057a9b6387fd2248537a3db82b691", "run_seconds": 0.7089002132415771, "seed": 1903721786, "started_at": "2026-11-22T04:01:27Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-155e7a44a4a4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "86517ad462685173687a7d76962cac9c9fb8f399dc0a9ea5172d78c0597e3a6b", "primary_control": "e141_best", "result_hash": "1677bbedf27d4ff268bbe809416a59b5b43057a9b6387fd2248537a3db82b691", "signature": "4829509075287efa1d5ca96d2617595dc492709d7590f52bd0b1d16f765f1c5494aef7839a2f735af0cd9ea97c2cfe68f00efe82558f08406ee44cd9efc20c05", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:23Z", "cpu_seconds": 0.1457728020000002, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-56c855b5d603", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "bacc7cd922f48b6c2a0bbc15d41ac9c48a2e1d17c6af16f83eccec8bba9a3bd8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-56c855b5d603", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:01:19Z", "seed": 1822121342, "treatment": "full"}, "priority": 0.26123548458265433, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "010494dc2ba67afb5e7fa77751a5b475d765173c5fff3b09c610bc6014b4f9a3", "run_seconds": 0.14881014823913574, "seed": 1822121342, "started_at": "2026-11-22T04:01:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.733549, 1.016451], "effect": 0.875, "evaluated_at": "2026-11-22T04:02:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-56c855b5d603", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.25, 0.0], "fixed_policy": [0.5, 0.5, 0.25, 0.75], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "bacc7cd922f48b6c2a0bbc15d41ac9c48a2e1d17c6af16f83eccec8bba9a3bd8", "primary_control": "e141_best", "result_hash": "010494dc2ba67afb5e7fa77751a5b475d765173c5fff3b09c610bc6014b4f9a3", "signature": "06613086af1f74108c2acaf147402183e11f01a9a942f706648ba63b1f309a5c54d449f61cd158a90525fc9d2fa64d4a2cb27ebab84b35e81080efecf7fdba08", "state": "VERIFIED", "summary": {"e141_best": 0.0625, "fixed_policy": 0.5, "full": 0.9375, "no_world_model": 0.875, "random": 0.75}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:02:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:22Z", "cpu_seconds": 0.26832138799999994, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-85220c017346", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "392f8f0aad7a199252e2c7a8e50496faee5ad689b387142a0d38c02b4e579eec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-85220c017346", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:01:19Z", "seed": 1795783834, "treatment": "full"}, "priority": 0.2630874227553082, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "24d39bf7d8e9c010eb3dbfd5b41856aaf31390c10c709df906b362753ad4bf75", "run_seconds": 0.2782711982727051, "seed": 1795783834, "started_at": "2026-11-22T04:01:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.094139, 0.844139], "effect": 0.375, "evaluated_at": "2026-11-22T04:02:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-85220c017346", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 0.0], "fixed_policy": [0.0, 1.0, 0.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "392f8f0aad7a199252e2c7a8e50496faee5ad689b387142a0d38c02b4e579eec", "primary_control": "e141_best", "result_hash": "24d39bf7d8e9c010eb3dbfd5b41856aaf31390c10c709df906b362753ad4bf75", "signature": "4fd5619533a826ca32c67cfe1d5d3132522c96820adfcaa37d70fa3a32d13ed667abe4a991d858647ea2724aa518d19f811692589746b3bc4aa79477f9bd7f0b", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:02:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:33Z", "cpu_seconds": 0.2677456869999997, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a527058ba2ea", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "9cd9cbb9f6dc5f46a20352c5aecdd7c561b4a125844c16acad86cd0901256515", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a527058ba2ea", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:01:19Z", "seed": 956811875, "treatment": "full"}, "priority": 0.25611584180625196, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e9d69eb7387cd70793cc12cbf362e8ec574064cd6106aed64eb9c39577f8ec80", "run_seconds": 0.30644965171813965, "seed": 956811875, "started_at": "2026-11-22T04:01:32Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a527058ba2ea", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.25, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9cd9cbb9f6dc5f46a20352c5aecdd7c561b4a125844c16acad86cd0901256515", "primary_control": "e141_best", "result_hash": "e9d69eb7387cd70793cc12cbf362e8ec574064cd6106aed64eb9c39577f8ec80", "signature": "836e9ba138e7c86e8c31c53a746e04039fa363c9f2d171a514b0f6d27409e4df5c874ddfeb1d11d0321e90c1033a1e5f90021c6bc31e68c9dbcb9706317a9b01", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:28Z", "cpu_seconds": 0.10717776700000003, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-dd95e308f8de", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "9b5f5b184fba2fb61954504a6e0c1ceb521751057dfc1ad0f85ba173d7eb6087", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-dd95e308f8de", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:01:19Z", "seed": 276477013, "treatment": "full"}, "priority": 0.2575702225842249, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "059b87d423c2d166e921bca1689a6014aaa8ab4594eab187e4377529c9654f09", "run_seconds": 0.10860133171081543, "seed": 276477013, "started_at": "2026-11-22T04:01:28Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:02:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-dd95e308f8de", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.75, 0.5, 0.25], "fixed_policy": [0.5, 0.0, 0.25, 0.25], "full": [0.25, 0.75, 0.5, 0.25], "no_world_model": [0.25, 0.75, 0.5, 0.25], "random": [0.25, 0.75, 0.5, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9b5f5b184fba2fb61954504a6e0c1ceb521751057dfc1ad0f85ba173d7eb6087", "primary_control": "e141_best", "result_hash": "059b87d423c2d166e921bca1689a6014aaa8ab4594eab187e4377529c9654f09", "signature": "84f9220e90f2d57551ddd9b7aea38b29ab503d01f476c69cbfe1f9c55125ea12cf6de4f5ab396a985e5c1d9a782a907a5bbaf92174ee1b909291006c0891ae06", "state": "PARTIAL", "summary": {"e141_best": 0.4375, "fixed_policy": 0.25, "full": 0.4375, "no_world_model": 0.4375, "random": 0.4375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:02:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:09Z", "cpu_seconds": 0.351959996999998, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-38eeec17e2a4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "8e3707c0a879959ab2f85c8eae799b8692f5fe959e69621020e848e1908befa6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-38eeec17e2a4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:01:19Z", "seed": 302110800, "treatment": "full"}, "priority": 0.25379067098442726, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1ceb15ea06d8639954082e987809ac76d966f5644a4fdff97fc53f5829f1fce4", "run_seconds": 0.4040074348449707, "seed": 302110800, "started_at": "2026-11-22T04:03:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.011105, 0.051105], "effect": 0.02, "evaluated_at": "2026-11-22T04:03:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-38eeec17e2a4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 1.0, 0.986666667, 0.92], "frozen": [0.2, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 0.986666667], "retrieval": [1.0, 1.0, 0.986666667, 0.946666667], "rules_nodetect": [1.0, 1.0, 1.0, 0.986666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "8e3707c0a879959ab2f85c8eae799b8692f5fe959e69621020e848e1908befa6", "primary_control": "facts", "result_hash": "1ceb15ea06d8639954082e987809ac76d966f5644a4fdff97fc53f5829f1fce4", "signature": "f5400e6887974704919b311b91a6cddd119a20000b27a93fbf6bce2c92b3b37f9ee07d7fce6e2f4f253863c766c92baa55ae57522fa66abdbe703802e98f0404", "state": "VERIFIED", "summary": {"facts": 0.976667, "frozen": 0.05, "full": 0.996667, "retrieval": 0.983333, "rules_nodetect": 0.996667}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:12Z", "cpu_seconds": 0.5903045789999997, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5d023b6c533b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "89eede327488eaa53809bbabc6d3a09dc5a16b1a7c46a1658fcfb6f5048a2de8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5d023b6c533b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:01:19Z", "seed": 1633144428, "treatment": "full"}, "priority": 0.25169577470398685, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "39a74fffae0cd84385ad8b2dcc28696364a0898da44a733e85583b27b73f28cc", "run_seconds": 0.8880996704101562, "seed": 1633144428, "started_at": "2026-11-22T04:03:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.079909, 0.191397], "effect": 0.055744, "evaluated_at": "2026-11-22T04:03:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5d023b6c533b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.878205128, 0.692307692, 0.163398693, 0.698717949], "frozen": [0.076923077, 0.083333333, 0.091503268, 0.076923077], "full": [0.858974359, 0.705128205, 0.424836601, 0.666666667], "retrieval": [0.891025641, 0.730769231, 0.509803922, 0.756410256], "rules_nodetect": [0.858974359, 0.705128205, 0.424836601, 0.666666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "89eede327488eaa53809bbabc6d3a09dc5a16b1a7c46a1658fcfb6f5048a2de8", "primary_control": "facts", "result_hash": "39a74fffae0cd84385ad8b2dcc28696364a0898da44a733e85583b27b73f28cc", "signature": "bdba0d52ea868cb2bee7103cb6e991d83c5a69128738563b5d2c75f8aa5c2e6ee761b6dfff4434dcfa892959b28d457e5238070d14fd7b18fd977358ca7b520c", "state": "VERIFIED", "summary": {"facts": 0.608157, "frozen": 0.082171, "full": 0.663901, "retrieval": 0.722002, "rules_nodetect": 0.663901}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:38Z", "cpu_seconds": 182.265079862, "created_at": "2026-11-22T04:01:19Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-b2471cff136f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "b00e37f51b65ab8cefd6343bfd34488a5250b0fe0a5e535db41b159c7ee11469", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-b2471cff136f", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:01:19Z", "seed": 1750881146, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "3d13061d4f174df1ec08d86fa8cdf100fac1d07ca5fea55b63a7ea2639c265cd", "run_seconds": 193.57807683944702, "seed": 1750881146, "started_at": "2026-11-22T04:01:24Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 3, "inherit_mem": 3, "inherit_nomem": 1, "random_search": 1, "reset_mem": 3, "retrieval": 0, "rootcause": 3, "xfer_mem": 2}, "better": "higher", "ci95": [-0.049442, 0.105466], "effect": 0.028012, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:04:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-b2471cff136f", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 10, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 10, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 10, "goal_bias": 0.3, "history2": true, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 10, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 10, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.334821429, 1.0, 0.921875, 0.303294574, 0.256862745, 0.628620352, 0.852272727, 0.944444444, 0.791666667, 0.90625], "compressed": [0.348214286, 1.0, 0.9296875, 0.669573643, 0.258039216, 0.628620352, 0.892045455, 0.944444444, 0.744047619, 0.805555556], "inherit_mem": [0.348214286, 1.0, 0.9296875, 0.669573643, 0.258039216, 0.628620352, 0.892045455, 0.944444444, 0.744047619, 0.805555556], "inherit_nomem": [0.334821429, 1.0, 0.921875, 0.470930233, 0.256862745, 0.628620352, 0.852272727, 0.944444444, 0.68452381, 0.90625], "random_search": [0.334821429, 1.0, 0.921875, 0.470930233, 0.256862745, 0.628620352, 0.852272727, 0.944444444, 0.791666667, 0.90625], "reset_mem": [0.348214286, 1.0, 0.9296875, 0.669573643, 0.258039216, 0.628620352, 0.892045455, 0.944444444, 0.744047619, 0.805555556], "retrieval": [0.334821429, 1.0, 0.921875, 0.303294574, 0.256862745, 0.628620352, 0.852272727, 0.944444444, 0.791666667, 0.90625], "rootcause": [0.348214286, 1.0, 0.9296875, 0.669573643, 0.258039216, 0.628620352, 0.892045455, 0.944444444, 0.744047619, 0.805555556], "xfer_mem": [0.348214286, 1.0, 0.9296875, 0.669573643, 0.258039216, 0.628620352, 0.892045455, 0.944444444, 0.678571429, 0.805555556]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.6448, "losses": 2, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.227", "per_generation_sealed": [0.694011, 0.685395, 0.715475, 0.722023], "prereg_hash": "b00e37f51b65ab8cefd6343bfd34488a5250b0fe0a5e535db41b159c7ee11469", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "retrieval", "rootcause"], "result_hash": "3d13061d4f174df1ec08d86fa8cdf100fac1d07ca5fea55b63a7ea2639c265cd", "signature": "c12c4ed21144ffcf7944d4ac50cead9c7b410ce3deef80b554b9dcfbdaa16cbb75d708e92293a7f95c77d4ee33c46d3ad42f6114b316a983d571e077c2e92300", "state": "VERIFIED", "summary": {"G0": 0.694011, "compressed": 0.722023, "inherit_mem": 0.722023, "inherit_nomem": 0.70006, "random_search": 0.710774, "reset_mem": 0.722023, "retrieval": 0.694011, "rootcause": 0.722023, "xfer_mem": 0.715475}, "ties": 3, "treatment": "inherit_mem", "wins": 5}, "verified_at": "2026-11-22T04:07:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:38Z", "cpu_seconds": 0.3150865009999997, "created_at": "2026-11-22T04:01:36Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-376c6fbd6786", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "0ae3246eafd31f404ab63932adb5e6018ee4067d1eabb8c1d6f28d23dc96ab04", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-376c6fbd6786", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:01:36Z", "seed": 294673589, "treatment": "frontier"}, "priority": 0.26640083937603065, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "178f29a80fb8164ed354e5516548b2128d0165e2238061aba53a84b88c03447f", "run_seconds": 0.3469088077545166, "seed": 294673589, "started_at": "2026-11-22T04:01:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.022911, 0.128604], "effect": 0.075758, "evaluated_at": "2026-11-22T04:03:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-376c6fbd6786", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.636363636, 0.454545455, 0.318181818, 0.545454545], "frontier": [0.772727273, 0.696969697, 0.636363636, 0.651515152], "goal": [0.46969697, 0.621212121, 0.621212121, 0.545454545], "hypdisc": [0.606060606, 0.515151515, 0.651515152, 0.651515152], "infogain": [0.818181818, 0.636363636, 0.712121212, 0.742424242], "prederr": [0.833333333, 0.303030303, 0.712121212, 0.742424242], "random": [0.621212121, 0.621212121, 0.606060606, 0.606060606], "risk": [0.772727273, 0.696969697, 0.636363636, 0.651515152]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "0ae3246eafd31f404ab63932adb5e6018ee4067d1eabb8c1d6f28d23dc96ab04", "primary_control": "random", "result_hash": "178f29a80fb8164ed354e5516548b2128d0165e2238061aba53a84b88c03447f", "signature": "31f95dd4c54849d82f1c6f14c977bc531958e1b4e9326f1e6ebf0855833dad28111c6abf2900182765f9f6853d508142ff5a41b5f2201a97a097af68b693280b", "state": "VERIFIED", "summary": {"count": 0.488636, "frontier": 0.689394, "goal": 0.564394, "hypdisc": 0.606061, "infogain": 0.727273, "prederr": 0.647727, "random": 0.613636, "risk": 0.689394}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:03:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:50Z", "cpu_seconds": 0.6121956340000003, "created_at": "2026-11-22T04:01:36Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-47c0670979f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "8b2d0ad14c67ac6cf880a83f89fba5bf97f314022444177260a1daed094dcd19", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-47c0670979f0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:01:36Z", "seed": 92120540, "treatment": "frontier"}, "priority": 0.2554574910543945, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "87374819bb5bf7c55ab7de8476b1406574cbebaa5c086ee079165bbb43eb3525", "run_seconds": 0.6954081058502197, "seed": 92120540, "started_at": "2026-11-22T04:01:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.146208, 0.206064], "effect": 0.176136, "evaluated_at": "2026-11-22T04:03:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-47c0670979f0", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.787878788, 0.901515152, 0.863636364, 0.803030303], "frontier": [0.833333333, 0.909090909, 0.893939394, 0.856060606], "goal": [0.659090909, 0.787878788, 0.71969697, 0.666666667], "hypdisc": [0.537878788, 0.803030303, 0.757575758, 0.704545455], "infogain": [0.833333333, 0.909090909, 0.893939394, 0.856060606], "prederr": [0.833333333, 0.909090909, 0.893939394, 0.856060606], "random": [0.613636364, 0.75, 0.71969697, 0.704545455], "risk": [0.833333333, 0.909090909, 0.893939394, 0.856060606]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "8b2d0ad14c67ac6cf880a83f89fba5bf97f314022444177260a1daed094dcd19", "primary_control": "random", "result_hash": "87374819bb5bf7c55ab7de8476b1406574cbebaa5c086ee079165bbb43eb3525", "signature": "4eb68556297af282cc87e40a7997ed9fa38b14ecc32d21f3bd73711c65c594fe7dafa8670dc42eef717ad1f02cd5e78a5a288b8196c4cabdce8d147eb4e8a102", "state": "VERIFIED", "summary": {"count": 0.839015, "frontier": 0.873106, "goal": 0.708333, "hypdisc": 0.700758, "infogain": 0.873106, "prederr": 0.873106, "random": 0.69697, "risk": 0.873106}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:03:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:44Z", "cpu_seconds": 1.1822194189999995, "created_at": "2026-11-22T04:01:36Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b0b9b57782bd", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "ccb507028e1a3256e0a85feebdcc1aa8afc19949bfcb77cc2267b3c77db936f7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b0b9b57782bd", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:01:36Z", "seed": 138403957, "treatment": "frontier"}, "priority": 0.26040665152557096, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-b4d8d0cb8ae9", "result_hash": "c7bb6d910361c26773a3becbeefcbeead39c60049d5d40f4151d678db812046a", "run_seconds": 1.3153576850891113, "seed": 138403957, "started_at": "2026-11-22T04:01:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.091315, 0.014815], "effect": -0.03825, "evaluated_at": "2026-11-22T04:03:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b0b9b57782bd", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.726351351, 0.928571429, 0.81779661, 0.645833333], "frontier": [0.64527027, 0.928571429, 0.669491525, 0.791666667], "goal": [0.64527027, 0.928571429, 0.665254237, 0.791666667], "hypdisc": [0.631756757, 0.928571429, 0.665254237, 0.791666667], "infogain": [0.64527027, 0.928571429, 0.669491525, 0.791666667], "prederr": [0.64527027, 0.928571429, 0.56779661, 0.791666667], "random": [0.760135135, 0.928571429, 0.707627119, 0.791666667], "risk": [0.64527027, 0.928571429, 0.669491525, 0.791666667]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "ccb507028e1a3256e0a85feebdcc1aa8afc19949bfcb77cc2267b3c77db936f7", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-b4d8d0cb8ae9", "result_hash": "c7bb6d910361c26773a3becbeefcbeead39c60049d5d40f4151d678db812046a", "signature": "fab05c34847d1ba123941d76f5b1567a4269f5656ce605dbacade0739c0fb9479034aa3d8911b2daab89aed03b4fae26ab01f7e579163992d417e0c4ca227608", "state": "VERIFIED", "summary": {"count": 0.779638, "frontier": 0.75875, "goal": 0.757691, "hypdisc": 0.754312, "infogain": 0.75875, "prederr": 0.733326, "random": 0.797, "risk": 0.75875}, "ties": 2, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:03:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:50Z", "cpu_seconds": 0.6219202500000005, "created_at": "2026-11-22T04:01:36Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-fb6e70252d06", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "9b18ecda5e09556ebe9ade1d70c6d2dd2da92da1e3d957b60ff74e924a2610f7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-fb6e70252d06", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:01:36Z", "seed": 220478508, "treatment": "full"}, "priority": 0.2550026583068819, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3adf460c5755b1a5089b11953eb5e4e85d6e24a29c94944761de22d783c40b05", "run_seconds": 0.7057461738586426, "seed": 220478508, "started_at": "2026-11-22T04:01:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018585, 0.021951], "effect": 0.001683, "evaluated_at": "2026-11-22T04:03:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-fb6e70252d06", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.77294686, 0.738562092, 0.846153846, 0.793939394], "frozen": [0.299516908, 0.320261438, 0.41025641, 0.321212121], "full": [0.777777778, 0.764705882, 0.846153846, 0.76969697], "retrieval": [0.777777778, 0.738562092, 0.846153846, 0.793939394], "rules_nodetect": [0.758454106, 0.790849673, 0.846153846, 0.76969697]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "9b18ecda5e09556ebe9ade1d70c6d2dd2da92da1e3d957b60ff74e924a2610f7", "primary_control": "facts", "result_hash": "3adf460c5755b1a5089b11953eb5e4e85d6e24a29c94944761de22d783c40b05", "signature": "31964a5c7519349ec86435c40e95db041d2ad4c70071db44b929baacc0fef89b540d03d7949cca3b7267722f2f611eb9c876f8702f1ede0ac558554d67112f09", "state": "VERIFIED", "summary": {"facts": 0.787901, "frozen": 0.337812, "full": 0.789584, "retrieval": 0.789108, "rules_nodetect": 0.791289}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:44Z", "cpu_seconds": 0.21220956599999985, "created_at": "2026-11-22T04:01:36Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-09c7f1c09160", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "prereg_hash": "4599541033cbd0c42a28fbc2fe97e45711476200a39e054bd983c29bdaab6aff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-09c7f1c09160", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:01:36Z", "seed": 262314264, "treatment": "gated"}, "priority": 0.25561387358463483, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "bbdc0f8b9f2147bfd776d64f6eb58232c65f0d9b8d79e26e03b6028f7de02fc0", "run_seconds": 0.24192190170288086, "seed": 262314264, "started_at": "2026-11-22T04:01:44Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-09c7f1c09160", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.439393939, 0.363636364, 0.484848485, 0.166666667], "gated": [0.439393939, 0.363636364, 0.484848485, 0.166666667], "naive": [0.439393939, 0.363636364, 0.212121212, 0.166666667], "scratch": [0.439393939, 0.363636364, 0.484848485, 0.166666667], "ungated": [0.439393939, 0.363636364, 0.212121212, 0.166666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4599541033cbd0c42a28fbc2fe97e45711476200a39e054bd983c29bdaab6aff", "primary_control": "scratch", "result_hash": "bbdc0f8b9f2147bfd776d64f6eb58232c65f0d9b8d79e26e03b6028f7de02fc0", "signature": "eddce4f0973114f4ef91498c0376e10fb4ac7e66377d2b031031990685ffd7fc43b28ffbd33f2b855b050c38ddcc28904decb2b174c971879d71c62b8c4e7e0d", "state": "PARTIAL", "summary": {"compressed": 0.363636, "gated": 0.363636, "naive": 0.295455, "scratch": 0.363636, "ungated": 0.295455}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:03:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:39Z", "cpu_seconds": 0.18665409900000007, "created_at": "2026-11-22T04:01:36Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a2e17c00a18e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "prereg_hash": "c982089429e3340ad3f05bd69e598a09548a237e3512a96756a07c0c9f4af783", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a2e17c00a18e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:01:36Z", "seed": 116141835, "treatment": "gated"}, "priority": 0.26322491308268364, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "b71c0268caf171575d5e222a48ce89ab9ab8acc98e181f30836f00950f72934f", "run_seconds": 0.2280123233795166, "seed": 116141835, "started_at": "2026-11-22T04:01:38Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a2e17c00a18e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.181818182, 0.272727273, 0.348484848, 0.303030303], "gated": [0.181818182, 0.272727273, 0.348484848, 0.303030303], "naive": [0.090909091, 0.272727273, 0.348484848, 0.090909091], "scratch": [0.181818182, 0.272727273, 0.348484848, 0.303030303], "ungated": [0.090909091, 0.272727273, 0.348484848, 0.090909091]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c982089429e3340ad3f05bd69e598a09548a237e3512a96756a07c0c9f4af783", "primary_control": "scratch", "result_hash": "b71c0268caf171575d5e222a48ce89ab9ab8acc98e181f30836f00950f72934f", "signature": "8f79c83f57c6792bc22259b228fe37321ae9ac0c93fa5b0e405cbfe4c81d0a537a56350ae8ab3be3f45c1b8f308f7b0e84cf5fda167f58c4309e278713fd0206", "state": "PARTIAL", "summary": {"compressed": 0.276515, "gated": 0.276515, "naive": 0.200758, "scratch": 0.276515, "ungated": 0.200758}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:03:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:07Z", "cpu_seconds": 0.5649569459999997, "created_at": "2026-11-22T04:01:52Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d8ee38f1d0f3", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "7cc69d9bd4ec0adcf58a4c9f07ebb40f9d745d27c51852cb3b5d95eafc560198", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d8ee38f1d0f3", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:01:52Z", "seed": 1439520736, "treatment": "frontier"}, "priority": 0.2575512944225674, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "dfadf49f33442df9b4e60a7d6b3ff242996f4d34b8c28ea72dcc2a78abae5a46", "run_seconds": 0.6367559432983398, "seed": 1439520736, "started_at": "2026-11-22T04:02:06Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:03:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d8ee38f1d0f3", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.898065], "frontier": [1.0, 0.924107], "goal": [1.0, 0.84375], "hypdisc": [1.0, 0.921875], "infogain": [1.0, 0.943452], "prederr": [1.0, 0.749256], "random": [1.0, 0.875744], "risk": [1.0, 0.924107]}, "prereg_hash": "7cc69d9bd4ec0adcf58a4c9f07ebb40f9d745d27c51852cb3b5d95eafc560198", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "dfadf49f33442df9b4e60a7d6b3ff242996f4d34b8c28ea72dcc2a78abae5a46", "signature": "e3f1615dc87004e9f55d763ad6417a6edd773e03891bc2e7c6355606782335401ec1b12913662423dee0092da98047c20b90bfb32747cef43f947eefcd3b5d05", "state": "REJECTED", "summary": {"count": 0.898065, "frontier": 0.924107, "goal": 0.84375, "hypdisc": 0.921875, "infogain": 0.943452, "prederr": 0.749256, "random": 0.875744, "risk": 0.924107}}, "verified_at": "2026-11-22T04:03:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:59Z", "cpu_seconds": 3.1964683780000005, "created_at": "2026-11-22T04:01:52Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-da02425004dd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "b9225956749cb4819fe8925597e28c083daa9ce10d521a3f7cb08b2d2b59e495", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-da02425004dd", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:01:52Z", "seed": 1891335600, "treatment": "frontier"}, "priority": 0.25960054548486683, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "8077a47d7ba1944813cd8237b6f8796f089e6bc629c1413c10135510c8785281", "run_seconds": 3.686079740524292, "seed": 1891335600, "started_at": "2026-11-22T04:01:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.229147, 0.130239], "effect": -0.049454, "evaluated_at": "2026-11-22T04:03:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-da02425004dd", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.451388889, 0.366935484, 0.982142857, 0.33213141], "frontier": [0.4921875, 0.400345622, 0.589285714, 0.338942308], "goal": [0.410590278, 0.343894009, 0.607142857, 0.309695513], "hypdisc": [0.421006944, 0.361175115, 0.857142857, 0.330929487], "infogain": [0.4921875, 0.400345622, 0.589285714, 0.338942308], "prederr": [0.4921875, 0.400345622, 0.589285714, 0.338942308], "random": [0.414930556, 0.365207373, 0.910714286, 0.327724359], "risk": [0.4921875, 0.400345622, 0.589285714, 0.338942308]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "b9225956749cb4819fe8925597e28c083daa9ce10d521a3f7cb08b2d2b59e495", "primary_control": "random", "result_hash": "8077a47d7ba1944813cd8237b6f8796f089e6bc629c1413c10135510c8785281", "signature": "6a74360132d9d37f2600b95a1b80678304711297c57faece8d1d59edaacdc0d218f0530f8ab8a6c09f22000d0f6e83c4f8e6d3c017288596d67b298b8046810b", "state": "VERIFIED", "summary": {"count": 0.53315, "frontier": 0.45519, "goal": 0.417831, "hypdisc": 0.492564, "infogain": 0.45519, "prederr": 0.45519, "random": 0.504644, "risk": 0.45519}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:03:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:10Z", "cpu_seconds": 0.6249847549999998, "created_at": "2026-11-22T04:01:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-10247be50e80", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "3d5249aa3defee8cc63d21cd69f076de0a5c827e0dd3b38fdc31e291da80060f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-10247be50e80", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:01:52Z", "seed": 377540444, "treatment": "full"}, "priority": 0.2562843038091918, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a611e3bfaa85e10f87906188a51d0783ed422c15f27ea5f19b8b0313c587290a", "run_seconds": 0.8675315380096436, "seed": 377540444, "started_at": "2026-11-22T04:02:09Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-10247be50e80", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.929824561, 0.907407407, 0.944444444, 0.923076923], "frozen": [0.473684211, 0.481481481, 0.5, 0.538461538], "full": [0.929824561, 0.907407407, 0.944444444, 0.923076923], "retrieval": [0.929824561, 0.907407407, 0.944444444, 0.923076923], "rules_nodetect": [0.894736842, 0.944444444, 0.944444444, 0.923076923]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3d5249aa3defee8cc63d21cd69f076de0a5c827e0dd3b38fdc31e291da80060f", "primary_control": "facts", "result_hash": "a611e3bfaa85e10f87906188a51d0783ed422c15f27ea5f19b8b0313c587290a", "signature": "37dafeebc238d157d8fee83a4b34709f6a1c512460d1e9c632186e9d4fb641fd4da8be437c069e6f914fe0d3b87b8a315ded396f843d334fae904a5a160e9007", "state": "PARTIAL", "summary": {"facts": 0.926188, "frozen": 0.498407, "full": 0.926188, "retrieval": 0.926188, "rules_nodetect": 0.926676}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:03:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:01:56Z", "cpu_seconds": 0.3405217880000002, "created_at": "2026-11-22T04:01:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-27b536a3889e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "161a04bd5e629d10a88dfee92e3a2ed97acff2dc5f7ff2c55f0f73dd73c7f73b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-27b536a3889e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:01:52Z", "seed": 1354358534, "treatment": "full"}, "priority": 0.25890933083911505, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e80a0aa0d377f4e33a39e2f1d885e48c913a2c63cd71c09af5fe59c0c2aae979", "run_seconds": 0.5182666778564453, "seed": 1354358534, "started_at": "2026-11-22T04:01:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.046415, 0.193169], "effect": 0.119792, "evaluated_at": "2026-11-22T04:03:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-27b536a3889e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.770833333, 0.9375, 0.916666667, 0.875], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 0.979166667], "retrieval": [0.895833333, 0.9375, 1.0, 0.958333333], "rules_nodetect": [1.0, 1.0, 1.0, 0.979166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "161a04bd5e629d10a88dfee92e3a2ed97acff2dc5f7ff2c55f0f73dd73c7f73b", "primary_control": "facts", "result_hash": "e80a0aa0d377f4e33a39e2f1d885e48c913a2c63cd71c09af5fe59c0c2aae979", "signature": "feb2b411556319c02851a11f133eb2f5dfa1902614993090832340f1bbe4534e256d9989f79b7c42a20671ef9a8aba0ac239c1851d82ef7ebff81af890f70809", "state": "VERIFIED", "summary": {"facts": 0.875, "frozen": 0.0, "full": 0.994792, "retrieval": 0.947917, "rules_nodetect": 0.994792}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:03:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:04Z", "cpu_seconds": 0.6030493060000008, "created_at": "2026-11-22T04:01:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-77b3137bc021", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "5159efb68202ffe197c0c8e8a0d03d027cc6fe26ee4042ac0f1c40b854098bdc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-77b3137bc021", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:01:52Z", "seed": 1121847638, "treatment": "gated"}, "priority": 0.2582036739239758, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7b6d92e064cbb3c0d2807e74a55fa9242557cef2bc5018afb35e17bfe5a9b17e", "run_seconds": 0.6067605018615723, "seed": 1121847638, "started_at": "2026-11-22T04:02:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004706, 0.01451], "effect": 0.004902, "evaluated_at": "2026-11-22T04:03:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-77b3137bc021", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.647058824, 0.614379085, 0.666666667, 0.666666667], "gated": [0.647058824, 0.614379085, 0.666666667, 0.666666667], "naive": [0.549019608, 0.62745098, 0.733333333, 0.532163743], "scratch": [0.62745098, 0.614379085, 0.666666667, 0.666666667], "ungated": [0.607843137, 0.614379085, 0.666666667, 0.666666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "5159efb68202ffe197c0c8e8a0d03d027cc6fe26ee4042ac0f1c40b854098bdc", "primary_control": "scratch", "result_hash": "7b6d92e064cbb3c0d2807e74a55fa9242557cef2bc5018afb35e17bfe5a9b17e", "signature": "e576f031bf889c198619dda517f917529a1ae366e720033738a86f5b033333ddbf652d26a8440a183288d4f0e410fe4738710f1afed40689b8fb933e187c7601", "state": "VERIFIED", "summary": {"compressed": 0.648693, "gated": 0.648693, "naive": 0.610492, "scratch": 0.643791, "ungated": 0.638889}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:03:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:01Z", "cpu_seconds": 0.6006183540000007, "created_at": "2026-11-22T04:01:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-81a45483ab66", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "4510c93bf492a3efe92a79d7b3d39b472a8befe1aed296377bec6c67eb9cdb51", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-81a45483ab66", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:01:52Z", "seed": 470231926, "treatment": "gated"}, "priority": 0.25860700742407894, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "69c96ce88d6b76f1c8cbf13a3f4a77aa79d1fb5df14770a21306e3cef0e9dc1e", "run_seconds": 0.6975123882293701, "seed": 470231926, "started_at": "2026-11-22T04:02:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0025, 0.007708], "effect": 0.002604, "evaluated_at": "2026-11-22T04:03:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-81a45483ab66", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.434108527, 0.483660131, 0.489583333, 0.341085271], "gated": [0.434108527, 0.483660131, 0.489583333, 0.341085271], "naive": [0.217054264, 0.333333333, 0.489583333, 0.286821705], "scratch": [0.434108527, 0.483660131, 0.479166667, 0.341085271], "ungated": [0.356589147, 0.483660131, 0.489583333, 0.286821705]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "4510c93bf492a3efe92a79d7b3d39b472a8befe1aed296377bec6c67eb9cdb51", "primary_control": "scratch", "result_hash": "69c96ce88d6b76f1c8cbf13a3f4a77aa79d1fb5df14770a21306e3cef0e9dc1e", "signature": "e71a2d5138e659e0437e6db338b9633d5cd965187f8b5bb031def61433e1d3fae452d430ea886b1ac4597a661f3cbbc4e84a331807b69818f0bfec04304e420b", "state": "VERIFIED", "summary": {"compressed": 0.437109, "gated": 0.437109, "naive": 0.331698, "scratch": 0.434505, "ungated": 0.404164}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:03:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:12Z", "cpu_seconds": 0.2554941660000001, "created_at": "2026-11-22T04:02:09Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-46e9a2bce674", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "3ce06c654fc49e9fba09d38106a02f258fc73e711fb12d928041ad763a8b6a10", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-46e9a2bce674", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:02:09Z", "seed": 858602808, "treatment": "full"}, "priority": 0.26442213341184057, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d32664ee9c2e4482f39b1adc94815cfae866c5b13189e5c5394b69e22e98781f", "run_seconds": 0.27309679985046387, "seed": 858602808, "started_at": "2026-11-22T04:02:12Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-46e9a2bce674", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3ce06c654fc49e9fba09d38106a02f258fc73e711fb12d928041ad763a8b6a10", "primary_control": "e141_best", "result_hash": "d32664ee9c2e4482f39b1adc94815cfae866c5b13189e5c5394b69e22e98781f", "signature": "16f5dfb6884887c69780d99ffc7c605af25dc13e763261e0e8da65741cdc1d92cb479b6fc21d76b21534d2c7dd757d3cac5a9884d632179f9a8221475305100d", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:03:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:16Z", "cpu_seconds": 0.5661846170000011, "created_at": "2026-11-22T04:02:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7d634ea0cb11", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "e5d86a8d01b0803154b721339607175306e476e13e7c58cf0c5815e7ed387bbb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7d634ea0cb11", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:09Z", "seed": 1562465979, "treatment": "full"}, "priority": 0.2581366539512216, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "47959b48c462bfef6c3e26c45c72f225f1966ed028049ce76c5f86a2a10cd742", "run_seconds": 0.6191129684448242, "seed": 1562465979, "started_at": "2026-11-22T04:02:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.047643, 0.08716], "effect": 0.067402, "evaluated_at": "2026-11-22T04:03:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7d634ea0cb11", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.661764706, 0.696078431, 0.769607843, 0.397058824], "frozen": [0.073529412, 0.073529412, 0.068627451, 0.068627451], "full": [0.715686275, 0.789215686, 0.818627451, 0.470588235], "retrieval": [0.838235294, 0.852941176, 0.852941176, 0.671568627], "rules_nodetect": [0.715686275, 0.789215686, 0.818627451, 0.470588235]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "e5d86a8d01b0803154b721339607175306e476e13e7c58cf0c5815e7ed387bbb", "primary_control": "facts", "result_hash": "47959b48c462bfef6c3e26c45c72f225f1966ed028049ce76c5f86a2a10cd742", "signature": "b1413735b50759d8dafb8f45d7a47202d92cbaf8b282ec9477fec3119cc5d75d7d497d39aac0423cd5241e1453b45ec1ca3944f9e03d03fb7c99d04ec024b500", "state": "VERIFIED", "summary": {"facts": 0.631127, "frozen": 0.071078, "full": 0.698529, "retrieval": 0.803922, "rules_nodetect": 0.698529}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:24Z", "cpu_seconds": 0.6322257679999996, "created_at": "2026-11-22T04:02:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8d72cd281d94", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "78e8d0bb3c2ab0f4eef961cf3beb4c087742d16e1fbd9cfb7691578eb9022b46", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8d72cd281d94", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:09Z", "seed": 8506458, "treatment": "full"}, "priority": 0.2541181523695314, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5d1aa183bca220fab104c9bb3bd08ae939526c7b1f8e41a4b3a73d74d15c7480", "run_seconds": 0.6332478523254395, "seed": 8506458, "started_at": "2026-11-22T04:02:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.057363, 0.086509], "effect": 0.014573, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8d72cd281d94", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.525, 0.78125, 0.852941176, 0.464285714], "frozen": [0.45, 0.25, 0.338235294, 0.25], "full": [0.45, 0.8125, 0.852941176, 0.566326531], "retrieval": [0.475, 0.75, 0.852941176, 0.658163265], "rules_nodetect": [0.925, 0.8125, 0.852941176, 0.566326531]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "78e8d0bb3c2ab0f4eef961cf3beb4c087742d16e1fbd9cfb7691578eb9022b46", "primary_control": "facts", "result_hash": "5d1aa183bca220fab104c9bb3bd08ae939526c7b1f8e41a4b3a73d74d15c7480", "signature": "879e462d0ebaa6d48313b08ccc115dbc4b90a02b8af69be0bb399a2c4398b58af3890e2fe6f6842a2bb549de32aaff83af43956d71d005733475af4b39cdfa08", "state": "VERIFIED", "summary": {"facts": 0.655869, "frozen": 0.322059, "full": 0.670442, "retrieval": 0.684026, "rules_nodetect": 0.789192}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:19Z", "cpu_seconds": 1.0708133530000001, "created_at": "2026-11-22T04:02:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f2bf98c8ddb4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "8fb51f4ef98da4458f0cd6323d8ef33e8fcb5422681cc80250b3b845f859a5a7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f2bf98c8ddb4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:09Z", "seed": 1835541133, "treatment": "full"}, "priority": 0.25513821950501503, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c438b72f01ab915bd51aa4d476efdc3de6adb447868e78c546e01ff913c23776", "run_seconds": 1.2144801616668701, "seed": 1835541133, "started_at": "2026-11-22T04:02:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.03868, 0.061228], "effect": 0.011274, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f2bf98c8ddb4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.565714286, 0.477419355, 0.242553191, 0.570666667], "frozen": [0.137142857, 0.135483871, 0.127659574, 0.128], "full": [0.517142857, 0.535483871, 0.291489362, 0.557333333], "retrieval": [0.562857143, 0.529032258, 0.19787234, 0.554666667], "rules_nodetect": [0.225714286, 0.458064516, 0.393617021, 0.266666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "8fb51f4ef98da4458f0cd6323d8ef33e8fcb5422681cc80250b3b845f859a5a7", "primary_control": "facts", "result_hash": "c438b72f01ab915bd51aa4d476efdc3de6adb447868e78c546e01ff913c23776", "signature": "89d04874c5d871936fa584f880ed97016c87a0d4cf7bf7bfa93cedcb06a67584344cafa5665be22aa8cf660daa051bf5d42aa849303c1183797ef7dd8da78a0e", "state": "VERIFIED", "summary": {"facts": 0.464088, "frozen": 0.132072, "full": 0.475362, "retrieval": 0.461107, "rules_nodetect": 0.336016}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:21Z", "cpu_seconds": 0.4734318670000004, "created_at": "2026-11-22T04:02:09Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1b6528aac1ad", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "prereg_hash": "b5f721eebac22c527a7d25357b5701b65d8e2ade9432ca7d846310d3ba101ff4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1b6528aac1ad", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:02:09Z", "seed": 425921852, "treatment": "gated"}, "priority": 0.2541839438408585, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7ba1678482e88a91c12216d486cdd92676edc04b071022b337945570be1e0bc6", "run_seconds": 0.5198562145233154, "seed": 425921852, "started_at": "2026-11-22T04:02:21Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1b6528aac1ad", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.875, 0.8125, 0.675, 0.833333333], "gated": [0.875, 0.8125, 0.675, 0.833333333], "naive": [0.208333333, 0.375, 0.275, 0.333333333], "scratch": [0.875, 0.8125, 0.675, 0.833333333], "ungated": [0.5, 0.479166667, 0.6, 0.625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b5f721eebac22c527a7d25357b5701b65d8e2ade9432ca7d846310d3ba101ff4", "primary_control": "scratch", "result_hash": "7ba1678482e88a91c12216d486cdd92676edc04b071022b337945570be1e0bc6", "signature": "10e95bf28fecb6e182d6db8332c1381b233f26590dac645e10d2e47576d1b91eeee74074598ea6430f1e7cb90bf000b8ab1ee76cbd3993b325a5d07804219d05", "state": "PARTIAL", "summary": {"compressed": 0.798958, "gated": 0.798958, "naive": 0.297917, "scratch": 0.798958, "ungated": 0.551042}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:38Z", "cpu_seconds": 6.230820818, "created_at": "2026-11-22T04:02:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-20e5b549e995", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "db1ac75d6db5ea9c6c4912533004f449d4aaef9382ae81b5a096e3c7af97f863", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-20e5b549e995", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:02:26Z", "seed": 436259628, "treatment": "frontier"}, "priority": 0.2608977139237181, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d78185a463c1dfdab06b3a5788c1412490809f9b149a8b5fa84ca16e6eb75e71", "run_seconds": 6.427042245864868, "seed": 436259628, "started_at": "2026-11-22T04:02:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.096787, 0.09774], "effect": 0.000477, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-20e5b549e995", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.71875, 0.962962963, 0.776470588, 0.84375], "frontier": [0.71875, 0.814814815, 0.929411765, 0.78125], "goal": [0.71875, 0.962962963, 0.929411765, 0.78125], "hypdisc": [0.53125, 0.962962963, 0.952941176, 0.84375], "infogain": [0.71875, 0.962962963, 0.623529412, 0.78125], "prederr": [0.71875, 0.342592593, 0.505882353, 0.78125], "random": [0.71875, 0.805555556, 0.811764706, 0.90625], "risk": [0.71875, 0.814814815, 0.929411765, 0.84375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "db1ac75d6db5ea9c6c4912533004f449d4aaef9382ae81b5a096e3c7af97f863", "primary_control": "random", "result_hash": "d78185a463c1dfdab06b3a5788c1412490809f9b149a8b5fa84ca16e6eb75e71", "signature": "1c4f755d194f7356c6a08a1050c950fce1b46ec3b09144b5cba45e81d248013634f40bb404d2f324ab9349514376891f1cbe5c8d46311a64ad523ae249afdc03", "state": "VERIFIED", "summary": {"count": 0.825483, "frontier": 0.811057, "goal": 0.848094, "hypdisc": 0.822726, "infogain": 0.771623, "prederr": 0.587119, "random": 0.81058, "risk": 0.826682}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:27Z", "cpu_seconds": 0.4045875809999995, "created_at": "2026-11-22T04:02:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-2f857173df39", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "64244b71265e8d78bc3ee738c4141379fbff17734a8d1d0b4ced835bca9ddeb4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2f857173df39", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:02:26Z", "seed": 1473553619, "treatment": "frontier"}, "priority": 0.26286447026130993, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4c835fa8dcfa9eb7e4fc0f2be6accd2c8bf6667e1baf1b0e156db42237e7f68b", "run_seconds": 0.5600779056549072, "seed": 1473553619, "started_at": "2026-11-22T04:02:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.031264, 0.15813], "effect": 0.094697, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-2f857173df39", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.96969697, 1.0], "frontier": [0.984848485, 0.984848485, 0.96969697, 0.818181818], "goal": [0.772727273, 0.772727273, 0.863636364, 0.666666667], "hypdisc": [0.893939394, 0.939393939, 0.984848485, 0.893939394], "infogain": [0.984848485, 0.984848485, 0.96969697, 0.818181818], "prederr": [0.984848485, 0.984848485, 0.96969697, 0.818181818], "random": [0.878787879, 0.848484848, 0.833333333, 0.818181818], "risk": [0.984848485, 0.984848485, 0.96969697, 0.818181818]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "64244b71265e8d78bc3ee738c4141379fbff17734a8d1d0b4ced835bca9ddeb4", "primary_control": "random", "result_hash": "4c835fa8dcfa9eb7e4fc0f2be6accd2c8bf6667e1baf1b0e156db42237e7f68b", "signature": "4d6a4085d3b7477627569aeff1b3150293e09301c5ff607878f4ca360397a55c7f8e560b0cce1d8bd660cd634d829998269aa18bd42345eb8d20007ad8e0190f", "state": "VERIFIED", "summary": {"count": 0.992424, "frontier": 0.939394, "goal": 0.768939, "hypdisc": 0.92803, "infogain": 0.939394, "prederr": 0.939394, "random": 0.844697, "risk": 0.939394}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:30Z", "cpu_seconds": 0.3596633229999995, "created_at": "2026-11-22T04:02:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1a9945072230", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "ce7e4fb5eefbba923aa2404ebbc972636fe8ff831eb3370f721ed0e4ff2f06c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1a9945072230", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:02:26Z", "seed": 2015269625, "treatment": "full"}, "priority": 0.26254872150375674, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-5c49646123af", "result_hash": "8debf58c069acfa19709585dda83f2f131f16dc564e0a5090b489222c15cc1c4", "run_seconds": 0.38900089263916016, "seed": 2015269625, "started_at": "2026-11-22T04:02:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.549352, 0.299352], "effect": -0.125, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1a9945072230", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 0.75, 0.0], "fixed_policy": [0.0, 0.5, 0.25, 0.5], "full": [0.25, 0.5, 0.5, 0.5], "no_world_model": [0.0, 0.5, 0.0, 0.5], "random": [0.75, 0.75, 0.75, 0.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "ce7e4fb5eefbba923aa2404ebbc972636fe8ff831eb3370f721ed0e4ff2f06c6", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-5c49646123af", "result_hash": "8debf58c069acfa19709585dda83f2f131f16dc564e0a5090b489222c15cc1c4", "signature": "6f450adf846e2b746be2ae2e579687b97e7517460479c3b48f6e0b024ca0c810345fb751f87b7169888488ed950769e8aaa9294c696ad971ea93bb1f5a39ee0c", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.3125, "full": 0.4375, "no_world_model": 0.25, "random": 0.5625}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:55Z", "cpu_seconds": 0.5923938559999993, "created_at": "2026-11-22T04:02:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-99562203d045", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "359ecd0c4d3283751a8e80ed0d42b080681f984b9c52144a467a3d0345734331", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-99562203d045", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:02:26Z", "seed": 437009545, "treatment": "full"}, "priority": 0.25453499820102415, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "afa7161e2fb2a48deea3f42830fce832d53a22808e166012143589c7929da41d", "run_seconds": 0.8127994537353516, "seed": 437009545, "started_at": "2026-11-22T04:02:54Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-99562203d045", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.5, 0.0, 0.75], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 1.0, 1.0], "random": [0.0, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "359ecd0c4d3283751a8e80ed0d42b080681f984b9c52144a467a3d0345734331", "primary_control": "e141_best", "result_hash": "afa7161e2fb2a48deea3f42830fce832d53a22808e166012143589c7929da41d", "signature": "283859acdc8e6891d5ba8fe6b65962b562129e851c4331b2ad2b8f43a3686ec6b52886593f37493696f693cd8f1fc915a54447d13c381696f19c6e3bf370670b", "state": "PARTIAL", "summary": {"e141_best": 0.75, "fixed_policy": 0.3125, "full": 0.75, "no_world_model": 0.75, "random": 0.6875}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:03:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:35Z", "cpu_seconds": 0.14888738700000026, "created_at": "2026-11-22T04:02:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1f85aca0a25c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "b20a608a1e8148541e80666fe4cfa73bf0d3c36fe6b122702959db8eee76bd06", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1f85aca0a25c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:26Z", "seed": 1955478628, "treatment": "full"}, "priority": 0.26006974764373425, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3b5c69ed5dfe2a16aef3da6c44e671395e6ee5a0833f498061db28d504217403", "run_seconds": 0.15928077697753906, "seed": 1955478628, "started_at": "2026-11-22T04:02:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002727, 0.008409], "effect": 0.002841, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1f85aca0a25c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.988636364, 0.977272727, 1.0], "frozen": [0.034090909, 0.034090909, 0.022727273, 0.034090909], "full": [1.0, 0.988636364, 0.988636364, 1.0], "retrieval": [0.988636364, 0.977272727, 0.988636364, 1.0], "rules_nodetect": [1.0, 0.988636364, 0.988636364, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "b20a608a1e8148541e80666fe4cfa73bf0d3c36fe6b122702959db8eee76bd06", "primary_control": "facts", "result_hash": "3b5c69ed5dfe2a16aef3da6c44e671395e6ee5a0833f498061db28d504217403", "signature": "5e65d03300b82aae7086fa917a75f124f71fa7970b8aca1265ac5d2297b8b6012f090dd0124e79ae1abb1d39f23bef78107c653c32929a8a26fac8fc1ff29b0c", "state": "VERIFIED", "summary": {"facts": 0.991477, "frozen": 0.03125, "full": 0.994318, "retrieval": 0.988636, "rules_nodetect": 0.994318}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:41Z", "cpu_seconds": 0.5656651159999999, "created_at": "2026-11-22T04:02:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ef83999d88b6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "a67201c696f081cfffaf5133d8017ba05c9bc65e8a4dc13f69be34781a71c96a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ef83999d88b6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:26Z", "seed": 1881894431, "treatment": "full"}, "priority": 0.25818616710787645, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d73ba9be6624ed63e1df5ab23dd8e2a85758026263a3484b3922da567905cabb", "run_seconds": 0.8274569511413574, "seed": 1881894431, "started_at": "2026-11-22T04:02:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026392, 0.081375], "effect": 0.027491, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ef83999d88b6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.8125, 0.820512821, 0.642611684, 0.717948718], "frozen": [0.354166667, 0.358974359, 0.281786942, 0.41025641], "full": [0.8125, 0.820512821, 0.75257732, 0.717948718], "retrieval": [0.8125, 0.820512821, 0.670103093, 0.743589744], "rules_nodetect": [0.395833333, 0.666666667, 0.649484536, 0.58974359]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a67201c696f081cfffaf5133d8017ba05c9bc65e8a4dc13f69be34781a71c96a", "primary_control": "facts", "result_hash": "d73ba9be6624ed63e1df5ab23dd8e2a85758026263a3484b3922da567905cabb", "signature": "7ed4005792a74ccd3de5c7e67ecd0b732991ea16d06df08cd95e35fa41aff430c7fb3e1a429abfa4117e1ee08e18da9850f67b7e8c3486588b80175f00243604", "state": "VERIFIED", "summary": {"facts": 0.748393, "frozen": 0.351296, "full": 0.775885, "retrieval": 0.761676, "rules_nodetect": 0.575432}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:03:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:57Z", "cpu_seconds": 11.118073438, "created_at": "2026-11-22T04:02:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5a25caa36a9a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "9da2822ae4041b911fa6a19ddcb4a3e3ce91328ac42cb2548a6e1dd29908bb5c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5a25caa36a9a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:02:43Z", "seed": 1104298993, "treatment": "frontier"}, "priority": 0.323576439204156, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-e02b7f210fc8", "result_hash": "ccd06dfcb663e78d8c94ccc1cac7f51c81c8ea8cf15ddeafac22a46c300a7067", "run_seconds": 13.056984901428223, "seed": 1104298993, "started_at": "2026-11-22T04:02:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.034139, 0.146676], "effect": 0.056269, "evaluated_at": "2026-11-22T04:03:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5a25caa36a9a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.788135593, 0.7575, 0.821428571, 1.0], "frontier": [0.733898305, 0.8825, 0.85, 1.0], "goal": [0.740677966, 0.839166667, 0.785714286, 1.0], "hypdisc": [0.737288136, 0.8125, 0.8125, 1.0], "infogain": [0.733898305, 0.8775, 0.85, 1.0], "prederr": [0.730508475, 0.819166667, 0.85, 1.0], "random": [0.750847458, 0.695833333, 0.794642857, 1.0], "risk": [0.733898305, 0.8825, 0.85, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "9da2822ae4041b911fa6a19ddcb4a3e3ce91328ac42cb2548a6e1dd29908bb5c", "primary_control": "random", "replicated": false, "replication_of": "E141-ow_explore-e02b7f210fc8", "result_hash": "ccd06dfcb663e78d8c94ccc1cac7f51c81c8ea8cf15ddeafac22a46c300a7067", "signature": "29b2d1c8d11352b9d3541a19fa536def925461097335b86aec23cae857d618f045f5d24d2922408e3a98142dd551b6fce49044d6b0baf3cacb095e0ca3b05206", "state": "VERIFIED", "summary": {"count": 0.841766, "frontier": 0.8666, "goal": 0.84139, "hypdisc": 0.840572, "infogain": 0.86535, "prederr": 0.849919, "random": 0.810331, "risk": 0.8666}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:03:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:15Z", "cpu_seconds": 0.5363861939999985, "created_at": "2026-11-22T04:02:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2bbe743211b1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "d2783fc61d95fbf99ef14c5e9d1283fa141533694b8b94e08431adfb47496118", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2bbe743211b1", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:43Z", "seed": 1094216128, "treatment": "full"}, "priority": 0.25158639777225555, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1187962504d1afa4185d9be123c21755173d480eb06b218e6793797e5d71a4ac", "run_seconds": 0.7890734672546387, "seed": 1094216128, "started_at": "2026-11-22T04:03:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.087908, 0.138697], "effect": 0.113302, "evaluated_at": "2026-11-22T04:03:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2bbe743211b1", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.719298246, 0.488888889, 0.685897436, 0.662790698], "frozen": [0.118421053, 0.111111111, 0.102564103, 0.108527132], "full": [0.807017544, 0.614814815, 0.782051282, 0.80620155], "retrieval": [0.903508772, 0.696296296, 0.801282051, 0.891472868], "rules_nodetect": [0.807017544, 0.614814815, 0.782051282, 0.80620155]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "d2783fc61d95fbf99ef14c5e9d1283fa141533694b8b94e08431adfb47496118", "primary_control": "facts", "result_hash": "1187962504d1afa4185d9be123c21755173d480eb06b218e6793797e5d71a4ac", "signature": "e6a5bb7e9b0cf77203a690bc28fc4329d485e763a0e6e90c7bd3975762ac11e27665756d70101108addfe34cbe6a98780d6fc4fcccb67ed80bf5f91fd0eabf0a", "state": "VERIFIED", "summary": {"facts": 0.639219, "frozen": 0.110156, "full": 0.752521, "retrieval": 0.82314, "rules_nodetect": 0.752521}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:03:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:02:49Z", "cpu_seconds": 2.930297412, "created_at": "2026-11-22T04:02:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4d542d372def", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "eb3df92be66bdced945d6b93c62387f8a3e3e51bed97b8d38b9a576a339158ab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4d542d372def", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:43Z", "seed": 318665852, "treatment": "full"}, "priority": 0.3122609082173468, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "53b96be8eaeffb04aa2c1dc881906474bcda9fab45beb9aa15495ef0476f882a", "run_seconds": 3.5029518604278564, "seed": 318665852, "started_at": "2026-11-22T04:02:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.011562, 0.3087], "effect": 0.160131, "evaluated_at": "2026-11-22T04:03:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4d542d372def", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.528571429, 1.0, 0.403846154, 0.260162602], "frozen": [0.308571429, 0.482758621, 0.178846154, 0.124661247], "full": [0.825714286, 1.0, 0.465384615, 0.54200542], "retrieval": [0.842857143, 1.0, 0.444230769, 0.577235772], "rules_nodetect": [0.292380952, 1.0, 0.155769231, 0.134146341]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "eb3df92be66bdced945d6b93c62387f8a3e3e51bed97b8d38b9a576a339158ab", "primary_control": "facts", "result_hash": "53b96be8eaeffb04aa2c1dc881906474bcda9fab45beb9aa15495ef0476f882a", "signature": "d831ad95546b4a675f00285923c378a9ee0d4f3c83fdcb33112074af7994bd3faa9d93ae3a5ef2639c3e9cda4047eecb9964d266e6c5b2e6a315f4929bbce506", "state": "VERIFIED", "summary": {"facts": 0.548145, "frozen": 0.273709, "full": 0.708276, "retrieval": 0.716081, "rules_nodetect": 0.395574}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:03:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:35Z", "cpu_seconds": 490.118075661, "created_at": "2026-11-22T04:02:43Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-91bcf07c5977", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "2a6c4cc1de790e56e6d42394b13d7e64e9f88280640ab37d70dc7f4cfa4436a4", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-91bcf07c5977", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:02:43Z", "seed": 1301485900, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "af3f9a1b19fc2275ed14265438a6191184d887df7ada9b78ef91b7e378cda163", "run_seconds": 527.7159161567688, "seed": 1301485900, "started_at": "2026-11-22T04:02:47Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"compressed": 1, "inherit_mem": 1, "inherit_nomem": 1, "random_search": 1, "reset_mem": 1, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:11:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-91bcf07c5977", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.59, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "compressed": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "inherit_mem": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "inherit_nomem": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "random_search": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "reset_mem": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "retrieval": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "rootcause": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345], "xfer_mem": [0.609090909, 1.0, 0.98, 0.830065359, 0.244554455, 0.713815789, 0.803030303, 0.266843972, 0.909090909, 0.179310345]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "per_generation_sealed": [0.65358, 0.65358, 0.65358, 0.65358], "prereg_hash": "2a6c4cc1de790e56e6d42394b13d7e64e9f88280640ab37d70dc7f4cfa4436a4", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "compressed", "reset_mem"], "result_hash": "af3f9a1b19fc2275ed14265438a6191184d887df7ada9b78ef91b7e378cda163", "signature": "741662e0d12d19a2e8596e1d178b4b54f10d179e47663309c03a8a94881b8c6a3d5fac3d59d57dca9891e1878994c14f06bf2d8a2658afdc50eaf28dfe39d602", "state": "PARTIAL", "summary": {"G0": 0.65358, "compressed": 0.65358, "inherit_mem": 0.65358, "inherit_nomem": 0.65358, "random_search": 0.65358, "reset_mem": 0.65358, "retrieval": 0.65358, "rootcause": 0.65358, "xfer_mem": 0.65358}, "ties": 10, "treatment": "inherit_mem", "wins": 0}, "verified_at": "2026-11-22T04:15:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:21Z", "cpu_seconds": 0.5543901089999999, "created_at": "2026-11-22T04:02:43Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-300abddd1e1d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "prereg_hash": "6088a06848ec381c4f2a1dcc0a550616a97c14b3e186146640d0f7bab026779f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-300abddd1e1d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:02:43Z", "seed": 567604308, "treatment": "gated"}, "priority": 0.2514585179818866, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "716be8646c7ff5f9d84eccd2bfb6375391c07775befca703ad1227eff3ddcc97", "run_seconds": 0.6350560188293457, "seed": 567604308, "started_at": "2026-11-22T04:03:20Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-300abddd1e1d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [1.0, 1.0, 0.85483871, 1.0], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [1.0, 1.0, 0.784946237, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6088a06848ec381c4f2a1dcc0a550616a97c14b3e186146640d0f7bab026779f", "primary_control": "scratch", "result_hash": "716be8646c7ff5f9d84eccd2bfb6375391c07775befca703ad1227eff3ddcc97", "signature": "1efffa1c1cfc7bc8befe7ff9acfcd36b8c4fc7e2b2586b95c073eb1a8a336395bbb9c43ce365a8d56b43afb45a85da13f36713fd0bb7ed021c4fc46307c6440c", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.96371, "scratch": 1.0, "ungated": 0.946237}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:03:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:26Z", "cpu_seconds": 0.753551256999998, "created_at": "2026-11-22T04:02:43Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-f271f535cbc7", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "2eb0ae3b5fba0e869d4a35e26df0349f5bbd3b4176b7aa9620de444183cf3f9d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-f271f535cbc7", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:02:43Z", "seed": 1859958583, "treatment": "gated"}, "priority": 0.2507095944806273, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-9d363dd0f5ed", "result_hash": "9003e07dd9fece4cb649d0469a826c7576a163e79a31a3449395b82fb48a64a0", "run_seconds": 0.759427547454834, "seed": 1859958583, "started_at": "2026-11-22T04:03:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.021449, 0.006957], "effect": -0.007246, "evaluated_at": "2026-11-22T04:03:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-f271f535cbc7", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.61, 0.656565657, 0.61352657, 0.645833333], "gated": [0.61, 0.656565657, 0.61352657, 0.645833333], "naive": [0.59, 0.757575758, 0.584541063, 0.5625], "scratch": [0.61, 0.656565657, 0.642512077, 0.645833333], "ungated": [0.61, 0.656565657, 0.61352657, 0.645833333]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2eb0ae3b5fba0e869d4a35e26df0349f5bbd3b4176b7aa9620de444183cf3f9d", "primary_control": "scratch", "replicated": false, "replication_of": "E141-ow_transfer-9d363dd0f5ed", "result_hash": "9003e07dd9fece4cb649d0469a826c7576a163e79a31a3449395b82fb48a64a0", "signature": "7d63094e0d9194c67e72b039513774250ed46d1aa4b416f32ad4f63765a9c91f0146887c2ecddd0dd0148782b4a9d6594782cd7cad3f262ac9803c2e1dd7bb01", "state": "VERIFIED", "summary": {"compressed": 0.631481, "gated": 0.631481, "naive": 0.623654, "scratch": 0.638728, "ungated": 0.631481}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:03:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:00Z", "cpu_seconds": 0.1994994959999996, "created_at": "2026-11-22T04:02:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4f0e6efeb3a9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "f64bb531387d9c45414a809df14605048c00e9a014372d6dc37945779ef81397", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4f0e6efeb3a9", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:02:59Z", "seed": 1049697995, "treatment": "full"}, "priority": 0.32491370314912815, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2efb4a1251b1f765ca620befa877fe074e584272b71156fb3fcba47253734d72", "run_seconds": 0.20309925079345703, "seed": 1049697995, "started_at": "2026-11-22T04:03:00Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4f0e6efeb3a9", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.0, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f64bb531387d9c45414a809df14605048c00e9a014372d6dc37945779ef81397", "primary_control": "e141_best", "result_hash": "2efb4a1251b1f765ca620befa877fe074e584272b71156fb3fcba47253734d72", "signature": "b48dd131a2132afa46a4d239cf54ee4b524213addffb946dc2e443eaf7adcfc1395b629379f9680f20bda16e7e6dca25e4b672e710e180d2ea84de5877ebad02", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:03:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:06Z", "cpu_seconds": 0.36132668700000004, "created_at": "2026-11-22T04:02:59Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d1d17d38375f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "98ee35146dbe9d9d4b19e583f33fb8b44706f321df6b3307bbde03696bdcf273", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d1d17d38375f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:02:59Z", "seed": 1748568933, "treatment": "full"}, "priority": 0.25991284065948245, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "22350756b5ee5aa5e29d864b5cbbd4608efb86906884e76eb12b475ac7d00db1", "run_seconds": 0.4042198657989502, "seed": 1748568933, "started_at": "2026-11-22T04:03:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.057115, 0.159552], "effect": 0.108333, "evaluated_at": "2026-11-22T04:03:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d1d17d38375f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.85, 0.9375, 0.8125, 0.875], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.975, 0.979166667, 0.979166667, 0.975], "retrieval": [0.925, 0.979166667, 0.9375, 1.0], "rules_nodetect": [0.975, 0.979166667, 0.979166667, 0.975]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "98ee35146dbe9d9d4b19e583f33fb8b44706f321df6b3307bbde03696bdcf273", "primary_control": "facts", "result_hash": "22350756b5ee5aa5e29d864b5cbbd4608efb86906884e76eb12b475ac7d00db1", "signature": "c25eeef5397e8418892412ed9d00cf9c75ee0f577a751352af1db9e2ba7e0a4c2cd75d7f98d082a251901eacfdf06a74d406c6669e46a2db82942f13c80dc000", "state": "VERIFIED", "summary": {"facts": 0.86875, "frozen": 0.0, "full": 0.977083, "retrieval": 0.960417, "rules_nodetect": 0.977083}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:03:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:04Z", "cpu_seconds": 1.813586463, "created_at": "2026-11-22T04:02:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d2d93635e3f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "0cce70e40f7b3beaba524a013a64d312ebeb32b30a47ae0dc14a592ade8e829f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d2d93635e3f0", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:02:59Z", "seed": 1240043203, "treatment": "gated"}, "priority": 0.30300738228323154, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "0b867c4037fc058991072e62583575faef41b8dd569c3e8af9111b3dd766a178", "run_seconds": 2.145007371902466, "seed": 1240043203, "started_at": "2026-11-22T04:03:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007708, 0.0025], "effect": -0.002604, "evaluated_at": "2026-11-22T04:03:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d2d93635e3f0", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.61, 0.553571429, 0.75, 0.53125], "gated": [0.61, 0.553571429, 0.75, 0.53125], "naive": [0.66, 0.678571429, 0.681818182, 0.583333333], "scratch": [0.61, 0.553571429, 0.75, 0.541666667], "ungated": [0.59, 0.607142857, 0.681818182, 0.552083333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "0cce70e40f7b3beaba524a013a64d312ebeb32b30a47ae0dc14a592ade8e829f", "primary_control": "scratch", "result_hash": "0b867c4037fc058991072e62583575faef41b8dd569c3e8af9111b3dd766a178", "signature": "ac02461ee7ddbf0d012a39bd98e299617e1a035c680590ec78572d7a062f131621aa78ea7af9266744b9376cfc5d8b7dd852e07768b1c9ea1dc89c113ae2cf0f", "state": "VERIFIED", "summary": {"compressed": 0.611205, "gated": 0.611205, "naive": 0.650931, "scratch": 0.61381, "ungated": 0.607761}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:03:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:29Z", "cpu_seconds": 0.30708820199999565, "created_at": "2026-11-22T04:03:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f83013c13fbe", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "7d9982fd899b5020ad998140137dec34ab07ddd94316617e79c6159ca2d21b08", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f83013c13fbe", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:03:16Z", "seed": 2044479887, "treatment": "frontier"}, "priority": 0.24777050893509808, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-47cb67d3ac2f", "result_hash": "2774bd764fbb055d43b7c405af5cd32a40f999fff7f777c90dab47d689a6c0f2", "run_seconds": 0.34891414642333984, "seed": 2044479887, "started_at": "2026-11-22T04:05:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.150752, 0.258339], "effect": 0.204545, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f83013c13fbe", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.954545455, 0.977272727, 0.988636364], "frontier": [1.0, 0.920454545, 1.0, 1.0], "goal": [0.670454545, 0.625, 0.840909091, 0.681818182], "hypdisc": [0.795454545, 0.772727273, 0.840909091, 0.784090909], "infogain": [1.0, 0.920454545, 1.0, 1.0], "prederr": [1.0, 0.920454545, 1.0, 1.0], "random": [0.875, 0.704545455, 0.772727273, 0.75], "risk": [1.0, 0.920454545, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "7d9982fd899b5020ad998140137dec34ab07ddd94316617e79c6159ca2d21b08", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-47cb67d3ac2f", "result_hash": "2774bd764fbb055d43b7c405af5cd32a40f999fff7f777c90dab47d689a6c0f2", "signature": "4ea2a7e804b9613fdc39787e32e83aecb15b49eb7e7349dc6018f5f703a0667c6af9c4827bcee9086a0eb3af55e53005a0934f0a2ac92451ce4eb75702e90005", "state": "VERIFIED", "summary": {"count": 0.980114, "frontier": 0.980114, "goal": 0.704545, "hypdisc": 0.798295, "infogain": 0.980114, "prederr": 0.980114, "random": 0.775568, "risk": 0.980114}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:17Z", "cpu_seconds": 0.8147462109999992, "created_at": "2026-11-22T04:03:16Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2c276c20e04d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "48e359fd490a52b0fa4ffbce65b8b354403ae28adf19ca48c169752854973b2a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2c276c20e04d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:03:16Z", "seed": 569182046, "treatment": "full"}, "priority": 0.25273074431247905, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "3b76bf0a78945947323d238d0909e325db6a759579ff066f8b52045bb5be974e", "run_seconds": 0.9456839561462402, "seed": 569182046, "started_at": "2026-11-22T04:03:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019083, 0.855917], "effect": 0.4375, "evaluated_at": "2026-11-22T04:03:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2c276c20e04d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.75, 0.5, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 0.75], "no_world_model": [0.75, 1.0, 1.0, 0.75], "random": [0.0, 0.0, 0.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "48e359fd490a52b0fa4ffbce65b8b354403ae28adf19ca48c169752854973b2a", "primary_control": "e141_best", "result_hash": "3b76bf0a78945947323d238d0909e325db6a759579ff066f8b52045bb5be974e", "signature": "cfc7382dfec4e76679973725c4c2d8c47483e0b417127a6aa3e3bd84433cb1498de58bb82b11a9aff8912db390f58aa2914bf5cae82cb95b364d3481ed15ee0d", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.0625, "full": 0.9375, "no_world_model": 0.875, "random": 0.1875}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:03:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:06Z", "cpu_seconds": 0.11155363499999993, "created_at": "2026-11-22T04:03:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-12b932d2f458", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "467e2b5f1b3c02acff478ebe4fb3732a1a09c3304eaa8f3121b41b0b0e84818a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-12b932d2f458", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:16Z", "seed": 1495170417, "treatment": "full"}, "priority": 0.2480102054097684, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2479fcaa3161ce1e85847dde9692421cc2eb2f3f27aa5658e9c4e4f33d09799c", "run_seconds": 0.15653681755065918, "seed": 1495170417, "started_at": "2026-11-22T04:04:05Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:04:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-12b932d2f458", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 1.0, 1.0, 1.0], "frozen": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 1.0, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "467e2b5f1b3c02acff478ebe4fb3732a1a09c3304eaa8f3121b41b0b0e84818a", "primary_control": "facts", "result_hash": "2479fcaa3161ce1e85847dde9692421cc2eb2f3f27aa5658e9c4e4f33d09799c", "signature": "d8523566e9f4448ed16c1f01e4a474de23a16d830ecedc0a4443242958bab87ea99414d8a4e657a02a3f215c750f632147bfca573b7133c154cb807f4d8a8e0d", "state": "PARTIAL", "summary": {"facts": 1.0, "frozen": 1.0, "full": 1.0, "retrieval": 1.0, "rules_nodetect": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:04:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:29Z", "cpu_seconds": 0.8017214270000022, "created_at": "2026-11-22T04:03:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c09a7fb9b21c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "164a32bb31517d4afbfe6b05c33c424ec2f8a36160e9c840dcdcaeef284d80fe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c09a7fb9b21c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:16Z", "seed": 1190522600, "treatment": "full"}, "priority": 0.2501459726631332, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b85b46d7d815c0e8ad3c91418a005befd08846c8aeed517042aaf9afc178fb5c", "run_seconds": 0.8805077075958252, "seed": 1190522600, "started_at": "2026-11-22T04:03:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01451, 0.004706], "effect": -0.004902, "evaluated_at": "2026-11-22T04:03:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c09a7fb9b21c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.862745098, 0.770114943, 0.824561404, 0.833333333], "frozen": [0.352941176, 0.32183908, 0.403508772, 0.416666667], "full": [0.843137255, 0.770114943, 0.824561404, 0.833333333], "retrieval": [0.862745098, 0.770114943, 0.824561404, 0.833333333], "rules_nodetect": [0.843137255, 0.770114943, 0.824561404, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "164a32bb31517d4afbfe6b05c33c424ec2f8a36160e9c840dcdcaeef284d80fe", "primary_control": "facts", "result_hash": "b85b46d7d815c0e8ad3c91418a005befd08846c8aeed517042aaf9afc178fb5c", "signature": "3cb38b3768f4bf3323ae444e2c20c29153d64f3161fd07606f5b3a3219c030f3c5b0358afc65939e369064e159a1df3cc24eb22478b73f5e5f6a59c3cb7aae08", "state": "VERIFIED", "summary": {"facts": 0.822689, "frozen": 0.373739, "full": 0.817787, "retrieval": 0.822689, "rules_nodetect": 0.817787}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:03:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:32Z", "cpu_seconds": 0.8718773499999983, "created_at": "2026-11-22T04:03:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f4c19e7b9214", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "d95b2dea7b1cce314b1776c1eb00b7d3c5878a2e40b9feae5536bd43490e4823", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f4c19e7b9214", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:16Z", "seed": 63865265, "treatment": "full"}, "priority": 0.2495052331582659, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "8930c20b6d9de90df1907180894a94909872876c1f229032f77bcf7c4e2f7e97", "run_seconds": 0.8900225162506104, "seed": 63865265, "started_at": "2026-11-22T04:03:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.025244, 0.070669], "effect": 0.022712, "evaluated_at": "2026-11-22T04:03:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f4c19e7b9214", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.44248366, 0.481045752, 0.429411765, 0.460784314], "frozen": [0.328104575, 0.374509804, 0.335947712, 0.32745098], "full": [0.459477124, 0.475163399, 0.415686275, 0.554248366], "retrieval": [0.756862745, 0.746405229, 0.705882353, 0.735294118], "rules_nodetect": [0.459477124, 0.475163399, 0.415686275, 0.554248366]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "d95b2dea7b1cce314b1776c1eb00b7d3c5878a2e40b9feae5536bd43490e4823", "primary_control": "facts", "result_hash": "8930c20b6d9de90df1907180894a94909872876c1f229032f77bcf7c4e2f7e97", "signature": "02aa9c353b4a74ca69a207c06b115def0fe92434c1d22d5526c926dd87be6e56145306043657210543205a2999199119a0490dcb323bfab90aa4cfa29bf36c07", "state": "VERIFIED", "summary": {"facts": 0.453431, "frozen": 0.341503, "full": 0.476144, "retrieval": 0.736111, "rules_nodetect": 0.476144}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:49Z", "cpu_seconds": 615.127774996, "created_at": "2026-11-22T04:03:16Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-153f360a82e2", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "d83fd16be2f0c6b7ae73f6fb581df82b31d48fa550c4e49e52ea5c6d5804ce77", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-153f360a82e2", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:03:16Z", "seed": 52605744, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "8118d3b44b7b7c9afb913bfd15058ad560a757b994946aa71e82f833b19b4739", "run_seconds": 665.9887342453003, "seed": 52605744, "started_at": "2026-11-22T04:04:43Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"compressed": 0, "inherit_mem": 0, "inherit_nomem": 1, "random_search": 1, "reset_mem": 0, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:18:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-153f360a82e2", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"goal": 0.877, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.97}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.822916667, 0.681818182, 0.839473684, 0.898989899, 0.30952381], "compressed": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.822916667, 0.681818182, 0.839473684, 0.898989899, 0.30952381], "inherit_mem": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.822916667, 0.681818182, 0.839473684, 0.898989899, 0.30952381], "inherit_nomem": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.875, 0.59469697, 0.839473684, 0.898989899, 0.30952381], "random_search": [0.95, 1.0, 0.966666667, 0.371212121, 0.246666667, 0.786458333, 0.181818182, 0.881578947, 0.898989899, 0.49047619], "reset_mem": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.822916667, 0.681818182, 0.839473684, 0.898989899, 0.30952381], "retrieval": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.822916667, 0.681818182, 0.839473684, 0.898989899, 0.30952381], "rootcause": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.875, 0.59469697, 0.839473684, 0.898989899, 0.30952381], "xfer_mem": [0.916666667, 1.0, 0.975, 0.333333333, 0.252156863, 0.875, 0.59469697, 0.839473684, 0.898989899, 0.30952381]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "per_generation_sealed": [0.702988, 0.702988, 0.702988, 0.702988], "prereg_hash": "d83fd16be2f0c6b7ae73f6fb581df82b31d48fa550c4e49e52ea5c6d5804ce77", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "compressed", "random_search"], "result_hash": "8118d3b44b7b7c9afb913bfd15058ad560a757b994946aa71e82f833b19b4739", "signature": "0f9b912cbc16f611113e77c6a7ca810d8a684ca5b4c5ddd50c2d67dd47d5bcadc3f4ab381a1d11ed1576964ce42917cf7e2bd95b189d86b1aecfe6a08770ec01", "state": "PARTIAL", "summary": {"G0": 0.702988, "compressed": 0.702988, "inherit_mem": 0.702988, "inherit_nomem": 0.699484, "random_search": 0.677387, "reset_mem": 0.702988, "retrieval": 0.702988, "rootcause": 0.699484, "xfer_mem": 0.699484}, "ties": 10, "treatment": "inherit_mem", "wins": 0}, "verified_at": "2026-11-22T04:22:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:58Z", "cpu_seconds": 0.937538141999994, "created_at": "2026-11-22T04:03:16Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-10514a2ce9b3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "prereg_hash": "ab456208cf94f40d99a1b68b9bca41927187a9ab68566d4a30742aa547b6d01f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-10514a2ce9b3", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:03:16Z", "seed": 1530096804, "treatment": "gated"}, "priority": 0.24666360098648754, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7e6330c805a4f2bc99d7bb53bb2842cddaf6799fe0b7458a086010e9410f289b", "run_seconds": 0.9846971035003662, "seed": 1530096804, "started_at": "2026-11-22T04:05:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.047557, 0.106585], "effect": 0.029514, "evaluated_at": "2026-11-22T04:07:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-10514a2ce9b3", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.5, 0.611111111, 0.515625, 0.322916667], "gated": [0.5, 0.611111111, 0.515625, 0.46875], "naive": [0.443181818, 0.555555556, 0.421875, 0.53125], "scratch": [0.5, 0.638888889, 0.515625, 0.322916667], "ungated": [0.659090909, 0.555555556, 0.421875, 0.53125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "ab456208cf94f40d99a1b68b9bca41927187a9ab68566d4a30742aa547b6d01f", "primary_control": "scratch", "result_hash": "7e6330c805a4f2bc99d7bb53bb2842cddaf6799fe0b7458a086010e9410f289b", "signature": "395b852f8ff09f37a0160d2b04fa6730a55c19f95b9da6d1508c6e0701de85786316118564d61a9370276c8c2a338c943eeb7b33c3d97c54af5d7338fff5fa06", "state": "VERIFIED", "summary": {"compressed": 0.487413, "gated": 0.523872, "naive": 0.487966, "scratch": 0.494358, "ungated": 0.541943}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:07:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:46Z", "cpu_seconds": 0.5862967000000019, "created_at": "2026-11-22T04:03:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8f97f90cc534", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "37f6743fd64a87588be1d08b16b4148aca640c43437db1e2646aeb601b637ece", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8f97f90cc534", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:33Z", "seed": 1180037046, "treatment": "full"}, "priority": 0.27406717356390375, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-6380482e7fab", "result_hash": "c5fa1abbe177ad9e27349602c4fb91310d12e90e4922572c0fc1a86b43a294be", "run_seconds": 0.6177787780761719, "seed": 1180037046, "started_at": "2026-11-22T04:03:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007976, 0.004503], "effect": -0.001736, "evaluated_at": "2026-11-22T04:03:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8f97f90cc534", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.709219858, 0.869281046, 0.803921569, 0.423611111], "frozen": [0.09929078, 0.08496732, 0.08496732, 0.097222222], "full": [0.709219858, 0.862745098, 0.810457516, 0.416666667], "retrieval": [0.737588652, 0.875816993, 0.823529412, 0.576388889], "rules_nodetect": [0.709219858, 0.862745098, 0.810457516, 0.416666667]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "37f6743fd64a87588be1d08b16b4148aca640c43437db1e2646aeb601b637ece", "primary_control": "facts", "replicated": false, "replication_of": "E141-ow_model-6380482e7fab", "result_hash": "c5fa1abbe177ad9e27349602c4fb91310d12e90e4922572c0fc1a86b43a294be", "signature": "9f45beb0d945d44f4d45ff1ceb0dea1fbc9ef6f93dc2beaf5b9a0390ff13f4c76fe4bf5ea70bbf5b68b695fc8b62cf2491d1a10da15c616a275fdebc094d550d", "state": "VERIFIED", "summary": {"facts": 0.701508, "frozen": 0.091612, "full": 0.699772, "retrieval": 0.753331, "rules_nodetect": 0.699772}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:03:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:50Z", "cpu_seconds": 4.526329019000002, "created_at": "2026-11-22T04:03:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a3ad01ffa53e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "41ca1cb2afb320f37d01930733f3849f7049820e9c679ee762a4047248753657", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a3ad01ffa53e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:33Z", "seed": 1017404046, "treatment": "full"}, "priority": 0.30643569115260094, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e11b0d0e9c42a1f83ddc7ef179d7c077eaa556476c6d7c30575b759ba5a0b93a", "run_seconds": 4.626175165176392, "seed": 1017404046, "started_at": "2026-11-22T04:03:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024038, 0.08527], "effect": 0.030616, "evaluated_at": "2026-11-22T04:03:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a3ad01ffa53e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.534090909, 0.948320413, 0.398148148], "frozen": [0.0, 0.170454545, 0.275193798, 0.168209877], "full": [1.0, 0.539772727, 0.950904393, 0.512345679], "retrieval": [1.0, 0.539772727, 0.954780362, 0.617283951], "rules_nodetect": [0.7, 0.384469697, 0.950904393, 0.512345679]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "41ca1cb2afb320f37d01930733f3849f7049820e9c679ee762a4047248753657", "primary_control": "facts", "result_hash": "e11b0d0e9c42a1f83ddc7ef179d7c077eaa556476c6d7c30575b759ba5a0b93a", "signature": "01c41c98cbb927279d991b8525d3209911d67bc2c97238a93f475488c0a3cbb99074b47ed2812a67795f9f407a29c3024554c224e2f92855cdaf3b1ba117b806", "state": "VERIFIED", "summary": {"facts": 0.72014, "frozen": 0.153465, "full": 0.750756, "retrieval": 0.777959, "rules_nodetect": 0.63693}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:03:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:35Z", "cpu_seconds": 0.6743219799999984, "created_at": "2026-11-22T04:03:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a9bf348dbf33", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "4af412c0c535cc0aad74204e2c9c6d85e33c830c1592c12a6c75d01b13bc4b23", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a9bf348dbf33", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:33Z", "seed": 199447659, "treatment": "full"}, "priority": 0.4067106644208264, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "de28319beada1f2c1f938f2ceb932768008bef54ea83751f39a6f39e6987f4c7", "run_seconds": 0.892805814743042, "seed": 199447659, "started_at": "2026-11-22T04:03:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035, 0.107917], "effect": 0.036458, "evaluated_at": "2026-11-22T04:03:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a9bf348dbf33", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.583333333, 0.780068729, 0.888888889, 0.816993464], "frozen": [0.354166667, 0.281786942, 0.481481481, 0.320261438], "full": [0.729166667, 0.780068729, 0.888888889, 0.816993464], "retrieval": [0.708333333, 0.780068729, 0.888888889, 0.816993464], "rules_nodetect": [0.729166667, 0.780068729, 0.888888889, 0.816993464]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "4af412c0c535cc0aad74204e2c9c6d85e33c830c1592c12a6c75d01b13bc4b23", "primary_control": "facts", "result_hash": "de28319beada1f2c1f938f2ceb932768008bef54ea83751f39a6f39e6987f4c7", "signature": "4a5a14d494762dd5a55a8f7012d4df7b961b0e755ec1ed68b546397db3dd4d8769c72211366de4cc21f4262f2ad93851d36bbad904229fec77e5700e7ff87606", "state": "VERIFIED", "summary": {"facts": 0.767321, "frozen": 0.359424, "full": 0.803779, "retrieval": 0.798571, "rules_nodetect": 0.803779}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:03:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:40Z", "cpu_seconds": 0.6490080779999978, "created_at": "2026-11-22T04:03:33Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d956475c4a79", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "9fee6ac9dd0cadae3502618f419cfee3080b290f281a65e8c6ddea33df6375fd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d956475c4a79", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:03:33Z", "seed": 66448254, "treatment": "gated"}, "priority": 0.32601071975716694, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6870478718acf01123fca0669159e3420c239c560577d2048f2b617d9df7573a", "run_seconds": 0.6748018264770508, "seed": 66448254, "started_at": "2026-11-22T04:03:40Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:03:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d956475c4a79", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.75, 0.707070707, 0.61352657, 0.727272727], "gated": [0.75, 0.707070707, 0.61352657, 0.727272727], "naive": [0.666666667, 0.595959596, 0.47826087, 0.757575758], "scratch": [0.75, 0.707070707, 0.61352657, 0.727272727], "ungated": [0.666666667, 0.787878788, 0.285024155, 0.727272727]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9fee6ac9dd0cadae3502618f419cfee3080b290f281a65e8c6ddea33df6375fd", "primary_control": "scratch", "result_hash": "6870478718acf01123fca0669159e3420c239c560577d2048f2b617d9df7573a", "signature": "90e0bd9ba802c295a500b61ff5a1fab6bad5df7f84c28f6a6a6bf8e5012abb52bfd5f71fdea3b8a868b667b61e1ee58cdd39ebcacb01ba6d871816e3bfedc302", "state": "PARTIAL", "summary": {"compressed": 0.699468, "gated": 0.699468, "naive": 0.624616, "scratch": 0.699468, "ungated": 0.616711}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:03:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:40Z", "cpu_seconds": 2.5464740159999977, "created_at": "2026-11-22T04:03:33Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-dc94add5b95f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "c4fa8095ad50b3aafa5035540ad510a03769e5b83854c6db55b0329d35617aec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-dc94add5b95f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:03:33Z", "seed": 1822896577, "treatment": "gated"}, "priority": 0.3456196725055329, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "306852d42445d447a33837a7ec03768d45edd918481e9a05c7b2b88f0cdad4e3", "run_seconds": 2.5851306915283203, "seed": 1822896577, "started_at": "2026-11-22T04:03:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026492, 0.064854], "effect": 0.019181, "evaluated_at": "2026-11-22T04:03:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-dc94add5b95f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.4875, 0.553571429, 0.5, 0.543103448], "gated": [0.4875, 0.616071429, 0.5, 0.543103448], "naive": [0.6, 0.741071429, 0.630952381, 0.698275862], "scratch": [0.525, 0.553571429, 0.5, 0.49137931], "ungated": [0.6125, 0.616071429, 0.630952381, 0.49137931]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "c4fa8095ad50b3aafa5035540ad510a03769e5b83854c6db55b0329d35617aec", "primary_control": "scratch", "result_hash": "306852d42445d447a33837a7ec03768d45edd918481e9a05c7b2b88f0cdad4e3", "signature": "93af229d5487ee921042daa725866b22ea4af0af9375afb317fd89a5cfa21206fb15ec0b439374606553b6a84919080d817d7d0cb6a53aae0a78470ef16ef00e", "state": "VERIFIED", "summary": {"compressed": 0.521044, "gated": 0.536669, "naive": 0.667575, "scratch": 0.517488, "ungated": 0.587726}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:03:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:00Z", "cpu_seconds": 0.1716178049999968, "created_at": "2026-11-22T04:03:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3a6e3123567b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "6a8d5f6cd58dc8b8a3e8c325c29e50869bbf8dfbd5fe6c857d0fb5f2e83e1eb4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3a6e3123567b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:03:50Z", "seed": 1115235034, "treatment": "full"}, "priority": 0.2833583486988927, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c77f91ea7a5b03cbd89c05891c7243d2df926cedee6842035cd962866f546ad3", "run_seconds": 0.17546653747558594, "seed": 1115235034, "started_at": "2026-11-22T04:04:00Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:04:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3a6e3123567b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.25, 0.0, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6a8d5f6cd58dc8b8a3e8c325c29e50869bbf8dfbd5fe6c857d0fb5f2e83e1eb4", "primary_control": "e141_best", "result_hash": "c77f91ea7a5b03cbd89c05891c7243d2df926cedee6842035cd962866f546ad3", "signature": "79b04b857f5c3dbb8d3b137a105ce4ff72c2bcb30d70ea10da36814f2b70cdb9fec154c077f992832e463859fe0c2c65042d4a9907116edef21e8f30be51190b", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:04:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:52Z", "cpu_seconds": 0.5359256900000027, "created_at": "2026-11-22T04:03:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-19c7ed3308fd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "ba963ad1735f173ab64308ea08995a40a2e343e380a996a060f9f5dc3a85b895", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-19c7ed3308fd", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:50Z", "seed": 651840601, "treatment": "full"}, "priority": 0.33885407969245407, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "bdc14e78656300967b61d7f28ed0048414e9aa5a5043ae520261a0ca6de4e6a1", "run_seconds": 0.591461181640625, "seed": 651840601, "started_at": "2026-11-22T04:03:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026409, 0.077997], "effect": 0.025794, "evaluated_at": "2026-11-22T04:03:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-19c7ed3308fd", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.788617886, 0.837398374, 0.851851852, 0.595238095], "frozen": [0.308943089, 0.341463415, 0.444444444, 0.357142857], "full": [0.804878049, 0.821138211, 0.851851852, 0.698412698], "retrieval": [0.756097561, 0.837398374, 0.888888889, 0.706349206], "rules_nodetect": [0.804878049, 0.821138211, 0.851851852, 0.785714286]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "ba963ad1735f173ab64308ea08995a40a2e343e380a996a060f9f5dc3a85b895", "primary_control": "facts", "result_hash": "bdc14e78656300967b61d7f28ed0048414e9aa5a5043ae520261a0ca6de4e6a1", "signature": "3832e32d345f13cc1fe7b524799810a4dc843a6aa1da1c4f43884adb2834aff4c69da394dfb206d953fa38e17d3de0fc248a8e77e5af74499b456c9103196d03", "state": "VERIFIED", "summary": {"facts": 0.768277, "frozen": 0.362998, "full": 0.79407, "retrieval": 0.797184, "rules_nodetect": 0.815896}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:55Z", "cpu_seconds": 0.6056261080000027, "created_at": "2026-11-22T04:03:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-92dbeb94c675", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "53221cb01e00cf826a6fa7e609eb5574cb0d39ea81204efff3fde353c3a845f0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-92dbeb94c675", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:03:50Z", "seed": 1406012081, "treatment": "full"}, "priority": 0.3332552334249973, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e471a7699005876df3ef18737b7f7c23a9ef3ee2fd51e394f4d5bc1d2deb4b94", "run_seconds": 0.6472840309143066, "seed": 1406012081, "started_at": "2026-11-22T04:03:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.048222, 0.156932], "effect": 0.054355, "evaluated_at": "2026-11-22T04:03:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-92dbeb94c675", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.862745098, 0.292517007, 0.725490196, 0.843137255], "frozen": [0.08496732, 0.088435374, 0.091503268, 0.091503268], "full": [0.85620915, 0.503401361, 0.725490196, 0.85620915], "retrieval": [0.91503268, 0.673469388, 0.803921569, 0.888888889], "rules_nodetect": [0.85620915, 0.503401361, 0.725490196, 0.85620915]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "53221cb01e00cf826a6fa7e609eb5574cb0d39ea81204efff3fde353c3a845f0", "primary_control": "facts", "result_hash": "e471a7699005876df3ef18737b7f7c23a9ef3ee2fd51e394f4d5bc1d2deb4b94", "signature": "898ca29c83f7bb92eec14693c35128b9516c7abe6fe2d7d9488ddd71212d7b1322871f2474d5016f8ab74d7e0580fddeb0d2ac2560fbb40fc5f532a381d3d002", "state": "VERIFIED", "summary": {"facts": 0.680972, "frozen": 0.089102, "full": 0.735327, "retrieval": 0.820328, "rules_nodetect": 0.735327}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:03:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:03:59Z", "cpu_seconds": 2.6027132329999993, "created_at": "2026-11-22T04:03:50Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2b75ac93c397", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "d5ac0a605a4e6eda25aa38c880e08419fbff2a9edb4b064bd8a1919e650be27a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2b75ac93c397", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:03:50Z", "seed": 255641482, "treatment": "gated"}, "priority": 0.33078843932552476, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "fc05a9529516cd4bac3bafffbf35c0d19847bd1f6812344a45ad91bd34298c8c", "run_seconds": 2.809471607208252, "seed": 255641482, "started_at": "2026-11-22T04:03:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.086069, 0.106902], "effect": 0.010417, "evaluated_at": "2026-11-22T04:04:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2b75ac93c397", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.586538462, 0.511904762, 0.583333333, 0.5625], "gated": [0.586538462, 0.511904762, 0.583333333, 0.5625], "naive": [0.817307692, 0.678571429, 0.685185185, 0.625], "scratch": [0.711538462, 0.428571429, 0.583333333, 0.479166667], "ungated": [0.721153846, 0.642857143, 0.685185185, 0.614583333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d5ac0a605a4e6eda25aa38c880e08419fbff2a9edb4b064bd8a1919e650be27a", "primary_control": "scratch", "result_hash": "fc05a9529516cd4bac3bafffbf35c0d19847bd1f6812344a45ad91bd34298c8c", "signature": "d829e4f37ae0daed8285dddf0a8b721a4d0df9c88eb30e5a68402f09564edcccc82b8c1d7f2aa9b29bc9a1c4b58d2235d651e518a2bc7a27c72de22976980003", "state": "VERIFIED", "summary": {"compressed": 0.561069, "gated": 0.561069, "naive": 0.701516, "scratch": 0.550652, "ungated": 0.665945}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:04:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:05Z", "cpu_seconds": 0.3181753709999988, "created_at": "2026-11-22T04:03:50Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-623826e83ba9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "prereg_hash": "c4970c47608ec7f268dee0f7cab81ff2074d3ce4f54238c5ef03248c971eb246", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-623826e83ba9", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:03:50Z", "seed": 77800152, "treatment": "gated"}, "priority": 0.26191250731999954, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "ed4815b99c1d8a3b626462eef153d58c53f1c751907f34bec25777943cf0b3f6", "run_seconds": 0.32082319259643555, "seed": 77800152, "started_at": "2026-11-22T04:04:04Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:04:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-623826e83ba9", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.783333333, 0.783333333, 0.6, 0.783333333], "gated": [0.783333333, 0.783333333, 0.6, 0.783333333], "naive": [0.466666667, 0.583333333, 0.666666667, 0.383333333], "scratch": [0.783333333, 0.783333333, 0.6, 0.783333333], "ungated": [0.516666667, 0.583333333, 0.666666667, 0.383333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c4970c47608ec7f268dee0f7cab81ff2074d3ce4f54238c5ef03248c971eb246", "primary_control": "scratch", "result_hash": "ed4815b99c1d8a3b626462eef153d58c53f1c751907f34bec25777943cf0b3f6", "signature": "6182f066b927c835519f9533a7097da501fad9862475a23b9134a96e375ae4458739cf9b6a0566d0c63b7360af402c13d1568de993a863020179de774c6c9e04", "state": "PARTIAL", "summary": {"compressed": 0.7375, "gated": 0.7375, "naive": 0.525, "scratch": 0.7375, "ungated": 0.5375}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:04:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:21Z", "cpu_seconds": 0.3455171270000008, "created_at": "2026-11-22T04:04:06Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5ed315c5ff55", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "1137f5cd8b55e0486901b47fc11761a35570c934affebbf8bb78a6fa21483830", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5ed315c5ff55", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:04:06Z", "seed": 1838054997, "treatment": "full"}, "priority": 0.2608323093271384, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "77c2d1f13b6c2574a8cda439a4fc859c7828d5c2c7436831c790df9d3d0dfca6", "run_seconds": 0.3533968925476074, "seed": 1838054997, "started_at": "2026-11-22T04:04:20Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:04:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5ed315c5ff55", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 1.0, 0.5, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1137f5cd8b55e0486901b47fc11761a35570c934affebbf8bb78a6fa21483830", "primary_control": "e141_best", "result_hash": "77c2d1f13b6c2574a8cda439a4fc859c7828d5c2c7436831c790df9d3d0dfca6", "signature": "0d42cdbb91a2beecba8d3d45c459e8e1d956208e0b1f694cf5ab40f2dc82deffb4ad4b11c06e1ff308b704ed46b2cebda075106134e4aee35a0f9c387de32008", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5625, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:04:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:23Z", "cpu_seconds": 0.6157087690000012, "created_at": "2026-11-22T04:04:06Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-727d63f64ce3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "dd6f1bb68a2399bc826af9408bd7aba5d0b608dd0198d0e64bd14669638711c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-727d63f64ce3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:04:06Z", "seed": 2017553546, "treatment": "full"}, "priority": 0.2568001356421914, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "afd9de842cfaec6e6b59edf27be970542a2531c4d71e2f76179963c42497dfc2", "run_seconds": 0.6182420253753662, "seed": 2017553546, "started_at": "2026-11-22T04:04:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:04:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-727d63f64ce3", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.0, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 0.75], "random": [1.0, 0.5, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "dd6f1bb68a2399bc826af9408bd7aba5d0b608dd0198d0e64bd14669638711c3", "primary_control": "e141_best", "result_hash": "afd9de842cfaec6e6b59edf27be970542a2531c4d71e2f76179963c42497dfc2", "signature": "6f05273a80149cfa4420c736626fdb72a48eb025fa34c266177364078c14a82ed1ef51a71b3a8b11d65c4aa93a22aafef18fe5c9ce20cc3425e0647a28da0e0c", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 0.875, "random": 0.8125}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:04:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:12Z", "cpu_seconds": 0.4450477740000025, "created_at": "2026-11-22T04:04:06Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-425a5b99c6e5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "7899ac94957a37384f0dfd33e27a3302805a65aa18e6a9e54a7db3ace11b15eb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-425a5b99c6e5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:06Z", "seed": 1725663919, "treatment": "full"}, "priority": 0.2547165909567554, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fb36d520e3fdd0d7e6425a316a25faa05bdafbb15292340b5edd9fe2945a561f", "run_seconds": 0.6205790042877197, "seed": 1725663919, "started_at": "2026-11-22T04:05:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0072, 0.0222], "effect": 0.0075, "evaluated_at": "2026-11-22T04:07:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-425a5b99c6e5", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.45, 0.921875, 0.97, 0.95], "frozen": [0.133333333, 0.125, 0.0, 0.1], "full": [0.45, 0.921875, 1.0, 0.95], "retrieval": [0.45, 0.921875, 0.97, 0.95], "rules_nodetect": [0.166666667, 0.421875, 1.0, 0.95]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7899ac94957a37384f0dfd33e27a3302805a65aa18e6a9e54a7db3ace11b15eb", "primary_control": "facts", "result_hash": "fb36d520e3fdd0d7e6425a316a25faa05bdafbb15292340b5edd9fe2945a561f", "signature": "752e1359bbbc011e1e80d134ed892409c97c6e0ee1b761ef8dc648b02798be079678145ac6f18c14a7013c5025fcb706ebb87c4fbe5e68e699b04a17bd346002", "state": "VERIFIED", "summary": {"facts": 0.822969, "frozen": 0.089583, "full": 0.830469, "retrieval": 0.822969, "rules_nodetect": 0.634635}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:18Z", "cpu_seconds": 0.5532088669999986, "created_at": "2026-11-22T04:04:06Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-95b8d2844d4b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "fa3e8b647f84d7f3f52d32455baabc1e35f34bc0598cab9a8c291f36dfaaf2d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-95b8d2844d4b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:06Z", "seed": 512502590, "treatment": "full"}, "priority": 0.27508499331567404, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cd68978381f81f2b42a6234f693f345dff9c6aa34b5a8a0e69a0988d607e765b", "run_seconds": 0.5669946670532227, "seed": 512502590, "started_at": "2026-11-22T04:04:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024885, 0.092396], "effect": 0.033756, "evaluated_at": "2026-11-22T04:04:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-95b8d2844d4b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.921568627, 0.823529412, 0.784313725, 0.268292683], "frozen": [0.091503268, 0.091503268, 0.078431373, 0.089430894], "full": [0.91503268, 0.843137255, 0.784313725, 0.390243902], "retrieval": [0.954248366, 0.849673203, 0.875816993, 0.390243902], "rules_nodetect": [0.91503268, 0.843137255, 0.784313725, 0.390243902]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "fa3e8b647f84d7f3f52d32455baabc1e35f34bc0598cab9a8c291f36dfaaf2d2", "primary_control": "facts", "result_hash": "cd68978381f81f2b42a6234f693f345dff9c6aa34b5a8a0e69a0988d607e765b", "signature": "1921d393634abd108db6fecfeddfe305e0968dcb865006049b926540f59b17f446c70ef0e14d7c3d234129f04be0dc235af0e4991813181146e81deb66faad0a", "state": "VERIFIED", "summary": {"facts": 0.699426, "frozen": 0.087717, "full": 0.733182, "retrieval": 0.767496, "rules_nodetect": 0.733182}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:04:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:12Z", "cpu_seconds": 1.9868815840000025, "created_at": "2026-11-22T04:04:06Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c6b3657b8d29", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "1558763fb221d3c51af3ee55e64de7c1edc626eb64c9732f70372b1217f2c695", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c6b3657b8d29", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:06Z", "seed": 2004560677, "treatment": "full"}, "priority": 0.30710696156818545, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3a6084e68f478a3b392b70e660dd35d152d4b064ec59cee25ab544a2824ddbb1", "run_seconds": 2.1640686988830566, "seed": 2004560677, "started_at": "2026-11-22T04:04:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006252, 0.072062], "effect": 0.032905, "evaluated_at": "2026-11-22T04:04:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c6b3657b8d29", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.87037037, 0.823809524, 0.333333333, 0.576], "frozen": [0.0, 0.352380952, 0.0, 0.157333333], "full": [0.87037037, 0.904761905, 0.333333333, 0.626666667], "retrieval": [0.972222222, 0.971428571, 0.505128205, 0.765333333], "rules_nodetect": [0.296296296, 0.295238095, 0.333333333, 0.237333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "1558763fb221d3c51af3ee55e64de7c1edc626eb64c9732f70372b1217f2c695", "primary_control": "facts", "result_hash": "3a6084e68f478a3b392b70e660dd35d152d4b064ec59cee25ab544a2824ddbb1", "signature": "c3fcc769934ab170c2e92af115354bb508a72fd012f3358fbc1a3392e6955a68f0f97e30f4d0e95dbee7fd3e9203df69c949be2e16ad11b43cc3ade209640a09", "state": "VERIFIED", "summary": {"facts": 0.650878, "frozen": 0.127429, "full": 0.683783, "retrieval": 0.803528, "rules_nodetect": 0.29055}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:04:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:15Z", "cpu_seconds": 3.786200858000001, "created_at": "2026-11-22T04:04:06Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-709201c95c69", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "b0e7beefa84acd2496cabbfbc21650c7827b11bf91df41067e13e72f1ccd59aa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-709201c95c69", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:04:06Z", "seed": 489537892, "treatment": "gated"}, "priority": 0.29797690432737467, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "0af5933bc206412253f7d02ca1600376b0b7ffa36eb89f5c33fe739fcb949939", "run_seconds": 4.56777548789978, "seed": 489537892, "started_at": "2026-11-22T04:04:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.151601, 0.045285], "effect": -0.053158, "evaluated_at": "2026-11-22T04:04:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-709201c95c69", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.461538462, 0.553571429, 0.821428571, 0.490740741], "gated": [0.461538462, 0.553571429, 0.821428571, 0.490740741], "naive": [0.730769231, 0.642857143, 0.785714286, 0.638888889], "scratch": [0.461538462, 0.5625, 0.821428571, 0.694444444], "ungated": [0.625, 0.517857143, 0.821428571, 0.638888889]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "b0e7beefa84acd2496cabbfbc21650c7827b11bf91df41067e13e72f1ccd59aa", "primary_control": "scratch", "result_hash": "0af5933bc206412253f7d02ca1600376b0b7ffa36eb89f5c33fe739fcb949939", "signature": "0d398ee3d00c7e0799f640a76dd682b6752d47ae5d0ea35e04a2e5735bd882f3b680ee0c6e2f66bda514fc96468ef70965d4b9f692fde390bae4721ee947ea03", "state": "VERIFIED", "summary": {"compressed": 0.58182, "gated": 0.58182, "naive": 0.699557, "scratch": 0.634978, "ungated": 0.650794}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:04:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:44Z", "cpu_seconds": 4.563113176999998, "created_at": "2026-11-22T04:04:23Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6830db8470f2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "7976078f1fade55ff1f2e1741c21b31648467252c67d321119f055e99a30c7e5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6830db8470f2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:04:23Z", "seed": 1905512273, "treatment": "frontier"}, "priority": 0.254748639277104, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e160c8a36d5a20dbcdf9323573aead8251ff6c39c39c2b9e8b06ba54c9e54c62", "run_seconds": 5.074157238006592, "seed": 1905512273, "started_at": "2026-11-22T04:04:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.060832, 0.08254], "effect": 0.010854, "evaluated_at": "2026-11-22T04:07:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6830db8470f2", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.538834951, 0.418154762, 0.386904762, 0.229166667], "frontier": [0.469255663, 0.430059524, 0.398809524, 0.345238095], "goal": [0.234627832, 0.380952381, 0.412202381, 0.345238095], "hypdisc": [0.511326861, 0.421130952, 0.410714286, 0.456845238], "infogain": [0.381877023, 0.400297619, 0.398809524, 0.194940476], "prederr": [0.511326861, 0.168154762, 0.398809524, 0.473214286], "random": [0.466019417, 0.389880952, 0.3125, 0.431547619], "risk": [0.194174757, 0.430059524, 0.398809524, 0.345238095]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "7976078f1fade55ff1f2e1741c21b31648467252c67d321119f055e99a30c7e5", "primary_control": "random", "result_hash": "e160c8a36d5a20dbcdf9323573aead8251ff6c39c39c2b9e8b06ba54c9e54c62", "signature": "63bced28c673f3096f8bdc498e2a234abfe8c02e747e40c8ed0ff98e45be6615918ae5609580b41fd5498003db158d7a36d2b368bfb1ea5139a744bddae79100", "state": "VERIFIED", "summary": {"count": 0.393265, "frontier": 0.410841, "goal": 0.343255, "hypdisc": 0.450004, "infogain": 0.343981, "prederr": 0.387876, "random": 0.399987, "risk": 0.34207}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:07:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:34Z", "cpu_seconds": 0.24398565399999939, "created_at": "2026-11-22T04:04:23Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-6a0c82952958", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "8ba144a6f147d8aaf7a59a80575733735cdd7c709b2ca4e14fa415452e9d4411", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-6a0c82952958", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:04:23Z", "seed": 1805004482, "treatment": "full"}, "priority": 0.25682476141541866, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4431c6729e749f1db701e8fcd6a4ae744a74b8dc1569cd064c61c1bb9455835a", "run_seconds": 0.24694323539733887, "seed": 1805004482, "started_at": "2026-11-22T04:04:34Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:04:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-6a0c82952958", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8ba144a6f147d8aaf7a59a80575733735cdd7c709b2ca4e14fa415452e9d4411", "primary_control": "e141_best", "result_hash": "4431c6729e749f1db701e8fcd6a4ae744a74b8dc1569cd064c61c1bb9455835a", "signature": "1ea38432fad5c452fe6b8db2cd4b1a552d78b4db1be0cffaa3c47ad09c93cc542be7692c6c2bcf12e7cffc9751f6d0fda6a8022870050d3e2c7c648301d88f09", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:04:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:37Z", "cpu_seconds": 2.8477914280000007, "created_at": "2026-11-22T04:04:23Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-32ee914d2a2b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "c42039d81cdf2d8f2adc6e38a8c7916dbcf5ba90c4814d45b2009bcd8d082592", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-32ee914d2a2b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:23Z", "seed": 267696238, "treatment": "full"}, "priority": 0.2615967106776017, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ed41a6a11299dcea1d06b045b8f62c056d32a09881df8c9523c62295f49921c1", "run_seconds": 2.9334845542907715, "seed": 267696238, "started_at": "2026-11-22T04:04:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.07668, 0.251521], "effect": 0.087421, "evaluated_at": "2026-11-22T04:04:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-32ee914d2a2b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.328571429, 0.92, 0.294444444], "frozen": [0.433333333, 0.152380952, 0.346666667, 0.223809524], "full": [1.0, 0.314285714, 0.946666667, 0.631746032], "retrieval": [1.0, 0.304761905, 0.973333333, 0.777777778], "rules_nodetect": [1.0, 0.314285714, 0.466666667, 0.297619048]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "c42039d81cdf2d8f2adc6e38a8c7916dbcf5ba90c4814d45b2009bcd8d082592", "primary_control": "facts", "result_hash": "ed41a6a11299dcea1d06b045b8f62c056d32a09881df8c9523c62295f49921c1", "signature": "06b78a12607be0a88d756b0e962dce4b5cd469a4147afd6c6316046cf6756f65cc0ee72c9317c9ef9b2a52fd3074137cf1670e87a82c986c122664c8d72b3408", "state": "VERIFIED", "summary": {"facts": 0.635754, "frozen": 0.289048, "full": 0.723175, "retrieval": 0.763968, "rules_nodetect": 0.519643}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:04:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:29Z", "cpu_seconds": 3.231875675000005, "created_at": "2026-11-22T04:04:23Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-326d77bd97e7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "337ac6ed4aa987ad829ceb9d1cfe242c4f63c3f5674c107f2241382ac25c1573", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-326d77bd97e7", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:04:23Z", "seed": 1956921161, "treatment": "gated"}, "priority": 0.3248236740950613, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "cdebf79f719aa4769a499e3a602fdf0a15d150760dde663939dc84b6d1cff307", "run_seconds": 3.280958890914917, "seed": 1956921161, "started_at": "2026-11-22T04:04:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.080367, 0.057297], "effect": -0.011535, "evaluated_at": "2026-11-22T04:04:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-326d77bd97e7", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.61, 0.461538462, 0.516666667, 0.543103448], "gated": [0.61, 0.461538462, 0.516666667, 0.543103448], "naive": [0.62, 0.586538462, 0.641666667, 0.724137931], "scratch": [0.62, 0.394230769, 0.516666667, 0.646551724], "ungated": [0.55, 0.586538462, 0.575, 0.646551724]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "337ac6ed4aa987ad829ceb9d1cfe242c4f63c3f5674c107f2241382ac25c1573", "primary_control": "scratch", "result_hash": "cdebf79f719aa4769a499e3a602fdf0a15d150760dde663939dc84b6d1cff307", "signature": "c5b6da6b4eee9eda0e81e971e1f0f8b48feae0ce77eb18df02ea5f66bbe92c1385ffa2fd206ec8d7c734f8a48490b75f59cf64b57b80288b02124ed0985d5a0b", "state": "VERIFIED", "summary": {"compressed": 0.532827, "gated": 0.532827, "naive": 0.643086, "scratch": 0.544362, "ungated": 0.589523}, "ties": 1, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:04:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:29Z", "cpu_seconds": 0.5615573090000012, "created_at": "2026-11-22T04:04:23Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-b98adb2ae995", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "27c1837c253b29f597521ad38772977064157f4b92132bc9c4339f6d41125244", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-b98adb2ae995", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:04:23Z", "seed": 1007330405, "treatment": "gated"}, "priority": 0.2744250260127158, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "902f8a76125d474e7d921c8188b8aae24aae57991aa70ccc35ce894a5b6d246d", "run_seconds": 0.6183505058288574, "seed": 1007330405, "started_at": "2026-11-22T04:04:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.061119, 0.005829], "effect": -0.027645, "evaluated_at": "2026-11-22T04:04:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-b98adb2ae995", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.649122807, 0.777777778, 0.676767677, 0.743589744], "gated": [0.649122807, 0.777777778, 0.676767677, 0.743589744], "naive": [0.666666667, 0.851851852, 0.757575758, 0.666666667], "scratch": [0.719298246, 0.777777778, 0.717171717, 0.743589744], "ungated": [0.666666667, 0.777777778, 0.414141414, 0.717948718]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "27c1837c253b29f597521ad38772977064157f4b92132bc9c4339f6d41125244", "primary_control": "scratch", "result_hash": "902f8a76125d474e7d921c8188b8aae24aae57991aa70ccc35ce894a5b6d246d", "signature": "ca1ad345bba355626f3e683803043dba1d6cd64fdf4efe86b8951f7815cc7f4d8dc23372041c2df09d2d53109b2949622ad404e1f3a82c1deac4c8d64894d40a", "state": "VERIFIED", "summary": {"compressed": 0.711815, "gated": 0.711815, "naive": 0.73569, "scratch": 0.739459, "ungated": 0.644134}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:04:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:09Z", "cpu_seconds": 11.417740288000005, "created_at": "2026-11-22T04:04:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6c7e98718d4a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "f9a8fc456ebe8283b256136d9c1191e155067ced94533c698c5d19873e6ef91a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6c7e98718d4a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:04:40Z", "seed": 1541204512, "treatment": "frontier"}, "priority": 0.2716403712300218, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-212e9e2ec1f5", "result_hash": "2ddf247b873143ada6d1598f06646f6b7b39b51c3e8b89b5c8fea4e1b2e58a60", "run_seconds": 16.306110382080078, "seed": 1541204512, "started_at": "2026-11-22T04:04:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.039847, 0.078584], "effect": 0.059216, "evaluated_at": "2026-11-22T04:07:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6c7e98718d4a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.530980392, 0.639215686, 0.592156863, 0.638431373], "frontier": [0.575686275, 0.682352941, 0.618039216, 0.712156863], "goal": [0.570196078, 0.694117647, 0.618039216, 0.661960784], "hypdisc": [0.584313725, 0.61254902, 0.680784314, 0.687843137], "infogain": [0.575686275, 0.682352941, 0.618039216, 0.712156863], "prederr": [0.575686275, 0.682352941, 0.618039216, 0.712156863], "random": [0.519215686, 0.618823529, 0.583529412, 0.629803922], "risk": [0.575686275, 0.682352941, 0.618039216, 0.712156863]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f9a8fc456ebe8283b256136d9c1191e155067ced94533c698c5d19873e6ef91a", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-212e9e2ec1f5", "result_hash": "2ddf247b873143ada6d1598f06646f6b7b39b51c3e8b89b5c8fea4e1b2e58a60", "signature": "02797c71526337b700ad31c09b01704ce41d94e35c351c6a33e23ffee050237d61b54f0ad770b89827bdf055cff7a7f1942577952625ebf36afd04b041909a0e", "state": "VERIFIED", "summary": {"count": 0.600196, "frontier": 0.647059, "goal": 0.636078, "hypdisc": 0.641373, "infogain": 0.647059, "prederr": 0.647059, "random": 0.587843, "risk": 0.647059}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:07:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:17Z", "cpu_seconds": 0.2992887119999992, "created_at": "2026-11-22T04:04:40Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-599526bf9c18", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "5f8e93958279324c016d3eb976584ea7d6271ace8adf23712ea27ede26354be1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-599526bf9c18", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:04:40Z", "seed": 1053533962, "treatment": "full"}, "priority": 0.2516189549720727, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b47cf38866de35513784628d6e24293e121088b892d3737e249906bcdc5e4772", "run_seconds": 0.31816577911376953, "seed": 1053533962, "started_at": "2026-11-22T04:05:17Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-599526bf9c18", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.75, 0.5, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "5f8e93958279324c016d3eb976584ea7d6271ace8adf23712ea27ede26354be1", "primary_control": "e141_best", "result_hash": "b47cf38866de35513784628d6e24293e121088b892d3737e249906bcdc5e4772", "signature": "fdcc315d39fb119d0e4890453dbc70ae293715423c8f48ddf8d1508b7a2692f2e531e7e867f4432ad70407053d48c190386b0abd0ff44e3ae6530c7f5ec1bf02", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:50Z", "cpu_seconds": 0.654825541000001, "created_at": "2026-11-22T04:04:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3a74d457af6c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "ae54d324cb1b6d35c0811f8f6d70338cbdf0a04dc1a161c216612d1bc4cce465", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3a74d457af6c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:40Z", "seed": 1222346190, "treatment": "full"}, "priority": 0.28419520235186785, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "250735974dff9fae259d1ccf8845067551e50b8a50468b06954eb3c9e4b30250", "run_seconds": 0.837120532989502, "seed": 1222346190, "started_at": "2026-11-22T04:04:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007273, 0.022424], "effect": 0.007576, "evaluated_at": "2026-11-22T04:07:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3a74d457af6c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.772727273, 0.628787879, 0.820512821, 0.882352941], "frozen": [0.333333333, 0.340909091, 0.358974359, 0.352941176], "full": [0.772727273, 0.659090909, 0.820512821, 0.882352941], "retrieval": [0.772727273, 0.71969697, 0.820512821, 0.882352941], "rules_nodetect": [0.772727273, 0.810606061, 0.820512821, 0.882352941]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "ae54d324cb1b6d35c0811f8f6d70338cbdf0a04dc1a161c216612d1bc4cce465", "primary_control": "facts", "result_hash": "250735974dff9fae259d1ccf8845067551e50b8a50468b06954eb3c9e4b30250", "signature": "996bdd94c746774e18f7f330ab300c33e08f71ba7c9912e04647ea0b083542cc05104a6760381e1c60a20c888675d536ca0083ba861812a509cacd8bca59cd02", "state": "VERIFIED", "summary": {"facts": 0.776095, "frozen": 0.346539, "full": 0.783671, "retrieval": 0.798823, "rules_nodetect": 0.82155}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:47Z", "cpu_seconds": 4.293772564000001, "created_at": "2026-11-22T04:04:40Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8c16c61f2afe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "5ec6a8d0b097ccc232fb9c77aa4a19970327cec92e2663af4f80b29fbb7ac782", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8c16c61f2afe", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:04:40Z", "seed": 1985960343, "treatment": "gated"}, "priority": 0.3306210475648928, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9843ed492bb3862c56f0049f373dc8b4a2f30af5fc16418e5fda4770c0798cb0", "run_seconds": 5.480926990509033, "seed": 1985960343, "started_at": "2026-11-22T04:04:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0185, 0.006], "effect": -0.00625, "evaluated_at": "2026-11-22T04:07:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-8c16c61f2afe", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.37, 0.548076923, 0.543103448, 0.533333333], "gated": [0.37, 0.548076923, 0.543103448, 0.533333333], "naive": [0.73, 0.634615385, 0.724137931, 0.591666667], "scratch": [0.37, 0.548076923, 0.543103448, 0.558333333], "ungated": [0.64, 0.634615385, 0.655172414, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5ec6a8d0b097ccc232fb9c77aa4a19970327cec92e2663af4f80b29fbb7ac782", "primary_control": "scratch", "result_hash": "9843ed492bb3862c56f0049f373dc8b4a2f30af5fc16418e5fda4770c0798cb0", "signature": "11ecddb1e9757514e5c2085ab27e668d96113bc41087c00c3d00128355f20c22920d50266a8a6b64a8d8a80546fbabaa36c4a74a48b816b5fd2b8c511215d208", "state": "VERIFIED", "summary": {"compressed": 0.498628, "gated": 0.498628, "naive": 0.670105, "scratch": 0.504878, "ungated": 0.607447}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:07:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:04:56Z", "cpu_seconds": 0.8004243009999996, "created_at": "2026-11-22T04:04:40Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-cd78780449f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|crafting|crafting|1|1"}, "prereg_hash": "ffa7ae690a063e779a80d83bffb707c3066a3d71013a6559520a576eb718e71d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-cd78780449f0", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|crafting|crafting|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:04:40Z", "seed": 650614760, "treatment": "gated"}, "priority": 0.25570595531266527, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "44dd231b1b6374f1373f769a1cda8aaf5841c6ef789549d675e63fb4ec93d8f1", "run_seconds": 1.2103285789489746, "seed": 650614760, "started_at": "2026-11-22T04:04:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006588, 0.106588], "effect": 0.05, "evaluated_at": "2026-11-22T04:07:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-cd78780449f0", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.355294118, 0.541176471, 0.451764706, 0.520784314], "gated": [0.454117647, 0.541176471, 0.552941176, 0.520784314], "naive": [0.519215686, 0.645490196, 0.480784314, 0.548235294], "scratch": [0.355294118, 0.541176471, 0.451764706, 0.520784314], "ungated": [0.530196078, 0.491764706, 0.479215686, 0.373333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "ffa7ae690a063e779a80d83bffb707c3066a3d71013a6559520a576eb718e71d", "primary_control": "scratch", "result_hash": "44dd231b1b6374f1373f769a1cda8aaf5841c6ef789549d675e63fb4ec93d8f1", "signature": "7f4254bbf3c3fa0a893a009b0cea96daa395ac2826354237ba7d4aab1fd0bf011c7eab47d50484ccd00220dcad1abd994ff158439849bd1c3d17cfbe5b230b0f", "state": "VERIFIED", "summary": {"compressed": 0.467255, "gated": 0.517255, "naive": 0.548431, "scratch": 0.467255, "ungated": 0.468627}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:07:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:26Z", "cpu_seconds": 1.407193331000002, "created_at": "2026-11-22T04:04:57Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-44adde5520a8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "a3cd1bdf2e305ae97354137d4ff3a6ad57a2be217750afe627e650a1c3cfc8f2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-44adde5520a8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:04:57Z", "seed": 144498754, "treatment": "frontier"}, "priority": 0.24825748676315856, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "cc6efa32e2de8133a2686ad717434bbd12b78aa458734a5d2576e4d142b877a0", "run_seconds": 1.5415406227111816, "seed": 144498754, "started_at": "2026-11-22T04:05:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01, 0.030833], "effect": 0.010417, "evaluated_at": "2026-11-22T04:07:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-44adde5520a8", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.958333333, 0.966666667, 0.793650794, 1.0], "frontier": [0.958333333, 0.966666667, 0.793650794, 1.0], "goal": [0.875, 0.9, 0.722222222, 1.0], "hypdisc": [0.875, 0.966666667, 0.698412698, 1.0], "infogain": [0.958333333, 0.966666667, 0.793650794, 1.0], "prederr": [0.958333333, 0.966666667, 0.761904762, 1.0], "random": [0.958333333, 0.966666667, 0.793650794, 0.958333333], "risk": [0.958333333, 0.966666667, 0.793650794, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a3cd1bdf2e305ae97354137d4ff3a6ad57a2be217750afe627e650a1c3cfc8f2", "primary_control": "random", "result_hash": "cc6efa32e2de8133a2686ad717434bbd12b78aa458734a5d2576e4d142b877a0", "signature": "88e5e734c3a89b7f6dc5df4cefc4d14d67c5aecf32f0c0c2c670d654f23f7b5d7036c3e90c985bd3bdceacc87a0c706117c401bf3986764f71bc63f54aa4ab0a", "state": "VERIFIED", "summary": {"count": 0.929663, "frontier": 0.929663, "goal": 0.874306, "hypdisc": 0.88502, "infogain": 0.929663, "prederr": 0.921726, "random": 0.919246, "risk": 0.929663}, "ties": 3, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:46Z", "cpu_seconds": 0.3351936809999998, "created_at": "2026-11-22T04:04:57Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5bf786117821", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "c4d10055648efd999c5c6bc6f3cc003465fcbc0ff29d199cc8e2f99b12fc373f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5bf786117821", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:04:57Z", "seed": 1133999852, "treatment": "full"}, "priority": 0.24737364736193404, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "35bbf6f77b14db5da0e24c66ec041f1652aa19b814d8c98ff3207bb7f4f58613", "run_seconds": 0.33628058433532715, "seed": 1133999852, "started_at": "2026-11-22T04:05:45Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5bf786117821", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c4d10055648efd999c5c6bc6f3cc003465fcbc0ff29d199cc8e2f99b12fc373f", "primary_control": "e141_best", "result_hash": "35bbf6f77b14db5da0e24c66ec041f1652aa19b814d8c98ff3207bb7f4f58613", "signature": "35a8ffc2b136dfff035dbf6701aee7f3cace266a68eb1fac7b819ded223aa9590f45b59e3e94764f2dff0bf8e2f42e1a1f9b5fc3429124981b77516d719a2101", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.75, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:56Z", "cpu_seconds": 0.7045413569999965, "created_at": "2026-11-22T04:04:57Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a617a8381d31", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "bd1dbcd1265c0d64061ecb44963e863afe9cc180c8991f63005d20b1bf7fef19", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a617a8381d31", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:57Z", "seed": 1953250064, "treatment": "full"}, "priority": 0.24736237284811669, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9c2abdf775d7016cb0e3c9b8d80e8275a50f06b07331cb031ceeb804c82f9da7", "run_seconds": 0.755516529083252, "seed": 1953250064, "started_at": "2026-11-22T04:05:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.029937, 0.018546], "effect": -0.005695, "evaluated_at": "2026-11-22T04:07:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a617a8381d31", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.796296296, 0.379487179, 0.777777778, 0.543209877], "frozen": [0.37037037, 0.317948718, 0.287581699, 0.358024691], "full": [0.814814815, 0.348717949, 0.754901961, 0.555555556], "retrieval": [0.796296296, 0.430769231, 0.754901961, 0.543209877], "rules_nodetect": [0.814814815, 0.753846154, 0.732026144, 0.802469136]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "bd1dbcd1265c0d64061ecb44963e863afe9cc180c8991f63005d20b1bf7fef19", "primary_control": "facts", "result_hash": "9c2abdf775d7016cb0e3c9b8d80e8275a50f06b07331cb031ceeb804c82f9da7", "signature": "e5bd086ec62edad1495c5e65decdc19c8b5d2987cf113ef7ac6cffb2cc225e9824cd4084da16099584e9229dee425d95393cb293f815d6a97252a07415bcf50d", "state": "VERIFIED", "summary": {"facts": 0.624193, "frozen": 0.333481, "full": 0.618498, "retrieval": 0.631294, "rules_nodetect": 0.775789}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:24Z", "cpu_seconds": 0.9765149139999991, "created_at": "2026-11-22T04:04:57Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-bc6679d41771", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "e22f378efa6959a658da3952f7b3087c3c1ff341ce081d9c50471e1001fe34ee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bc6679d41771", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:57Z", "seed": 1315889539, "treatment": "full"}, "priority": 0.24884633804526748, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "baf4ee10285008d1e97039d608990f1b4374e47293a376929e2d53a22185acad", "run_seconds": 1.1892576217651367, "seed": 1315889539, "started_at": "2026-11-22T04:05:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.028384, 0.060473], "effect": 0.016045, "evaluated_at": "2026-11-22T04:07:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-bc6679d41771", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.52, 0.299029126, 0.383636364, 0.481355932], "frozen": [0.114285714, 0.132038835, 0.103636364, 0.13559322], "full": [0.571428571, 0.248543689, 0.416363636, 0.511864407], "retrieval": [0.594285714, 0.262135922, 0.42, 0.511864407], "rules_nodetect": [0.571428571, 0.248543689, 0.638181818, 0.511864407]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "e22f378efa6959a658da3952f7b3087c3c1ff341ce081d9c50471e1001fe34ee", "primary_control": "facts", "result_hash": "baf4ee10285008d1e97039d608990f1b4374e47293a376929e2d53a22185acad", "signature": "5f178fbf4bb4135b1011bd1af31658dba3aa9506d7209e377118d9bcf93a6b9a3a70a6a5d595a5e676d208b7f8636a5936522f7bff7927fc3a3d259302a4e101", "state": "VERIFIED", "summary": {"facts": 0.421005, "frozen": 0.121389, "full": 0.43705, "retrieval": 0.447072, "rules_nodetect": 0.492505}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:06Z", "cpu_seconds": 2.7942660519999976, "created_at": "2026-11-22T04:04:57Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-cc6d592fafd3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "6538d90883330a41ab2f8f726b5b7da75e5862d292803044daa8027104bac90f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-cc6d592fafd3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:04:57Z", "seed": 314896320, "treatment": "full"}, "priority": 0.2568873960329362, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f799ed941a6ea7d0a457268c81e4d14fbb92194cf0996da808007c35ecfb4def", "run_seconds": 5.151009798049927, "seed": 314896320, "started_at": "2026-11-22T04:05:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017618, 0.164145], "effect": 0.073264, "evaluated_at": "2026-11-22T04:07:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-cc6d592fafd3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.86, 0.99047619, 0.420833333, 0.993333333], "frozen": [0.33, 0.304761905, 0.138888889, 0.333333333], "full": [0.96, 0.99047619, 0.613888889, 0.993333333], "retrieval": [0.9, 1.0, 0.640277778, 0.993333333], "rules_nodetect": [0.58, 0.99047619, 0.613888889, 0.34]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "6538d90883330a41ab2f8f726b5b7da75e5862d292803044daa8027104bac90f", "primary_control": "facts", "result_hash": "f799ed941a6ea7d0a457268c81e4d14fbb92194cf0996da808007c35ecfb4def", "signature": "9e9db0d778bd36390afd80b2c833bd29acda238c3b6ac51ec78b46971811ff9d8112859d86bd7c793abfccfde98404c53c03239fd5aac397d1f03a72c81c1206", "state": "VERIFIED", "summary": {"facts": 0.816161, "frozen": 0.276746, "full": 0.889425, "retrieval": 0.883403, "rules_nodetect": 0.631091}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:20Z", "cpu_seconds": 0.7531271399999966, "created_at": "2026-11-22T04:05:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2356ab663a26", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "685a297ffe4c488fa51e1e29426903e31391d4edbb11949fe58346198c112e79", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2356ab663a26", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:05:14Z", "seed": 2017803862, "treatment": "full"}, "priority": 0.2498472815786726, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "630e48342ad323c08189924cb616e460c6a3624f5c8a203e266f4ddccaaf4f6a", "run_seconds": 0.8044922351837158, "seed": 2017803862, "started_at": "2026-11-22T04:05:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.370785, 0.245785], "effect": -0.0625, "evaluated_at": "2026-11-22T04:07:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2356ab663a26", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.25, 1.0, 0.75], "fixed_policy": [0.5, 0.5, 0.0, 0.0], "full": [0.5, 0.5, 1.0, 0.25], "no_world_model": [0.5, 0.75, 1.0, 0.75], "random": [0.0, 1.0, 1.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "685a297ffe4c488fa51e1e29426903e31391d4edbb11949fe58346198c112e79", "primary_control": "e141_best", "result_hash": "630e48342ad323c08189924cb616e460c6a3624f5c8a203e266f4ddccaaf4f6a", "signature": "9909550e2d6b5ff6b036610b1cc3f884679771e1bb93851fa7b4bbe6b911f3ee40be064b2e6b5840cd9800df8dae5f21730e87e4358f75eca5d21d6ddc856e0f", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.25, "full": 0.5625, "no_world_model": 0.75, "random": 0.5625}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:14Z", "cpu_seconds": 0.2828009780000045, "created_at": "2026-11-22T04:05:14Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-aa17454b1a13", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "e1a95d9ae3006a9b282d991019ddecc86888bc42d54117d1b10a3426064d5452", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-aa17454b1a13", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:05:14Z", "seed": 1428759838, "treatment": "full"}, "priority": 0.2541240456449343, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f1fc37a39740e89ae5385b2f9b1af28aa13b1de8ebe6f0bdae1829d5736f6f15", "run_seconds": 0.3639397621154785, "seed": 1428759838, "started_at": "2026-11-22T04:05:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001023, 0.009175], "effect": 0.004076, "evaluated_at": "2026-11-22T04:07:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-aa17454b1a13", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.913043478, 0.989130435, 0.989130435], "frozen": [0.016304348, 0.016304348, 0.016304348, 0.010869565], "full": [1.0, 0.923913043, 0.989130435, 0.994565217], "retrieval": [1.0, 0.913043478, 0.989130435, 0.97826087], "rules_nodetect": [1.0, 0.923913043, 0.989130435, 0.994565217]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "e1a95d9ae3006a9b282d991019ddecc86888bc42d54117d1b10a3426064d5452", "primary_control": "facts", "result_hash": "f1fc37a39740e89ae5385b2f9b1af28aa13b1de8ebe6f0bdae1829d5736f6f15", "signature": "b62344f60d85b410203a321c7f1cf3c3339dce4a13ab52e5c59d4aa47a1de0b273bfa61ad3d52b1f224e6e9870b5308ebff5cf7208d7f575482509909952a50a", "state": "VERIFIED", "summary": {"facts": 0.972826, "frozen": 0.014946, "full": 0.976902, "retrieval": 0.970109, "rules_nodetect": 0.976902}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:38Z", "cpu_seconds": 0.6904625539999998, "created_at": "2026-11-22T04:05:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4f2936fc9b74", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "f2c505ebe02561efc9893a021e3b79147248287ad242b33696ef7920add9ff83", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4f2936fc9b74", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:05:30Z", "seed": 22325277, "treatment": "full"}, "priority": 0.2518320150549794, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7ede3b3a8198be511231656a3ee57d0825cb9d76bbb9714c1b6561c0d301edaf", "run_seconds": 0.8014538288116455, "seed": 22325277, "started_at": "2026-11-22T04:05:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4f2936fc9b74", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 1.0], "fixed_policy": [0.5, 0.25, 1.0, 1.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 0.5, 1.0, 1.0], "random": [1.0, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "f2c505ebe02561efc9893a021e3b79147248287ad242b33696ef7920add9ff83", "primary_control": "e141_best", "result_hash": "7ede3b3a8198be511231656a3ee57d0825cb9d76bbb9714c1b6561c0d301edaf", "signature": "613564c34ab08c236e095a9bd69158c9a31fced45514db049d5752d8b3b9e6a1fdf947c97b988302d3737825467f2a11d4c2411054eae36b22217e9948c98b05", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.6875, "full": 0.9375, "no_world_model": 0.875, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:04Z", "cpu_seconds": 1.8813056750000001, "created_at": "2026-11-22T04:05:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9334171f48ef", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "1104804aa4eede0c7de08ff9dede3fda6153430689636f7b1888aaf8a8fa7100", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9334171f48ef", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:05:30Z", "seed": 645750824, "treatment": "full"}, "priority": 0.24620667587023068, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "53404921f3c77de0728d43c7bf23c0c7ac64e461b1b1e84584e7b45d7d1b6729", "run_seconds": 2.392916440963745, "seed": 645750824, "started_at": "2026-11-22T04:06:01Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9334171f48ef", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 1.0, 0.0], "fixed_policy": [0.5, 0.25, 0.25, 0.0], "full": [1.0, 0.0, 1.0, 0.0], "no_world_model": [1.0, 0.0, 1.0, 0.0], "random": [1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1104804aa4eede0c7de08ff9dede3fda6153430689636f7b1888aaf8a8fa7100", "primary_control": "e141_best", "result_hash": "53404921f3c77de0728d43c7bf23c0c7ac64e461b1b1e84584e7b45d7d1b6729", "signature": "14d581a3234c332dd430d3cc98722f71601f999ef85c58f8a25e8588b704b7fbc07469f5ec43812539f8c57b0aebb29c7748a0313884593dd5e997ef6667a40a", "state": "PARTIAL", "summary": {"e141_best": 0.5, "fixed_policy": 0.25, "full": 0.5, "no_world_model": 0.5, "random": 0.5}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:45Z", "cpu_seconds": 1.3800726020000056, "created_at": "2026-11-22T04:05:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a511a7eb5180", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 4, "n": 4}, "prereg_hash": "73670f095f8765ead34293367096b00994729adb261921ec53196593c17aebb0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a511a7eb5180", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:05:30Z", "seed": 1245803725, "treatment": "full"}, "priority": 0.2479715252682477, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8ec5cf7e9427b72502a09e2c953c3194ed145bedb2db255d55006401874fc7c1", "run_seconds": 1.7209100723266602, "seed": 1245803725, "started_at": "2026-11-22T04:05:43Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a511a7eb5180", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.75, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "73670f095f8765ead34293367096b00994729adb261921ec53196593c17aebb0", "primary_control": "e141_best", "result_hash": "8ec5cf7e9427b72502a09e2c953c3194ed145bedb2db255d55006401874fc7c1", "signature": "1abc8d00d463d876d10c99a2e39ec77d614e197ff32ca56fcb3643b4b07891efd02102400b02806ce1492d60631b326e36c8a6d863b6e09ffda335531d640b0d", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.9375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:32Z", "cpu_seconds": 0.6777022349999982, "created_at": "2026-11-22T04:05:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bed6c97c989d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "84512bfc2b55e262ea259f5212afd2846090fb3349a3b361d373e0dce69c0d0b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bed6c97c989d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:05:30Z", "seed": 1373922980, "treatment": "full"}, "priority": 0.25442797687491436, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8514c8ef998198fc9a49c512b655e62e19063e07fb5056f53b712bf7abe1e591", "run_seconds": 0.7902319431304932, "seed": 1373922980, "started_at": "2026-11-22T04:05:31Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-bed6c97c989d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.25, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 0.75, 0.75], "random": [0.5, 0.75, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "84512bfc2b55e262ea259f5212afd2846090fb3349a3b361d373e0dce69c0d0b", "primary_control": "e141_best", "result_hash": "8514c8ef998198fc9a49c512b655e62e19063e07fb5056f53b712bf7abe1e591", "signature": "3bfdf224730e57b51ef3bd6101f1c2129eec9a8e8ec01670e9e294e7e140eec1edf5113cb65e9d8bb14a15f93cf8ff9a5a2ad051eecfbae06e712391ba2fab07", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 0.8125, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:40Z", "cpu_seconds": 0.46837619600000124, "created_at": "2026-11-22T04:05:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d913475e7fe9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "8f88d1e1e9e69cf901f2b4bc664ba93b4f55aecfc0476ab531ea993af76abb44", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d913475e7fe9", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:05:30Z", "seed": 1173539945, "treatment": "full"}, "priority": 0.2502691045884054, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a2ccee1c2211029a5919e637831f6a34c72e0603e9cace498e6b485eaa0cf4dd", "run_seconds": 0.4716365337371826, "seed": 1173539945, "started_at": "2026-11-22T04:05:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d913475e7fe9", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 1.0, 0.25, 1.0], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 0.5, 0.75, 1.0], "random": [0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "8f88d1e1e9e69cf901f2b4bc664ba93b4f55aecfc0476ab531ea993af76abb44", "primary_control": "e141_best", "result_hash": "a2ccee1c2211029a5919e637831f6a34c72e0603e9cace498e6b485eaa0cf4dd", "signature": "4014f021c28f9350df9149381251e8c7cdc0d688adaf3c7e3ad4d2eb24f16297e1794927b010f57c52c01d3ea1a218c8a914770aa192f41573f7024fee51a10f", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.5625, "full": 0.75, "no_world_model": 0.5625, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:35Z", "cpu_seconds": 0.724764553, "created_at": "2026-11-22T04:05:30Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e9e4c2f9db7d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "ed2176124056accf69c62e2883ec55daa56d8a6607a2a66dc39ba043cffdf390", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e9e4c2f9db7d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:05:30Z", "seed": 441336562, "treatment": "gated"}, "priority": 0.2530978074488002, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "28917dd19740141a7810bcaa92d29f9fd3584b1cb3ec2ca84faf9b9dcab85800", "run_seconds": 0.7847723960876465, "seed": 441336562, "started_at": "2026-11-22T04:05:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017376, 0.026462], "effect": 0.004543, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-e9e4c2f9db7d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.807017544, 0.773809524, 0.851851852, 0.621212121], "gated": [0.807017544, 0.773809524, 0.851851852, 0.621212121], "naive": [0.807017544, 0.726190476, 0.851851852, 0.575757576], "scratch": [0.824561404, 0.738095238, 0.851851852, 0.621212121], "ungated": [0.789473684, 0.773809524, 0.888888889, 0.689393939]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "ed2176124056accf69c62e2883ec55daa56d8a6607a2a66dc39ba043cffdf390", "primary_control": "scratch", "result_hash": "28917dd19740141a7810bcaa92d29f9fd3584b1cb3ec2ca84faf9b9dcab85800", "signature": "61f60d6d41e2ccb9f7c6054e36bc8626a34cec7ea003542cba395e0fa0819cf5f60438e2a089edf5c819e4fd8ce7456f4c40e2a88a334d15ae9af4c97f5c5204", "state": "VERIFIED", "summary": {"compressed": 0.763473, "gated": 0.763473, "naive": 0.740204, "scratch": 0.75893, "ungated": 0.785392}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:07:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:12Z", "cpu_seconds": 0.8078276280000001, "created_at": "2026-11-22T04:05:47Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-210527c2505e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "149e7897805603a477f727deb64543811b5f59c6d9eedf3ffbbfe5755fb33cd2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-210527c2505e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:05:47Z", "seed": 2046585222, "treatment": "frontier"}, "priority": 0.2459162359878265, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3f3964cb258fbbde2e3d2d54a81c3b10f8fdcd4f5fc637e93c246c7985c76419", "run_seconds": 0.8750231266021729, "seed": 2046585222, "started_at": "2026-11-22T04:08:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.191566, 0.160725], "effect": -0.015421, "evaluated_at": "2026-11-22T04:08:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-210527c2505e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.790849673, 0.758169935, 0.916666667, 0.588235294], "frontier": [0.764705882, 0.882352941, 0.572916667, 0.790849673], "goal": [0.823529412, 0.849673203, 0.8125, 0.836601307], "hypdisc": [0.68627451, 0.823529412, 0.65625, 0.784313725], "infogain": [0.588235294, 0.673202614, 0.8125, 0.679738562], "prederr": [0.823529412, 0.594771242, 0.8125, 0.679738562], "random": [0.745098039, 0.732026144, 0.84375, 0.751633987], "risk": [0.699346405, 0.777777778, 0.572916667, 0.483660131]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "149e7897805603a477f727deb64543811b5f59c6d9eedf3ffbbfe5755fb33cd2", "primary_control": "random", "result_hash": "3f3964cb258fbbde2e3d2d54a81c3b10f8fdcd4f5fc637e93c246c7985c76419", "signature": "4b20bb369cba9589b1f87059bb5bafefd7c05a45c6028c21486e2176b439ef60dd19257cdf3cf35017b22c63aaf792a6ad46cee8b7472746cd1c78e2072aaf09", "state": "VERIFIED", "summary": {"count": 0.76348, "frontier": 0.752706, "goal": 0.830576, "hypdisc": 0.737592, "infogain": 0.688419, "prederr": 0.727635, "random": 0.768127, "risk": 0.633425}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:08:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:07Z", "cpu_seconds": 3.284384808999995, "created_at": "2026-11-22T04:05:47Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-70e9ba67aea2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "3194fa6117671b7b67ebf984f7612d068d433c3b848c2a35ad1d8bbe924b8eff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-70e9ba67aea2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:05:47Z", "seed": 1769902383, "treatment": "frontier"}, "priority": 0.2461511708967307, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "0604a361200555caefd156cc97a5403adae4d8e0d2e5adf4a8d48c31323c5a11", "run_seconds": 3.9261443614959717, "seed": 1769902383, "started_at": "2026-11-22T04:06:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00881, 0.002857], "effect": -0.002976, "evaluated_at": "2026-11-22T04:07:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-70e9ba67aea2", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.986111111, 1.0, 1.0, 1.0], "frontier": [0.986111111, 1.0, 0.988095238, 1.0], "goal": [0.986111111, 1.0, 0.988095238, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [0.986111111, 1.0, 0.988095238, 1.0], "prederr": [0.986111111, 1.0, 0.988095238, 1.0], "random": [0.986111111, 1.0, 1.0, 1.0], "risk": [0.986111111, 1.0, 0.988095238, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "3194fa6117671b7b67ebf984f7612d068d433c3b848c2a35ad1d8bbe924b8eff", "primary_control": "random", "result_hash": "0604a361200555caefd156cc97a5403adae4d8e0d2e5adf4a8d48c31323c5a11", "signature": "b0f514b65a20b6cf4da376545b9f16889780f97b378a54b26535069c5ae4ca3c497cb96d0f196cea44940f700de1574d5abb7d8635e40138931aa852d5628304", "state": "VERIFIED", "summary": {"count": 0.996528, "frontier": 0.993552, "goal": 0.993552, "hypdisc": 1.0, "infogain": 0.993552, "prederr": 0.993552, "random": 0.996528, "risk": 0.993552}, "ties": 3, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:07:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:15Z", "cpu_seconds": 1.2408276299999983, "created_at": "2026-11-22T04:05:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-58d27d4a052c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "66460e308ecce00d74f4218f0c995e2992cb03a6d822c988484d0985a8c1da91", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-58d27d4a052c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:05:47Z", "seed": 2139490120, "treatment": "full"}, "priority": 0.24574803720827196, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5e27854cd8548d9d7f48171c980bc9282231c517af769f06d98055990429cc2c", "run_seconds": 1.325911045074463, "seed": 2139490120, "started_at": "2026-11-22T04:08:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.04707, 0.42207], "effect": 0.1875, "evaluated_at": "2026-11-22T04:08:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-58d27d4a052c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 0.5], "fixed_policy": [0.0, 0.0, 0.5, 0.0], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [0.25, 1.0, 1.0, 0.0], "random": [0.75, 0.75, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "66460e308ecce00d74f4218f0c995e2992cb03a6d822c988484d0985a8c1da91", "primary_control": "e141_best", "result_hash": "5e27854cd8548d9d7f48171c980bc9282231c517af769f06d98055990429cc2c", "signature": "a35dd09cd0528a45a7b9f12e07cb5f571012e9e8bc64ef064880b1726d4b10ae25afa4d86880cb0bc8cb92b93fb8a284211956cd9163a989fe552ad999161508", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.125, "full": 0.9375, "no_world_model": 0.5625, "random": 0.8125}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:08:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:26Z", "cpu_seconds": 2.298130420000007, "created_at": "2026-11-22T04:05:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-263d51b4f357", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "40a6e665a2f7c3db671bebdfffdb696f426ec883da99f09eb66b5163bfb0f8f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-263d51b4f357", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:05:47Z", "seed": 1167408820, "treatment": "full"}, "priority": 0.24602534578521001, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-a56a76034979", "result_hash": "fa07b2412fe1c863bed7ebebe7054663cca029f7292da59673da87de3cac8590", "run_seconds": 3.1956613063812256, "seed": 1167408820, "started_at": "2026-11-22T04:07:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002055, 0.046905], "effect": 0.022425, "evaluated_at": "2026-11-22T04:07:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-263d51b4f357", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.777777778, 0.306878307, 0.910798122, 0.828125], "frozen": [0.08994709, 0.08994709, 0.084507042, 0.088541667], "full": [0.793650794, 0.365079365, 0.910798122, 0.84375], "retrieval": [0.894179894, 0.544973545, 0.943661972, 0.854166667], "rules_nodetect": [0.793650794, 0.365079365, 0.910798122, 0.84375]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "40a6e665a2f7c3db671bebdfffdb696f426ec883da99f09eb66b5163bfb0f8f4", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-a56a76034979", "result_hash": "fa07b2412fe1c863bed7ebebe7054663cca029f7292da59673da87de3cac8590", "signature": "d35a39055994bc51f3e31f7cfd653432158d96a5e69d78aefc3a2fe4e016d8652151e68f247df326a7da368356148ccebf27dc81ee774e8c0d09d5e8ac97490a", "state": "VERIFIED", "summary": {"facts": 0.705895, "frozen": 0.088236, "full": 0.72832, "retrieval": 0.809246, "rules_nodetect": 0.72832}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:51Z", "cpu_seconds": 0.7319241200000022, "created_at": "2026-11-22T04:05:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-db15477a369f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 6, "n": 4}, "prereg_hash": "bc8dbe6a513760c86fd474756e4591c38bd77c026864db37114c02f3b7eed2b2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-db15477a369f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:05:47Z", "seed": 1951546536, "treatment": "full"}, "priority": 0.24772779248718585, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c3a452ddc47335e74e4c4ceda790093fd1f3b0a3f192c4271c372c5eecaba440", "run_seconds": 0.7679522037506104, "seed": 1951546536, "started_at": "2026-11-22T04:05:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002493, 0.108828], "effect": 0.053167, "evaluated_at": "2026-11-22T04:07:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-db15477a369f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.81875, 0.441666667, 0.500847458], "frozen": [0.65, 0.5375, 0.366666667, 0.41440678], "full": [1.0, 0.90625, 0.556666667, 0.511016949], "retrieval": [1.0, 0.94375, 0.779166667, 0.78559322], "rules_nodetect": [0.875, 0.68125, 0.306666667, 0.466101695]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "bc8dbe6a513760c86fd474756e4591c38bd77c026864db37114c02f3b7eed2b2", "primary_control": "facts", "result_hash": "c3a452ddc47335e74e4c4ceda790093fd1f3b0a3f192c4271c372c5eecaba440", "signature": "8ebed54e2603c9f659a9b3e31efc7a1adf5af94495a7acf0d32470610db9da608079af238f817b471516d6c3b7215c2b523682afe45cc59114f444fe537d6f03", "state": "VERIFIED", "summary": {"facts": 0.690316, "frozen": 0.492143, "full": 0.743483, "retrieval": 0.877127, "rules_nodetect": 0.582255}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:05:52Z", "cpu_seconds": 0.9789913019999972, "created_at": "2026-11-22T04:05:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ed5bf3ee0daf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 10, "n": 4}, "prereg_hash": "928b7ff803295a8855752718b21d8057b45815ab8dfea1236496d1288adac66e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ed5bf3ee0daf", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:05:47Z", "seed": 965880430, "treatment": "full"}, "priority": 0.24743988864449354, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "63439a509dfad5291eac515886be2d1b9013affc0515d2218e3406dbd99ebe81", "run_seconds": 1.0541510581970215, "seed": 965880430, "started_at": "2026-11-22T04:05:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.023138, 0.133871], "effect": 0.078505, "evaluated_at": "2026-11-22T04:07:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ed5bf3ee0daf", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.629861111, 0.463848039, 0.6, 0.682352941], "frozen": [0.419444444, 0.394607843, 0.54137931, 0.529411765], "full": [0.657638889, 0.615808824, 0.693103448, 0.723529412], "retrieval": [0.642361111, 0.725490196, 0.751724138, 0.820588235], "rules_nodetect": [0.657638889, 0.615808824, 0.510344828, 0.502941176]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "928b7ff803295a8855752718b21d8057b45815ab8dfea1236496d1288adac66e", "primary_control": "facts", "result_hash": "63439a509dfad5291eac515886be2d1b9013affc0515d2218e3406dbd99ebe81", "signature": "bd08543976edccf7db4736ee33c0916ff3e51d626fbc90d9ca9feee3ee77d7c9b5d868bc4651c3290e07250e376287ede82d413e30f1ab14a921f898e96c6902", "state": "VERIFIED", "summary": {"facts": 0.594016, "frozen": 0.471211, "full": 0.67252, "retrieval": 0.735041, "rules_nodetect": 0.571683}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:07:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:18Z", "cpu_seconds": 1.1467823300000077, "created_at": "2026-11-22T04:06:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d07d1dc2e299", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "699d17c22a02332abd1e3c8295b5040fe0f00643d2971637c47d924ad40637a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d07d1dc2e299", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:06:04Z", "seed": 2006825659, "treatment": "frontier"}, "priority": 0.24864322399997307, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "0d7e07b2293199a8e96bb69797e4dab7da951443afbed1ad1a70d25edf1360b4", "run_seconds": 1.6084587574005127, "seed": 2006825659, "started_at": "2026-11-22T04:06:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.162925, 0.214418], "effect": 0.025747, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d07d1dc2e299", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.80952381, 0.888888889, 0.804878049, 0.552677029], "frontier": [0.857142857, 0.888888889, 0.804878049, 0.552677029], "goal": [0.857142857, 0.851851852, 0.804878049, 0.552677029], "hypdisc": [0.833333333, 0.851851852, 0.804878049, 0.499136442], "infogain": [0.857142857, 0.888888889, 0.804878049, 0.552677029], "prederr": [0.857142857, 0.888888889, 0.804878049, 0.497409326], "random": [0.904761905, 0.851851852, 0.520325203, 0.723661485], "risk": [0.857142857, 0.888888889, 0.804878049, 0.552677029]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "699d17c22a02332abd1e3c8295b5040fe0f00643d2971637c47d924ad40637a2", "primary_control": "random", "result_hash": "0d7e07b2293199a8e96bb69797e4dab7da951443afbed1ad1a70d25edf1360b4", "signature": "03aec7be87bc7f69643669d4dd4125269daa4e5131bcb7f9eddc65446fa4a7dc854c7edba57ed8ef521c0afad92166775fcf082c3f01245adaf46fdd4b3bec08", "state": "VERIFIED", "summary": {"count": 0.763992, "frontier": 0.775897, "goal": 0.766637, "hypdisc": 0.7473, "infogain": 0.775897, "prederr": 0.76208, "random": 0.75015, "risk": 0.775897}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:09Z", "cpu_seconds": 0.4978684290000004, "created_at": "2026-11-22T04:06:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1f7a6b1b6b59", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "a923788a65dd754bfe6813cd06f07fc8019fd5f42358e1a9817958ef88850976", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1f7a6b1b6b59", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:06:04Z", "seed": 1306459420, "treatment": "full"}, "priority": 0.25628662471529234, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4293618acdfa6ba86a11f6c2ac2e049a0f6622e60e2481b6f8e58eddca96c5a2", "run_seconds": 0.5256621837615967, "seed": 1306459420, "started_at": "2026-11-22T04:06:09Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1f7a6b1b6b59", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.25, 0.25, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a923788a65dd754bfe6813cd06f07fc8019fd5f42358e1a9817958ef88850976", "primary_control": "e141_best", "result_hash": "4293618acdfa6ba86a11f6c2ac2e049a0f6622e60e2481b6f8e58eddca96c5a2", "signature": "2051d774705dd7a2caf912d2e64821dfb3edcadad93a163737b4efca3825b55b107cce3369842d602d5fbbaca5ff80624935ecc816ee06ca461fbdffe3c12e04", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 0.8125, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:12Z", "cpu_seconds": 0.16997872400000347, "created_at": "2026-11-22T04:06:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-62cfa4e2a6fe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "40a2314fc86ed954df5415dfb69cde4e1238cf811b70322837c2edf1f670eac4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-62cfa4e2a6fe", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:06:04Z", "seed": 1374512056, "treatment": "full"}, "priority": 0.24971037448114175, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "04eb561216096f4480e9d82c528c74eb55a255a606e6fb8cb762aa60e93d7d06", "run_seconds": 0.1793508529663086, "seed": 1374512056, "started_at": "2026-11-22T04:06:12Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-62cfa4e2a6fe", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.75, 0.25, 0.75, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "40a2314fc86ed954df5415dfb69cde4e1238cf811b70322837c2edf1f670eac4", "primary_control": "e141_best", "result_hash": "04eb561216096f4480e9d82c528c74eb55a255a606e6fb8cb762aa60e93d7d06", "signature": "9f84d60005c1eb34faaeef110d05489e6f5339d0e648a9ca5d8293fdd58c0d667fd1fae263bf660b13c750922707290171779bd44ae87785159b23716daaf70e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.5, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:15Z", "cpu_seconds": 0.598651928999999, "created_at": "2026-11-22T04:06:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f68d0eb4e5ab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "352f6c23d513bd181a687b83c79b82d867e4e0e5cf9bbc754da2434802d76da5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f68d0eb4e5ab", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:06:04Z", "seed": 1739718129, "treatment": "full"}, "priority": 0.24931653801189121, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8da157a9974469047076bbfb1e686925613a3ec7aa8fcb8491a57b730d6aba68", "run_seconds": 0.7466304302215576, "seed": 1739718129, "started_at": "2026-11-22T04:06:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T04:07:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f68d0eb4e5ab", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 0.75], "no_world_model": [0.5, 0.5, 1.0, 1.0], "random": [1.0, 0.5, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "352f6c23d513bd181a687b83c79b82d867e4e0e5cf9bbc754da2434802d76da5", "primary_control": "e141_best", "result_hash": "8da157a9974469047076bbfb1e686925613a3ec7aa8fcb8491a57b730d6aba68", "signature": "21e548ae36469aaed9a9dfbc07110a9591c04488d22bdb3a10b14aff264636a28a20411c00f38a9e6986b1d5dec02e4111403e700cf5b00233d826f961185500", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.75, "random": 0.8125}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:38Z", "cpu_seconds": 0.5018790060000029, "created_at": "2026-11-22T04:06:04Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0cef5d6a8750", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "a7954d5cee514656b3562cc6976184762dbf9ae7fb4e59c7be3a06c4c0fc90dc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0cef5d6a8750", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:06:04Z", "seed": 2117243975, "treatment": "full"}, "priority": 0.2466409821240343, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "74c42648251fbdbdb9e03dc35f6fcc65b985fac0783a90ea6ecf1716279dcfbb", "run_seconds": 0.645897626876831, "seed": 2117243975, "started_at": "2026-11-22T04:06:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006316, 0.019474], "effect": 0.006579, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0cef5d6a8750", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.842105263, 0.828947368, 0.95, 0.863636364], "frozen": [0.592105263, 0.671052632, 0.625, 0.659090909], "full": [0.842105263, 0.855263158, 0.95, 0.863636364], "retrieval": [0.842105263, 0.855263158, 0.95, 0.863636364], "rules_nodetect": [0.842105263, 0.828947368, 0.95, 0.863636364]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a7954d5cee514656b3562cc6976184762dbf9ae7fb4e59c7be3a06c4c0fc90dc", "primary_control": "facts", "result_hash": "74c42648251fbdbdb9e03dc35f6fcc65b985fac0783a90ea6ecf1716279dcfbb", "signature": "892cbe02bc1f1716556d0b161cf7976e22bccf94d699e6ce61cb72f242563f186a13da748ae8f3b959e7014f93cf121c61f1847062a2219840610f31ae3a8f0e", "state": "VERIFIED", "summary": {"facts": 0.871172, "frozen": 0.636812, "full": 0.877751, "retrieval": 0.877751, "rules_nodetect": 0.871172}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:21Z", "cpu_seconds": 0.7315333170000002, "created_at": "2026-11-22T04:06:04Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e7c09da346c7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "fbf6c0c6b7207d82e7dbf6cc63b8ba89f849aaa52c7f9a42ccad020390ad2c77", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e7c09da346c7", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:06:04Z", "seed": 1777822174, "treatment": "gated"}, "priority": 0.24786425365759251, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "23aaa6d7d135dd5e68b580eb736643ed781a43c2732a7f83778636db1fd219b3", "run_seconds": 1.0707957744598389, "seed": 1777822174, "started_at": "2026-11-22T04:06:20Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-e7c09da346c7", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.72972973, 0.804878049, 0.538888889, 0.711610487], "gated": [0.72972973, 0.804878049, 0.538888889, 0.711610487], "naive": [0.72972973, 0.772357724, 0.627777778, 0.741573034], "scratch": [0.72972973, 0.804878049, 0.538888889, 0.711610487], "ungated": [0.72972973, 0.772357724, 0.627777778, 0.741573034]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fbf6c0c6b7207d82e7dbf6cc63b8ba89f849aaa52c7f9a42ccad020390ad2c77", "primary_control": "scratch", "result_hash": "23aaa6d7d135dd5e68b580eb736643ed781a43c2732a7f83778636db1fd219b3", "signature": "b6a96d9f86532b9b1163ed0f00ad56befde86e47e7cb4ab8205d0353ec081de2e6c1620edf3b3f41f704f9936733431d23f8d62885eb4696f899d382556fc405", "state": "PARTIAL", "summary": {"compressed": 0.696277, "gated": 0.696277, "naive": 0.71786, "scratch": 0.696277, "ungated": 0.71786}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:32Z", "cpu_seconds": 0.7169838109999986, "created_at": "2026-11-22T04:06:21Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-90fc148c5a83", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "56bebbab0fa4e405be6e9422d355c01d7e1caa698902d30663924eb1cb23b0be", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-90fc148c5a83", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:06:21Z", "seed": 1738440636, "treatment": "frontier"}, "priority": 0.2518472577545528, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d0e30e1d07807ff196d219361a0001c25d542a97fc4bc40c6e9f373cb0bf9b13", "run_seconds": 0.8492834568023682, "seed": 1738440636, "started_at": "2026-11-22T04:06:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.030548, 0.047214], "effect": 0.008333, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-90fc148c5a83", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 0.983333333, 0.983333333, 1.0], "goal": [0.933333333, 0.983333333, 0.966666667, 0.9], "hypdisc": [1.0, 0.9, 1.0, 1.0], "infogain": [1.0, 0.983333333, 0.983333333, 1.0], "prederr": [1.0, 0.983333333, 0.983333333, 1.0], "random": [0.933333333, 1.0, 1.0, 1.0], "risk": [1.0, 0.983333333, 0.983333333, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "56bebbab0fa4e405be6e9422d355c01d7e1caa698902d30663924eb1cb23b0be", "primary_control": "random", "result_hash": "d0e30e1d07807ff196d219361a0001c25d542a97fc4bc40c6e9f373cb0bf9b13", "signature": "c312f4263c754d88a7576c583c21261a78dcd2a529eaac78d10c06dc3d086c1ff9665f6bac665a07578866ce68ff0fbf94f6432f8839afe550d23c57ef79bc00", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.991667, "goal": 0.945833, "hypdisc": 0.975, "infogain": 0.991667, "prederr": 0.991667, "random": 0.983333, "risk": 0.991667}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:24Z", "cpu_seconds": 0.5784095939999929, "created_at": "2026-11-22T04:06:21Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c3f93c3ba450", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "9909f86ae731a62090cccd9cc3c94397c6cefd625ff6435322a2a7c5aa69e003", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c3f93c3ba450", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:06:21Z", "seed": 1864507679, "treatment": "frontier"}, "priority": 0.25580940467070806, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "712e82fe89f64ef5c092d5723ed6c3f390e28741b90422714b16bb65dd29a0b8", "run_seconds": 0.7351105213165283, "seed": 1864507679, "started_at": "2026-11-22T04:06:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.010768, 0.062149], "effect": 0.036458, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c3f93c3ba450", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.854166667, 0.9375, 0.916666667, 1.0], "frontier": [0.9375, 0.875, 0.958333333, 0.979166667], "goal": [0.729166667, 0.791666667, 0.895833333, 0.979166667], "hypdisc": [0.854166667, 0.833333333, 0.916666667, 1.0], "infogain": [0.916666667, 0.8125, 0.979166667, 0.979166667], "prederr": [0.895833333, 0.770833333, 0.916666667, 0.979166667], "random": [0.895833333, 0.8125, 0.916666667, 0.979166667], "risk": [0.9375, 0.875, 0.958333333, 0.979166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "9909f86ae731a62090cccd9cc3c94397c6cefd625ff6435322a2a7c5aa69e003", "primary_control": "random", "result_hash": "712e82fe89f64ef5c092d5723ed6c3f390e28741b90422714b16bb65dd29a0b8", "signature": "c6c7447ea8a2a92bb7bc547bcfab507105833da73e291bb6950a95a7397b67ef30a86dae1f54ba47577ea78b15887ff58511e43ddfb0dd6d579f04eba747c705", "state": "VERIFIED", "summary": {"count": 0.927083, "frontier": 0.9375, "goal": 0.848958, "hypdisc": 0.901042, "infogain": 0.921875, "prederr": 0.890625, "random": 0.901042, "risk": 0.9375}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:30Z", "cpu_seconds": 0.7826753249999996, "created_at": "2026-11-22T04:06:21Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2fd5b3afc0b5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "d5069c5d897dde299998b593bb6197dd58c23fc1a85b3c620bb0596513e4e215", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2fd5b3afc0b5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:06:21Z", "seed": 271819270, "treatment": "full"}, "priority": 0.252259072421736, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "353adcb0b35eb2099600dd1886a764dbfc2164e3148e1f551dc3e9c5b8074fa1", "run_seconds": 1.1399273872375488, "seed": 271819270, "started_at": "2026-11-22T04:06:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [4.8e-05, 0.127028], "effect": 0.063538, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2fd5b3afc0b5", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.779411765, 0.75, 0.368421053, 0.8125], "frozen": [0.264705882, 0.358695652, 0.276315789, 0.234375], "full": [0.808823529, 0.739130435, 0.486842105, 0.9296875], "retrieval": [0.75, 0.760869565, 0.486842105, 0.90625], "rules_nodetect": [0.308823529, 0.663043478, 0.421052632, 0.390625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "d5069c5d897dde299998b593bb6197dd58c23fc1a85b3c620bb0596513e4e215", "primary_control": "facts", "result_hash": "353adcb0b35eb2099600dd1886a764dbfc2164e3148e1f551dc3e9c5b8074fa1", "signature": "4ecb239572a731b3657216e853001a71726e31a2d334619d1e0171138764e7cd4a0e6e8c8d4e8e9b42ddd6685ea44ec21ad405919f902cd1c455492afc5a2f08", "state": "VERIFIED", "summary": {"facts": 0.677583, "frozen": 0.283523, "full": 0.741121, "retrieval": 0.72599, "rules_nodetect": 0.445886}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:36Z", "cpu_seconds": 0.8669973519999985, "created_at": "2026-11-22T04:06:21Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-bddff363a977", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "27db7dff3b6ab4adfbd6ab0650f1849c59ee3e2a79c5b5265e3db9ba73a35a86", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bddff363a977", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:06:21Z", "seed": 1374619381, "treatment": "full"}, "priority": 0.2510273894478454, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a3d03b03bcab2dc5816388293f11f93b654cb005e5d79ed59146c137abcb92ef", "run_seconds": 0.9418659210205078, "seed": 1374619381, "started_at": "2026-11-22T04:06:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023871, 0.073132], "effect": 0.02463, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-bddff363a977", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.903846154, 0.921875, 0.708333333, 0.659090909], "frozen": [0.269230769, 0.3125, 0.291666667, 0.340909091], "full": [0.875, 0.9375, 0.729166667, 0.75], "retrieval": [0.913461538, 0.96875, 0.8125, 0.840909091], "rules_nodetect": [0.875, 0.9375, 0.729166667, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "27db7dff3b6ab4adfbd6ab0650f1849c59ee3e2a79c5b5265e3db9ba73a35a86", "primary_control": "facts", "result_hash": "a3d03b03bcab2dc5816388293f11f93b654cb005e5d79ed59146c137abcb92ef", "signature": "a1cb0526f80980c105e49a1f552bb82dd3b0c5fb2edc94631b5dc93608bcddd80e4d1c2ede8bad5935f5497f70d7c738c28569c49d2e768b6af001c8b5e94f00", "state": "VERIFIED", "summary": {"facts": 0.798286, "frozen": 0.303577, "full": 0.822917, "retrieval": 0.883905, "rules_nodetect": 0.822917}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:26Z", "cpu_seconds": 0.3536259830000006, "created_at": "2026-11-22T04:06:21Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c1ab46da71cb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "5f501ed5f5c62cc82653bebb1acc8fa92fac137d68b0e355c35dc1a5c50e950d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c1ab46da71cb", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:06:21Z", "seed": 904683034, "treatment": "full"}, "priority": 0.252543900705658, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cf18960f11765bb08e17a9844baf0d8016ebcb6763767a55207640211a5b8d6d", "run_seconds": 0.3617367744445801, "seed": 904683034, "started_at": "2026-11-22T04:06:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.045397, 0.162064], "effect": 0.058333, "evaluated_at": "2026-11-22T04:07:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c1ab46da71cb", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.983333333, 1.0, 0.683333333, 1.0], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.9, 1.0], "retrieval": [1.0, 1.0, 0.9, 1.0], "rules_nodetect": [1.0, 1.0, 0.9, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "5f501ed5f5c62cc82653bebb1acc8fa92fac137d68b0e355c35dc1a5c50e950d", "primary_control": "facts", "result_hash": "cf18960f11765bb08e17a9844baf0d8016ebcb6763767a55207640211a5b8d6d", "signature": "4e58b293d5739a45d32f65ef153cf0eadad1ee6d5bec724b7b86cf2b4714f294d357f457e1f9c3f70523a70859bb6225ae7e6daff1d558f0a01e68b6f4ded209", "state": "VERIFIED", "summary": {"facts": 0.916667, "frozen": 0.0, "full": 0.975, "retrieval": 0.975, "rules_nodetect": 0.975}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:50Z", "cpu_seconds": 0.7807668510000028, "created_at": "2026-11-22T04:06:37Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-18357fd14eb0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "359d62d78c8f6cf55b50db25a897b2567f273cbf4db1bb0aaab12ba7300b19cd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-18357fd14eb0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:06:37Z", "seed": 1617075668, "treatment": "frontier"}, "priority": 0.25187409621833456, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "0050178426d2810aaf6c0a084944f552f337959bc059835a33d3b8d8771188a1", "run_seconds": 1.3324863910675049, "seed": 1617075668, "started_at": "2026-11-22T04:06:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.010667, 0.076], "effect": 0.043333, "evaluated_at": "2026-11-22T04:07:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-18357fd14eb0", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.986666667, 0.973333333, 1.0, 1.0], "goal": [0.946666667, 0.76, 0.92, 0.973333333], "hypdisc": [0.946666667, 0.946666667, 0.946666667, 1.0], "infogain": [0.986666667, 0.973333333, 1.0, 1.0], "prederr": [0.986666667, 0.973333333, 1.0, 1.0], "random": [0.933333333, 0.933333333, 0.92, 1.0], "risk": [0.986666667, 0.973333333, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "359d62d78c8f6cf55b50db25a897b2567f273cbf4db1bb0aaab12ba7300b19cd", "primary_control": "random", "result_hash": "0050178426d2810aaf6c0a084944f552f337959bc059835a33d3b8d8771188a1", "signature": "1e20ca9ea84c63a1a0ee653830083386163c750bff31a929cefe8ab281270d22c69bbd356f7aedf37d7ba39aa0b4c1268f4474fc0f74cc605c834d795235a505", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.99, "goal": 0.9, "hypdisc": 0.96, "infogain": 0.99, "prederr": 0.99, "random": 0.946667, "risk": 0.99}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:07:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:06Z", "cpu_seconds": 0.25055957200000023, "created_at": "2026-11-22T04:06:37Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0517e76c3188", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "937af8e060c34c200a2763cee38a8f123ae30de7e508178308cf1b39ed96594c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0517e76c3188", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:06:37Z", "seed": 1318932830, "treatment": "full"}, "priority": 0.24832725480829548, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4ba30d4ef9554c3105a1624237124f3dd356cf1fc94feb32c4c92eefb16c3dca", "run_seconds": 0.320284366607666, "seed": 1318932830, "started_at": "2026-11-22T04:07:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.12, 0.37], "effect": 0.125, "evaluated_at": "2026-11-22T04:07:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0517e76c3188", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.0], "fixed_policy": [0.25, 0.5, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 0.5], "no_world_model": [1.0, 1.0, 1.0, 0.5], "random": [0.75, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "937af8e060c34c200a2763cee38a8f123ae30de7e508178308cf1b39ed96594c", "primary_control": "e141_best", "result_hash": "4ba30d4ef9554c3105a1624237124f3dd356cf1fc94feb32c4c92eefb16c3dca", "signature": "360622d2e91c679e8e9f963861914edbd3f8d9807fde08b1103242d044a2be3ae720a02e310c8121de3739f30d492aba16a56f346ccb76a184db353f642abb02", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.25, "full": 0.875, "no_world_model": 0.875, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:00Z", "cpu_seconds": 10.946322261000006, "created_at": "2026-11-22T04:06:37Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1b013d18dd05", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "6b7574d9a6b4dcbe702b9df6825116248af797b8bdd6e5a9e5018ae674801d22", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1b013d18dd05", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:06:37Z", "seed": 2100335463, "treatment": "full"}, "priority": 0.25210605336291114, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b4965fac516ba73c37074f4d0c0debb12b28b5c51b4f1fed093d4f9145b1d259", "run_seconds": 13.028197050094604, "seed": 2100335463, "started_at": "2026-11-22T04:06:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.526278, 0.401278], "effect": -0.0625, "evaluated_at": "2026-11-22T04:07:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1b013d18dd05", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 0.5, 0.0], "fixed_policy": [0.5, 0.0, 0.5, 0.0], "full": [0.25, 0.75, 0.75, 0.0], "no_world_model": [0.75, 0.5, 0.75, 0.0], "random": [1.0, 0.5, 0.75, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6b7574d9a6b4dcbe702b9df6825116248af797b8bdd6e5a9e5018ae674801d22", "primary_control": "e141_best", "result_hash": "b4965fac516ba73c37074f4d0c0debb12b28b5c51b4f1fed093d4f9145b1d259", "signature": "056f737f1de1f6aa245a7326be3562ca194efc6c7c77ae1dd8d587cd2db4b99d03ae4e0a30fe0850828d8b2c2b21ff5b5b77f2507a2d0beedf5c54b1436ddf0b", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.25, "full": 0.4375, "no_world_model": 0.5, "random": 0.5625}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:44Z", "cpu_seconds": 0.7001661870000007, "created_at": "2026-11-22T04:06:37Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2cd81cb31021", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "fb43c643034b04b92744777d7746ea5aa07762bfca970d09e0d3ca646ee6dd13", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2cd81cb31021", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:06:37Z", "seed": 175303580, "treatment": "full"}, "priority": 0.25214851805333305, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1e1fbf60708159a9882ac3f8510a86bd4fec04da14a052551a3e28aadcab7cbb", "run_seconds": 0.7323827743530273, "seed": 175303580, "started_at": "2026-11-22T04:06:43Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:07:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2cd81cb31021", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.75], "fixed_policy": [1.0, 0.0], "full": [1.0, 0.9375], "no_world_model": [1.0, 0.75], "random": [1.0, 0.75]}, "prereg_hash": "fb43c643034b04b92744777d7746ea5aa07762bfca970d09e0d3ca646ee6dd13", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "1e1fbf60708159a9882ac3f8510a86bd4fec04da14a052551a3e28aadcab7cbb", "signature": "fe1c57b52e7324e7e54ce4f86d0f09865fbda99189ebb66bc9f726e4ad7f78e3a335100586368fe38985901cb71c3242c05ce84b15544f2583f35eeef51f250f", "state": "REJECTED", "summary": {"e141_best": 0.75, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.75, "random": 0.75}}, "verified_at": "2026-11-22T04:07:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:02Z", "cpu_seconds": 0.3419386099999997, "created_at": "2026-11-22T04:06:37Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3b0d740dc788", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "c4b9eefb2f77d746133a9187ddbb46305f6686cac7f0c0dfbe0a6ac71203c74d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3b0d740dc788", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:06:37Z", "seed": 1094974691, "treatment": "full"}, "priority": 0.2516220911944098, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-c9cdda06711d", "result_hash": "39387c72c854df1ccb18ca42dfeac55340af65fdc70de8d324c65279e00fdf2c", "run_seconds": 0.3601686954498291, "seed": 1094974691, "started_at": "2026-11-22T04:07:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003138, 0.028138], "effect": 0.0125, "evaluated_at": "2026-11-22T04:07:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3b0d740dc788", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.966666667, 1.0, 1.0, 0.966666667], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.983333333, 1.0, 1.0, 1.0], "retrieval": [0.983333333, 1.0, 1.0, 0.966666667], "rules_nodetect": [0.983333333, 1.0, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "c4b9eefb2f77d746133a9187ddbb46305f6686cac7f0c0dfbe0a6ac71203c74d", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-c9cdda06711d", "result_hash": "39387c72c854df1ccb18ca42dfeac55340af65fdc70de8d324c65279e00fdf2c", "signature": "5d8a7a66154f8dc59060f7e35da1bffc4a52334883cb5f639744c9b38db1913ddd10703564d9bb1fb9d9eb4fe39d55afdda99d7d53e9880bfa81e0c2ab1fed0e", "state": "VERIFIED", "summary": {"facts": 0.983333, "frozen": 0.0, "full": 0.995833, "retrieval": 0.9875, "rules_nodetect": 0.995833}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:42Z", "cpu_seconds": 1.0441760900000077, "created_at": "2026-11-22T04:06:37Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8a97376a3b71", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "dafc3e060c16d143fc3e35beb24a212d360b7662d2d9e8e3a1621e9e47175c74", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8a97376a3b71", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:06:37Z", "seed": 914205947, "treatment": "gated"}, "priority": 0.25231910332402646, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6d065317b3abb8b87a6b6f5320c9578f51ecd57a34182aeae69bd8e6be224d20", "run_seconds": 1.057234287261963, "seed": 914205947, "started_at": "2026-11-22T04:06:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.008633, 0.0028], "effect": -0.002917, "evaluated_at": "2026-11-22T04:07:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-8a97376a3b71", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.499122807, 0.549166667, 0.568333333, 0.607017544], "gated": [0.499122807, 0.549166667, 0.556666667, 0.607017544], "naive": [0.335964912, 0.64, 0.566666667, 0.415789474], "scratch": [0.499122807, 0.549166667, 0.568333333, 0.607017544], "ungated": [0.511403509, 0.653333333, 0.5175, 0.522807018]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "dafc3e060c16d143fc3e35beb24a212d360b7662d2d9e8e3a1621e9e47175c74", "primary_control": "scratch", "result_hash": "6d065317b3abb8b87a6b6f5320c9578f51ecd57a34182aeae69bd8e6be224d20", "signature": "9d2d3b207713fd016e16b92e046925e80c336318b9728712f564dbaf495d90dcb49bad745e4650211c6fd3f344757ac01309d008108f39857f53eb13a7170f02", "state": "VERIFIED", "summary": {"compressed": 0.55591, "gated": 0.552993, "naive": 0.489605, "scratch": 0.55591, "ungated": 0.551261}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:07:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:08Z", "cpu_seconds": 1.0000530140000023, "created_at": "2026-11-22T04:06:52Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b244a3422592", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "a385ef8eea4f09505f28fff363152fe65bfb2e547e3d262f173b0a97f390ca94", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b244a3422592", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:06:52Z", "seed": 952442436, "treatment": "frontier"}, "priority": 0.2481125540012569, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d573e8ae5c21d4c51c54151cc38f0b4e796377453232a0ffe9779ac490bd6891", "run_seconds": 1.2483887672424316, "seed": 952442436, "started_at": "2026-11-22T04:07:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.015813, 0.116999], "effect": 0.066406, "evaluated_at": "2026-11-22T04:07:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b244a3422592", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 0.975, 1.0], "goal": [1.0, 0.9, 0.95, 1.0], "hypdisc": [1.0, 0.9, 0.975, 1.0], "infogain": [1.0, 1.0, 0.975, 1.0], "prederr": [1.0, 1.0, 0.975, 1.0], "random": [0.9375, 0.875, 0.975, 0.921875], "risk": [1.0, 1.0, 0.975, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "a385ef8eea4f09505f28fff363152fe65bfb2e547e3d262f173b0a97f390ca94", "primary_control": "random", "result_hash": "d573e8ae5c21d4c51c54151cc38f0b4e796377453232a0ffe9779ac490bd6891", "signature": "c5f269b5b4c331395c88ea1363cb70436f4800b98922402cfe8ae459308522cb75ed9fcf03cd76443d13ffe5ac29ee933dde5ccc09cfc373bb0ef018fe03b20e", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.99375, "goal": 0.9625, "hypdisc": 0.96875, "infogain": 0.99375, "prederr": 0.99375, "random": 0.927344, "risk": 0.99375}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:07:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:15Z", "cpu_seconds": 1.427889497999999, "created_at": "2026-11-22T04:06:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-df1b942c0604", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "1a84da11d20da242e047260ac59e2afa405929fee522ec5c92e50d48dd3997de", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-df1b942c0604", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:06:52Z", "seed": 1195045352, "treatment": "full"}, "priority": 0.2478019244852431, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f2b3b6e244db9c5da519c20167dd2b3cf96418a807bb535152cd9bc733372b90", "run_seconds": 1.655174970626831, "seed": 1195045352, "started_at": "2026-11-22T04:07:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.526278, 0.401278], "effect": -0.0625, "evaluated_at": "2026-11-22T04:07:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-df1b942c0604", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.0, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.75, 0.25, 1.0, 0.0], "no_world_model": [0.25, 0.0, 1.0, 0.0], "random": [0.25, 0.0, 1.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "1a84da11d20da242e047260ac59e2afa405929fee522ec5c92e50d48dd3997de", "primary_control": "e141_best", "result_hash": "f2b3b6e244db9c5da519c20167dd2b3cf96418a807bb535152cd9bc733372b90", "signature": "d131a54c11db4e3b06965e73b43b518ce2e1bfffa0c6d3ac2bd94b5114e46e6b58c09b22825aa64c3ea07f9a041f00aa1a93441d78bde20c30d428971067a90f", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.0, "full": 0.5, "no_world_model": 0.3125, "random": 0.375}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:07:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:06:56Z", "cpu_seconds": 1.0510443519999981, "created_at": "2026-11-22T04:06:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-47e998d438d9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "abac61cc414d3a88d7b1a86ea88002ca39fe4ee543a5392f0104f70fce1c3148", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-47e998d438d9", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:06:52Z", "seed": 1285002105, "treatment": "full"}, "priority": 0.25467868508559943, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7c1cfce65d7872d2bf80d9d2e46e2580f0dda4eed15675d51e26fd9f8d3fb964", "run_seconds": 1.2272984981536865, "seed": 1285002105, "started_at": "2026-11-22T04:06:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.009168, 0.062213], "effect": 0.035691, "evaluated_at": "2026-11-22T04:07:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-47e998d438d9", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.5, 0.85, 1.0, 0.815789474], "frozen": [0.289473684, 0.2375, 0.307692308, 0.263157895], "full": [0.565789474, 0.8875, 1.0, 0.855263158], "retrieval": [0.736842105, 0.9, 1.0, 0.815789474], "rules_nodetect": [0.565789474, 0.8875, 1.0, 0.855263158]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "abac61cc414d3a88d7b1a86ea88002ca39fe4ee543a5392f0104f70fce1c3148", "primary_control": "facts", "result_hash": "7c1cfce65d7872d2bf80d9d2e46e2580f0dda4eed15675d51e26fd9f8d3fb964", "signature": "15db9570046f2a6310963f16d0a5368dc8fa95a969583e99383cf70456d5091c1dacbfc196e3e11e852cedbcaec2df51794e321be715d56b834a62bbe7919900", "state": "VERIFIED", "summary": {"facts": 0.791447, "frozen": 0.274456, "full": 0.827138, "retrieval": 0.863158, "rules_nodetect": 0.827138}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:21Z", "cpu_seconds": 0.9365504980000026, "created_at": "2026-11-22T04:06:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a9cf309fafaf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|crafting|crafting|1|1"}, "prereg_hash": "7aa352686a874f94b7e3d3ab15de1334468ef7ede1c2ec5eb2d834a541233799", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a9cf309fafaf", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|crafting|crafting|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:06:52Z", "seed": 1047947012, "treatment": "gated"}, "priority": 0.24731839182763907, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "62bfc625392f0606d50d0b0b2cb9aae505eab105010a77dfc3e2449a76863c3f", "run_seconds": 1.511782169342041, "seed": 1047947012, "started_at": "2026-11-22T04:07:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035953, 0.110855], "effect": 0.037451, "evaluated_at": "2026-11-22T04:07:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a9cf309fafaf", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.550588235, 0.385882353, 0.592941176, 0.574117647], "gated": [0.550588235, 0.535686275, 0.592941176, 0.574117647], "naive": [0.461176471, 0.601568627, 0.644705882, 0.662745098], "scratch": [0.550588235, 0.385882353, 0.592941176, 0.574117647], "ungated": [0.534901961, 0.589019608, 0.471372549, 0.613333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7aa352686a874f94b7e3d3ab15de1334468ef7ede1c2ec5eb2d834a541233799", "primary_control": "scratch", "result_hash": "62bfc625392f0606d50d0b0b2cb9aae505eab105010a77dfc3e2449a76863c3f", "signature": "0f5a1ff955f74905ec91253eb6d3b5c058c5638b66de5b5006e13e76bc0cdad4c0e820fd032b3f8213e2b013bc541d9a63852506118e0a1d6b684b902a298506", "state": "VERIFIED", "summary": {"compressed": 0.525882, "gated": 0.563333, "naive": 0.592549, "scratch": 0.525882, "ungated": 0.552157}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:07:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:28Z", "cpu_seconds": 0.7041806119999947, "created_at": "2026-11-22T04:07:09Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-872ffb2330b1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "2f4b3bba733749db22b297eb10bf07d7928a113231c4984542eb054752a89d70", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-872ffb2330b1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:07:09Z", "seed": 1757998454, "treatment": "frontier"}, "priority": 0.24028576650675942, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "cbc0710d2d596b6f7ebedc5fd7cd902982ee9edd56c9b90ec5f5a1a2e86901dc", "run_seconds": 0.7624309062957764, "seed": 1757998454, "started_at": "2026-11-22T04:10:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06985, 0.074016], "effect": 0.002083, "evaluated_at": "2026-11-22T04:10:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-872ffb2330b1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.916666667, 1.0, 0.9, 0.854166667], "frontier": [0.783333333, 1.0, 0.666666667, 1.0], "goal": [0.8, 1.0, 1.0, 0.916666667], "hypdisc": [0.866666667, 1.0, 1.0, 0.9375], "infogain": [0.816666667, 1.0, 0.666666667, 0.979166667], "prederr": [0.083333333, 1.0, 0.733333333, 0.916666667], "random": [0.716666667, 1.0, 0.766666667, 0.958333333], "risk": [0.783333333, 1.0, 0.666666667, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "2f4b3bba733749db22b297eb10bf07d7928a113231c4984542eb054752a89d70", "primary_control": "random", "result_hash": "cbc0710d2d596b6f7ebedc5fd7cd902982ee9edd56c9b90ec5f5a1a2e86901dc", "signature": "f2275cc1138b6a80a83c7fbcf160645bdd752976e8d44d25b8a08108dfc7018cf18dbe506740d486ad894eb7f3a7713b1981af01c042abfb3260084a7ab33906", "state": "VERIFIED", "summary": {"count": 0.917708, "frontier": 0.8625, "goal": 0.929167, "hypdisc": 0.951042, "infogain": 0.865625, "prederr": 0.683333, "random": 0.860417, "risk": 0.8625}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:10:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:18Z", "cpu_seconds": 0.7824161959999998, "created_at": "2026-11-22T04:07:09Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1ee88322dfdd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "1c3faeae4272f07663d7742910a6fdba5643c4a6a0fcf859bfa2d3dfc7b5dce6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1ee88322dfdd", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:07:09Z", "seed": 1971008637, "treatment": "full"}, "priority": 0.2475695189243013, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "dd8dda31ff6a88e98dd15a615a61259bdca71e7873a476c5b2b9fbaf0887cc18", "run_seconds": 1.0778210163116455, "seed": 1971008637, "started_at": "2026-11-22T04:07:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.200648, 1.049352], "effect": 0.625, "evaluated_at": "2026-11-22T04:07:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1ee88322dfdd", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.0, 0.75, 0.0], "fixed_policy": [0.5, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "1c3faeae4272f07663d7742910a6fdba5643c4a6a0fcf859bfa2d3dfc7b5dce6", "primary_control": "e141_best", "result_hash": "dd8dda31ff6a88e98dd15a615a61259bdca71e7873a476c5b2b9fbaf0887cc18", "signature": "c7afef952069b8930779b729245ca6f0bd435e9a2477a9d9001604e08420a13da46eb3d76800b4fe460e084f8cddfcf671788ebf0562e235b77ea1b5a728a605", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.125, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:07:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:12Z", "cpu_seconds": 0.6796871290000013, "created_at": "2026-11-22T04:07:09Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4bcb8b4d7783", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "c62ba4b48fe8c973183cfc0b59e614bee2be6b221c8759a4a9bd9d245c570c09", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4bcb8b4d7783", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:07:09Z", "seed": 793628244, "treatment": "full"}, "priority": 0.25277049916412553, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2c912d70f99c05a278aabc92b18923a2971b01073d0fb842c349f00a6743f2f1", "run_seconds": 0.8087294101715088, "seed": 793628244, "started_at": "2026-11-22T04:07:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T04:07:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4bcb8b4d7783", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.75, 0.75, 1.0, 1.0], "no_world_model": [0.5, 0.75, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c62ba4b48fe8c973183cfc0b59e614bee2be6b221c8759a4a9bd9d245c570c09", "primary_control": "e141_best", "result_hash": "2c912d70f99c05a278aabc92b18923a2971b01073d0fb842c349f00a6743f2f1", "signature": "0c9348d3c869e1823f6b94c0a86a8fbab5d23843b26c2c3c2444f4825bed53f9a10f101f8dabb09c4de7833c9e894f6ac21697700f2894dedc093d5ee8942c07", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 0.875, "no_world_model": 0.8125, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:32Z", "cpu_seconds": 0.8256157529999992, "created_at": "2026-11-22T04:07:09Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d36229399f5d", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "c36e55421a7c76febc110b57bf13cdb88ca2b202d8562e0dd3a4b794c4445903", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d36229399f5d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:07:09Z", "seed": 1496902039, "treatment": "full"}, "priority": 0.24246761607690637, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "058b764069800c67d6fe1674f6bb3b2bd263661505c5e2c12567706a9de4578c", "run_seconds": 0.8398709297180176, "seed": 1496902039, "started_at": "2026-11-22T04:08:31Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:08:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d36229399f5d", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.522981], "frozen": [1.0, 0.390959], "full": [1.0, 0.58778], "retrieval": [1.0, 0.836523], "rules_nodetect": [1.0, 0.58778]}, "prereg_hash": "c36e55421a7c76febc110b57bf13cdb88ca2b202d8562e0dd3a4b794c4445903", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "058b764069800c67d6fe1674f6bb3b2bd263661505c5e2c12567706a9de4578c", "signature": "e857999457350a4b8b89c12018f0dbdc46e4863b025c1c676bbb9d78b14f1e071ed478a403ec343f0545728b1a3ba28e4b0e995c558e82d4573762c316305b09", "state": "REJECTED", "summary": {"facts": 0.522981, "frozen": 0.390959, "full": 0.58778, "retrieval": 0.836523, "rules_nodetect": 0.58778}}, "verified_at": "2026-11-22T04:08:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:27Z", "cpu_seconds": 0.8316335719999941, "created_at": "2026-11-22T04:07:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7aa2d649c0ad", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "ab1e3565a33ab17c6b1464ccc2dc167021893bfbf03d41e2015c16260eb1f7ed", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7aa2d649c0ad", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:07:26Z", "seed": 1268978760, "treatment": "frontier"}, "priority": 0.4166638315530862, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-0be5c77c883a", "result_hash": "be1016eb9e0235994a5bc80a182d3e26ee6f0c5653fe5f92fc9e78e0d193cbf5", "run_seconds": 1.2143199443817139, "seed": 1268978760, "started_at": "2026-11-22T04:07:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024274, 0.099782], "effect": 0.037754, "evaluated_at": "2026-11-22T04:07:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7aa2d649c0ad", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.986928105, 0.568627451, 0.67251462, 0.941176471], "frontier": [0.575163399, 0.764705882, 0.695906433, 0.901960784], "goal": [0.575163399, 0.764705882, 0.701754386, 0.941176471], "hypdisc": [0.934640523, 0.68627451, 0.701754386, 0.803921569], "infogain": [0.764705882, 0.470588235, 0.701754386, 0.921568627], "prederr": [0.882352941, 0.745098039, 0.736842105, 0.901960784], "random": [0.516339869, 0.647058824, 0.701754386, 0.921568627], "risk": [0.575163399, 0.764705882, 0.695906433, 0.901960784]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "ab1e3565a33ab17c6b1464ccc2dc167021893bfbf03d41e2015c16260eb1f7ed", "primary_control": "random", "replicated": false, "replication_of": "E141-ow_explore-0be5c77c883a", "result_hash": "be1016eb9e0235994a5bc80a182d3e26ee6f0c5653fe5f92fc9e78e0d193cbf5", "signature": "5797c7ff4fdc3ac9e9600834aec149be585ba9e5d6699ac3da93373fad9d91e39c8cba2df38bfa3b7e69506ae689504a3dad1faaf592600df1297a806d3efd00", "state": "VERIFIED", "summary": {"count": 0.792312, "frontier": 0.734434, "goal": 0.7457, "hypdisc": 0.781648, "infogain": 0.714654, "prederr": 0.816563, "random": 0.69668, "risk": 0.734434}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:07:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:32Z", "cpu_seconds": 0.671287575000008, "created_at": "2026-11-22T04:07:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f03e6e9d0197", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "084d8c5ccd79c58a31dcaaa8c48f74f3eaf61cadcf5c7ffa3ae91fd2ad771181", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f03e6e9d0197", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:07:26Z", "seed": 630878032, "treatment": "frontier"}, "priority": 0.37365112987754323, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6a99f609bf36718a0b0db3dd32ba02efa01fd3c57c331eaf562e7e1c941d6173", "run_seconds": 0.7112433910369873, "seed": 630878032, "started_at": "2026-11-22T04:07:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.060318, 0.076985], "effect": 0.008333, "evaluated_at": "2026-11-22T04:07:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f03e6e9d0197", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 0.9, 1.0], "goal": [0.933333333, 0.966666667, 0.95, 0.983333333], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 0.9, 1.0], "prederr": [1.0, 1.0, 0.9, 1.0], "random": [0.966666667, 0.916666667, 0.983333333, 1.0], "risk": [1.0, 1.0, 0.9, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "084d8c5ccd79c58a31dcaaa8c48f74f3eaf61cadcf5c7ffa3ae91fd2ad771181", "primary_control": "random", "result_hash": "6a99f609bf36718a0b0db3dd32ba02efa01fd3c57c331eaf562e7e1c941d6173", "signature": "bbed80fb7b673a300afeefd5564e46bdd6580914686755626a2756f1a754d34a3157ac4f5a33f9f9a183ebb4e0a33734402f74ae1533e2f96c25ec0d51df6f0a", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.975, "goal": 0.958333, "hypdisc": 1.0, "infogain": 0.975, "prederr": 0.975, "random": 0.966667, "risk": 0.975}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:07:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:43Z", "cpu_seconds": 0.8973121289999995, "created_at": "2026-11-22T04:07:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8bdec6f19e5f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "47ab77c6dc7d35ed6889516c5ab5f6b74d58f7c07abadf7ea5a36f19b8352874", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8bdec6f19e5f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:07:26Z", "seed": 1400354968, "treatment": "full"}, "priority": 0.24959539031784422, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8d11a12be22c6dca293fc353a3cc486004516fd7a8a4561872d3eeaec1058345", "run_seconds": 1.7663013935089111, "seed": 1400354968, "started_at": "2026-11-22T04:07:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:07:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8bdec6f19e5f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 1.0], "fixed_policy": [0.25, 0.25, 0.25, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.5, 1.0], "random": [0.5, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "47ab77c6dc7d35ed6889516c5ab5f6b74d58f7c07abadf7ea5a36f19b8352874", "primary_control": "e141_best", "result_hash": "8d11a12be22c6dca293fc353a3cc486004516fd7a8a4561872d3eeaec1058345", "signature": "2614b42ef16aaf263e9cde7bf62148af32274d85289679ca378000d2d1550cccfeae9e00f352907e92a5898522699841679e9bc25c3e8d94437b133042b75f02", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 0.875, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:07:56Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:43Z", "cpu_seconds": 2.215323965, "created_at": "2026-11-22T04:07:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-be0cd2844d2b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "0e108ca4bbc3458e695212ae1d755aa56989de957306fb8002dca1af5bb5bc6f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-be0cd2844d2b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:07:26Z", "seed": 832694376, "treatment": "full"}, "priority": 0.2632343172800576, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a45d54773b7c737322c055b56c45baf7ffdf6baf0c2a10971c78887ce7574b29", "run_seconds": 4.089068651199341, "seed": 832694376, "started_at": "2026-11-22T04:07:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.020475, 0.280257], "effect": 0.129891, "evaluated_at": "2026-11-22T04:07:56Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-be0cd2844d2b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.410769231, 0.728888889, 0.503968254, 0.475260417], "frozen": [0.0, 0.296666667, 0.269047619, 0.134114583], "full": [0.410769231, 0.896666667, 0.834920635, 0.49609375], "retrieval": [0.856923077, 0.956666667, 0.877777778, 0.630208333], "rules_nodetect": [0.412307692, 0.068888889, 0.353968254, 0.6875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "0e108ca4bbc3458e695212ae1d755aa56989de957306fb8002dca1af5bb5bc6f", "primary_control": "facts", "result_hash": "a45d54773b7c737322c055b56c45baf7ffdf6baf0c2a10971c78887ce7574b29", "signature": "a5560260ce6072e907226fb7ed783bb9235995c8bb9766958fd0cfd479fe6a126b0aa08529de2606d0a6fd805baac053173e796a829b699a8462506d7334ed03", "state": "VERIFIED", "summary": {"facts": 0.529722, "frozen": 0.174957, "full": 0.659613, "retrieval": 0.830394, "rules_nodetect": 0.380666}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:00Z", "cpu_seconds": 19.044440457000064, "created_at": "2026-11-22T04:07:26Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-499e318628da", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "660c183da9e61f5c4ef35e39da6f6cc5a0c706bf38a29e7e1a6b8a0c1a938c5a", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-499e318628da", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:07:26Z", "seed": 1213441732, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "4e70e941e7bca1f4751ab34a15cf480c9c1db2cd6fa0d9b6fcb715d7011b0049", "run_seconds": 20.513302087783813, "seed": 1213441732, "started_at": "2026-11-22T04:11:40Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 3, "inherit_mem": 3, "inherit_nomem": 1, "random_search": 1, "reset_mem": 2, "retrieval": 0, "rootcause": 3, "xfer_mem": 3}, "better": "higher", "ci95": [-0.03728, 0.194289], "effect": 0.078505, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:15:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-499e318628da", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.45, "history2": false, "mix": {"hypdisc": 0.8, "random": 1.0}, "planner_depth": 20, "reset_patience": 0, "rule_threshold": 0.75}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"hypdisc": 0.6, "random": 1.0}, "planner_depth": 18, "reset_patience": 20, "rule_threshold": 0.75}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "random_search": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": true, "mix": {"random": 1.0}, "planner_depth": 14, "reset_patience": 20, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.45, "history2": false, "mix": {"hypdisc": 0.8, "random": 1.0}, "planner_depth": 20, "reset_patience": 0, "rule_threshold": 0.75}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"hypdisc": 0.6, "random": 1.0}, "planner_depth": 18, "reset_patience": 20, "rule_threshold": 0.75}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.919642857, 1.0, 0.375, 0.836734694, 0.373076923, 0.310176125, 0.772727273, 0.626724138, 0.375, 0.236160714], "compressed": [0.848214286, 1.0, 0.958333333, 0.863945578, 0.497115385, 0.679549902, 0.78030303, 0.787931034, 0.5, 0.327232143], "inherit_mem": [0.803571429, 1.0, 0.9375, 0.901360544, 0.390384615, 0.305039139, 0.818181818, 0.801724138, 0.333333333, 0.319196429], "inherit_nomem": [0.919642857, 1.0, 0.395833333, 0.850340136, 0.376923077, 0.310176125, 0.772727273, 0.626724138, 0.375, 0.322321429], "random_search": [0.919642857, 1.0, 0.854166667, 0.836734694, 0.371153846, 0.310176125, 0.75, 0.619827586, 0.5, 0.23125], "reset_mem": [0.839285714, 1.0, 0.395833333, 0.891156463, 0.548076923, 0.306751468, 0.727272727, 0.806896552, 0.375, 0.380357143], "retrieval": [0.919642857, 1.0, 0.375, 0.836734694, 0.373076923, 0.310176125, 0.772727273, 0.626724138, 0.375, 0.236160714], "rootcause": [0.848214286, 1.0, 0.958333333, 0.863945578, 0.497115385, 0.679549902, 0.78030303, 0.787931034, 0.5, 0.327232143], "xfer_mem": [0.803571429, 1.0, 0.9375, 0.901360544, 0.390384615, 0.305039139, 0.818181818, 0.801724138, 0.333333333, 0.319196429]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -0.5953, "losses": 3, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.254", "per_generation_sealed": [0.582524, 0.594969, 0.669779, 0.661029], "prereg_hash": "660c183da9e61f5c4ef35e39da6f6cc5a0c706bf38a29e7e1a6b8a0c1a938c5a", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "xfer_mem", "rootcause"], "result_hash": "4e70e941e7bca1f4751ab34a15cf480c9c1db2cd6fa0d9b6fcb715d7011b0049", "signature": "8852a74be2af69e51d7e6cdaa611b01594a6fe5a82f8328df20ce3d20416a2e7a9a532dcd2de1417a8bc1ae0e16a7c5a446e97f24f3be9cebc3b2352199f5207", "state": "VERIFIED", "summary": {"G0": 0.582524, "compressed": 0.724262, "inherit_mem": 0.661029, "inherit_nomem": 0.594969, "random_search": 0.639295, "reset_mem": 0.627063, "retrieval": 0.582524, "rootcause": 0.724262, "xfer_mem": 0.661029}, "ties": 1, "treatment": "inherit_mem", "wins": 6}, "verified_at": "2026-11-22T04:17:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:36Z", "cpu_seconds": 2.967711003000005, "created_at": "2026-11-22T04:07:26Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-b52688cd3082", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "2db177c22ac640b5b3944043296ea34a615c6be67b398dbeb511e4d090a2e03c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-b52688cd3082", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:07:26Z", "seed": 624819935, "treatment": "gated"}, "priority": 0.31578907639249487, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "8e942730a42c38245d744cbcd568aa8038a950f2ce3f643543821c452490dc32", "run_seconds": 3.51072359085083, "seed": 624819935, "started_at": "2026-11-22T04:07:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006852, 0.002222], "effect": -0.002315, "evaluated_at": "2026-11-22T04:07:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-b52688cd3082", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.509259259, 0.556818182, 0.79, 0.4375], "gated": [0.509259259, 0.556818182, 0.79, 0.4375], "naive": [0.666666667, 0.625, 0.79, 0.609375], "scratch": [0.518518519, 0.556818182, 0.79, 0.4375], "ungated": [0.611111111, 0.625, 0.79, 0.609375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2db177c22ac640b5b3944043296ea34a615c6be67b398dbeb511e4d090a2e03c", "primary_control": "scratch", "result_hash": "8e942730a42c38245d744cbcd568aa8038a950f2ce3f643543821c452490dc32", "signature": "37cecb83a0472d5ea8dfcb1f345b736739f63ca3dd9d8ed3d27360c1d210af22a4c2414664616a11659ad5799088c09f9e80f89a0c8bb8603576107d56bc180e", "state": "VERIFIED", "summary": {"compressed": 0.573394, "gated": 0.573394, "naive": 0.67276, "scratch": 0.575709, "ungated": 0.658872}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:07:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:48Z", "cpu_seconds": 0.4256911099999998, "created_at": "2026-11-22T04:07:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-76e94fcae7a0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "a75dbef47dd4d1def07ef22eb6fc30c30b4525198665be7577d26197d2b7f848", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-76e94fcae7a0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:07:43Z", "seed": 838637835, "treatment": "frontier"}, "priority": 0.25783125097866494, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e941e4833908353d37252ad27b263b1978d27a1f7d66df945e9ff96eb651f8d1", "run_seconds": 0.5923240184783936, "seed": 838637835, "started_at": "2026-11-22T04:07:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.015298, 0.184965], "effect": 0.100132, "evaluated_at": "2026-11-22T04:07:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-76e94fcae7a0", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.984848485, 0.984848485, 1.0, 1.0], "frontier": [0.863636364, 0.984848485, 1.0, 1.0], "goal": [0.863636364, 0.787878788, 0.96969697, 1.0], "hypdisc": [0.818181818, 0.818181818, 0.96969697, 0.913043478], "infogain": [0.863636364, 0.984848485, 1.0, 1.0], "prederr": [0.863636364, 0.984848485, 1.0, 1.0], "random": [0.878787879, 0.848484848, 0.909090909, 0.811594203], "risk": [0.863636364, 0.984848485, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "a75dbef47dd4d1def07ef22eb6fc30c30b4525198665be7577d26197d2b7f848", "primary_control": "random", "result_hash": "e941e4833908353d37252ad27b263b1978d27a1f7d66df945e9ff96eb651f8d1", "signature": "fb44d898fe69ac0e216ffd77f9a6ca3947b2be99388939c5829f2fcc3477cf7f1d44b9d0dfc5e71bcd295e03352f9c0e5fb379f22843a1a1be49c809ebccec05", "state": "VERIFIED", "summary": {"count": 0.992424, "frontier": 0.962121, "goal": 0.905303, "hypdisc": 0.879776, "infogain": 0.962121, "prederr": 0.962121, "random": 0.861989, "risk": 0.962121}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:07:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:48Z", "cpu_seconds": 0.5098421739999992, "created_at": "2026-11-22T04:07:43Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4ee3eaa12fe6", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "3785a100f65de9689787cdccc116ba48582a7f882e922051bccc05e569950f85", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4ee3eaa12fe6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:07:43Z", "seed": 1030202499, "treatment": "full"}, "priority": 0.38949101903948485, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-456b318f1ed2", "result_hash": "b1d9fde12db1b32098c3a5c15467c02d67c61e8d2d094d21700c2e11ad911c3f", "run_seconds": 0.7771928310394287, "seed": 1030202499, "started_at": "2026-11-22T04:07:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:07:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4ee3eaa12fe6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 0.5], "no_world_model": [1.0, 1.0, 1.0, 0.0], "random": [0.75, 0.0, 0.25, 0.5]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "3785a100f65de9689787cdccc116ba48582a7f882e922051bccc05e569950f85", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-456b318f1ed2", "result_hash": "b1d9fde12db1b32098c3a5c15467c02d67c61e8d2d094d21700c2e11ad911c3f", "signature": "1b0a065be8f8270d061ef6b0bcb098a3aa593ffbacaf0a38e9e6b3252921587322ee959019b515b091212439e2c192ef61eccc57e549d9be5a67f925c7508001", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 0.875, "no_world_model": 0.75, "random": 0.375}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:07:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:55Z", "cpu_seconds": 0.9488037380000005, "created_at": "2026-11-22T04:07:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-95c0a9e36f46", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "4d90c98682feb75371a6dd1f2329f1a3e44d2f243a86b356e5e635a073e28211", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-95c0a9e36f46", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:07:43Z", "seed": 2015551147, "treatment": "full"}, "priority": 0.25216560243221514, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "57d13ddc4ce021514b6e33835dd820711a9cedb4b880bfca57d1839f945531d8", "run_seconds": 1.2923939228057861, "seed": 2015551147, "started_at": "2026-11-22T04:07:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0029, 0.174832], "effect": 0.085966, "evaluated_at": "2026-11-22T04:07:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-95c0a9e36f46", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.53201581, 0.512941176, 0.539607843, 0.550588235], "frozen": [0.38972332, 0.419607843, 0.399215686, 0.436078431], "full": [0.574703557, 0.491764706, 0.701960784, 0.710588235], "retrieval": [0.843478261, 0.801568627, 0.860392157, 0.829803922], "rules_nodetect": [0.574703557, 0.491764706, 0.701960784, 0.710588235]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "4d90c98682feb75371a6dd1f2329f1a3e44d2f243a86b356e5e635a073e28211", "primary_control": "facts", "result_hash": "57d13ddc4ce021514b6e33835dd820711a9cedb4b880bfca57d1839f945531d8", "signature": "3e4be42851fcd34dc936851c3e36b1b9991216ef64b7f7aed966b77fcc3ef00a82d12056b9082c58e59314b6e1b3d10c505a6650477457b122a13692e7b4d90e", "state": "VERIFIED", "summary": {"facts": 0.533788, "frozen": 0.411156, "full": 0.619754, "retrieval": 0.833811, "rules_nodetect": 0.619754}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:08:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:07:55Z", "cpu_seconds": 1.2331318089999996, "created_at": "2026-11-22T04:07:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-fe66afdbaa3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "48beb24f2fad94ab266c7df1b680aa70c0c7f7a770698366f77708cb96cc31f8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-fe66afdbaa3b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:07:43Z", "seed": 1765928584, "treatment": "full"}, "priority": 0.25173894935557084, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "638a6cdc91ac51974b6aa9b42bbbe132a8c3ef98a7a931aac5f37446418adf75", "run_seconds": 1.9365942478179932, "seed": 1765928584, "started_at": "2026-11-22T04:07:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.002451, 0.134686], "effect": 0.068569, "evaluated_at": "2026-11-22T04:07:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-fe66afdbaa3b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.635294118, 0.819354839, 0.641176471, 0.36], "frozen": [0.035294118, 0.135483871, 0.035294118, 0.229333333], "full": [0.758823529, 0.819354839, 0.770588235, 0.381333333], "retrieval": [0.682352941, 0.819354839, 0.694117647, 0.306666667], "rules_nodetect": [0.758823529, 0.819354839, 0.770588235, 0.4]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "48beb24f2fad94ab266c7df1b680aa70c0c7f7a770698366f77708cb96cc31f8", "primary_control": "facts", "result_hash": "638a6cdc91ac51974b6aa9b42bbbe132a8c3ef98a7a931aac5f37446418adf75", "signature": "f6b399fc0e2cef92d42f7ade4950eab9fcb056fc84c8bf061f57afe1f7e89f32b54e4a3f579ae4080f33bde30e6c92cfc64fad29068f5daf7d6ff5483e9d2209", "state": "VERIFIED", "summary": {"facts": 0.613956, "frozen": 0.108851, "full": 0.682525, "retrieval": 0.625623, "rules_nodetect": 0.687192}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:07:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:06Z", "cpu_seconds": 0.7543319469999998, "created_at": "2026-11-22T04:07:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-175c6b931dfb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "164a20ba89ef98e2ca809cb5357d0c9b636e86db2aea4854100a4f974dc6a751", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-175c6b931dfb", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:07:59Z", "seed": 871187540, "treatment": "frontier"}, "priority": 0.30007753310474516, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "992d0893c85e130272d64c4d1fd45c1455fb3daa6ea07b9b9b1f677ad81f2b0b", "run_seconds": 0.7650508880615234, "seed": 871187540, "started_at": "2026-11-22T04:08:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.05, 0.05], "effect": 0.05, "evaluated_at": "2026-11-22T04:08:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-175c6b931dfb", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.95, 0.9, 0.983333333, 0.966666667], "hypdisc": [1.0, 0.95, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.95, 0.95, 0.95, 0.95], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "164a20ba89ef98e2ca809cb5357d0c9b636e86db2aea4854100a4f974dc6a751", "primary_control": "random", "result_hash": "992d0893c85e130272d64c4d1fd45c1455fb3daa6ea07b9b9b1f677ad81f2b0b", "signature": "66ceaaeb53095b5994bedde541899fcb750d2f0d43ad9400a95b15ad3cc484ff25065d7a2ce28ab685ac11430bed95eafa8a377589177a827dbc7eef62cc980b", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 1.0, "goal": 0.95, "hypdisc": 0.9875, "infogain": 1.0, "prederr": 1.0, "random": 0.95, "risk": 1.0}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:08:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:01Z", "cpu_seconds": 0.5743006459999975, "created_at": "2026-11-22T04:07:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d84561d96cc4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "69c1c0d8175e9191282b071d282e124409d8da215ee685f3f40800e9e84ed495", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d84561d96cc4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:07:59Z", "seed": 748785798, "treatment": "full"}, "priority": 0.38796396822617774, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a01699e906a5be311e46da940007476adf84f607af958d098551bc6e943c6dd0", "run_seconds": 0.8044736385345459, "seed": 748785798, "started_at": "2026-11-22T04:08:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T04:08:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d84561d96cc4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 0.5], "fixed_policy": [0.25, 0.0, 0.25, 0.5], "full": [1.0, 0.75, 0.75, 0.75], "no_world_model": [1.0, 0.75, 0.5, 0.75], "random": [1.0, 0.75, 0.75, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "69c1c0d8175e9191282b071d282e124409d8da215ee685f3f40800e9e84ed495", "primary_control": "e141_best", "result_hash": "a01699e906a5be311e46da940007476adf84f607af958d098551bc6e943c6dd0", "signature": "a87b54917d72007fec9ef6a8d5dbe9fa6b3a76dccc64449026e03b74e95341180d29fd0b41955c017488377f8344a74723992eaef431f14916ca30572826ec0d", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.25, "full": 0.8125, "no_world_model": 0.75, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:08:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:09Z", "cpu_seconds": 2.713783688999996, "created_at": "2026-11-22T04:07:59Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0fe2cfa423a5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "1d1a51e1100461854e6097e43d65c356fa578558f30f5762ad6f9afc7b01079f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0fe2cfa423a5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:07:59Z", "seed": 703942728, "treatment": "full"}, "priority": 0.2985949365355252, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "6a5ffea13ace1e6d2bec6f80f37cded310eb1bbe0feb04da05606355a91c3df4", "run_seconds": 2.79481840133667, "seed": 703942728, "started_at": "2026-11-22T04:08:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.009123, 0.310539], "effect": 0.159831, "evaluated_at": "2026-11-22T04:08:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0fe2cfa423a5", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.367436744, 0.816666667, 0.396491228, 1.0], "frozen": [0.235423542, 0.483333333, 0.224561404, 0.333333333], "full": [0.721672167, 0.9, 0.598245614, 1.0], "retrieval": [0.824532453, 0.95, 0.619298246, 1.0], "rules_nodetect": [0.800880088, 0.9, 0.336842105, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "1d1a51e1100461854e6097e43d65c356fa578558f30f5762ad6f9afc7b01079f", "primary_control": "facts", "result_hash": "6a5ffea13ace1e6d2bec6f80f37cded310eb1bbe0feb04da05606355a91c3df4", "signature": "bfe225a26c35f9bd3fdb359dac274d9d81cff94238792335b666b6d34bcfc323078aa712498972861a0677053a22b26a3955592d6a84869716c8d806b0889408", "state": "VERIFIED", "summary": {"facts": 0.645149, "frozen": 0.319163, "full": 0.804979, "retrieval": 0.848458, "rules_nodetect": 0.759431}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:08:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:01Z", "cpu_seconds": 0.7854670219999997, "created_at": "2026-11-22T04:07:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-61e6f3e1028f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "66e77a5c6bc880dde7664b89fb7f7e36cd8c2fb4510d5f1b7f069f4c5bcbc71d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-61e6f3e1028f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:07:59Z", "seed": 1029517697, "treatment": "gated"}, "priority": 0.3906924605296107, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7c19d5e43f03aa2329e7d852542587b6bc85684e64a27ea02e3c402c4b91302a", "run_seconds": 0.9307944774627686, "seed": 1029517697, "started_at": "2026-11-22T04:08:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0235, 0.072458], "effect": 0.024479, "evaluated_at": "2026-11-22T04:08:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-61e6f3e1028f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.448076923, 0.528125, 0.588333333, 0.329166667], "gated": [0.448076923, 0.528125, 0.588333333, 0.427083333], "naive": [0.541346154, 0.4203125, 0.5375, 0.263541667], "scratch": [0.448076923, 0.528125, 0.588333333, 0.329166667], "ungated": [0.280769231, 0.3078125, 0.440833333, 0.446875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "66e77a5c6bc880dde7664b89fb7f7e36cd8c2fb4510d5f1b7f069f4c5bcbc71d", "primary_control": "scratch", "result_hash": "7c19d5e43f03aa2329e7d852542587b6bc85684e64a27ea02e3c402c4b91302a", "signature": "3d68cec7b5dec8e828e794970af1d66e4833ed6f0b43651a3189741572d2bfa751f6bf9af6ecc893e157857b243edd7f0b42f804b72b92a24345941ba3dbf90e", "state": "VERIFIED", "summary": {"compressed": 0.473425, "gated": 0.497905, "naive": 0.440675, "scratch": 0.473425, "ungated": 0.369073}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:08:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:56Z", "cpu_seconds": 1.4575709009999969, "created_at": "2026-11-22T04:08:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1f7bcce38492", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "2cc9d97797ce0df1943b51d13a4d52b09f0b56fe61738df9c0454ad83975faaa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1f7bcce38492", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:08:16Z", "seed": 1091772538, "treatment": "frontier"}, "priority": 0.24227964418800102, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "536fae63ada33b9af510f6629e4336bce0d218e4a4ce903d5e3f0d866a0e9c4e", "run_seconds": 1.4794573783874512, "seed": 1091772538, "started_at": "2026-11-22T04:09:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.158843, 0.249653], "effect": 0.204248, "evaluated_at": "2026-11-22T04:09:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1f7bcce38492", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.816993464, 0.830065359, 0.803921569, 0.843137255], "frontier": [0.91503268, 0.947712418, 0.882352941, 0.895424837], "goal": [0.836601307, 0.816993464, 0.830065359, 0.875816993], "hypdisc": [0.830065359, 0.764705882, 0.810457516, 0.732026144], "infogain": [0.91503268, 0.947712418, 0.882352941, 0.895424837], "prederr": [0.91503268, 0.947712418, 0.882352941, 0.895424837], "random": [0.745098039, 0.679738562, 0.712418301, 0.68627451], "risk": [0.91503268, 0.947712418, 0.882352941, 0.895424837]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2cc9d97797ce0df1943b51d13a4d52b09f0b56fe61738df9c0454ad83975faaa", "primary_control": "random", "result_hash": "536fae63ada33b9af510f6629e4336bce0d218e4a4ce903d5e3f0d866a0e9c4e", "signature": "922ff2aad660b39b4505a6d39e5a690add495dceea7760bcd7f9d4a092cdf7008d22f655fac0579e0ad3520a8a09eaccca7e936c5f102b1cc0d96fb563b4d705", "state": "VERIFIED", "summary": {"count": 0.823529, "frontier": 0.910131, "goal": 0.839869, "hypdisc": 0.784314, "infogain": 0.910131, "prederr": 0.910131, "random": 0.705882, "risk": 0.910131}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:09:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:50Z", "cpu_seconds": 0.8754957890000128, "created_at": "2026-11-22T04:08:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-809d4f30332c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "884fea32a7b38d1721b4ff239e943c5d0484b7407d21460597b5559304680242", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-809d4f30332c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:08:16Z", "seed": 421711251, "treatment": "frontier"}, "priority": 0.24228267007876947, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6ad0140912b16efd1153d64a3179f45880db23d5f12f700c29b045a48c6ca7c3", "run_seconds": 0.9780499935150146, "seed": 421711251, "started_at": "2026-11-22T04:08:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002193, 0.035527], "effect": 0.016667, "evaluated_at": "2026-11-22T04:08:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-809d4f30332c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.933333333, 0.95, 1.0, 1.0], "hypdisc": [0.966666667, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.966666667, 0.966666667, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "884fea32a7b38d1721b4ff239e943c5d0484b7407d21460597b5559304680242", "primary_control": "random", "result_hash": "6ad0140912b16efd1153d64a3179f45880db23d5f12f700c29b045a48c6ca7c3", "signature": "eac84de14db1667b9603389dd6be807cead76edef5f90b4b8861f9705fe829a40fe53241ffe5256ab80afebeb401adb05e3a16ed017059c605d534e1bd0a1b0b", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 1.0, "goal": 0.970833, "hypdisc": 0.991667, "infogain": 1.0, "prederr": 1.0, "random": 0.983333, "risk": 1.0}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:08:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:21Z", "cpu_seconds": 0.6787601070000022, "created_at": "2026-11-22T04:08:16Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a569b452c422", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "f32eff1d53a28a1a70f6b184b736128a3692f6b35bdce87e9d13fc3a4aa106d6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a569b452c422", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:08:16Z", "seed": 2110663737, "treatment": "full"}, "priority": 0.3138471213849761, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "30a5fc33a4d865bddc35d9c71b0f48713b2c0c79a92ca9c41d97b471210d80a3", "run_seconds": 0.7027440071105957, "seed": 2110663737, "started_at": "2026-11-22T04:08:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:08:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a569b452c422", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f32eff1d53a28a1a70f6b184b736128a3692f6b35bdce87e9d13fc3a4aa106d6", "primary_control": "e141_best", "result_hash": "30a5fc33a4d865bddc35d9c71b0f48713b2c0c79a92ca9c41d97b471210d80a3", "signature": "8830493e8e4f5d20e88f6cba05a785f6cd9ecf37016cd55b18c8a6cb50a3fb9949f57961903a29968194a755d30f6f6c3cb98d50e3491170c81469036e5c670a", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.9375, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:08:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:28Z", "cpu_seconds": 2.040507122000008, "created_at": "2026-11-22T04:08:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d5fe5939322c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "d113692905104176e2b3e72d1b83c4cb1d1330670991e3e18b8326b19d4f17a0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d5fe5939322c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:08:16Z", "seed": 1528449490, "treatment": "full"}, "priority": 0.2514145735661503, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "685833cb8924a641b885122986b02ccabf9b80d9c0d1f89ca9f122c9e59641e0", "run_seconds": 2.0755574703216553, "seed": 1528449490, "started_at": "2026-11-22T04:08:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.054881, 0.264429], "effect": 0.159655, "evaluated_at": "2026-11-22T04:08:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d5fe5939322c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.935185185, 0.28030303, 0.520930233, 0.626984127], "frozen": [0.0, 0.09469697, 0.223255814, 0.364285714], "full": [0.935185185, 0.494318182, 0.720930233, 0.851587302], "retrieval": [0.944444444, 0.486742424, 0.674418605, 0.866666667], "rules_nodetect": [0.407407407, 0.123106061, 0.451162791, 0.533333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "d113692905104176e2b3e72d1b83c4cb1d1330670991e3e18b8326b19d4f17a0", "primary_control": "facts", "result_hash": "685833cb8924a641b885122986b02ccabf9b80d9c0d1f89ca9f122c9e59641e0", "signature": "cd9f91be450eaa28eb1a43c3ade5a84de219ac0cfcf29f63f4cc945dde7ba6ae412e05b5fd721b2a8fcd26d904a76360c86e698bd74429754056109b9e754b0e", "state": "VERIFIED", "summary": {"facts": 0.590851, "frozen": 0.17056, "full": 0.750505, "retrieval": 0.743068, "rules_nodetect": 0.378752}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:08:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:26Z", "cpu_seconds": 3.0913340830000005, "created_at": "2026-11-22T04:08:16Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-7bc5fccd3ebe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "6bd3723d3e2fa5e257b6d2d6e11d3ab3f53d770cacf8f4a4b52fb440e629c9a6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-7bc5fccd3ebe", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:08:16Z", "seed": 204726212, "treatment": "gated"}, "priority": 0.3043158962089015, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c175b90d9312d0491774d7495eb9e143ff0a5bf4e96e8090711c8ee8d1587c6a", "run_seconds": 3.2274653911590576, "seed": 204726212, "started_at": "2026-11-22T04:08:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.040988, 0.104735], "effect": 0.031873, "evaluated_at": "2026-11-22T04:08:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-7bc5fccd3ebe", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.574074074, 0.711538462, 0.537037037, 0.567307692], "gated": [0.574074074, 0.711538462, 0.62037037, 0.567307692], "naive": [0.787037037, 0.711538462, 0.731481481, 0.625], "scratch": [0.472222222, 0.711538462, 0.537037037, 0.625], "ungated": [0.657407407, 0.673076923, 0.62037037, 0.538461538]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6bd3723d3e2fa5e257b6d2d6e11d3ab3f53d770cacf8f4a4b52fb440e629c9a6", "primary_control": "scratch", "result_hash": "c175b90d9312d0491774d7495eb9e143ff0a5bf4e96e8090711c8ee8d1587c6a", "signature": "081123891a0cd799d098177b38cad6ceb29baeec3f2748a6dbb551c652e58f91f657d115cd1aff3163e8260303a5460f71224f59833c5908b94f2800f9be390f", "state": "VERIFIED", "summary": {"compressed": 0.597489, "gated": 0.618323, "naive": 0.713764, "scratch": 0.586449, "ungated": 0.622329}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:08:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:18Z", "cpu_seconds": 0.6987802780000001, "created_at": "2026-11-22T04:08:16Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ec5826ceab70", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "8eba603a9d53576f2032423e1bfecf10d7938850b1fb8ffd9b3f6272f62175b2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ec5826ceab70", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:08:16Z", "seed": 1115548195, "treatment": "gated"}, "priority": 0.33073352421158164, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "419a2ce71f024c5f245cb4faced8c2eeed39b0156e78049c9a7da54846ce7f22", "run_seconds": 0.7796802520751953, "seed": 1115548195, "started_at": "2026-11-22T04:08:17Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:08:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-ec5826ceab70", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.424576271, 0.454166667, 0.421666667, 0.615], "gated": [0.424576271, 0.454166667, 0.421666667, 0.615], "naive": [0.456779661, 0.653333333, 0.4175, 0.38], "scratch": [0.424576271, 0.454166667, 0.421666667, 0.615], "ungated": [0.348305085, 0.556666667, 0.4525, 0.389166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8eba603a9d53576f2032423e1bfecf10d7938850b1fb8ffd9b3f6272f62175b2", "primary_control": "scratch", "result_hash": "419a2ce71f024c5f245cb4faced8c2eeed39b0156e78049c9a7da54846ce7f22", "signature": "8bf75dea188d081d31edf8f96f3a1a3314cd5b345141dd19b836a65bd072f62669181f4d111859b7191f7d35ee42f05af9dc58481da7ac90ed1ecb206289380e", "state": "PARTIAL", "summary": {"compressed": 0.478852, "gated": 0.478852, "naive": 0.476903, "scratch": 0.478852, "ungated": 0.43666}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:08:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:38Z", "cpu_seconds": 0.610368952, "created_at": "2026-11-22T04:08:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d9769e375bc2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "c8d199271ffbc703bfb7fc4b099627371b60d1c5e19eb8af9e4929a2e692cde4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d9769e375bc2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:08:33Z", "seed": 1862667562, "treatment": "full"}, "priority": 0.28040187430622465, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0875a2550f362172e1d81904cbb85d0f6607482a75e82f850372b8166e635798", "run_seconds": 0.6134226322174072, "seed": 1862667562, "started_at": "2026-11-22T04:08:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T04:08:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d9769e375bc2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.75, 1.0, 0.5], "fixed_policy": [0.5, 0.0, 0.0, 0.25], "full": [0.5, 1.0, 1.0, 0.75], "no_world_model": [0.25, 1.0, 1.0, 0.5], "random": [1.0, 0.75, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "c8d199271ffbc703bfb7fc4b099627371b60d1c5e19eb8af9e4929a2e692cde4", "primary_control": "e141_best", "result_hash": "0875a2550f362172e1d81904cbb85d0f6607482a75e82f850372b8166e635798", "signature": "6fb3b2bea0636171b0f5861ecc181867e2da7ad01f32a419e29aaa25171de0867d204d367af4493b41794cd2cd7b25256397dd2e332608f5a85be0576815f702", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.1875, "full": 0.8125, "no_world_model": 0.6875, "random": 0.875}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:08:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:50Z", "cpu_seconds": 0.9281913519999989, "created_at": "2026-11-22T04:08:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-39d1c8ea32c6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "aacfd1326161958ae3670a984a7950896394d405c79cb5aa0b0ce6e7a45c5e0c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-39d1c8ea32c6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:08:33Z", "seed": 495902078, "treatment": "full"}, "priority": 0.24614243453756093, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9b76d6748f96266965d68fa0f92c4f97eaa8c196b0e721a6e0d0f5046e052b66", "run_seconds": 1.0884928703308105, "seed": 495902078, "started_at": "2026-11-22T04:08:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022953, 0.136043], "effect": 0.056545, "evaluated_at": "2026-11-22T04:08:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-39d1c8ea32c6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.5, 0.604166667, 0.685483871, 0.947916667], "frozen": [0.333333333, 0.364583333, 0.282258065, 0.270833333], "full": [0.601851852, 0.614583333, 0.830645161, 0.916666667], "retrieval": [0.62962963, 0.6875, 0.85483871, 0.947916667], "rules_nodetect": [0.601851852, 0.614583333, 0.830645161, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "aacfd1326161958ae3670a984a7950896394d405c79cb5aa0b0ce6e7a45c5e0c", "primary_control": "facts", "result_hash": "9b76d6748f96266965d68fa0f92c4f97eaa8c196b0e721a6e0d0f5046e052b66", "signature": "962afbe4cabb3c2a8b355de9ea4d4b95d56312db04bc30ca12648172c046e91f3e4416f162f03c0c29dff4068aa70e5547ef5d80d513c88e6a0899c6eb8a8a0e", "state": "VERIFIED", "summary": {"facts": 0.684392, "frozen": 0.312752, "full": 0.740937, "retrieval": 0.779971, "rules_nodetect": 0.740937}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:08:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:44Z", "cpu_seconds": 2.8914218540000007, "created_at": "2026-11-22T04:08:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-79e84d5f28bc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "02c73194f170c86a23bd456378a875cbde08becf7aa912b1d4c34a464552d30a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-79e84d5f28bc", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:08:33Z", "seed": 2056165063, "treatment": "full"}, "priority": 0.27002241833418394, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "aad8c21613984ad8ee2ffa91f98059b30b23469b3233d96653cffbbdf3c1c743", "run_seconds": 3.0115864276885986, "seed": 2056165063, "started_at": "2026-11-22T04:08:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.00315, 0.17564], "effect": 0.089395, "evaluated_at": "2026-11-22T04:08:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-79e84d5f28bc", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.677777778, 0.60380117, 0.98, 0.399705015], "frozen": [0.251851852, 0.14619883, 0.38, 0.162241888], "full": [0.837962963, 0.631578947, 0.98, 0.569321534], "retrieval": [0.92962963, 0.706140351, 0.96, 0.631268437], "rules_nodetect": [0.561111111, 0.631578947, 0.56, 0.569321534]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "02c73194f170c86a23bd456378a875cbde08becf7aa912b1d4c34a464552d30a", "primary_control": "facts", "result_hash": "aad8c21613984ad8ee2ffa91f98059b30b23469b3233d96653cffbbdf3c1c743", "signature": "3d1e82e6404694c9582128f2fb22aac181bcb7d8bae8d048c1dbf5ec39c23ee4c7d9529af6a63afc2a45f6c2533f523db903fc6d2b652b1c18c4afa2f05e1609", "state": "VERIFIED", "summary": {"facts": 0.665321, "frozen": 0.235073, "full": 0.754716, "retrieval": 0.80676, "rules_nodetect": 0.580503}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:08:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:44Z", "cpu_seconds": 0.8101540269999994, "created_at": "2026-11-22T04:08:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a841344c7cd7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 4, "n": 4}, "prereg_hash": "a292b6233dc494999e085088d42b107db6db5ec3f848dbafaddea0f34e5d441a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a841344c7cd7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:08:33Z", "seed": 2038993369, "treatment": "full"}, "priority": 0.24821491282929048, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0f44e91fbf56b2cdec3fe439b52831cf38e9516fbd511b8235f5aed936175a1f", "run_seconds": 0.9285900592803955, "seed": 2038993369, "started_at": "2026-11-22T04:08:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.051744, 0.174688], "effect": 0.113216, "evaluated_at": "2026-11-22T04:08:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a841344c7cd7", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.5171875, 0.78125, 0.5234375, 0.44375], "frozen": [0.38125, 0.5375, 0.353125, 0.3625], "full": [0.58125, 0.875, 0.61328125, 0.648958333], "retrieval": [0.8, 0.9375, 0.82734375, 0.813541667], "rules_nodetect": [0.58125, 0.875, 0.61328125, 0.648958333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a292b6233dc494999e085088d42b107db6db5ec3f848dbafaddea0f34e5d441a", "primary_control": "facts", "result_hash": "0f44e91fbf56b2cdec3fe439b52831cf38e9516fbd511b8235f5aed936175a1f", "signature": "21c6189607d06bce1db61823bbb9c7ed75723ac894f707f57915e378a256732baeb88652fcedb6098bc707b3a951a1a8eb87f00ec7bde4d52e4461d858a26508", "state": "VERIFIED", "summary": {"facts": 0.566406, "frozen": 0.408594, "full": 0.679622, "retrieval": 0.844596, "rules_nodetect": 0.679622}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:08:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:36Z", "cpu_seconds": 2.888032924000001, "created_at": "2026-11-22T04:08:33Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-410264ecdd1e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "616ed051b1d96d7765ed480cd38a7810615faa4e02ea450974bfecfde155ffcf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-410264ecdd1e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:08:33Z", "seed": 354867796, "treatment": "gated"}, "priority": 0.3310382824236112, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "4934d5e64967ac6bbd17f44ed3e624a816e7a3e102c32394179d323b5f4d8099", "run_seconds": 3.070275068283081, "seed": 354867796, "started_at": "2026-11-22T04:08:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016364, 0.050455], "effect": 0.017045, "evaluated_at": "2026-11-22T04:08:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-410264ecdd1e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.568965517, 0.42, 0.384615385, 0.556818182], "gated": [0.568965517, 0.42, 0.384615385, 0.556818182], "naive": [0.775862069, 0.62, 0.615384615, 0.647727273], "scratch": [0.568965517, 0.42, 0.384615385, 0.488636364], "ungated": [0.706896552, 0.53, 0.519230769, 0.556818182]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "616ed051b1d96d7765ed480cd38a7810615faa4e02ea450974bfecfde155ffcf", "primary_control": "scratch", "result_hash": "4934d5e64967ac6bbd17f44ed3e624a816e7a3e102c32394179d323b5f4d8099", "signature": "af0fc175b28c9035d85478d4ae9680deafd1a5a30029fbb27db0077e624c9518e594d7d1c4b1c41fc754d98cfb5c8f63be071fd968e7baedbb2dc75f64d43205", "state": "VERIFIED", "summary": {"compressed": 0.4826, "gated": 0.4826, "naive": 0.664743, "scratch": 0.465554, "ungated": 0.578236}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:08:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:24Z", "cpu_seconds": 1.432725743000006, "created_at": "2026-11-22T04:08:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c1c7d645bcfc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "becffd40b3411877ecddb1436bb696660b670b9956ca6da69c7dcc99012c267e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c1c7d645bcfc", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:08:50Z", "seed": 1326737123, "treatment": "full"}, "priority": 0.24462962866393506, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "12b5f53b974e88269f28d0294ba3375256161cd1ca92e620625510ab31ec1231", "run_seconds": 1.446007490158081, "seed": 1326737123, "started_at": "2026-11-22T04:09:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.07, 0.805], "effect": 0.4375, "evaluated_at": "2026-11-22T04:09:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c1c7d645bcfc", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 0.0, 0.75], "fixed_policy": [0.25, 0.25, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 1.0, 1.0], "random": [0.75, 1.0, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "becffd40b3411877ecddb1436bb696660b670b9956ca6da69c7dcc99012c267e", "primary_control": "e141_best", "result_hash": "12b5f53b974e88269f28d0294ba3375256161cd1ca92e620625510ab31ec1231", "signature": "e2787336530193139829ca6eb187e2f4eaae0d0097894220e72c66225b42ee32c2ce3c9d15e869fde316442f1689745b2af6dc426d48eea709845aed957edc0c", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 0.875, "random": 0.8125}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:09:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:22Z", "cpu_seconds": 0.49570558399999953, "created_at": "2026-11-22T04:08:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c6f96ef00cd6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "962972498e84f919b3434a5c016d641efb918ce91c26405ea3bf1f756a16ba75", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c6f96ef00cd6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:08:50Z", "seed": 168041311, "treatment": "full"}, "priority": 0.245718294274003, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "926af9cecc168f6734a64e849f8993be87a4b98559ee8f97901b102092d5c2e8", "run_seconds": 0.5239920616149902, "seed": 168041311, "started_at": "2026-11-22T04:09:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.052693, 0.947307], "effect": 0.5, "evaluated_at": "2026-11-22T04:09:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-c6f96ef00cd6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.0, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [1.0, 0.0, 1.0, 1.0], "no_world_model": [0.0, 0.0, 0.5, 1.0], "random": [0.25, 0.5, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "962972498e84f919b3434a5c016d641efb918ce91c26405ea3bf1f756a16ba75", "primary_control": "e141_best", "result_hash": "926af9cecc168f6734a64e849f8993be87a4b98559ee8f97901b102092d5c2e8", "signature": "0d7331c3377501734a9b51ddd830989175675c6e77e67aa852cda057f4712c1390b62fea408f1119c4e8ddf4e7990151bd79469c9c5d523a238ce45d82037206", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 0.75, "no_world_model": 0.375, "random": 0.6875}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:09:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:02Z", "cpu_seconds": 0.8782556889999995, "created_at": "2026-11-22T04:08:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d4597930aa6c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "27df79992fdf52f4d8a4ba38066fa2634651cb525b6391b64d7020cbe068b562", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d4597930aa6c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:08:50Z", "seed": 1678115069, "treatment": "full"}, "priority": 0.24777659163531898, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-8ad50116d430", "result_hash": "0832831499365e5b8e48dfa50ad80737e809e1182c7c93db67bac7c55c21afd6", "run_seconds": 0.9016773700714111, "seed": 1678115069, "started_at": "2026-11-22T04:09:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.855917, -0.019083], "effect": -0.4375, "evaluated_at": "2026-11-22T04:09:04Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d4597930aa6c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [0.0, 1.0, 0.5, 0.5], "no_world_model": [1.0, 1.0, 1.0, 0.25], "random": [1.0, 1.0, 1.0, 0.25]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "27df79992fdf52f4d8a4ba38066fa2634651cb525b6391b64d7020cbe068b562", "primary_control": "e141_best", "replicated": false, "replication_of": "E141-ow_goal-8ad50116d430", "result_hash": "0832831499365e5b8e48dfa50ad80737e809e1182c7c93db67bac7c55c21afd6", "signature": "09159743a035453237cc5b13a38e612b21459eee589d31b017be1500db4d62cb8f98ec819020419861289c48bff29b150851cc58f78d845459943458d9c74209", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0625, "full": 0.5, "no_world_model": 0.8125, "random": 0.8125}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:09:05Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:58Z", "cpu_seconds": 2.420063142999993, "created_at": "2026-11-22T04:08:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c2419fed8634", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "aa6f0dffb64b0b869d50814a169e84a0f6a565a18eda767c06921cfe6599650d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c2419fed8634", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:08:50Z", "seed": 59202177, "treatment": "full"}, "priority": 0.2736393819237545, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "945cf93efcaf5c922e0e9cc896b2cbd23d6d3c20d9ec25e96202daee273bc81a", "run_seconds": 2.7717068195343018, "seed": 59202177, "started_at": "2026-11-22T04:08:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.049039, 0.270775], "effect": 0.159907, "evaluated_at": "2026-11-22T04:08:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c2419fed8634", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.601626016, 0.664166667, 0.904, 0.482222222], "frozen": [0.136178862, 0.305, 0.36, 0.259259259], "full": [0.727642276, 0.826666667, 0.944, 0.793333333], "retrieval": [0.715447154, 0.918333333, 0.92, 0.837037037], "rules_nodetect": [0.727642276, 0.26, 0.424, 0.395555556]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "aa6f0dffb64b0b869d50814a169e84a0f6a565a18eda767c06921cfe6599650d", "primary_control": "facts", "result_hash": "945cf93efcaf5c922e0e9cc896b2cbd23d6d3c20d9ec25e96202daee273bc81a", "signature": "1454fef4174d25554852fae714c43a55afb93fb2d397921cd79d288332abcf3a1b06ef451c65511f3841587b3f632231be650e6bc55749a066367285f7312b0a", "state": "VERIFIED", "summary": {"facts": 0.663004, "frozen": 0.26511, "full": 0.822911, "retrieval": 0.847704, "rules_nodetect": 0.451799}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:09:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:05Z", "cpu_seconds": 2.1061971720000088, "created_at": "2026-11-22T04:08:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d59d14772610", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 10, "n": 4}, "prereg_hash": "6c2808f6de84064a5d642a81eed3fd7976cab61e0c45ca04bfdd484d42d09612", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d59d14772610", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:08:50Z", "seed": 1255846421, "treatment": "full"}, "priority": 0.24665884250538464, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f0478311c7708564cebfbe8e3cd08bb8f10e2b72e44935361816b9e7d0ea8066", "run_seconds": 2.341510772705078, "seed": 1255846421, "started_at": "2026-11-22T04:09:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.044855, 0.35378], "effect": 0.199318, "evaluated_at": "2026-11-22T04:09:07Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d59d14772610", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.256076389, 0.932291667, 0.232835821, 0.073684211], "frozen": [0.211805556, 0.25, 0.047761194, 0.028947368], "full": [0.53125, 0.942708333, 0.373134328, 0.445065789], "retrieval": [0.557725694, 0.921875, 0.365671642, 0.419407895], "rules_nodetect": [0.046006944, 0.651041667, 0.373134328, 0.445065789]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "6c2808f6de84064a5d642a81eed3fd7976cab61e0c45ca04bfdd484d42d09612", "primary_control": "facts", "result_hash": "f0478311c7708564cebfbe8e3cd08bb8f10e2b72e44935361816b9e7d0ea8066", "signature": "0fceb2720fdd4f53bf35c50e59496d0946b269c7d3a2506178277a71351cc5ee300feba5c7c2e52ce50aa200782df6c034f76147fb4661d396285e27d5b3e30c", "state": "VERIFIED", "summary": {"facts": 0.373722, "frozen": 0.134629, "full": 0.57304, "retrieval": 0.56617, "rules_nodetect": 0.378812}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:09:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:08:56Z", "cpu_seconds": 1.0039422899999995, "created_at": "2026-11-22T04:08:50Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e22818ad1b21", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "0b5607a9c34c48a97512ab7974732c991c51b3e4a45a36d1a1c243587b89ccf6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e22818ad1b21", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:08:50Z", "seed": 74659099, "treatment": "gated"}, "priority": 0.3036901680845045, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "e070c4c25795119744c40fd115a42442746902bbfaca5b6780938608c371b980", "run_seconds": 1.2888267040252686, "seed": 74659099, "started_at": "2026-11-22T04:08:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037045, 0.037045], "effect": 0.0, "evaluated_at": "2026-11-22T04:08:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-e22818ad1b21", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.425925926, 0.6, 0.574074074, 0.54], "gated": [0.425925926, 0.6, 0.574074074, 0.54], "naive": [0.574074074, 0.67, 0.675925926, 0.71], "scratch": [0.472222222, 0.6, 0.527777778, 0.54], "ungated": [0.574074074, 0.65, 0.509259259, 0.63]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "0b5607a9c34c48a97512ab7974732c991c51b3e4a45a36d1a1c243587b89ccf6", "primary_control": "scratch", "result_hash": "e070c4c25795119744c40fd115a42442746902bbfaca5b6780938608c371b980", "signature": "7e47f902d20fded1a17792729f0c04e1df6dddf49778f6214a52dde72721a2e785c633373b91a465a4d6b31859ee520d46101e2521a5483d2b563e638057750c", "state": "VERIFIED", "summary": {"compressed": 0.535, "gated": 0.535, "naive": 0.6575, "scratch": 0.535, "ungated": 0.590833}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:08:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:11Z", "cpu_seconds": 0.9166526279999943, "created_at": "2026-11-22T04:09:06Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-220911192f1e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "d9abcf8439b7ebace60d3d34e4754047c3dd53b859e16787c7d7b317f2ba698e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-220911192f1e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:09:06Z", "seed": 1735608640, "treatment": "full"}, "priority": 0.2846225351205448, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e00a39f6f7ce3bac5916b431070313d94855e32bbd0e2fe354a9948f2288075e", "run_seconds": 0.9674472808837891, "seed": 1735608640, "started_at": "2026-11-22T04:09:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.18, 0.555], "effect": 0.1875, "evaluated_at": "2026-11-22T04:09:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-220911192f1e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 0.25, 1.0], "fixed_policy": [0.25, 1.0, 1.0, 0.75], "full": [0.0, 0.25, 1.0, 1.0], "no_world_model": [0.0, 0.25, 0.5, 0.75], "random": [0.5, 0.25, 0.5, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d9abcf8439b7ebace60d3d34e4754047c3dd53b859e16787c7d7b317f2ba698e", "primary_control": "e141_best", "result_hash": "e00a39f6f7ce3bac5916b431070313d94855e32bbd0e2fe354a9948f2288075e", "signature": "f2b4fb5dcfee3b0d60e5bde25b18831092939bb5cf11db6a03021eac0be80857b9b34303d473d7924aaaacbcb3c18686f7df4d48204d583dfee3b3dbd7e2af0a", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.75, "full": 0.5625, "no_world_model": 0.375, "random": 0.5625}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:09:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:16Z", "cpu_seconds": 1.9873677630000017, "created_at": "2026-11-22T04:09:06Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ec3046e0f5fa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "9b130ec17510b275bd36140efb68194ee8bc971a622d778e58b840379d436521", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ec3046e0f5fa", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:09:06Z", "seed": 782877341, "treatment": "full"}, "priority": 0.27378800441661566, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c80045b17807cce2c9d37ee5e1f16f0d658a352db9dbd1748ee72d814b902ac3", "run_seconds": 2.0738086700439453, "seed": 782877341, "started_at": "2026-11-22T04:09:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.000226, 0.135015], "effect": 0.067395, "evaluated_at": "2026-11-22T04:09:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ec3046e0f5fa", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.717647059, 0.287545788, 1.0, 0.936], "frozen": [0.137254902, 0.0, 0.303529412, 0.36], "full": [0.876470588, 0.366300366, 1.0, 0.968], "retrieval": [0.845098039, 0.723443223, 1.0, 0.976], "rules_nodetect": [0.876470588, 0.366300366, 1.0, 0.968]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "9b130ec17510b275bd36140efb68194ee8bc971a622d778e58b840379d436521", "primary_control": "facts", "result_hash": "c80045b17807cce2c9d37ee5e1f16f0d658a352db9dbd1748ee72d814b902ac3", "signature": "f6c8a99affa619c10747e1abea4853f0139162b81f62a3712b12f6470681056e8c53592d982ffcc2525bc193717dfb79cefde3fe4c30e9c1118b4e2ccfaf440c", "state": "VERIFIED", "summary": {"facts": 0.735298, "frozen": 0.200196, "full": 0.802693, "retrieval": 0.886135, "rules_nodetect": 0.802693}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:09:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:09Z", "cpu_seconds": 2.0797019920000004, "created_at": "2026-11-22T04:09:06Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-734d8819dd28", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "60a74370dd12300500614ecac513ee6df4299248474d6776c025f2491a78aa9d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-734d8819dd28", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:09:06Z", "seed": 507057533, "treatment": "gated"}, "priority": 0.319409082521128, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f2eb6ed1d365826d8cd4e47364bf3b51de71d28c3129370da7ed809f0fe251c7", "run_seconds": 2.129964828491211, "seed": 507057533, "started_at": "2026-11-22T04:09:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.082222, 0.026667], "effect": -0.027778, "evaluated_at": "2026-11-22T04:09:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-734d8819dd28", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.608333333, 0.625, 0.565217391, 0.398148148], "gated": [0.608333333, 0.625, 0.565217391, 0.398148148], "naive": [0.616666667, 0.625, 0.663043478, 0.601851852], "scratch": [0.608333333, 0.625, 0.565217391, 0.509259259], "ungated": [0.566666667, 0.625, 0.663043478, 0.462962963]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "60a74370dd12300500614ecac513ee6df4299248474d6776c025f2491a78aa9d", "primary_control": "scratch", "result_hash": "f2eb6ed1d365826d8cd4e47364bf3b51de71d28c3129370da7ed809f0fe251c7", "signature": "c96e9c78376267fe65f65c8cbdd34a04dd31ccf200e338ed84d9568bc22f23d84808c95376f005f0caa5ca14319491027f9369ce26803e23f1c74d33db461204", "state": "VERIFIED", "summary": {"compressed": 0.549175, "gated": 0.549175, "naive": 0.62664, "scratch": 0.576952, "ungated": 0.579418}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:09:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:17Z", "cpu_seconds": 0.7869251219999995, "created_at": "2026-11-22T04:09:06Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a920fc8e08f7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "4a61b639cc4f506d2638cc3b345ed1a758dbc65a2a885852ed3b91ff0696b541", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a920fc8e08f7", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:09:06Z", "seed": 306987947, "treatment": "gated"}, "priority": 0.2518838968341427, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "897f2287d48eadc06ea3b992a904bb059071c67872c5f1709838434eba357588", "run_seconds": 0.7976036071777344, "seed": 306987947, "started_at": "2026-11-22T04:09:16Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:09:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a920fc8e08f7", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.428723404, 0.875, 0.9125, 0.640625], "gated": [0.428723404, 0.875, 0.9125, 0.640625], "naive": [0.384042553, 0.79375, 0.775, 0.433333333], "scratch": [0.428723404, 0.875, 0.9125, 0.640625], "ungated": [0.337234043, 0.79375, 0.85, 0.301041667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4a61b639cc4f506d2638cc3b345ed1a758dbc65a2a885852ed3b91ff0696b541", "primary_control": "scratch", "result_hash": "897f2287d48eadc06ea3b992a904bb059071c67872c5f1709838434eba357588", "signature": "b3134f980fc5e42b8476d81cb6462a8c3ed4705acea06a89cb4db2858773ef590db715f3eb4c96587918f6ab220ea85a7a4911e81adc9c2bab3a4546da64690b", "state": "PARTIAL", "summary": {"compressed": 0.714212, "gated": 0.714212, "naive": 0.596531, "scratch": 0.714212, "ungated": 0.570506}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:09:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:37Z", "cpu_seconds": 0.6313343529999997, "created_at": "2026-11-22T04:09:23Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0208dd0c5fa6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "175a6ecbd45ac95162445ecc9cd855d3084cc2cc1338f83f8e74afc1ddce65b0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0208dd0c5fa6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:09:23Z", "seed": 1019658743, "treatment": "frontier"}, "priority": 0.24371761592234734, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5f75019b53e90ce4174c58c204e08a88f44a76f5b6cdcb022632ece12836b2b7", "run_seconds": 0.6533520221710205, "seed": 1019658743, "started_at": "2026-11-22T04:09:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.043209, 0.059876], "effect": 0.008333, "evaluated_at": "2026-11-22T04:09:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0208dd0c5fa6", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.95, 1.0], "frontier": [0.958333333, 1.0, 0.9, 1.0], "goal": [0.9375, 0.958333333, 0.95, 0.9375], "hypdisc": [0.958333333, 1.0, 0.9, 0.979166667], "infogain": [0.958333333, 1.0, 0.9, 1.0], "prederr": [0.958333333, 1.0, 0.9, 1.0], "random": [0.979166667, 0.958333333, 0.95, 0.9375], "risk": [0.958333333, 1.0, 0.9, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "175a6ecbd45ac95162445ecc9cd855d3084cc2cc1338f83f8e74afc1ddce65b0", "primary_control": "random", "result_hash": "5f75019b53e90ce4174c58c204e08a88f44a76f5b6cdcb022632ece12836b2b7", "signature": "b608e3e308d65bd3f46fff7234be55461fa2c8ccc9fae9476086963e897b4c526fa01873ed1e89f5e972210ffb0a9186c97ea516a7173600afceca6d9e8d8d06", "state": "VERIFIED", "summary": {"count": 0.9875, "frontier": 0.964583, "goal": 0.945833, "hypdisc": 0.959375, "infogain": 0.964583, "prederr": 0.964583, "random": 0.95625, "risk": 0.964583}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:09:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:36Z", "cpu_seconds": 0.9272667890000008, "created_at": "2026-11-22T04:09:23Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3723ba704f59", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "89c44e0740667b964cc8054e53b3d8c7fdd418217541c57ca1263485e6df9710", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3723ba704f59", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:09:23Z", "seed": 193285454, "treatment": "frontier"}, "priority": 0.24744784224196223, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "46bc3efcbdc9c85559de14d1f0e5497795cc5e82184b18b0509fd8b33c7ab861", "run_seconds": 0.9301002025604248, "seed": 193285454, "started_at": "2026-11-22T04:09:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.047542, 0.035961], "effect": -0.005791, "evaluated_at": "2026-11-22T04:09:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-3723ba704f59", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.833333333, 0.646666667, 0.791666667, 0.461538462], "frontier": [0.833333333, 0.593333333, 0.722222222, 0.820512821], "goal": [0.833333333, 0.493333333, 0.722222222, 0.820512821], "hypdisc": [0.833333333, 0.293333333, 0.736111111, 0.820512821], "infogain": [0.833333333, 0.546666667, 0.722222222, 0.820512821], "prederr": [0.833333333, 0.666666667, 0.722222222, 0.820512821], "random": [0.833333333, 0.64, 0.75, 0.769230769], "risk": [0.833333333, 0.593333333, 0.722222222, 0.820512821]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "89c44e0740667b964cc8054e53b3d8c7fdd418217541c57ca1263485e6df9710", "primary_control": "random", "result_hash": "46bc3efcbdc9c85559de14d1f0e5497795cc5e82184b18b0509fd8b33c7ab861", "signature": "7495f0a3796f29133de977a14ab305cbce545b2c75385d79621ff5754545b14656cb8bb5f2505dd60a6e652212c27daac1c6dbb68cea36512652309364bc830a", "state": "VERIFIED", "summary": {"count": 0.683301, "frontier": 0.74235, "goal": 0.71735, "hypdisc": 0.670823, "infogain": 0.730684, "prederr": 0.760684, "random": 0.748141, "risk": 0.74235}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:09:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:31Z", "cpu_seconds": 2.039665278000001, "created_at": "2026-11-22T04:09:23Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a728527c1ea8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "1d697bf35d6a169483771aa3a8b4ae8c110dc6d713bfc3bd8286a9e5162524f0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a728527c1ea8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:09:23Z", "seed": 880723778, "treatment": "full"}, "priority": 0.2790478120429367, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5f5d82f08c85586ead0b52cdd3a8129503a8889eacb8a263b5145259539ec179", "run_seconds": 2.1146328449249268, "seed": 880723778, "started_at": "2026-11-22T04:09:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.005228, 0.270253], "effect": 0.13774, "evaluated_at": "2026-11-22T04:09:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a728527c1ea8", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.449275362, 0.432098765, 0.979166667, 0.845714286], "frozen": [0.12173913, 0.281481481, 0.4375, 0.365714286], "full": [0.592753623, 0.749382716, 0.972222222, 0.942857143], "retrieval": [0.669565217, 0.867901235, 0.979166667, 0.971428571], "rules_nodetect": [0.592753623, 0.325925926, 0.791666667, 0.491428571]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "1d697bf35d6a169483771aa3a8b4ae8c110dc6d713bfc3bd8286a9e5162524f0", "primary_control": "facts", "result_hash": "5f5d82f08c85586ead0b52cdd3a8129503a8889eacb8a263b5145259539ec179", "signature": "ae05ec4604a64afc4a9697297ba9e1519ca053bdb8bbfa412bb43b2c7776c3b717197a58d181f08285897163fc7111d488dabc16a6e62d2522da8105ea8ff30b", "state": "VERIFIED", "summary": {"facts": 0.676564, "frozen": 0.301609, "full": 0.814304, "retrieval": 0.872015, "rules_nodetect": 0.550444}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:09:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:30Z", "cpu_seconds": 2.947476424999998, "created_at": "2026-11-22T04:09:23Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5e37a13c059f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "8b26aed9e41fd0656348011974e05a3c8b3fe3916e20f7391fa5af6e8d28bc2f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5e37a13c059f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:09:23Z", "seed": 832136343, "treatment": "gated"}, "priority": 0.30624962745600803, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "a0cd19470404fb757ac5a32ab97ee309f619bd1808d871a7665f1d9635a8bdc0", "run_seconds": 3.265307903289795, "seed": 832136343, "started_at": "2026-11-22T04:09:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.122412, 0.163254], "effect": 0.020421, "evaluated_at": "2026-11-22T04:09:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-5e37a13c059f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.675925926, 0.615384615, 0.6875, 0.565789474], "gated": [0.675925926, 0.615384615, 0.6875, 0.565789474], "naive": [0.675925926, 0.692307692, 0.734375, 0.710526316], "scratch": [0.444444444, 0.673076923, 0.6875, 0.657894737], "ungated": [0.453703704, 0.634615385, 0.734375, 0.605263158]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "8b26aed9e41fd0656348011974e05a3c8b3fe3916e20f7391fa5af6e8d28bc2f", "primary_control": "scratch", "result_hash": "a0cd19470404fb757ac5a32ab97ee309f619bd1808d871a7665f1d9635a8bdc0", "signature": "9d994c466f242a20296ce9f17d40f1758a1c7dbc5413c53a808c16e3723f6fd5662b8104d2965ab2e4a3225950b7134a385860a2db50f3c7169ecfd9326fb50c", "state": "VERIFIED", "summary": {"compressed": 0.63615, "gated": 0.63615, "naive": 0.703284, "scratch": 0.615729, "ungated": 0.606989}, "ties": 1, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:09:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:25Z", "cpu_seconds": 1.1601240109999935, "created_at": "2026-11-22T04:09:23Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-9a78c70b9f09", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "475261bce8c93dd9bd41bcd10196ec562d2eacf8a454c73627a032c760a99e39", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-9a78c70b9f09", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:09:23Z", "seed": 358845336, "treatment": "gated"}, "priority": 0.24210053841644352, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "84e1e4d996cb234cdf2d40fb628005cb51d2948e1fd75ba3d919c62b29718899", "run_seconds": 1.2786800861358643, "seed": 358845336, "started_at": "2026-11-22T04:10:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.086951, 0.122784], "effect": 0.017917, "evaluated_at": "2026-11-22T04:10:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-9a78c70b9f09", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.579545455, 0.525, 0.43, 0.544642857], "gated": [0.579545455, 0.641666667, 0.51, 0.544642857], "naive": [0.727272727, 0.641666667, 0.59, 0.776785714], "scratch": [0.579545455, 0.525, 0.43, 0.669642857], "ungated": [0.727272727, 0.641666667, 0.58, 0.776785714]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "475261bce8c93dd9bd41bcd10196ec562d2eacf8a454c73627a032c760a99e39", "primary_control": "scratch", "result_hash": "84e1e4d996cb234cdf2d40fb628005cb51d2948e1fd75ba3d919c62b29718899", "signature": "37add03ceced7d511862e4961fef16f106d91dd0ff12e6f1d5f2c3a0c7179c096af68a9450322d1da2acc4af0af42a1d7f95c4a8ad0359d9369a540c6355eb03", "state": "VERIFIED", "summary": {"compressed": 0.519797, "gated": 0.568964, "naive": 0.683931, "scratch": 0.551047, "ungated": 0.681431}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:10:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:53Z", "cpu_seconds": 0.8724312559999987, "created_at": "2026-11-22T04:09:23Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a1b915b37abe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "prereg_hash": "795e14f1ecfa2d3dba46ff5d42db28bb890eaf88100ef59bb25c96b61786cfbb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a1b915b37abe", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:09:23Z", "seed": 1183491699, "treatment": "gated"}, "priority": 0.24239270857662942, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c5c4453623e9de49ce115c6f4d5faab36964b7c9762c0211430045f345baf577", "run_seconds": 0.879338264465332, "seed": 1183491699, "started_at": "2026-11-22T04:09:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005184, 0.082844], "effect": 0.03883, "evaluated_at": "2026-11-22T04:09:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a1b915b37abe", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.208510638, 0.268085106, 0.156097561, 0.261702128], "gated": [0.208510638, 0.268085106, 0.156097561, 0.261702128], "naive": [0.219148936, 0.187234043, 0.153658537, 0.172340426], "scratch": [0.208510638, 0.187234043, 0.156097561, 0.187234043], "ungated": [0.282978723, 0.229787234, 0.153658537, 0.242553191]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "795e14f1ecfa2d3dba46ff5d42db28bb890eaf88100ef59bb25c96b61786cfbb", "primary_control": "scratch", "result_hash": "c5c4453623e9de49ce115c6f4d5faab36964b7c9762c0211430045f345baf577", "signature": "514a7711934e783c11c61a7cd7d2aa62c0229f5c716590af0a07660aed4740ab6361a426936af22354068a77b1ae7a4350536424a3474834f08e96906e736207", "state": "VERIFIED", "summary": {"compressed": 0.223599, "gated": 0.223599, "naive": 0.183095, "scratch": 0.184769, "ungated": 0.227244}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:09:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:42Z", "cpu_seconds": 1.0019851919999994, "created_at": "2026-11-22T04:09:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1dff8c12c39c", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "e60cc868f581a633ecf9a9ce103bc011c92f68b48a1e88132772404e3a170bdc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1dff8c12c39c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:09:40Z", "seed": 445091800, "treatment": "frontier"}, "priority": 0.3719699627481029, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c969c8c420aed0eb2b475de482427660573b6ec41b5045f7383583f3ca478586", "run_seconds": 1.0333001613616943, "seed": 445091800, "started_at": "2026-11-22T04:09:41Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:09:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1dff8c12c39c", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.766616], "frontier": [1.0, 0.776779], "goal": [1.0, 0.807438], "hypdisc": [1.0, 0.765873], "infogain": [1.0, 0.776779], "prederr": [1.0, 0.776779], "random": [1.0, 0.798862], "risk": [1.0, 0.776779]}, "prereg_hash": "e60cc868f581a633ecf9a9ce103bc011c92f68b48a1e88132772404e3a170bdc", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "c969c8c420aed0eb2b475de482427660573b6ec41b5045f7383583f3ca478586", "signature": "757e2bc78dc204f1a5fa046580b90e5e222f7314d1572337cc66409fd729a0a0a21e3f649481c51d82217f8a4c587950fa3c025432f63b7a2f2153ff1ce10107", "state": "REJECTED", "summary": {"count": 0.766616, "frontier": 0.776779, "goal": 0.807438, "hypdisc": 0.765873, "infogain": 0.776779, "prederr": 0.776779, "random": 0.798862, "risk": 0.776779}}, "verified_at": "2026-11-22T04:09:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:50Z", "cpu_seconds": 0.9708134490000049, "created_at": "2026-11-22T04:09:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1ab4ec458c3e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "25745f26972845e37ecd6cea03015edd92fe7794de131a99c52387e066710f95", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1ab4ec458c3e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:09:40Z", "seed": 385024051, "treatment": "full"}, "priority": 0.24265395416846533, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b3b8355f37b896d3bef777721649bddc7cdeeb92b762ca4c3211c30c40122a4f", "run_seconds": 1.0279781818389893, "seed": 385024051, "started_at": "2026-11-22T04:09:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.101777, 0.056142], "effect": -0.022817, "evaluated_at": "2026-11-22T04:09:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1ab4ec458c3e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.444444444, 0.8, 0.871794872, 0.80952381], "frozen": [0.361111111, 0.316666667, 0.269230769, 0.30952381], "full": [0.305555556, 0.8, 0.871794872, 0.857142857], "retrieval": [0.472222222, 0.8, 0.871794872, 0.857142857], "rules_nodetect": [0.888888889, 0.8, 0.871794872, 0.857142857]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "25745f26972845e37ecd6cea03015edd92fe7794de131a99c52387e066710f95", "primary_control": "facts", "result_hash": "b3b8355f37b896d3bef777721649bddc7cdeeb92b762ca4c3211c30c40122a4f", "signature": "31bc990d7f5da122021191f83834bc504ef7d6142ed98eadac92eb490ce5c2002fb8ec037ceeebb1c755ac7817aa4d131ee321f72386e2896437267799953e04", "state": "VERIFIED", "summary": {"facts": 0.731441, "frozen": 0.314133, "full": 0.708623, "retrieval": 0.75029, "rules_nodetect": 0.854457}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:09:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:48Z", "cpu_seconds": 0.39251815700000137, "created_at": "2026-11-22T04:09:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-773f04e04f3c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "63b47430528c7c673f35e8ad9db4219420155230259b404ff2c789eee375d0a3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-773f04e04f3c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:09:40Z", "seed": 1443178442, "treatment": "full"}, "priority": 0.2442771634411126, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "bf78af1f3cc0d32a3fb57a6b1788364658e96ffdb19184280d74ed2018ae6dfc", "run_seconds": 0.39476609230041504, "seed": 1443178442, "started_at": "2026-11-22T04:09:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.072433, 0.143808], "effect": 0.10812, "evaluated_at": "2026-11-22T04:09:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-773f04e04f3c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.573643411, 0.19379845, 0.549019608, 0.590909091], "frozen": [0.100775194, 0.100775194, 0.08496732, 0.098484848], "full": [0.627906977, 0.325581395, 0.666666667, 0.71969697], "retrieval": [0.728682171, 0.379844961, 0.712418301, 0.803030303], "rules_nodetect": [0.503875969, 0.224806202, 0.594771242, 0.378787879]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "63b47430528c7c673f35e8ad9db4219420155230259b404ff2c789eee375d0a3", "primary_control": "facts", "result_hash": "bf78af1f3cc0d32a3fb57a6b1788364658e96ffdb19184280d74ed2018ae6dfc", "signature": "8260cda790849300421cf75b7737fbe0fc6ec74ea2e8fd399d7f3594aaf09d25545dfd8522810294121219e8d063543673f0ff3dc9dfef6f214308622e0c3f07", "state": "VERIFIED", "summary": {"facts": 0.476843, "frozen": 0.096251, "full": 0.584963, "retrieval": 0.655994, "rules_nodetect": 0.42556}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:09:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:09:44Z", "cpu_seconds": 2.033768968000004, "created_at": "2026-11-22T04:09:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ddb6efd48fe6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "46f0465fe0970b4fdcd69c525d202c28d3be6029a7dbe6314b3ea0ac6a66c9e7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ddb6efd48fe6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:09:40Z", "seed": 1678688986, "treatment": "full"}, "priority": 0.2521936224662738, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "dc7b2404813ce0a39a7133f17be3d33590ec25fa460a870a595e32d4d5306cb8", "run_seconds": 2.092481851577759, "seed": 1678688986, "started_at": "2026-11-22T04:09:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.016967, 0.091181], "effect": 0.054074, "evaluated_at": "2026-11-22T04:09:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ddb6efd48fe6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.654166667, 0.875925926, 0.373333333, 0.886666667], "frozen": [0.0, 0.261111111, 0.19, 0.333333333], "full": [0.654166667, 0.947222222, 0.431666667, 0.973333333], "retrieval": [0.910416667, 0.984259259, 0.44, 0.98], "rules_nodetect": [0.654166667, 0.947222222, 0.216666667, 0.973333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "46f0465fe0970b4fdcd69c525d202c28d3be6029a7dbe6314b3ea0ac6a66c9e7", "primary_control": "facts", "result_hash": "dc7b2404813ce0a39a7133f17be3d33590ec25fa460a870a595e32d4d5306cb8", "signature": "549639bc10ec23f99c446094dd50666a80df137f8604c0af677715800b277e13bf5519172798c871830c908386dbc4da7536e90217e9d3fef83545d5322fb507", "state": "VERIFIED", "summary": {"facts": 0.697523, "frozen": 0.196111, "full": 0.751597, "retrieval": 0.828669, "rules_nodetect": 0.697847}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:09:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:52Z", "cpu_seconds": 0.6451553780000054, "created_at": "2026-11-22T04:09:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8a80b6767ccc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "c5bf72f4f2072e7241b0edd61a247317b92844a60bc8619344b8fbf9f682cc3c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8a80b6767ccc", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:09:56Z", "seed": 278462160, "treatment": "frontier"}, "priority": 0.23947938050141088, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "aaa6a9a1f3aa8656837433a5523e347fcb965712d8b325d23dae07debc6f7206", "run_seconds": 0.7318110466003418, "seed": 278462160, "started_at": "2026-11-22T04:11:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.004238, 0.032964], "effect": 0.018601, "evaluated_at": "2026-11-22T04:15:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-8a80b6767ccc", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.982142857, 0.979166667, 0.982142857, 1.0], "goal": [0.928571429, 0.895833333, 0.946428571, 1.0], "hypdisc": [0.928571429, 1.0, 0.964285714, 1.0], "infogain": [0.982142857, 0.979166667, 0.982142857, 1.0], "prederr": [0.982142857, 0.979166667, 0.982142857, 1.0], "random": [0.964285714, 0.958333333, 0.946428571, 1.0], "risk": [0.982142857, 0.979166667, 0.982142857, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c5bf72f4f2072e7241b0edd61a247317b92844a60bc8619344b8fbf9f682cc3c", "primary_control": "random", "result_hash": "aaa6a9a1f3aa8656837433a5523e347fcb965712d8b325d23dae07debc6f7206", "signature": "be0215f963c34bc26b1f525460dd5edf0e7587bc9e69b7ecae9cd9f47b4fd1ac4e8d003f555e6104636eea2dc2468ad09edb0f57889a886a2672c5eb233ce50b", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.985863, "goal": 0.942708, "hypdisc": 0.973214, "infogain": 0.985863, "prederr": 0.985863, "random": 0.967262, "risk": 0.985863}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:15:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:00Z", "cpu_seconds": 1.0686915360000029, "created_at": "2026-11-22T04:09:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d76fbed1bd6c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "d2366ebc3a31be7c8d0460f5781ccb22827bc765279487ae77edd4451b0662db", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d76fbed1bd6c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:09:56Z", "seed": 1643424194, "treatment": "frontier"}, "priority": 0.4098906044447772, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "38e26a39a030e8da4332281ce240d6c33e48cc85e592da56751e9319daabe92a", "run_seconds": 1.2266159057617188, "seed": 1643424194, "started_at": "2026-11-22T04:09:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.124938, 0.229858], "effect": 0.05246, "evaluated_at": "2026-11-22T04:10:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d76fbed1bd6c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.829457364, 0.775, 0.355555556, 0.862745098], "frontier": [0.689922481, 0.741666667, 0.488888889, 0.725490196], "goal": [0.720930233, 0.741666667, 0.377777778, 0.764705882], "hypdisc": [0.689922481, 0.5875, 0.733333333, 0.816993464], "infogain": [0.782945736, 0.741666667, 0.6, 0.816993464], "prederr": [0.689922481, 0.741666667, 0.333333333, 0.790849673], "random": [0.720930233, 0.425, 0.577777778, 0.712418301], "risk": [0.689922481, 0.741666667, 0.488888889, 0.725490196]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "d2366ebc3a31be7c8d0460f5781ccb22827bc765279487ae77edd4451b0662db", "primary_control": "random", "result_hash": "38e26a39a030e8da4332281ce240d6c33e48cc85e592da56751e9319daabe92a", "signature": "cae0f9bacef7143ce2273c8302aef7715f319cd9b5fb0b94f31195e872005b467a8bcf5389942aa25323e26e06d7b356766953d39bb90bfdd3eb49f50036d707", "state": "VERIFIED", "summary": {"count": 0.70569, "frontier": 0.661492, "goal": 0.65127, "hypdisc": 0.706937, "infogain": 0.735401, "prederr": 0.638943, "random": 0.609032, "risk": 0.661492}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:10:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:07Z", "cpu_seconds": 0.41128164600000616, "created_at": "2026-11-22T04:09:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-17be96875162", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "40efa6fb2f2d56ba1fb5f7945dbeb2fd4e75fce97cf28f9154e65feaa583d3aa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-17be96875162", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:09:56Z", "seed": 1084016940, "treatment": "full"}, "priority": 0.24803014029166054, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6eb1348e980cb71fa639144db7faef8bd9099b1837622cc7a011368b30038436", "run_seconds": 0.432248592376709, "seed": 1084016940, "started_at": "2026-11-22T04:10:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.68, 0.055], "effect": -0.3125, "evaluated_at": "2026-11-22T04:10:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-17be96875162", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.5, 0.25, 0.0], "full": [0.5, 1.0, 1.0, 0.0], "no_world_model": [0.0, 1.0, 1.0, 0.0], "random": [0.0, 1.0, 0.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "40efa6fb2f2d56ba1fb5f7945dbeb2fd4e75fce97cf28f9154e65feaa583d3aa", "primary_control": "e141_best", "result_hash": "6eb1348e980cb71fa639144db7faef8bd9099b1837622cc7a011368b30038436", "signature": "6c59fe5eef1b31230287529322732791a3290138a756213a055d9a3e8cdcdacbfa0b6ce29585559bdc721cdb1045a226d13f3a2e834f2b7701eabd175ba78208", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.1875, "full": 0.625, "no_world_model": 0.5, "random": 0.4375}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:10:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:22Z", "cpu_seconds": 1.1953540239999998, "created_at": "2026-11-22T04:09:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-317794a580d2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "60d1f3dc510fc2763699aa7367885361ae76025a3059813d624a4337f5a4d452", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-317794a580d2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:09:56Z", "seed": 707281625, "treatment": "full"}, "priority": 0.23912154787196238, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f8a37f410ba6d26582b52d3453e9e341da8e5ab0a6970b3522f34151f38c46d1", "run_seconds": 1.7695109844207764, "seed": 707281625, "started_at": "2026-11-22T04:12:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.055, 0.68], "effect": 0.3125, "evaluated_at": "2026-11-22T04:17:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-317794a580d2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 0.25], "fixed_policy": [0.75, 0.75, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "60d1f3dc510fc2763699aa7367885361ae76025a3059813d624a4337f5a4d452", "primary_control": "e141_best", "result_hash": "f8a37f410ba6d26582b52d3453e9e341da8e5ab0a6970b3522f34151f38c46d1", "signature": "d64e92360f71d85d6a4eff02feeb923b744f943e018ba3ab8c6c33d88b579d256af3df86f263d556b0ffede5606d87a9638f1263473e266a1b40e4ec44fc9f0d", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:17:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:02Z", "cpu_seconds": 0.6464322190000047, "created_at": "2026-11-22T04:09:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7a7dcf1b3d4a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "b45dbb8ca1d240847a6314b9ae5352b26a5f4258ffc3d372602c23d70967310e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7a7dcf1b3d4a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:09:56Z", "seed": 1796766144, "treatment": "full"}, "priority": 0.270621061433387, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "18fc0f5e95bb01b52e5bfda002cb580698a9c1635a838de7d414652250ee7066", "run_seconds": 0.7291984558105469, "seed": 1796766144, "started_at": "2026-11-22T04:10:01Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:10:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7a7dcf1b3d4a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.0, 0.25, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 0.25, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b45dbb8ca1d240847a6314b9ae5352b26a5f4258ffc3d372602c23d70967310e", "primary_control": "e141_best", "result_hash": "18fc0f5e95bb01b52e5bfda002cb580698a9c1635a838de7d414652250ee7066", "signature": "3cc42f493770f1e58bfaa181ee39ff398acc94858484fcf84d6b59f60f2cefb347b1dcd4b2cea60cbbe32d0c5b992b75bdf93e21fd929cdaa9c2ffaf46faf000", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 0.875, "random": 0.8125}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:10:03Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:08Z", "cpu_seconds": 2.9707688339999976, "created_at": "2026-11-22T04:09:56Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9adc0d9b6378", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "f4544d0dc0cb3f2518b5f927191338d0d7ad27a2bff737948b8e72d5e6501054", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9adc0d9b6378", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:09:56Z", "seed": 775311535, "treatment": "full"}, "priority": 0.2525214408427656, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "29a72fb7f3ac2a4847bb14b46e591e95e593ca86bb74ab16f706022fee6badf5", "run_seconds": 3.037627696990967, "seed": 775311535, "started_at": "2026-11-22T04:10:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.004726, 0.189501], "effect": 0.097113, "evaluated_at": "2026-11-22T04:10:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9adc0d9b6378", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.874285714, 0.264814815, 0.993333333, 0.408675799], "frozen": [0.365714286, 0.144444444, 0.433333333, 0.075342466], "full": [0.937142857, 0.366666667, 0.993333333, 0.632420091], "retrieval": [0.977142857, 0.442592593, 0.993333333, 0.655251142], "rules_nodetect": [0.937142857, 0.553703704, 0.533333333, 0.235159817]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "f4544d0dc0cb3f2518b5f927191338d0d7ad27a2bff737948b8e72d5e6501054", "primary_control": "facts", "result_hash": "29a72fb7f3ac2a4847bb14b46e591e95e593ca86bb74ab16f706022fee6badf5", "signature": "c4ea674a5ca478a833684774ce94196ad129393a6bd2e6d886277c1cf65e6afe214f010c81a95e21c2cb3f0ed2a007bd48801a10cd96ab53d94ed150dc289e0a", "state": "VERIFIED", "summary": {"facts": 0.635277, "frozen": 0.254709, "full": 0.732391, "retrieval": 0.76708, "rules_nodetect": 0.564835}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:10:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:13Z", "cpu_seconds": 0.8442906580000056, "created_at": "2026-11-22T04:10:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-39a0eef9defa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "bcdf80e4a919b41d42fd420d0fdacd743cc8a75b7ca3f08d5d14ca9f7e73d260", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-39a0eef9defa", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:10:11Z", "seed": 710558671, "treatment": "frontier"}, "priority": 0.3253189022669987, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fe4efe700475bfd285007f53d746fdbdb123e2ee0da14afe9309ffc568f424a1", "run_seconds": 0.8545913696289062, "seed": 710558671, "started_at": "2026-11-22T04:10:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035238, 0.011429], "effect": -0.011905, "evaluated_at": "2026-11-22T04:10:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-39a0eef9defa", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.717948718, 0.952380952, 0.925925926, 0.833333333], "frontier": [0.846153846, 0.904761905, 0.888888889, 0.833333333], "goal": [0.820512821, 0.952380952, 0.888888889, 0.833333333], "hypdisc": [0.846153846, 0.952380952, 0.888888889, 0.833333333], "infogain": [0.846153846, 0.904761905, 0.888888889, 0.833333333], "prederr": [0.794871795, 0.904761905, 0.888888889, 0.833333333], "random": [0.846153846, 0.952380952, 0.888888889, 0.833333333], "risk": [0.846153846, 0.904761905, 0.888888889, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "bcdf80e4a919b41d42fd420d0fdacd743cc8a75b7ca3f08d5d14ca9f7e73d260", "primary_control": "random", "result_hash": "fe4efe700475bfd285007f53d746fdbdb123e2ee0da14afe9309ffc568f424a1", "signature": "216f4acbcf4332a9efaaedbec635942a20e55652f10426555d99732c415c29e26c0a416a59cfbbd5b77dcaa916648f203019b9891b104e2d6d839a3a071fda0e", "state": "VERIFIED", "summary": {"count": 0.857397, "frontier": 0.868284, "goal": 0.873779, "hypdisc": 0.880189, "infogain": 0.868284, "prederr": 0.855464, "random": 0.880189, "risk": 0.868284}, "ties": 3, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:10:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:23Z", "cpu_seconds": 1.3458161339999961, "created_at": "2026-11-22T04:10:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4ca627699b68", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 2, "n": 4}, "prereg_hash": "68ad75713d33dba879c0dc8f4b95f9dcd042cd32b8304600ead1a5987e9b66f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4ca627699b68", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:10:11Z", "seed": 351983352, "treatment": "full"}, "priority": 0.24512462172814437, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0a092f1ccb3630d2509868e0fdbea9b28adb10ca33bd3ad52cf549d38f56b74c", "run_seconds": 1.397010087966919, "seed": 351983352, "started_at": "2026-11-22T04:10:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.06054, 0.192089], "effect": 0.126315, "evaluated_at": "2026-11-22T04:10:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4ca627699b68", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.219569472, 0.17964775, 0.221917808, 0.948387097], "frozen": [0.078277886, 0.03444227, 0.078277886, 0.2], "full": [0.384735812, 0.33776908, 0.378082192, 0.974193548], "retrieval": [0.446575342, 0.359686888, 0.450097847, 0.961290323], "rules_nodetect": [0.384735812, 0.33776908, 0.378082192, 0.974193548]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "68ad75713d33dba879c0dc8f4b95f9dcd042cd32b8304600ead1a5987e9b66f4", "primary_control": "facts", "result_hash": "0a092f1ccb3630d2509868e0fdbea9b28adb10ca33bd3ad52cf549d38f56b74c", "signature": "ef10a98602a436d19af76814583c1dd0aec23d6452cdb10cd996592a3c53788639da01f4bd00718406ecda49105336b2dd7d056eb12b5e8b8754593bab5bf703", "state": "VERIFIED", "summary": {"facts": 0.392381, "frozen": 0.09775, "full": 0.518695, "retrieval": 0.554413, "rules_nodetect": 0.518695}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:10:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:16Z", "cpu_seconds": 2.7580440690000003, "created_at": "2026-11-22T04:10:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-874131ac557f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "3a81d33b2e7a692f4e3cd817ae4ca5fc2ed56ffd83be1d0d1abb5636b1cd789f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-874131ac557f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:10:11Z", "seed": 138639200, "treatment": "full"}, "priority": 0.2660932529059659, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "957431c93db15c6da20b699523f8e6fb2813e51807beac531e272f679fd5bb39", "run_seconds": 2.9921817779541016, "seed": 138639200, "started_at": "2026-11-22T04:10:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.027762, 0.103772], "effect": 0.038005, "evaluated_at": "2026-11-22T04:10:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-874131ac557f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.933333333, 0.395454545, 0.983333333, 0.933333333], "frozen": [0.333333333, 0.198484848, 0.366666667, 0.455555556], "full": [1.0, 0.503030303, 0.933333333, 0.961111111], "retrieval": [1.0, 0.672727273, 0.941666667, 0.977777778], "rules_nodetect": [1.0, 0.356060606, 0.416666667, 0.961111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "3a81d33b2e7a692f4e3cd817ae4ca5fc2ed56ffd83be1d0d1abb5636b1cd789f", "primary_control": "facts", "result_hash": "957431c93db15c6da20b699523f8e6fb2813e51807beac531e272f679fd5bb39", "signature": "e4daa359c2c7122ae9e01ef29be42530cb33bb302671bfcb6340f5447f6af56f3bb986ff41f127db31bf54ce1ecd3404bb63dcc3ee914ff6474ee0f71858c90b", "state": "VERIFIED", "summary": {"facts": 0.811364, "frozen": 0.33851, "full": 0.849369, "retrieval": 0.898043, "rules_nodetect": 0.68346}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:10:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:19Z", "cpu_seconds": 0.6844049439999935, "created_at": "2026-11-22T04:10:11Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e0f148be9c48", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "94fb5e4fbd9fd69723c00188abc5e38080a92d715a636eec79555581c6a6ae21", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e0f148be9c48", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:10:11Z", "seed": 1311818434, "treatment": "gated"}, "priority": 0.24666295080872694, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "b11fe6480eb299ccc2aa28de66228d0297eca5ebfb86173dae23eff0dd5975f6", "run_seconds": 0.7186784744262695, "seed": 1311818434, "started_at": "2026-11-22T04:10:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0592, 0.0192], "effect": -0.02, "evaluated_at": "2026-11-22T04:10:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-e0f148be9c48", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.410852713, 0.274509804, 0.215686275, 0.326666667], "gated": [0.410852713, 0.274509804, 0.215686275, 0.326666667], "naive": [0.271317829, 0.261437908, 0.261437908, 0.266666667], "scratch": [0.410852713, 0.274509804, 0.215686275, 0.406666667], "ungated": [0.356589147, 0.431372549, 0.215686275, 0.266666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "94fb5e4fbd9fd69723c00188abc5e38080a92d715a636eec79555581c6a6ae21", "primary_control": "scratch", "result_hash": "b11fe6480eb299ccc2aa28de66228d0297eca5ebfb86173dae23eff0dd5975f6", "signature": "b21b6052f161c6f430a5082aa22135a20e9a14becfded13f7bd105cd09d0b23a0ea1b2c16aa3543d725c9844a3ecc4b6ca64b5e7e20da7c6ff301832b454cf00", "state": "VERIFIED", "summary": {"compressed": 0.306929, "gated": 0.306929, "naive": 0.265215, "scratch": 0.326929, "ungated": 0.317579}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:10:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:40Z", "cpu_seconds": 1.2690560369999986, "created_at": "2026-11-22T04:10:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e9ae22ad5e74", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "64cf2849b3a235fdd1576b200bab2219c1759b8c6fa07e89b27d28bfdf6bcce9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e9ae22ad5e74", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:10:28Z", "seed": 427397235, "treatment": "frontier"}, "priority": 0.2626445858652412, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-b4d8d0cb8ae9", "result_hash": "65edf2b5ed58b6300de562e252541ba7ba3e8d69916e33d8a9ea25ced9ed7352", "run_seconds": 1.3237335681915283, "seed": 427397235, "started_at": "2026-11-22T04:10:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.03011, 0.098414], "effect": 0.034152, "evaluated_at": "2026-11-22T04:10:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e9ae22ad5e74", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.566901408, 0.693181818, 0.928571429, 0.725490196], "frontier": [0.704225352, 0.698863636, 0.892857143, 0.764705882], "goal": [0.693661972, 0.676136364, 0.75, 0.764705882], "hypdisc": [0.563380282, 0.613636364, 0.660714286, 0.764705882], "infogain": [0.704225352, 0.210227273, 0.892857143, 0.666666667], "prederr": [0.704225352, 0.596590909, 0.892857143, 0.745098039], "random": [0.714788732, 0.715909091, 0.767857143, 0.725490196], "risk": [0.704225352, 0.698863636, 0.892857143, 0.764705882]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "64cf2849b3a235fdd1576b200bab2219c1759b8c6fa07e89b27d28bfdf6bcce9", "primary_control": "random", "replicated": false, "replication_of": "E141-ow_explore-b4d8d0cb8ae9", "result_hash": "65edf2b5ed58b6300de562e252541ba7ba3e8d69916e33d8a9ea25ced9ed7352", "signature": "868ce44a57c660d8942c4ab28b0551e03f81bfe35e3b8a7160d1c65e17f7339e8e98a084c6bdc6f7fe5f4536c4283f5c00e6c888378a891fac71ec883ddcb106", "state": "VERIFIED", "summary": {"count": 0.728536, "frontier": 0.765163, "goal": 0.721126, "hypdisc": 0.650609, "infogain": 0.618494, "prederr": 0.734693, "random": 0.731011, "risk": 0.765163}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:10:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:00Z", "cpu_seconds": 1.1089337500000056, "created_at": "2026-11-22T04:10:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4b63b40607c6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "3500c000a3c5c03b4a9fd478bf81562bf7903688b7b3ed9bf97ed994b973088b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4b63b40607c6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:10:28Z", "seed": 940641276, "treatment": "full"}, "priority": 0.24303191107327432, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7965920efc72e224e1cf7f8d7d83fa6d8b670c0de2ece880b7306dbf650571a2", "run_seconds": 1.1428883075714111, "seed": 940641276, "started_at": "2026-11-22T04:10:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013151, 0.203953], "effect": 0.095401, "evaluated_at": "2026-11-22T04:11:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4b63b40607c6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.828571429, 0.857142857, 0.195694716, 0.221624266], "frozen": [0.0, 0.142857143, 0.011741683, 0.039138943], "full": [0.828571429, 0.857142857, 0.400684932, 0.398238748], "retrieval": [0.85, 0.964285714, 0.393835616, 0.433953033], "rules_nodetect": [0.828571429, 0.857142857, 0.400684932, 0.398238748]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "3500c000a3c5c03b4a9fd478bf81562bf7903688b7b3ed9bf97ed994b973088b", "primary_control": "facts", "result_hash": "7965920efc72e224e1cf7f8d7d83fa6d8b670c0de2ece880b7306dbf650571a2", "signature": "36c0593e0e4205038f28f626dd4b705afe3cd995be0239b3590b2d2226c405dcc8a10fde97c9a4d94d07d18a437b4be58dc0731ddd4e98cb8fc3f38b2dc48a07", "state": "VERIFIED", "summary": {"facts": 0.525758, "frozen": 0.048434, "full": 0.621159, "retrieval": 0.660519, "rules_nodetect": 0.621159}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:11:04Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:39Z", "cpu_seconds": 3.2628077039999965, "created_at": "2026-11-22T04:10:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b88fc1e46d79", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "03949aba7d996715adbe5450722626bf6b0f0cd38adf9266c262d0d2a6878de1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b88fc1e46d79", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:10:28Z", "seed": 87782041, "treatment": "full"}, "priority": 0.28342816371628554, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "056acd55e91935602adf4c738c58d0183715d3194a13b24db0e237720151a1ad", "run_seconds": 3.3520188331604004, "seed": 87782041, "started_at": "2026-11-22T04:10:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.027594, 0.264017], "effect": 0.118211, "evaluated_at": "2026-11-22T04:10:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b88fc1e46d79", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.871794872, 0.524761905, 0.734848485, 0.593023256], "frozen": [0.0, 0.308571429, 0.325757576, 0.151162791], "full": [0.871794872, 0.836190476, 0.893939394, 0.595348837], "retrieval": [0.935897436, 0.85047619, 0.946969697, 0.595348837], "rules_nodetect": [0.871794872, 0.301904762, 0.503787879, 0.325581395]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "03949aba7d996715adbe5450722626bf6b0f0cd38adf9266c262d0d2a6878de1", "primary_control": "facts", "result_hash": "056acd55e91935602adf4c738c58d0183715d3194a13b24db0e237720151a1ad", "signature": "e1f83c94a4ed377488a6cf19b4631beb4aa381daba0b6c469264ae41ae0bc18371c475bd1d1bc42046ff83b59dd27b0b445302b0c8600d89ddafed648c306402", "state": "VERIFIED", "summary": {"facts": 0.681107, "frozen": 0.196373, "full": 0.799318, "retrieval": 0.832173, "rules_nodetect": 0.500767}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:10:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:34Z", "cpu_seconds": 0.6199837610000003, "created_at": "2026-11-22T04:10:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f63c80e6ae9d", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "43b1b7ff88e90d1b6c655a60d1b89ba9a5b3d5427637f0877fa055d00adbe7ca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f63c80e6ae9d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:10:28Z", "seed": 661524158, "treatment": "full"}, "priority": 0.28866884579574426, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-fb6e70252d06", "result_hash": "6bfd95a959a60118e8e84f3603ec405ed3930ea509a1ff6e0a95c1bafebefa64", "run_seconds": 0.6807351112365723, "seed": 661524158, "started_at": "2026-11-22T04:10:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004161, 0.064346], "effect": 0.030093, "evaluated_at": "2026-11-22T04:10:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f63c80e6ae9d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.846153846, 0.744444444, 0.574074074, 0.790262172], "frozen": [0.358974359, 0.311111111, 0.342592593, 0.307116105], "full": [0.846153846, 0.8, 0.638888889, 0.790262172], "retrieval": [0.846153846, 0.733333333, 0.675925926, 0.790262172], "rules_nodetect": [0.846153846, 0.8, 0.759259259, 0.790262172]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "43b1b7ff88e90d1b6c655a60d1b89ba9a5b3d5427637f0877fa055d00adbe7ca", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-fb6e70252d06", "result_hash": "6bfd95a959a60118e8e84f3603ec405ed3930ea509a1ff6e0a95c1bafebefa64", "signature": "48b3386673697aa41b7d5851cd09a6c51852a382750d3ca85f92c15bf768a68f59e8c27cd2adcf6e2463c57212625f60dab3d107dc63a71fc0ea0ddeca93080a", "state": "VERIFIED", "summary": {"facts": 0.738734, "frozen": 0.329949, "full": 0.768826, "retrieval": 0.761419, "rules_nodetect": 0.798919}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:10:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:45Z", "cpu_seconds": 0.6644563609999921, "created_at": "2026-11-22T04:10:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-c94d1823b792", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "bf175084dd35a993f950b8400d44ade70984a4d97ad3abb10b1fa589884abfca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-c94d1823b792", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:10:28Z", "seed": 552533667, "treatment": "gated"}, "priority": 0.24727733828862725, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-81a45483ab66", "result_hash": "910ac6a7dad7e1f4948340dae8070ab27c33239ca5bf2391b464d62d14592f27", "run_seconds": 0.6909277439117432, "seed": 552533667, "started_at": "2026-11-22T04:10:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.051105, 0.011105], "effect": -0.02, "evaluated_at": "2026-11-22T04:10:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-c94d1823b792", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.313333333, 0.503875969, 0.253333333, 0.333333333], "gated": [0.313333333, 0.503875969, 0.253333333, 0.333333333], "naive": [0.24, 0.503875969, 0.193333333, 0.271317829], "scratch": [0.326666667, 0.503875969, 0.32, 0.333333333], "ungated": [0.326666667, 0.503875969, 0.193333333, 0.356589147]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "bf175084dd35a993f950b8400d44ade70984a4d97ad3abb10b1fa589884abfca", "primary_control": "scratch", "replicated": false, "replication_of": "E141-ow_transfer-81a45483ab66", "result_hash": "910ac6a7dad7e1f4948340dae8070ab27c33239ca5bf2391b464d62d14592f27", "signature": "99ed387117a3fd054e62b6a452ed8e3d55eb0ba85bae9f0b25c7199aa02983a730f72695f4c039e21aa59377704304fdf79e500a0bb6cd5ffa2257baaa2dd708", "state": "VERIFIED", "summary": {"compressed": 0.350969, "gated": 0.350969, "naive": 0.302132, "scratch": 0.370969, "ungated": 0.345116}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:10:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:31Z", "cpu_seconds": 0.5460679139999911, "created_at": "2026-11-22T04:10:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-cc8ab9f79c80", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "107d7fcb534ba4afb54a7cb45e7aedd54a5a31330cf2eedabaaef4f82a6d58cc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-cc8ab9f79c80", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:10:28Z", "seed": 971306533, "treatment": "gated"}, "priority": 0.4052956377531799, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f8c0b0d687445fa7e2788c42861ea9b042330ea5e98da2549e7b446ddbf4ba34", "run_seconds": 0.5597357749938965, "seed": 971306533, "started_at": "2026-11-22T04:10:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.103059, -0.036709], "effect": -0.069884, "evaluated_at": "2026-11-22T04:10:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-cc8ab9f79c80", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.333333333, 0.411764706, 0.356589147, 0.209302326], "gated": [0.333333333, 0.392156863, 0.356589147, 0.209302326], "naive": [0.25170068, 0.130718954, 0.294573643, 0.147286822], "scratch": [0.414965986, 0.411764706, 0.449612403, 0.294573643], "ungated": [0.25170068, 0.392156863, 0.294573643, 0.147286822]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "107d7fcb534ba4afb54a7cb45e7aedd54a5a31330cf2eedabaaef4f82a6d58cc", "primary_control": "scratch", "result_hash": "f8c0b0d687445fa7e2788c42861ea9b042330ea5e98da2549e7b446ddbf4ba34", "signature": "9581751ba500301b885c25961e0619e97656e6ecb7d4bdbf43370a9824b78efcb12555062ac290e4da3f427664ceb62ab344bdc14b1e1f0f1640ab58d4481904", "state": "VERIFIED", "summary": {"compressed": 0.327747, "gated": 0.322845, "naive": 0.20607, "scratch": 0.392729, "ungated": 0.27143}, "ties": 0, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:10:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:59Z", "cpu_seconds": 1.0716220649999997, "created_at": "2026-11-22T04:10:45Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-63749236ec17", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "a2c511eda40240fe8450b3b4f00a28cb73e5789e3f2191aa2ff444fc81864803", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-63749236ec17", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:10:45Z", "seed": 560040621, "treatment": "frontier"}, "priority": 0.27284011378174877, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a5990daf15b0293fe7a2809c3f6d070eb4e55028f20dfb7021de42818452ecc9", "run_seconds": 1.1332859992980957, "seed": 560040621, "started_at": "2026-11-22T04:10:58Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:11:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-63749236ec17", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.661412], "frontier": [1.0, 0.737567], "goal": [1.0, 0.696934], "hypdisc": [1.0, 0.7273], "infogain": [1.0, 0.751456], "prederr": [1.0, 0.744512], "random": [1.0, 0.596233], "risk": [1.0, 0.737567]}, "prereg_hash": "a2c511eda40240fe8450b3b4f00a28cb73e5789e3f2191aa2ff444fc81864803", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "a5990daf15b0293fe7a2809c3f6d070eb4e55028f20dfb7021de42818452ecc9", "signature": "07f0f55332bfdbb635a1cc4fcf68ec898c809dacc5a7549fc2798628c4cddb0222680da46928e784cb0d44a73d210e536bf096d95f7ab56ed53e64d346226604", "state": "REJECTED", "summary": {"count": 0.661412, "frontier": 0.737567, "goal": 0.696934, "hypdisc": 0.7273, "infogain": 0.751456, "prederr": 0.744512, "random": 0.596233, "risk": 0.737567}}, "verified_at": "2026-11-22T04:11:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:34Z", "cpu_seconds": 0.5319121909999964, "created_at": "2026-11-22T04:10:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1ebd204da7a1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "5535ca4d0d5d21ffcf802097a0c1ab318dc95bba805d7305e994d7eada8e4b8c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1ebd204da7a1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:10:45Z", "seed": 1557113217, "treatment": "full"}, "priority": 0.24038932717325767, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "87e258184bb751b61239d02e49cbd390fb9c590e4eb3e717bafe4db9e09404d4", "run_seconds": 0.5415408611297607, "seed": 1557113217, "started_at": "2026-11-22T04:11:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:11:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1ebd204da7a1", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 1.0, 0.75], "fixed_policy": [0.25, 0.0, 0.25, 0.25], "full": [0.25, 0.5, 1.0, 1.0], "no_world_model": [0.25, 0.0, 1.0, 0.5], "random": [0.25, 0.5, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "5535ca4d0d5d21ffcf802097a0c1ab318dc95bba805d7305e994d7eada8e4b8c", "primary_control": "e141_best", "result_hash": "87e258184bb751b61239d02e49cbd390fb9c590e4eb3e717bafe4db9e09404d4", "signature": "aa0827606559fb6c8ebaa4c8ca9b44c49cbbdb3839842b6fe4017bdc6d3e72494de55cee613c927a73cdff198d77620e350c5ae24ae2128e3dcbcc20c9f0b407", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.1875, "full": 0.6875, "no_world_model": 0.4375, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:11:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:53Z", "cpu_seconds": 3.1820050460000004, "created_at": "2026-11-22T04:10:45Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3d5528d9d45b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "5f85cb89e8f5f29feeca3f308b16c7ffe8a8eb9c2c7eca1453ea8ec9c2708900", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3d5528d9d45b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:10:45Z", "seed": 1932233241, "treatment": "full"}, "priority": 0.2926118510953313, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2bf3e4355daf48e25245852329aaff07a82aa85566cd7452b05ae9cf600d05e3", "run_seconds": 3.2939531803131104, "seed": 1932233241, "started_at": "2026-11-22T04:10:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.065192, 0.24359], "effect": 0.154391, "evaluated_at": "2026-11-22T04:10:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3d5528d9d45b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.628148148, 0.474226804, 0.873333333, 0.841423948], "frozen": [0.366666667, 0.129896907, 0.38, 0.310679612], "full": [0.853333333, 0.713402062, 0.933333333, 0.934627832], "retrieval": [0.926666667, 0.740206186, 0.92, 0.980582524], "rules_nodetect": [0.853333333, 0.284536082, 0.486666667, 0.934627832]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "5f85cb89e8f5f29feeca3f308b16c7ffe8a8eb9c2c7eca1453ea8ec9c2708900", "primary_control": "facts", "result_hash": "2bf3e4355daf48e25245852329aaff07a82aa85566cd7452b05ae9cf600d05e3", "signature": "831d3dd2a93c3fde9dd12c53613e00463785701641977f741090cec7a0c8bac30dbc4e518ab57fc09dd80ac4f23c330ecee300057299ab722b81ab439567910b", "state": "VERIFIED", "summary": {"facts": 0.704283, "frozen": 0.296811, "full": 0.858674, "retrieval": 0.891864, "rules_nodetect": 0.639791}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:10:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:46Z", "cpu_seconds": 0.6255356400000025, "created_at": "2026-11-22T04:10:45Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-36c436685797", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "d6bfad3f24f6368235ac9c3d7364c0ce997ffe6530461c0a4b72d6ea39c2e8d7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-36c436685797", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:10:45Z", "seed": 1041536392, "treatment": "gated"}, "priority": 0.29906801424192186, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "1e12041fa2cdd9d85ab463a78a0bd371565a4d6248426f942979ffb28d4a28a4", "run_seconds": 0.664804220199585, "seed": 1041536392, "started_at": "2026-11-22T04:10:45Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:10:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-36c436685797", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.386363636, 0.405228758, 0.427083333, 0.34375], "gated": [0.386363636, 0.405228758, 0.427083333, 0.34375], "naive": [0.159090909, 0.405228758, 0.40625, 0.21875], "scratch": [0.386363636, 0.405228758, 0.427083333, 0.34375], "ungated": [0.393939394, 0.261437908, 0.322916667, 0.270833333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d6bfad3f24f6368235ac9c3d7364c0ce997ffe6530461c0a4b72d6ea39c2e8d7", "primary_control": "scratch", "result_hash": "1e12041fa2cdd9d85ab463a78a0bd371565a4d6248426f942979ffb28d4a28a4", "signature": "8534245568dd415530204c1195b41863b63d9cf6fbf47bd06704961aae7aa9bd30e7ae5c445637530fcc3db48d8139fd6dd19a6fa63da03cb1411d9e59c60207", "state": "PARTIAL", "summary": {"compressed": 0.390606, "gated": 0.390606, "naive": 0.29733, "scratch": 0.390606, "ungated": 0.312282}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:10:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:10:54Z", "cpu_seconds": 2.730457088999998, "created_at": "2026-11-22T04:10:45Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-66291aa9d71a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "d79f822fe901b755a03a49668ed02a470f33751d80e2733c5c3101f80b8e1e78", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-66291aa9d71a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:10:45Z", "seed": 581426718, "treatment": "gated"}, "priority": 0.28723857233724626, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "2cd036fc8f0a2848371cb6b327a61e9296a42b2cfeadc6f59a2ed85de4eac503", "run_seconds": 2.8969626426696777, "seed": 581426718, "started_at": "2026-11-22T04:10:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.058433, 0.095636], "effect": 0.018601, "evaluated_at": "2026-11-22T04:10:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-66291aa9d71a", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.45, 0.55952381, 0.5, 0.580357143], "gated": [0.45, 0.55952381, 0.5, 0.580357143], "naive": [0.691666667, 0.714285714, 0.770833333, 0.723214286], "scratch": [0.45, 0.464285714, 0.583333333, 0.517857143], "ungated": [0.558333333, 0.5, 0.770833333, 0.455357143]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d79f822fe901b755a03a49668ed02a470f33751d80e2733c5c3101f80b8e1e78", "primary_control": "scratch", "result_hash": "2cd036fc8f0a2848371cb6b327a61e9296a42b2cfeadc6f59a2ed85de4eac503", "signature": "eda7574a2d91b4102c6b2a657f8533d707322344aff61b8bce537f374194a9a9c81b63d002f93c859e6061b3f7bc9ed3981196990b60c38b6379db09db7e4503", "state": "VERIFIED", "summary": {"compressed": 0.52247, "gated": 0.52247, "naive": 0.725, "scratch": 0.503869, "ungated": 0.571131}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:10:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:17Z", "cpu_seconds": 4.322467493000005, "created_at": "2026-11-22T04:11:01Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8411f476b0a4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "4a3162063deceda5e4880a1e12a7afc2764a65d94733f81eba48233657e68a6c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8411f476b0a4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:11:01Z", "seed": 111572766, "treatment": "full"}, "priority": 0.2756058856116305, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "887d4ad6897ec62600562aa6df706a0272f6eaeaab01dcb166448be5a6e6d248", "run_seconds": 4.6005589962005615, "seed": 111572766, "started_at": "2026-11-22T04:11:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.156142, 0.413419], "effect": 0.284781, "evaluated_at": "2026-11-22T04:11:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8411f476b0a4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.455555556, 0.89, 0.394389439, 0.352336449], "frozen": [0.347619048, 0.33, 0.322332233, 0.309345794], "full": [0.794708995, 0.98, 0.727722772, 0.728971963], "retrieval": [0.811111111, 0.965, 0.792079208, 0.814953271], "rules_nodetect": [0.794708995, 0.27, 0.727722772, 0.408411215]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "4a3162063deceda5e4880a1e12a7afc2764a65d94733f81eba48233657e68a6c", "primary_control": "facts", "result_hash": "887d4ad6897ec62600562aa6df706a0272f6eaeaab01dcb166448be5a6e6d248", "signature": "5fe4a1a707abb816b97f631ffbc57d262dc5d1bdd82cf0e774191997e7d5bc442bd7e0add7ef3ef590ba696e2b8d06b87dceabc9134455b40554c5ac5693890c", "state": "VERIFIED", "summary": {"facts": 0.52307, "frozen": 0.327324, "full": 0.807851, "retrieval": 0.845786, "rules_nodetect": 0.550211}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:11:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:18Z", "cpu_seconds": 0.6372565759999986, "created_at": "2026-11-22T04:11:01Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3c5792e943ef", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "8cfa9331e83d50c9a04f803f92e971e303cfd71a0527afe7d9dcc622291fc1ce", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3c5792e943ef", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:01Z", "seed": 1332281859, "treatment": "gated"}, "priority": 0.24656648662284608, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-a574fa9240e5", "result_hash": "997987d1f7040f026f8143ab586e2f4f191e4b3d6965d1bc49e331c883e22b69", "run_seconds": 0.7133162021636963, "seed": 1332281859, "started_at": "2026-11-22T04:11:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.009673, 0.003137], "effect": -0.003268, "evaluated_at": "2026-11-22T04:11:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-3c5792e943ef", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.392156863, 0.248366013, 0.244897959, 0.202614379], "gated": [0.392156863, 0.248366013, 0.244897959, 0.202614379], "naive": [0.254901961, 0.248366013, 0.156462585, 0.254901961], "scratch": [0.392156863, 0.248366013, 0.244897959, 0.215686275], "ungated": [0.339869281, 0.202614379, 0.285714286, 0.169934641]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "8cfa9331e83d50c9a04f803f92e971e303cfd71a0527afe7d9dcc622291fc1ce", "primary_control": "scratch", "replicated": false, "replication_of": "E141-ow_transfer-a574fa9240e5", "result_hash": "997987d1f7040f026f8143ab586e2f4f191e4b3d6965d1bc49e331c883e22b69", "signature": "ccab10e2bb0e2376a794405a3700007e323bb0cf574da7bf9c102ee57cd3ce229165b745f08064186757a8f0fa7d7b35d958bc2359a4bf39a3bfd9c0c2ed750a", "state": "VERIFIED", "summary": {"compressed": 0.272009, "gated": 0.272009, "naive": 0.228658, "scratch": 0.275277, "ungated": 0.249533}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:11:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:12Z", "cpu_seconds": 0.6209323080000004, "created_at": "2026-11-22T04:11:01Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-44d6542bdbec", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "a4621d2566a9ca7344194e8e31c322320356c321e6c550766ceae0aa5c9ecf82", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-44d6542bdbec", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:01Z", "seed": 1483647741, "treatment": "gated"}, "priority": 0.2857647454704956, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "179ce4db0787ed087b9659235af559768d396466cc7de8439215c751a77128ae", "run_seconds": 0.6491572856903076, "seed": 1483647741, "started_at": "2026-11-22T04:11:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003137, 0.009673], "effect": 0.003268, "evaluated_at": "2026-11-22T04:11:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-44d6542bdbec", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.290780142, 0.302083333, 0.240310078, 0.196078431], "gated": [0.290780142, 0.302083333, 0.240310078, 0.196078431], "naive": [0.19858156, 0.302083333, 0.209302326, 0.183006536], "scratch": [0.290780142, 0.302083333, 0.240310078, 0.183006536], "ungated": [0.276595745, 0.302083333, 0.209302326, 0.183006536]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a4621d2566a9ca7344194e8e31c322320356c321e6c550766ceae0aa5c9ecf82", "primary_control": "scratch", "result_hash": "179ce4db0787ed087b9659235af559768d396466cc7de8439215c751a77128ae", "signature": "7559f07cbe8fe7766579b006f3fa2d9c315b8a57c2b1746c87cfa1114ba4033900e2217f9a989803cfa3b8dee776dd744956cad720d802fd302b06128666de00", "state": "VERIFIED", "summary": {"compressed": 0.257313, "gated": 0.257313, "naive": 0.223243, "scratch": 0.254045, "ungated": 0.242747}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:11:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:06Z", "cpu_seconds": 0.6616158200000015, "created_at": "2026-11-22T04:11:01Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-622940b58c49", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "0579583a6806672e91d103e14a5eff12db1fa8b1320343cae3bc28e83e72041a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-622940b58c49", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:01Z", "seed": 1750683223, "treatment": "gated"}, "priority": 0.3059659843920886, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "ab4d326c5e548dcb0f4ce88330d691830e8cd32fc539dad1b8e599c0bb606e68", "run_seconds": 0.7100057601928711, "seed": 1750683223, "started_at": "2026-11-22T04:11:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.011417, 0.006934], "effect": -0.002242, "evaluated_at": "2026-11-22T04:11:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-622940b58c49", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.356589147, 0.302325581, 0.254901961, 0.333333333], "gated": [0.356589147, 0.302325581, 0.254901961, 0.333333333], "naive": [0.356589147, 0.294573643, 0.254901961, 0.254901961], "scratch": [0.356589147, 0.317829457, 0.254901961, 0.326797386], "ungated": [0.333333333, 0.279069767, 0.307189542, 0.333333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "0579583a6806672e91d103e14a5eff12db1fa8b1320343cae3bc28e83e72041a", "primary_control": "scratch", "result_hash": "ab4d326c5e548dcb0f4ce88330d691830e8cd32fc539dad1b8e599c0bb606e68", "signature": "71cfacabcfae96c13448ca9a8a0e597c1c8c1449ab36c1b0550180b317617698a52d1a82450fb7e0b9bee4cb302a39fec8d3d705a398545225de191350833e07", "state": "VERIFIED", "summary": {"compressed": 0.311788, "gated": 0.311788, "naive": 0.290242, "scratch": 0.314029, "ungated": 0.313231}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:11:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:08Z", "cpu_seconds": 3.0103407809999965, "created_at": "2026-11-22T04:11:01Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a589a04a3ffa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "0825525c69440e0148f7ce14166d5e4a52d0fe756f40c874f7473346049b7df3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a589a04a3ffa", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:01Z", "seed": 1858040273, "treatment": "gated"}, "priority": 0.31302508016866254, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d2192212ce5704de6af3924eb9d3bafdb5228a356e2b521a1791c345925635ce", "run_seconds": 3.135878562927246, "seed": 1858040273, "started_at": "2026-11-22T04:11:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.098344, 0.061089], "effect": -0.018627, "evaluated_at": "2026-11-22T04:11:08Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a589a04a3ffa", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.705882353, 0.557692308, 0.580357143, 0.416666667], "gated": [0.705882353, 0.557692308, 0.580357143, 0.416666667], "naive": [0.705882353, 0.615384615, 0.767857143, 0.758333333], "scratch": [0.647058824, 0.557692308, 0.580357143, 0.55], "ungated": [0.647058824, 0.615384615, 0.651785714, 0.65]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "0825525c69440e0148f7ce14166d5e4a52d0fe756f40c874f7473346049b7df3", "primary_control": "scratch", "result_hash": "d2192212ce5704de6af3924eb9d3bafdb5228a356e2b521a1791c345925635ce", "signature": "1a2d5440b210a25a03b80cc3d34d26dcecb34d6744f927f2cb4f444aaae614b5f9d8fd5c6b8dd94218d4d940eb1821aaaba3b4faf55fd99809371c200f3a7207", "state": "VERIFIED", "summary": {"compressed": 0.56515, "gated": 0.56515, "naive": 0.711864, "scratch": 0.583777, "ungated": 0.641057}, "ties": 2, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:11:08Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:31Z", "cpu_seconds": 0.7079384249999947, "created_at": "2026-11-22T04:11:18Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9d8f78eabfa6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "e1dbb6e7fc2b45519d39a076a776849c9a9b600369621cebaa58243e8933b946", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9d8f78eabfa6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:11:18Z", "seed": 734458357, "treatment": "full"}, "priority": 0.24248869140662674, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e0c302667f8993245f8a440cd5d2018bae39da75cb88a1eda9a8294d9c2514b3", "run_seconds": 0.7858304977416992, "seed": 734458357, "started_at": "2026-11-22T04:11:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.12, 0.37], "effect": 0.125, "evaluated_at": "2026-11-22T04:11:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9d8f78eabfa6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "e1dbb6e7fc2b45519d39a076a776849c9a9b600369621cebaa58243e8933b946", "primary_control": "e141_best", "result_hash": "e0c302667f8993245f8a440cd5d2018bae39da75cb88a1eda9a8294d9c2514b3", "signature": "2459b4a683067cdd6690adb82bb202068d2057facabd7ec55c6e092d51348afae806e247ee0dba332e2d097592efa4e22c03895ca8a11abcdd19e43d7b3e4a02", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.0625, "full": 0.9375, "no_world_model": 1.0, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:11:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:29Z", "cpu_seconds": 0.44601091500000223, "created_at": "2026-11-22T04:11:18Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a10eb6f32cc9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "cabe51f8ecb697a5549b8b45c9c12871ab3664d2ace9d32a33470bc1ef3bc066", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a10eb6f32cc9", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:11:18Z", "seed": 75466431, "treatment": "full"}, "priority": 0.2450677088019574, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d507a99f1db3f8c373bd2cefec82ed3ca682dc42142ec5e12fab9a3786fffbed", "run_seconds": 0.4633004665374756, "seed": 75466431, "started_at": "2026-11-22T04:11:28Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:11:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a10eb6f32cc9", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.5, 0.75, 0.25], "full": [0.25, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 1.0, 1.0], "random": [0.25, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "cabe51f8ecb697a5549b8b45c9c12871ab3664d2ace9d32a33470bc1ef3bc066", "primary_control": "e141_best", "result_hash": "d507a99f1db3f8c373bd2cefec82ed3ca682dc42142ec5e12fab9a3786fffbed", "signature": "69402dd305c48260915246357368c2af3b8cdfbe6fa8abfc8602acdcf7d1548e35dd141ba5d0389d1ca07322ca9ed68f02f12eeb4987b8c5247027070f72e20c", "state": "PARTIAL", "summary": {"e141_best": 0.8125, "fixed_policy": 0.375, "full": 0.8125, "no_world_model": 0.75, "random": 0.8125}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:11:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:25Z", "cpu_seconds": 2.943172931999996, "created_at": "2026-11-22T04:11:18Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-0d07cb10f319", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "812d646f47d3720c9aabae83636e66163e17e0db496166c50da6d2c3bdd32cc8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-0d07cb10f319", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:18Z", "seed": 639355586, "treatment": "gated"}, "priority": 0.29482775765320357, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "59e5ccdec6fbe2a1f365b6d8531384f1cc07f52b40af8894e3e59ec09400e8f5", "run_seconds": 3.268051862716675, "seed": 639355586, "started_at": "2026-11-22T04:11:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00925, 0.003], "effect": -0.003125, "evaluated_at": "2026-11-22T04:11:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-0d07cb10f319", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.5, 0.62962963, 0.597222222, 0.425], "gated": [0.5, 0.62962963, 0.597222222, 0.425], "naive": [0.666666667, 0.685185185, 0.680555556, 0.5625], "scratch": [0.5, 0.62962963, 0.597222222, 0.4375], "ungated": [0.638888889, 0.685185185, 0.680555556, 0.5375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "812d646f47d3720c9aabae83636e66163e17e0db496166c50da6d2c3bdd32cc8", "primary_control": "scratch", "result_hash": "59e5ccdec6fbe2a1f365b6d8531384f1cc07f52b40af8894e3e59ec09400e8f5", "signature": "20edc678603e30a658842b876ecb478ff72796d3cdc88b03c0cda62aa19b94fdb3adb3911170e0ef4d3deb90c65364474f483c1f414d4b8200e3576a35a2680d", "state": "VERIFIED", "summary": {"compressed": 0.537963, "gated": 0.537963, "naive": 0.648727, "scratch": 0.541088, "ungated": 0.635532}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:11:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:21Z", "cpu_seconds": 0.7119728970000001, "created_at": "2026-11-22T04:11:18Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1c7251871187", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "920fd8bb02132488a1f09cd094498002c139d4c871660cb5ce4447134ca9eb98", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1c7251871187", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:18Z", "seed": 687469388, "treatment": "gated"}, "priority": 0.23924197164769212, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "33a4223ba90e23d4d69cc9671a7c8119ef498b57ed9cf0bc4c27087870f2908b", "run_seconds": 0.9721677303314209, "seed": 687469388, "started_at": "2026-11-22T04:12:20Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:17:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1c7251871187", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.469491525, 0.516949153, 0.665625, 0.534375], "gated": [0.469491525, 0.516949153, 0.665625, 0.534375], "naive": [0.365254237, 0.261016949, 0.373958333, 0.414583333], "scratch": [0.469491525, 0.516949153, 0.665625, 0.534375], "ungated": [0.491525424, 0.395762712, 0.373958333, 0.328125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "920fd8bb02132488a1f09cd094498002c139d4c871660cb5ce4447134ca9eb98", "primary_control": "scratch", "result_hash": "33a4223ba90e23d4d69cc9671a7c8119ef498b57ed9cf0bc4c27087870f2908b", "signature": "72ab0488534c083a96a022172a74915f33204303fb3503529f458fb0a210cec99baf4260c037f52ec68e1e96b6f1b828dd26c572de2f66188c3d884132aa0d00", "state": "PARTIAL", "summary": {"compressed": 0.54661, "gated": 0.54661, "naive": 0.353703, "scratch": 0.54661, "ungated": 0.397343}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:23Z", "cpu_seconds": 0.6288105079999937, "created_at": "2026-11-22T04:11:18Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-cc825b071a93", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "e73961e37fd35865312ffe536fa13674cd5c1b3c98c84caa718b054f9c9ae44a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-cc825b071a93", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:18Z", "seed": 1327517291, "treatment": "gated"}, "priority": 0.27787756693665583, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "27f56bf63ac8d1dc795199a1109171258d75ea3d22c96c86c031b936342f0c74", "run_seconds": 0.7287275791168213, "seed": 1327517291, "started_at": "2026-11-22T04:11:23Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:11:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-cc825b071a93", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.356589147, 0.226666667, 0.42, 0.450980392], "gated": [0.356589147, 0.253333333, 0.42, 0.450980392], "naive": [0.333333333, 0.206666667, 0.446666667, 0.431372549], "scratch": [0.356589147, 0.253333333, 0.42, 0.450980392], "ungated": [0.317829457, 0.28, 0.42, 0.392156863]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e73961e37fd35865312ffe536fa13674cd5c1b3c98c84caa718b054f9c9ae44a", "primary_control": "scratch", "result_hash": "27f56bf63ac8d1dc795199a1109171258d75ea3d22c96c86c031b936342f0c74", "signature": "295c9897359f4ce295e3b10232e676e1548cb9b386f2417c1206c6c6450f29c92ba522552cc4374b3be8178af801a972037aac34e9be68d900cd3e72ca82fc0c", "state": "PARTIAL", "summary": {"compressed": 0.363559, "gated": 0.370226, "naive": 0.35451, "scratch": 0.370226, "ungated": 0.352497}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:11:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:40Z", "cpu_seconds": 0.9888877789999952, "created_at": "2026-11-22T04:11:35Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-582f93f3dbd1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "d3883fa4cb45cdb08d6156ff71934760b9d77d6c38e1feaf7f47040a241a7966", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-582f93f3dbd1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:11:35Z", "seed": 5439792, "treatment": "frontier"}, "priority": 0.26195872482751736, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1518e08aea6e274345040200bc0093bef63fd508efd434ba82e24b6aedf30c36", "run_seconds": 1.0005004405975342, "seed": 5439792, "started_at": "2026-11-22T04:11:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035704, 0.321082], "effect": 0.142689, "evaluated_at": "2026-11-22T04:15:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-582f93f3dbd1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.833333333, 0.909090909, 0.729885057, 0.898550725], "frontier": [0.777777778, 0.909090909, 0.603448276, 0.826086957], "goal": [0.777777778, 0.878787879, 0.385057471, 0.826086957], "hypdisc": [0.722222222, 0.909090909, 0.626436782, 0.855072464], "infogain": [0.777777778, 0.878787879, 0.672413793, 0.826086957], "prederr": [0.777777778, 0.878787879, 0.637931034, 0.739130435], "random": [0.416666667, 0.909090909, 0.379310345, 0.84057971], "risk": [0.777777778, 0.909090909, 0.603448276, 0.826086957]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d3883fa4cb45cdb08d6156ff71934760b9d77d6c38e1feaf7f47040a241a7966", "primary_control": "random", "result_hash": "1518e08aea6e274345040200bc0093bef63fd508efd434ba82e24b6aedf30c36", "signature": "511b383edbb33cf4953e9664a2d6a2ddf8965d90a9a6a63e87dd910a9075539037c96740288cd9facd1035a775cec84f15136680c859e0162b202087e1d73107", "state": "VERIFIED", "summary": {"count": 0.842715, "frontier": 0.779101, "goal": 0.716928, "hypdisc": 0.778206, "infogain": 0.788767, "prederr": 0.758407, "random": 0.636412, "risk": 0.779101}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:15:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:47Z", "cpu_seconds": 1.2610512240000062, "created_at": "2026-11-22T04:11:35Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-76adf88be433", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "88ea4d2914f37e1a57720b027dfb21cf18f07ee8fb2919160c1934376b0ded1f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-76adf88be433", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:11:35Z", "seed": 836569928, "treatment": "full"}, "priority": 0.24480548260884086, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "324a7fb71691dd438425280a33fbd57819601c430c51d3f68351b286ef2ddc53", "run_seconds": 1.341911792755127, "seed": 836569928, "started_at": "2026-11-22T04:11:45Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:15:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-76adf88be433", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.5, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "88ea4d2914f37e1a57720b027dfb21cf18f07ee8fb2919160c1934376b0ded1f", "primary_control": "e141_best", "result_hash": "324a7fb71691dd438425280a33fbd57819601c430c51d3f68351b286ef2ddc53", "signature": "0a3323700a3cf89e803b47176a3619d61d8de1bbcff4240f8ce0f766e5783ab010967b0e6d52f782a972f2387385289d805fa0a40ce113137d3e6bbeb182cd0e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:15:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:55Z", "cpu_seconds": 4.638031007999999, "created_at": "2026-11-22T04:11:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-49628448aacf", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 7, "n": 4}, "prereg_hash": "8c5add39ccfd213c8ba5dfb499bcfbc645f4d19aba914e467d079c05c510462b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-49628448aacf", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:11:35Z", "seed": 997735893, "treatment": "full"}, "priority": 0.2435966763242851, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-872e4a3a5555", "result_hash": "64ad2c353de1d42928c40dfcac7d11532a80f775a686647295caee9e501592c3", "run_seconds": 5.15261435508728, "seed": 997735893, "started_at": "2026-11-22T04:11:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.345496, 0.50867], "effect": 0.427083, "evaluated_at": "2026-11-22T04:15:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-49628448aacf", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.453125, 0.5859375, 0.643229167, 0.609375], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.87890625, 0.91015625, 0.895833333, 0.893229167], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "8c5add39ccfd213c8ba5dfb499bcfbc645f4d19aba914e467d079c05c510462b", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-872e4a3a5555", "result_hash": "64ad2c353de1d42928c40dfcac7d11532a80f775a686647295caee9e501592c3", "signature": "162d20a3e17de29003b16252b288ff2d3bbf977a4b171daec9a82643cd28deb0d66d7766211b6f8991331e7a736357f1b542de567bb33d8c80d5b1b4162ccf05", "state": "VERIFIED", "summary": {"facts": 0.572917, "frozen": 0.0, "full": 1.0, "retrieval": 0.894531, "rules_nodetect": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:15:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:45Z", "cpu_seconds": 0.5939762410000071, "created_at": "2026-11-22T04:11:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4ac8a138dc12", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "245068e903b475a1c22d39c246c497a86b605375461e9880837ac319ce703dcf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4ac8a138dc12", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:11:35Z", "seed": 348115925, "treatment": "full"}, "priority": 0.24550817590949667, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "af297d96c33b7c36fc804c2bcb0229b8d3d3390c532965cef4493b9f53ef7345", "run_seconds": 0.5953118801116943, "seed": 348115925, "started_at": "2026-11-22T04:11:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024217, 0.024884], "effect": 0.000333, "evaluated_at": "2026-11-22T04:15:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-4ac8a138dc12", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.620915033, 0.882352941, 0.897959184, 0.908496732], "frozen": [0.08496732, 0.091503268, 0.081632653, 0.091503268], "full": [0.594771242, 0.875816993, 0.931972789, 0.908496732], "retrieval": [0.640522876, 0.888888889, 0.918367347, 0.921568627], "rules_nodetect": [0.594771242, 0.875816993, 0.931972789, 0.908496732]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "245068e903b475a1c22d39c246c497a86b605375461e9880837ac319ce703dcf", "primary_control": "facts", "result_hash": "af297d96c33b7c36fc804c2bcb0229b8d3d3390c532965cef4493b9f53ef7345", "signature": "f9bc24bf54255eca2a4359b7139273a8616f70c0282dfce7e4e123e7c952049c29aa870ea72cb125fbdd6df0bdd077d57181b177fe9ae8f753945b953fbab503", "state": "VERIFIED", "summary": {"facts": 0.827431, "frozen": 0.087402, "full": 0.827764, "retrieval": 0.842337, "rules_nodetect": 0.827764}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:15:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:39Z", "cpu_seconds": 2.6480268699999954, "created_at": "2026-11-22T04:11:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7f51049966f9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "14b0b23b759d890d47e5cba0dfced8b3f1ae7da70d14fb16554062b71bf5d478", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7f51049966f9", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:11:35Z", "seed": 1794227811, "treatment": "full"}, "priority": 0.275435255141791, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2866443e60578055881f793264481ebda81e40a5da3a70d24aaa3061c297e50a", "run_seconds": 2.8295676708221436, "seed": 1794227811, "started_at": "2026-11-22T04:11:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.061992, 0.334378], "effect": 0.136193, "evaluated_at": "2026-11-22T04:15:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7f51049966f9", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.34379085, 0.994444444, 0.295833333, 0.991666667], "frozen": [0.254901961, 0.3, 0.158333333, 0.358333333], "full": [0.771895425, 0.994444444, 0.4125, 0.991666667], "retrieval": [0.820915033, 1.0, 0.504166667, 0.991666667], "rules_nodetect": [0.258823529, 0.994444444, 0.429166667, 0.216666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "14b0b23b759d890d47e5cba0dfced8b3f1ae7da70d14fb16554062b71bf5d478", "primary_control": "facts", "result_hash": "2866443e60578055881f793264481ebda81e40a5da3a70d24aaa3061c297e50a", "signature": "ce4a839c83e331146fa498a9055ed7a114f5e568200437a542789583d59f3ee6f436f07020bf7fb8a378381130b8398708b92e1b6b57c0d98f4f2fe183772c0c", "state": "VERIFIED", "summary": {"facts": 0.656434, "frozen": 0.267892, "full": 0.792627, "retrieval": 0.829187, "rules_nodetect": 0.474775}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:15:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:40Z", "cpu_seconds": 0.9348034939999996, "created_at": "2026-11-22T04:11:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-88fd1bf04c7b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "73a7fe9f41c87cbe1612307ab0c0bf7ec32747b00861f274d45e204e357496ba", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-88fd1bf04c7b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:11:52Z", "seed": 1993837948, "treatment": "full"}, "priority": 0.2372828320182323, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ea344f51320932a7d17be18c7c9d468150d808db2e9d6609ec5a09f3f54daeeb", "run_seconds": 0.9609692096710205, "seed": 1993837948, "started_at": "2026-11-22T04:12:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T04:17:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-88fd1bf04c7b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 0.0, 0.75], "fixed_policy": [0.25, 1.0, 0.25, 0.0], "full": [0.25, 1.0, 0.0, 0.0], "no_world_model": [0.5, 1.0, 0.0, 0.0], "random": [0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "73a7fe9f41c87cbe1612307ab0c0bf7ec32747b00861f274d45e204e357496ba", "primary_control": "e141_best", "result_hash": "ea344f51320932a7d17be18c7c9d468150d808db2e9d6609ec5a09f3f54daeeb", "signature": "f99447038ddeedc68565a99f1b4ce1c7b9b3f781321eae52f7d4cdbc9be194437b3bebea3fb7bb353520550513cb81bcd675d2657cec9dd330c142f5ea0b5d07", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.375, "full": 0.3125, "no_world_model": 0.375, "random": 0.25}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:17:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:14Z", "cpu_seconds": 0.7099456070000008, "created_at": "2026-11-22T04:11:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ca0b88256698", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "a0a2821615f6bbbae0d042a8f545ec9cf1ab8bbfde24e927842ae00d1b1e22bd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ca0b88256698", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:11:52Z", "seed": 142574068, "treatment": "full"}, "priority": 0.23703249552234446, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9af726f2758b924e9f06299ccff4301a09f6f64449233d8f6c22b04f1f1db31e", "run_seconds": 0.7868087291717529, "seed": 142574068, "started_at": "2026-11-22T04:13:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.049958, 0.450042], "effect": 0.25, "evaluated_at": "2026-11-22T04:17:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ca0b88256698", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.25, 1.0, 0.25], "fixed_policy": [0.5, 0.25, 0.0, 0.0], "full": [0.5, 0.5, 1.0, 0.75], "no_world_model": [0.75, 0.25, 1.0, 0.75], "random": [0.0, 0.5, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "a0a2821615f6bbbae0d042a8f545ec9cf1ab8bbfde24e927842ae00d1b1e22bd", "primary_control": "e141_best", "result_hash": "9af726f2758b924e9f06299ccff4301a09f6f64449233d8f6c22b04f1f1db31e", "signature": "e3294bcf02deb34f30e88f96031e719cf24fea303451113a354db46707181cf97796fe14c495d843f77b8827a3687dfc33ea1b87ea258ec336d9db55c353710b", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.1875, "full": 0.6875, "no_world_model": 0.6875, "random": 0.5625}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:39Z", "cpu_seconds": 0.8984151840000001, "created_at": "2026-11-22T04:11:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0d42d181ac67", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "054eecfe12a20e3f4031dde4bad790da44af7c195ca99fa48ccf7f81c71b11c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0d42d181ac67", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:11:52Z", "seed": 1226015421, "treatment": "full"}, "priority": 0.2377511475448155, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "70beaf30adc989919dfece53d3de9ecac408919d0ab43debcd70130fab20854e", "run_seconds": 0.9260478019714355, "seed": 1226015421, "started_at": "2026-11-22T04:12:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.041922, 0.02165], "effect": -0.010136, "evaluated_at": "2026-11-22T04:17:35Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0d42d181ac67", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.75, 0.961538462, 0.975, 0.983333333], "frozen": [0.29375, 0.384615385, 0.325, 0.35], "full": [0.78125, 0.923076923, 0.975, 0.95], "retrieval": [0.83125, 0.961538462, 1.0, 0.966666667], "rules_nodetect": [0.78125, 0.923076923, 0.975, 0.95]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "054eecfe12a20e3f4031dde4bad790da44af7c195ca99fa48ccf7f81c71b11c3", "primary_control": "facts", "result_hash": "70beaf30adc989919dfece53d3de9ecac408919d0ab43debcd70130fab20854e", "signature": "ff3890bda7e55fec9901339797546ec34d21b6f23c879871b6de1c1d5651d78f3b1e54efb1822a27cb4ff8a479c90f99929401aeec06f3bfd9bb339720c2be0d", "state": "VERIFIED", "summary": {"facts": 0.917468, "frozen": 0.338341, "full": 0.907332, "retrieval": 0.939864, "rules_nodetect": 0.907332}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:17:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:11:59Z", "cpu_seconds": 0.8933654349999998, "created_at": "2026-11-22T04:11:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2136ab90e70a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "d89190f274840ecfc16cc26a7410c7f5b6dab3ecc5fd7183c2371b0138c82eba", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2136ab90e70a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:11:52Z", "seed": 90644613, "treatment": "full"}, "priority": 0.2446973209902979, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f1f75f4eb9b67d40755beade367d256f51dc296f47de2ef6c3c39825af4f8b37", "run_seconds": 1.251600980758667, "seed": 90644613, "started_at": "2026-11-22T04:11:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.066716, 0.197738], "effect": 0.065511, "evaluated_at": "2026-11-22T04:15:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2136ab90e70a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.892857143, 0.796511628, 0.9296875, 0.419642857], "frozen": [0.375, 0.377906977, 0.6484375, 0.214285714], "full": [0.892857143, 0.790697674, 0.9296875, 0.6875], "retrieval": [0.892857143, 0.76744186, 0.9296875, 0.6875], "rules_nodetect": [0.705357143, 0.790697674, 0.9296875, 0.857142857]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "d89190f274840ecfc16cc26a7410c7f5b6dab3ecc5fd7183c2371b0138c82eba", "primary_control": "facts", "result_hash": "f1f75f4eb9b67d40755beade367d256f51dc296f47de2ef6c3c39825af4f8b37", "signature": "5d1a68d722edb254cc6e7fb5c99491fcfbf408d77992cd05578553752987e7b5e438fbf5c38dabd53734dbc5aa13cdd4e516ced6998e632a7a564506811e6302", "state": "VERIFIED", "summary": {"facts": 0.759675, "frozen": 0.403908, "full": 0.825186, "retrieval": 0.819372, "rules_nodetect": 0.820721}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:15:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:57Z", "cpu_seconds": 0.4587184999999998, "created_at": "2026-11-22T04:11:52Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8b2c0f71075e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "24e8d32f8d5cf3213fcb788e3d2040e508bcc0091787273b8ccb04d6fca3a6cb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8b2c0f71075e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:11:52Z", "seed": 609480586, "treatment": "full"}, "priority": 0.23709261473719356, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c80669be83f554568a4708a4a86f0c18acd244f0ebb6688b645b975f6c9a8640", "run_seconds": 0.463026762008667, "seed": 609480586, "started_at": "2026-11-22T04:12:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.003167, 0.060107], "effect": 0.031637, "evaluated_at": "2026-11-22T04:17:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8b2c0f71075e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.614379085, 0.916666667, 0.906976744, 0.621212121], "frozen": [0.091503268, 0.08974359, 0.108527132, 0.098484848], "full": [0.633986928, 0.916666667, 0.945736434, 0.689393939], "retrieval": [0.764705882, 0.929487179, 0.937984496, 0.810606061], "rules_nodetect": [0.633986928, 0.916666667, 0.945736434, 0.689393939]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "24e8d32f8d5cf3213fcb788e3d2040e508bcc0091787273b8ccb04d6fca3a6cb", "primary_control": "facts", "result_hash": "c80669be83f554568a4708a4a86f0c18acd244f0ebb6688b645b975f6c9a8640", "signature": "a78228e919d595d651601cf796a585ed7d249fd49e21ceaf7bcad7313c93fedc22f4133ff223dcf34c54cc80d14a66c6c57a81fabecfc9bdecba230cfc6be40d", "state": "VERIFIED", "summary": {"facts": 0.764809, "frozen": 0.097065, "full": 0.796446, "retrieval": 0.860696, "rules_nodetect": 0.796446}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:30Z", "cpu_seconds": 0.7376991399999984, "created_at": "2026-11-22T04:11:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1a14d6c010ed", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "5c5c19e95858dbc7b6b7053cc316ed829d758a7f5ac12c01994d385434e566d7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1a14d6c010ed", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:52Z", "seed": 339673379, "treatment": "gated"}, "priority": 0.23545255571151422, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "1a988b35c28788f201adbd514c41008ba8dd9309251d41e4cb9d8bfac5ffed87", "run_seconds": 0.7959811687469482, "seed": 339673379, "started_at": "2026-11-22T04:13:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.03491, 0.002503], "effect": -0.016204, "evaluated_at": "2026-11-22T04:17:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1a14d6c010ed", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.777777778, 0.769230769, 0.638888889, 0.8], "gated": [0.777777778, 0.769230769, 0.638888889, 0.8], "naive": [0.814814815, 0.794871795, 0.75, 0.8], "scratch": [0.814814815, 0.769230769, 0.666666667, 0.8], "ungated": [0.814814815, 0.794871795, 0.75, 0.733333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5c5c19e95858dbc7b6b7053cc316ed829d758a7f5ac12c01994d385434e566d7", "primary_control": "scratch", "result_hash": "1a988b35c28788f201adbd514c41008ba8dd9309251d41e4cb9d8bfac5ffed87", "signature": "dfe3f45af5ec7483c9b8caf6e66b39c0eb6688dfe42fdf75246448c0fd656acbfd35a2177d3cc37bf5418276050bbc2b4f3f223220ad6f683bbe7f2442d5c302", "state": "VERIFIED", "summary": {"compressed": 0.746474, "gated": 0.746474, "naive": 0.789922, "scratch": 0.762678, "ungated": 0.773255}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:10Z", "cpu_seconds": 0.685972502, "created_at": "2026-11-22T04:11:52Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-c31218c0a8d9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "2341fd678f170cd864b75139b3c71f10e16870dedb37689f88b7fb15abc09daf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-c31218c0a8d9", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:11:52Z", "seed": 1976987591, "treatment": "gated"}, "priority": 0.24153724948768107, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "4c8e1b619a65f594fd9e29e8855dfd91e105cc4ce8bb4812620ca1830ab6c764", "run_seconds": 1.103773593902588, "seed": 1976987591, "started_at": "2026-11-22T04:12:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.030257, -0.001754], "effect": -0.016006, "evaluated_at": "2026-11-22T04:17:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-c31218c0a8d9", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.401515152, 0.248366013, 0.36, 0.196969697], "gated": [0.401515152, 0.248366013, 0.36, 0.196969697], "naive": [0.128787879, 0.150326797, 0.34, 0.204545455], "scratch": [0.409090909, 0.274509804, 0.36, 0.227272727], "ungated": [0.401515152, 0.248366013, 0.36, 0.257575758]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2341fd678f170cd864b75139b3c71f10e16870dedb37689f88b7fb15abc09daf", "primary_control": "scratch", "result_hash": "4c8e1b619a65f594fd9e29e8855dfd91e105cc4ce8bb4812620ca1830ab6c764", "signature": "7cd44a9acfd88b1748dff118818fc54e2f4b298db8491b5e3900fda693962893c7d1c224c9fa642cf1233bed266157ba5cffa23d21b39d67edc88836cad3720c", "state": "VERIFIED", "summary": {"compressed": 0.301713, "gated": 0.301713, "naive": 0.205915, "scratch": 0.317718, "ungated": 0.316864}, "ties": 1, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:12Z", "cpu_seconds": 2.389194091, "created_at": "2026-11-22T04:12:09Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-fa5c47699100", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "b06135e2c1c4d273e38ab5941c65f84d0fc87143ef39ad9615ee45df24b81d0e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-fa5c47699100", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:12:09Z", "seed": 41318332, "treatment": "gated"}, "priority": 0.290948125093941, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c8cf78e1b7ce1954280c31bfadc9b06f7f8dca992ef1b6766db961594574070a", "run_seconds": 2.63535737991333, "seed": 41318332, "started_at": "2026-11-22T04:12:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016288, 0.049051], "effect": 0.016381, "evaluated_at": "2026-11-22T04:17:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-fa5c47699100", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.551724138, 0.633928571, 0.642857143, 0.554347826], "gated": [0.551724138, 0.633928571, 0.642857143, 0.554347826], "naive": [0.663793103, 0.705357143, 0.714285714, 0.673913043], "scratch": [0.560344828, 0.633928571, 0.633928571, 0.489130435], "ungated": [0.594827586, 0.651785714, 0.642857143, 0.586956522]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "b06135e2c1c4d273e38ab5941c65f84d0fc87143ef39ad9615ee45df24b81d0e", "primary_control": "scratch", "result_hash": "c8cf78e1b7ce1954280c31bfadc9b06f7f8dca992ef1b6766db961594574070a", "signature": "d7910229fbb1eba0caddb819feae06613a92db93369cd1d7c79a4335f4ed9530f5b8b37f0992e703c1b5558a6922b3af8aba120859c1f3c4c8c9e18994c54c06", "state": "VERIFIED", "summary": {"compressed": 0.595714, "gated": 0.595714, "naive": 0.689337, "scratch": 0.579333, "ungated": 0.619107}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:17:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:34Z", "cpu_seconds": 1.765061105, "created_at": "2026-11-22T04:12:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1e2b01b66b3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "c0f3f8548d1d12b358b51c760213b24c3086460af52f0d1311fc14454d3cccfa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1e2b01b66b3b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:12:26Z", "seed": 1099121421, "treatment": "full"}, "priority": 0.24194376486859767, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "74452f6691952c203f2e853bfb25fa191277ab998ef2480f8c12b910f5941ce0", "run_seconds": 2.146807909011841, "seed": 1099121421, "started_at": "2026-11-22T04:12:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.049958, 0.450042], "effect": 0.25, "evaluated_at": "2026-11-22T04:17:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1e2b01b66b3b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.5, 0.25, 0.5], "fixed_policy": [1.0, 0.25, 1.0, 0.75], "full": [0.75, 0.5, 0.5, 1.0], "no_world_model": [0.5, 0.75, 0.25, 0.5], "random": [1.0, 1.0, 0.5, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c0f3f8548d1d12b358b51c760213b24c3086460af52f0d1311fc14454d3cccfa", "primary_control": "e141_best", "result_hash": "74452f6691952c203f2e853bfb25fa191277ab998ef2480f8c12b910f5941ce0", "signature": "79ab053b413d2232f273b5960277b55ba51791b51453e2de233083017016e216b2d8f77c7f90625114ea91581b5f3ee9052fcd43a3fba59bcbd139fbbdff3905", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.75, "full": 0.6875, "no_world_model": 0.5, "random": 0.75}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:35Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:33Z", "cpu_seconds": 0.4764683270000001, "created_at": "2026-11-22T04:12:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-01cb1ebcc310", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "8538533319cebac90aa2bf27c2ce89e74cf78287baf10f55d4c0fcbb815f2976", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-01cb1ebcc310", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:12:26Z", "seed": 528618643, "treatment": "full"}, "priority": 0.24141770666552392, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f773f3acacbd1735f059aca68fbab2d5a6960e15483e39735b99d4b86216cc6c", "run_seconds": 0.6831305027008057, "seed": 528618643, "started_at": "2026-11-22T04:12:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001869, 0.058975], "effect": 0.030422, "evaluated_at": "2026-11-22T04:17:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-01cb1ebcc310", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.451923077, 0.75, 0.294444444, 0.666666667], "frozen": [0.0625, 0.076923077, 0.227777778, 0.333333333], "full": [0.504807692, 0.807692308, 0.305555556, 0.666666667], "retrieval": [0.572115385, 0.783653846, 0.305555556, 0.666666667], "rules_nodetect": [0.466346154, 0.807692308, 0.305555556, 0.416666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "8538533319cebac90aa2bf27c2ce89e74cf78287baf10f55d4c0fcbb815f2976", "primary_control": "facts", "result_hash": "f773f3acacbd1735f059aca68fbab2d5a6960e15483e39735b99d4b86216cc6c", "signature": "0d533facda1adccdb0944cee532839b930d366dd2140a013b13648cfdab7925f603949670b8febcd80a72ed970ce28a21a0238184da341cd1d2063d6c3ff4700", "state": "VERIFIED", "summary": {"facts": 0.540759, "frozen": 0.175134, "full": 0.571181, "retrieval": 0.581998, "rules_nodetect": 0.499065}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:27Z", "cpu_seconds": 0.48063265899999985, "created_at": "2026-11-22T04:12:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-273989b63419", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "7cae33dfb79f01cdbff314dae99f43e4aae82587d97e75023f3f2220f310c006", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-273989b63419", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:12:26Z", "seed": 1879373863, "treatment": "full"}, "priority": 0.24450770645784115, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cf9bf070cc39a53fadccf24896f835882dad37abae5111d41e1da5a577b38d0a", "run_seconds": 0.5081830024719238, "seed": 1879373863, "started_at": "2026-11-22T04:12:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.046546, 0.125281], "effect": 0.039368, "evaluated_at": "2026-11-22T04:17:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-273989b63419", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.791666667, 0.635658915, 0.418300654, 0.281045752], "frozen": [0.395833333, 0.162790698, 0.418300654, 0.274509804], "full": [0.791666667, 0.80620155, 0.418300654, 0.267973856], "retrieval": [0.791666667, 0.720930233, 0.411764706, 0.281045752], "rules_nodetect": [0.791666667, 0.84496124, 0.45751634, 0.333333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "7cae33dfb79f01cdbff314dae99f43e4aae82587d97e75023f3f2220f310c006", "primary_control": "facts", "result_hash": "cf9bf070cc39a53fadccf24896f835882dad37abae5111d41e1da5a577b38d0a", "signature": "4c256355eead3a8fb2ee08dcd6964d14c3d4cd61bddfadb3e2cfcef3894095824af1a6e693c0d0d60507f65f1fb152b17c580934d65b38ea202294753b238306", "state": "VERIFIED", "summary": {"facts": 0.531668, "frozen": 0.312859, "full": 0.571036, "retrieval": 0.551352, "rules_nodetect": 0.606869}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:17:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:27Z", "cpu_seconds": 0.8082594039999997, "created_at": "2026-11-22T04:12:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ac859af1a930", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "27b9934ea00f290bc32835e6c8786b935ffef24b678f410f1f524d2d22ae3bb9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ac859af1a930", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:12:26Z", "seed": 1243150563, "treatment": "full"}, "priority": 0.24856104299680684, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-8df3f9bcccd6", "result_hash": "43550797e8fb29da87eabc0bb46effd8009ea29514ed08f385386443805ce77e", "run_seconds": 0.8087978363037109, "seed": 1243150563, "started_at": "2026-11-22T04:12:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.084725, 0.045392], "effect": -0.019667, "evaluated_at": "2026-11-22T04:17:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ac859af1a930", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.552941176, 0.380821918, 0.2275, 0.24], "frozen": [0.035294118, 0.131506849, 0.13, 0.13], "full": [0.629411765, 0.350684932, 0.155, 0.1875], "retrieval": [0.582352941, 0.342465753, 0.1225, 0.2475], "rules_nodetect": [0.629411765, 0.350684932, 0.155, 0.1875]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "27b9934ea00f290bc32835e6c8786b935ffef24b678f410f1f524d2d22ae3bb9", "primary_control": "facts", "replicated": false, "replication_of": "E141-ow_model-8df3f9bcccd6", "result_hash": "43550797e8fb29da87eabc0bb46effd8009ea29514ed08f385386443805ce77e", "signature": "04bf1a231b289d262a1e7bfc08eff4071eaa2e3e8d4dd2b75ca7d12e97c6700c0698018073b0ba0b96d5acf5f5780483c1f4cf388449737952acbcc53cb75c04", "state": "VERIFIED", "summary": {"facts": 0.350316, "frozen": 0.1067, "full": 0.330649, "retrieval": 0.323705, "rules_nodetect": 0.330649}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:17:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:19Z", "cpu_seconds": 2.3947231230000003, "created_at": "2026-11-22T04:12:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ee39c23d33f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "c699c15ee8894b51c0efb3e274deefefd7a21d678aae8f3a4be2920a158acfd2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ee39c23d33f0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:12:41Z", "seed": 894622810, "treatment": "frontier"}, "priority": 0.23696260907416541, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6108ce0474e34c8232acaad4c7d90d8ed9cc8a27d42fb4b2d9187663878cef3e", "run_seconds": 2.500678300857544, "seed": 894622810, "started_at": "2026-11-22T04:13:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.000914, 0.014803], "effect": 0.006944, "evaluated_at": "2026-11-22T04:17:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ee39c23d33f0", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.993055556, 0.993055556, 1.0], "frontier": [1.0, 1.0, 1.0, 0.993055556], "goal": [1.0, 1.0, 0.993055556, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [0.986111111, 1.0, 1.0, 0.993055556], "random": [0.986111111, 0.986111111, 1.0, 0.993055556], "risk": [1.0, 1.0, 1.0, 0.993055556]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "c699c15ee8894b51c0efb3e274deefefd7a21d678aae8f3a4be2920a158acfd2", "primary_control": "random", "result_hash": "6108ce0474e34c8232acaad4c7d90d8ed9cc8a27d42fb4b2d9187663878cef3e", "signature": "424cde84c400fac3f76c7f081d51aac96ac709477bd2505c5e6454337e8ccf6bcde6b29b85ae13fb67a02cb478a53eac9411fee10195cff8fc886e96263e3f07", "state": "VERIFIED", "summary": {"count": 0.996528, "frontier": 0.998264, "goal": 0.998264, "hypdisc": 1.0, "infogain": 1.0, "prederr": 0.994792, "random": 0.991319, "risk": 0.998264}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:17:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:51Z", "cpu_seconds": 1.4447226029999998, "created_at": "2026-11-22T04:12:41Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-42106630dbce", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "fc5999a0a8ab3f20ac2307c846f1616affebb3aa8d7b09ddf60dc0a0b27fd4c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-42106630dbce", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:12:41Z", "seed": 720600650, "treatment": "full"}, "priority": 0.23977157470736427, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1e7ed39a6ebfb81cc746385104f29bc8e31cb371d6c3f5eb433dd5e2458c9094", "run_seconds": 1.482114553451538, "seed": 720600650, "started_at": "2026-11-22T04:12:50Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:17:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-42106630dbce", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fc5999a0a8ab3f20ac2307c846f1616affebb3aa8d7b09ddf60dc0a0b27fd4c3", "primary_control": "e141_best", "result_hash": "1e7ed39a6ebfb81cc746385104f29bc8e31cb371d6c3f5eb433dd5e2458c9094", "signature": "471b207b6a1131a1b49843e00fe93ed8c600152fe13dbbee4ee93d130d5f265eea173c87654603b089431e50a34532d78af745e86a454ea6a7028356e1a4ea08", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:17:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:27Z", "cpu_seconds": 2.5025511520000006, "created_at": "2026-11-22T04:12:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-acdccb519c2d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 4, "n": 4}, "prereg_hash": "e8a5ad6e8dc65f1aceb43ada685cc97cfe199aa4fdb692b94812aebc5e090fa5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-acdccb519c2d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:12:41Z", "seed": 1242280498, "treatment": "full"}, "priority": 0.2356366135020633, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a7e876e427f662ac48bf213081f99a39b051445b0c62924a6157f185e6d2326a", "run_seconds": 2.816331148147583, "seed": 1242280498, "started_at": "2026-11-22T04:13:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.003654, 0.048242], "effect": 0.025948, "evaluated_at": "2026-11-22T04:17:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-acdccb519c2d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.973333333, 0.975, 0.9, 0.710344828], "frozen": [0.346666667, 0.366666667, 0.33, 0.065517241], "full": [0.986666667, 0.991666667, 0.96, 0.724137931], "retrieval": [0.986666667, 1.0, 0.97, 0.85862069], "rules_nodetect": [0.986666667, 0.991666667, 0.96, 0.724137931]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "e8a5ad6e8dc65f1aceb43ada685cc97cfe199aa4fdb692b94812aebc5e090fa5", "primary_control": "facts", "result_hash": "a7e876e427f662ac48bf213081f99a39b051445b0c62924a6157f185e6d2326a", "signature": "2c1e736ab97c9b82a13a171d7113eff3b6790ae46f6d4008980ac3b458cd265fb61ef9cd545cce080096bcbe53419d0898330925f6d511059e4cba5864c4400c", "state": "VERIFIED", "summary": {"facts": 0.88967, "frozen": 0.277213, "full": 0.915618, "retrieval": 0.953822, "rules_nodetect": 0.915618}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:17:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:46Z", "cpu_seconds": 1.3328362550000001, "created_at": "2026-11-22T04:12:41Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3e6c0bad4ac4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "9fee2a6ca32456dfcaefabdde703c97e9214caedbe14534742bf0baad04cac2b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3e6c0bad4ac4", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:12:41Z", "seed": 1472943663, "treatment": "gated"}, "priority": 0.23996716745604946, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "8ecbb77d81827a5f294769d9ff2d590ba3c5f926b388d08df5c0ffac76ed976c", "run_seconds": 1.3518321514129639, "seed": 1472943663, "started_at": "2026-11-22T04:12:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.039729, 0.030358], "effect": -0.004686, "evaluated_at": "2026-11-22T04:17:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-3e6c0bad4ac4", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.294736842, 0.273684211, 0.527272727, 0.165333333], "gated": [0.294736842, 0.273684211, 0.527272727, 0.165333333], "naive": [0.289473684, 0.263157895, 0.454545455, 0.162666667], "scratch": [0.294736842, 0.231578947, 0.545454545, 0.208], "ungated": [0.289473684, 0.268421053, 0.527272727, 0.157333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "9fee2a6ca32456dfcaefabdde703c97e9214caedbe14534742bf0baad04cac2b", "primary_control": "scratch", "result_hash": "8ecbb77d81827a5f294769d9ff2d590ba3c5f926b388d08df5c0ffac76ed976c", "signature": "835a016e64f80a60fb6efed1cfb2c1cfa7dc30ade649e1d0b9d689793651adefbf0b8b0fb71345319db4ffe5ff3f4344da9698d8b41b8daf162476f2610eba07", "state": "VERIFIED", "summary": {"compressed": 0.315257, "gated": 0.315257, "naive": 0.292461, "scratch": 0.319943, "ungated": 0.310625}, "ties": 1, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:17:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:52Z", "cpu_seconds": 1.0198979160000006, "created_at": "2026-11-22T04:12:41Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5352e9cf7f08", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "prereg_hash": "fe061740a8a35a5aab954b409620f600f6ee089d0317007ce2fba7a40be9a9e8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5352e9cf7f08", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:12:41Z", "seed": 1154324030, "treatment": "gated"}, "priority": 0.2376005374410287, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "baee9dd5bb5dbebbfc3d119a21dc39b88ce4592f52f2ead0fefe9784c4bd7bb7", "run_seconds": 1.0776164531707764, "seed": 1154324030, "started_at": "2026-11-22T04:12:51Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:17:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-5352e9cf7f08", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.24040404, 0.390804598, 0.197979798, 0.143434343], "gated": [0.24040404, 0.390804598, 0.197979798, 0.143434343], "naive": [0.206060606, 0.390804598, 0.129292929, 0.165656566], "scratch": [0.24040404, 0.390804598, 0.197979798, 0.143434343], "ungated": [0.206060606, 0.335632184, 0.129292929, 0.165656566]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fe061740a8a35a5aab954b409620f600f6ee089d0317007ce2fba7a40be9a9e8", "primary_control": "scratch", "result_hash": "baee9dd5bb5dbebbfc3d119a21dc39b88ce4592f52f2ead0fefe9784c4bd7bb7", "signature": "bbac655a2512a13f3d62dc9d93989a1c3640d52ec82d1abc0ca1c0605dd474f0b89d59bf2efd877c2453663c15dd8721ba5146ca3b7953f0129e7893357cc301", "state": "PARTIAL", "summary": {"compressed": 0.243156, "gated": 0.243156, "naive": 0.222954, "scratch": 0.243156, "ungated": 0.209161}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:45Z", "cpu_seconds": 0.7750664510000003, "created_at": "2026-11-22T04:12:41Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a2fe535807eb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "b5c5f43edeea30e3b6184f507668612c4bf844556ccb923dad0cfff1400a19b9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a2fe535807eb", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:12:41Z", "seed": 1826997960, "treatment": "gated"}, "priority": 0.24671120319043197, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "10bd56e47873a680cf05bb2beabc5563a551e243e624319043e74619241c7e8f", "run_seconds": 0.7870066165924072, "seed": 1826997960, "started_at": "2026-11-22T04:12:44Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:17:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a2fe535807eb", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.49, 0.717171717, 0.771929825, 0.656565657], "gated": [0.49, 0.717171717, 0.771929825, 0.656565657], "naive": [0.31, 0.616161616, 0.649122807, 0.616161616], "scratch": [0.49, 0.717171717, 0.771929825, 0.656565657], "ungated": [0.26, 0.575757576, 0.719298246, 0.656565657]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b5c5f43edeea30e3b6184f507668612c4bf844556ccb923dad0cfff1400a19b9", "primary_control": "scratch", "result_hash": "10bd56e47873a680cf05bb2beabc5563a551e243e624319043e74619241c7e8f", "signature": "7b640c43ce07efbd77cad680397e8497fff4063ff2eb932eb8a19a0ef556132e575351ed5385ec814b79f4549584d799c12d70cdb25d3feaeb0f190cbdfd0708", "state": "PARTIAL", "summary": {"compressed": 0.658917, "gated": 0.658917, "naive": 0.547862, "scratch": 0.658917, "ungated": 0.552905}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:06Z", "cpu_seconds": 2.3720405069999995, "created_at": "2026-11-22T04:12:57Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0a726f03460a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "d1cbbfcb1024c580be4b90602936798a8941b3ec9f111cb7702034fe57bbac8e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0a726f03460a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:12:57Z", "seed": 1105255653, "treatment": "full"}, "priority": 0.2397452206837124, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-2356ab663a26", "result_hash": "df6f747e5a3a6d3f32d77b82bd8dd56f6acf0e1f94e50b0740072d8ed26a6bd9", "run_seconds": 2.9238498210906982, "seed": 1105255653, "started_at": "2026-11-22T04:13:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.223722, 1.151278], "effect": 0.6875, "evaluated_at": "2026-11-22T04:17:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0a726f03460a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.0, 0.0, 1.0], "fixed_policy": [0.0, 0.0, 0.5, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.5, 1.0], "random": [1.0, 0.0, 0.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "d1cbbfcb1024c580be4b90602936798a8941b3ec9f111cb7702034fe57bbac8e", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-2356ab663a26", "result_hash": "df6f747e5a3a6d3f32d77b82bd8dd56f6acf0e1f94e50b0740072d8ed26a6bd9", "signature": "faea7d8393f531babfe82831879b9f2141b79674797839bcc75eee64a1e49d96a15ea5740be464e3fd7c45010b53c3e70759e4df7139130489fabf5d3e79e10c", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 0.875, "random": 0.5}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:12:58Z", "cpu_seconds": 0.3558897659999998, "created_at": "2026-11-22T04:12:57Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-624e675259e3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "b0f933d269e4ca1de9df7d57689a17ffa0d05d5385c56c97fcd3188f44afd66d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-624e675259e3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:12:57Z", "seed": 11010281, "treatment": "full"}, "priority": 0.2469550456386189, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "57868339de522ca41acf7827a8d3f6a421a05d9a54c04f8da0af8fbacb32856a", "run_seconds": 0.37052440643310547, "seed": 11010281, "started_at": "2026-11-22T04:12:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.746989, 0.871989], "effect": 0.0625, "evaluated_at": "2026-11-22T04:17:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-624e675259e3", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.5, 1.0], "fixed_policy": [0.0, 1.0, 0.5, 0.0], "full": [1.0, 0.0, 0.75, 0.0], "no_world_model": [0.5, 0.0, 0.75, 0.25], "random": [0.0, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "b0f933d269e4ca1de9df7d57689a17ffa0d05d5385c56c97fcd3188f44afd66d", "primary_control": "e141_best", "result_hash": "57868339de522ca41acf7827a8d3f6a421a05d9a54c04f8da0af8fbacb32856a", "signature": "4498efa964d950837d2ca54cb59b25a70eba8e148f4f8ad38c2665a4725804ea1c37de0e49a272e68895a44ec06394cc592ed035efe42a888d7a6b9036a42001", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.375, "full": 0.4375, "no_world_model": 0.375, "random": 0.0}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:17:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:08Z", "cpu_seconds": 1.0258938290000001, "created_at": "2026-11-22T04:12:57Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8225aa43043b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "b887377372833a053f4f5ed847c05a174a35ef1f7ebd7a47000ef99875f43ab4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8225aa43043b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:12:57Z", "seed": 926777613, "treatment": "full"}, "priority": 0.2392426578211388, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "944a24c62d9d24055dd0c0045b65fd6ef7fbc0a765d865c4e645ef100681c0ed", "run_seconds": 1.049516201019287, "seed": 926777613, "started_at": "2026-11-22T04:13:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.07, 0.805], "effect": 0.4375, "evaluated_at": "2026-11-22T04:17:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8225aa43043b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.25, 1.0, 0.75], "fixed_policy": [1.0, 0.5, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b887377372833a053f4f5ed847c05a174a35ef1f7ebd7a47000ef99875f43ab4", "primary_control": "e141_best", "result_hash": "944a24c62d9d24055dd0c0045b65fd6ef7fbc0a765d865c4e645ef100681c0ed", "signature": "f21f4d7781e9d50afa46a2365b8dff2a20d9c680022e718cf005ad11d918841a70e8eb418641f31f171b6a0dc147166dcb6d0ccb0607bcc113cb01bb130d7c01", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:12Z", "cpu_seconds": 0.8858994849999995, "created_at": "2026-11-22T04:12:57Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e496c5d68eaf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "141dcebe5c61b9ac9745f924e06fa10f7ce19bda0f05f3645823840dc054a59f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e496c5d68eaf", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:12:57Z", "seed": 756517614, "treatment": "full"}, "priority": 0.23712045467096177, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3a0603023dd8d7dcc70b178c95ea79bb488728644b4cfb67e3ac0a58d96cb974", "run_seconds": 0.922722339630127, "seed": 756517614, "started_at": "2026-11-22T04:13:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007742, 0.023871], "effect": 0.008065, "evaluated_at": "2026-11-22T04:17:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e496c5d68eaf", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.98, 0.911290323, 1.0], "frozen": [0.361111111, 0.27, 0.241935484, 0.325], "full": [1.0, 0.98, 0.943548387, 1.0], "retrieval": [1.0, 0.98, 0.951612903, 1.0], "rules_nodetect": [1.0, 0.98, 0.943548387, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "141dcebe5c61b9ac9745f924e06fa10f7ce19bda0f05f3645823840dc054a59f", "primary_control": "facts", "result_hash": "3a0603023dd8d7dcc70b178c95ea79bb488728644b4cfb67e3ac0a58d96cb974", "signature": "7ed1f0612ad6adb9b59f48ebb70d43d9a8c7af441dc074b120171ee4f95b446997005acd5d5d8aea6444ce629c307e0bbd65b6380d32576b7d34601f7cb09b04", "state": "VERIFIED", "summary": {"facts": 0.972823, "frozen": 0.299512, "full": 0.980887, "retrieval": 0.982903, "rules_nodetect": 0.980887}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:17:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:02Z", "cpu_seconds": 0.47166524299999946, "created_at": "2026-11-22T04:12:57Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-e77abc768943", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "c0065f8b23efd889f724798c65ab5a3b38eec04017ea3f5db6dda4b5265ca97b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e77abc768943", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:12:57Z", "seed": 1351481904, "treatment": "full"}, "priority": 0.24508877880246155, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "74314b6d64b5e273ed83bc352d1c2e47115e670981ca59c29bbe698714f506cc", "run_seconds": 0.5032455921173096, "seed": 1351481904, "started_at": "2026-11-22T04:13:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001993, 0.059577], "effect": 0.030785, "evaluated_at": "2026-11-22T04:17:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-e77abc768943", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.743589744, 0.634408602, 0.641025641, 0.683333333], "frozen": [0.314102564, 0.35483871, 0.41025641, 0.375], "full": [0.730769231, 0.677419355, 0.692307692, 0.725], "retrieval": [0.730769231, 0.698924731, 0.666666667, 0.7], "rules_nodetect": [0.730769231, 0.677419355, 0.794871795, 0.8]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "c0065f8b23efd889f724798c65ab5a3b38eec04017ea3f5db6dda4b5265ca97b", "primary_control": "facts", "result_hash": "74314b6d64b5e273ed83bc352d1c2e47115e670981ca59c29bbe698714f506cc", "signature": "93161d44e4fedda2ca248cab319c799f87854078ecc3c284a57bc59c6e51da1e0b21b2230bf2dbd3fdd3dbd099819bc9ba29140581787e34b0fc2f2c0836660a", "state": "VERIFIED", "summary": {"facts": 0.675589, "frozen": 0.363549, "full": 0.706374, "retrieval": 0.69909, "rules_nodetect": 0.750765}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:28Z", "cpu_seconds": 4.2405532820000005, "created_at": "2026-11-22T04:13:14Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9dab2471f769", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "ab62ff3cc6d709c75bb24b5e8d0ff131de0a57994bbc5e5a17e5b1dfa7f2b076", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9dab2471f769", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:13:14Z", "seed": 776827008, "treatment": "frontier"}, "priority": 0.23283401731407655, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "79a00ac29dd49c5f8a571eb6189fff100c8765bded29a5fccdb6f7f1fc14363a", "run_seconds": 4.833179235458374, "seed": 776827008, "started_at": "2026-11-22T04:15:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.096522, 0.096359], "effect": -8.2e-05, "evaluated_at": "2026-11-22T04:18:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9dab2471f769", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.582926829, 0.517391304, 0.538297872, 0.519148936], "frontier": [0.390243902, 0.513043478, 0.4, 0.410638298], "goal": [0.441463415, 0.2, 0.474468085, 0.368085106], "hypdisc": [0.502439024, 0.484782609, 0.482978723, 0.465957447], "infogain": [0.390243902, 0.258695652, 0.4, 0.410638298], "prederr": [0.390243902, 0.286956522, 0.4, 0.410638298], "random": [0.517073171, 0.467391304, 0.29787234, 0.431914894], "risk": [0.390243902, 0.210869565, 0.4, 0.410638298]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "ab62ff3cc6d709c75bb24b5e8d0ff131de0a57994bbc5e5a17e5b1dfa7f2b076", "primary_control": "random", "result_hash": "79a00ac29dd49c5f8a571eb6189fff100c8765bded29a5fccdb6f7f1fc14363a", "signature": "b8c453b0bc82635c66078b59995aa95e003c23db3160c8d2cbbe7258eff09c397dc96504313a67f007687db105b6365aaede6cc33e8f6807eae0f507b59e6c0f", "state": "VERIFIED", "summary": {"count": 0.539441, "frontier": 0.428481, "goal": 0.371004, "hypdisc": 0.484039, "infogain": 0.364894, "prederr": 0.37196, "random": 0.428563, "risk": 0.352938}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:18:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:49Z", "cpu_seconds": 2.2101727049999997, "created_at": "2026-11-22T04:13:14Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e8a4d716bd45", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "29fecbbd9f3016be35b4ac621a9c53ad3ead16172f131e1cb463709051d704ed", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e8a4d716bd45", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:13:14Z", "seed": 1121722238, "treatment": "frontier"}, "priority": 0.2337789720608309, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "313b18b7901cafe089d50b923dc81e077e367bfb57e1638ac9552e3c1a840738", "run_seconds": 2.2987513542175293, "seed": 1121722238, "started_at": "2026-11-22T04:13:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.322043, 0.444243], "effect": 0.383143, "evaluated_at": "2026-11-22T04:17:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e8a4d716bd45", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.584, 0.771428571, 0.806666667, 0.626666667], "frontier": [0.96, 0.985714286, 1.0, 0.88], "goal": [0.84, 0.892857143, 0.9, 0.653333333], "hypdisc": [0.712, 0.85, 0.833333333, 0.7], "infogain": [0.96, 0.985714286, 1.0, 0.88], "prederr": [0.96, 0.985714286, 1.0, 0.88], "random": [0.496, 0.657142857, 0.6, 0.54], "risk": [0.96, 0.985714286, 1.0, 0.88]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "29fecbbd9f3016be35b4ac621a9c53ad3ead16172f131e1cb463709051d704ed", "primary_control": "random", "result_hash": "313b18b7901cafe089d50b923dc81e077e367bfb57e1638ac9552e3c1a840738", "signature": "8c625d338bd4121e46ebeae70898a9841cb06ecee79cdff5a2cea2090369b7ebf80823b97ac8684490130ac46755de70a0f7d28026f6e7c6b3631f36c28f4103", "state": "VERIFIED", "summary": {"count": 0.69719, "frontier": 0.956429, "goal": 0.821548, "hypdisc": 0.773833, "infogain": 0.956429, "prederr": 0.956429, "random": 0.573286, "risk": 0.956429}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:17:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:35Z", "cpu_seconds": 1.5561104049999983, "created_at": "2026-11-22T04:13:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-68d971d78b33", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "3212252ffb8f34896e05b0a972d7d5322a5db3cadf8ab2683b1b9b2317ab5c61", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-68d971d78b33", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:13:14Z", "seed": 345205927, "treatment": "full"}, "priority": 0.2330523300688783, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9f12af5667725ebc47df34072e631e426bf2448161cfaef64e25d566a76b4466", "run_seconds": 2.3010783195495605, "seed": 345205927, "started_at": "2026-11-22T04:14:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:18:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-68d971d78b33", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.25, 0.0], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 1.0, 1.0, 1.0], "random": [0.25, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "3212252ffb8f34896e05b0a972d7d5322a5db3cadf8ab2683b1b9b2317ab5c61", "primary_control": "e141_best", "result_hash": "9f12af5667725ebc47df34072e631e426bf2448161cfaef64e25d566a76b4466", "signature": "10aed084f4f598987552e0e0d36d1e179d653b13551d8ad5fe1be22ee1486193d43933f76a6c6da0c5a1bbf2de168ecacaad837ab220809865ef771fba414205", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.125, "full": 0.75, "no_world_model": 0.75, "random": 0.8125}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:18:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:34Z", "cpu_seconds": 1.2200598059999948, "created_at": "2026-11-22T04:13:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9c066949a315", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "a9a125b867868e2ff7f0324340624fa1bdc9c9f843d2224f94b17cb2b63a0ba7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9c066949a315", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:13:14Z", "seed": 992172769, "treatment": "full"}, "priority": 0.23225902443927926, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d95ffc9cf8d654678bd83fc0eddf22e8f232832cf53863fe45d5552795da58c0", "run_seconds": 1.595482349395752, "seed": 992172769, "started_at": "2026-11-22T04:15:33Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:18:47Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9c066949a315", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.5, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 0.75, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a9a125b867868e2ff7f0324340624fa1bdc9c9f843d2224f94b17cb2b63a0ba7", "primary_control": "e141_best", "result_hash": "d95ffc9cf8d654678bd83fc0eddf22e8f232832cf53863fe45d5552795da58c0", "signature": "32b2474c702af32b90cab7759eeaeaafac881433668dd661f042d1feb1e5d7fd41cdf551251648fe79555955a28efcb28f848085b542e5370106243c8d50650f", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 0.875, "random": 0.9375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:18:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:46Z", "cpu_seconds": 1.404486557000002, "created_at": "2026-11-22T04:13:14Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-136593259db4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "a895d3d880a9d7e7fdc6286f6c554b5fe7fc16225436735a6f9b38c374c74d65", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-136593259db4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:13:14Z", "seed": 2003947687, "treatment": "full"}, "priority": 0.23400067908683064, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3f9dc41356f8ff56729f2e1064c1f11130bf86a78c4883f475a9e1bc84a2ddd7", "run_seconds": 1.4310195446014404, "seed": 2003947687, "started_at": "2026-11-22T04:13:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.053514, 0.286764], "effect": 0.170139, "evaluated_at": "2026-11-22T04:17:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-136593259db4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.9375, 0.854166667, 0.854166667, 0.659722222], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 0.993055556, 0.993055556, 1.0], "retrieval": [0.993055556, 0.986111111, 0.965277778, 0.944444444], "rules_nodetect": [0.541666667, 0.201388889, 0.375, 0.194444444]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a895d3d880a9d7e7fdc6286f6c554b5fe7fc16225436735a6f9b38c374c74d65", "primary_control": "facts", "result_hash": "3f9dc41356f8ff56729f2e1064c1f11130bf86a78c4883f475a9e1bc84a2ddd7", "signature": "e81d0cd2614618f19ba53b0d59d1d2966000c7870cad75c8da73af4733d59610f595538aa26e7cbed9dc0f4ec9ccf98c9f09fe256999f9ba96d106fa4f70d801", "state": "VERIFIED", "summary": {"facts": 0.826389, "frozen": 0.0, "full": 0.996528, "retrieval": 0.972222, "rules_nodetect": 0.328125}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:17:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:24Z", "cpu_seconds": 4.723477316, "created_at": "2026-11-22T04:13:14Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8efca12d0294", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "cb9a3570317bca53c7ea2698b58205d5029fdd9a1b3477f6c11e8194a38213c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8efca12d0294", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:13:14Z", "seed": 1071497940, "treatment": "gated"}, "priority": 0.23656590637106406, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "379f61c9f805de40ed49b39ee8bc517c6a8d6d091e6261472a46b7b21e353243", "run_seconds": 4.926120758056641, "seed": 1071497940, "started_at": "2026-11-22T04:13:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.011935, 0.003871], "effect": -0.004032, "evaluated_at": "2026-11-22T04:17:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-8efca12d0294", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.471153846, 0.5, 0.537037037, 0.39516129], "gated": [0.471153846, 0.5, 0.537037037, 0.39516129], "naive": [0.692307692, 0.590909091, 0.648148148, 0.508064516], "scratch": [0.471153846, 0.5, 0.537037037, 0.411290323], "ungated": [0.615384615, 0.534090909, 0.62037037, 0.508064516]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "cb9a3570317bca53c7ea2698b58205d5029fdd9a1b3477f6c11e8194a38213c6", "primary_control": "scratch", "result_hash": "379f61c9f805de40ed49b39ee8bc517c6a8d6d091e6261472a46b7b21e353243", "signature": "b2a44017b3d205244edf1b197414486eeaa7b64a6bcab2c5e9914e19fa8d194847151212f45fe2cd98a3dc0ce60ea4e564c0dce4655e5ae270f104b1312add04", "state": "VERIFIED", "summary": {"compressed": 0.475838, "gated": 0.475838, "naive": 0.609857, "scratch": 0.47987, "ungated": 0.569478}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:33Z", "cpu_seconds": 1.1797591889999985, "created_at": "2026-11-22T04:13:31Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1dbda77c8dc7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "b16e2934179c48138ff36958fbf0eb23aa043a6dd53aa98316677e025ac42d29", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1dbda77c8dc7", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:13:31Z", "seed": 1461116193, "treatment": "frontier"}, "priority": 0.24208542117652673, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "348783b93ca487c005cf86bae830eb3e08220dd18a6ac1c9a9f5b7600339bae4", "run_seconds": 1.2416865825653076, "seed": 1461116193, "started_at": "2026-11-22T04:13:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.103526, 0.177341], "effect": 0.036907, "evaluated_at": "2026-11-22T04:17:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1dbda77c8dc7", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.821428571, 0.8125, 0.844827586, 0.916666667], "frontier": [0.857142857, 0.78125, 0.99137931, 0.875], "goal": [0.75, 0.78125, 0.922413793, 0.875], "hypdisc": [0.642857143, 0.8125, 0.913793103, 0.916666667], "infogain": [0.785714286, 0.8125, 0.99137931, 0.875], "prederr": [0.642857143, 0.8125, 0.99137931, 0.875], "random": [0.857142857, 0.875, 0.75, 0.875], "risk": [0.857142857, 0.78125, 0.99137931, 0.875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "b16e2934179c48138ff36958fbf0eb23aa043a6dd53aa98316677e025ac42d29", "primary_control": "random", "result_hash": "348783b93ca487c005cf86bae830eb3e08220dd18a6ac1c9a9f5b7600339bae4", "signature": "cb1359cf19369f0cb7447bf2f461617b894ede2d63ba2c0bab6292399fa7562949033ce3442f30c82f12b36675352e42c01b38c2aaac7835d8e7711d02bd850f", "state": "VERIFIED", "summary": {"count": 0.848856, "frontier": 0.876193, "goal": 0.832166, "hypdisc": 0.821454, "infogain": 0.866148, "prederr": 0.830434, "random": 0.839286, "risk": 0.876193}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:17:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:36Z", "cpu_seconds": 0.7047265279999984, "created_at": "2026-11-22T04:13:31Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d3ccdb2c133a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "274aa595ee000f8172f92233794b0934f95561b2c2edeafbb66b51394468aa6f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d3ccdb2c133a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:13:31Z", "seed": 692899081, "treatment": "frontier"}, "priority": 0.23913267621176287, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c52587cc1d3f5fda44258d88266e4171a61d009c9365919c026b3c78361a5c12", "run_seconds": 0.7387306690216064, "seed": 692899081, "started_at": "2026-11-22T04:13:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.076142, 0.080309], "effect": 0.002083, "evaluated_at": "2026-11-22T04:17:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d3ccdb2c133a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.9375, 0.9875], "frontier": [1.0, 0.875, 0.927083333, 1.0], "goal": [0.9, 0.916666667, 0.71875, 0.8375], "hypdisc": [0.9875, 0.958333333, 0.9375, 0.925], "infogain": [1.0, 0.875, 0.927083333, 1.0], "prederr": [1.0, 0.875, 0.927083333, 1.0], "random": [0.975, 0.979166667, 0.927083333, 0.9125], "risk": [1.0, 0.875, 0.927083333, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "274aa595ee000f8172f92233794b0934f95561b2c2edeafbb66b51394468aa6f", "primary_control": "random", "result_hash": "c52587cc1d3f5fda44258d88266e4171a61d009c9365919c026b3c78361a5c12", "signature": "0248d91f67cdf3a5d47abb3fd4871a5a0ae8f4e40d617dd31f0e6836fbb00d929a6b6df872046d6535a2011d35e44fa39013fcdac8d6dc8c726c7368a09fd307", "state": "VERIFIED", "summary": {"count": 0.98125, "frontier": 0.950521, "goal": 0.843229, "hypdisc": 0.952083, "infogain": 0.950521, "prederr": 0.950521, "random": 0.948438, "risk": 0.950521}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:17:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:39Z", "cpu_seconds": 0.897263719999998, "created_at": "2026-11-22T04:13:31Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-27728d11f985", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "80a4168ee06aef990353b2ba62a4523ee7302b022b30996fe308465f2ffd5c7a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-27728d11f985", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:13:31Z", "seed": 850887756, "treatment": "full"}, "priority": 0.2353821783564069, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "bc85468f2317e013f136528d71e527115e0825a4bc9ea4d1c70e239c66ac67d0", "run_seconds": 0.920088529586792, "seed": 850887756, "started_at": "2026-11-22T04:13:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.008119, 0.225215], "effect": 0.116667, "evaluated_at": "2026-11-22T04:17:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-27728d11f985", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.741666667, 0.975, 0.975, 0.725], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.958333333, 1.0, 0.991666667, 0.933333333], "retrieval": [0.9, 0.991666667, 0.991666667, 0.9], "rules_nodetect": [0.958333333, 1.0, 0.991666667, 0.933333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "80a4168ee06aef990353b2ba62a4523ee7302b022b30996fe308465f2ffd5c7a", "primary_control": "facts", "result_hash": "bc85468f2317e013f136528d71e527115e0825a4bc9ea4d1c70e239c66ac67d0", "signature": "edbc4a270595fc95cbeb58cc8501c26ef7021866730d5b24b5f8c1c176b2f94d5e7a0695c3efdc17d30dceba0dd862fee96075e47ca5e0eacdc5e7919a175701", "state": "VERIFIED", "summary": {"facts": 0.854167, "frozen": 0.0, "full": 0.970833, "retrieval": 0.945833, "rules_nodetect": 0.970833}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:17:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:41Z", "cpu_seconds": 0.8851859439999998, "created_at": "2026-11-22T04:13:31Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f48ee6903ccf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "e0808f726f24219d16e39d2819bfda5542be182975ff0bae3e7bc2a598aeb24c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f48ee6903ccf", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:13:31Z", "seed": 2051362474, "treatment": "full"}, "priority": 0.23424294494041822, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9ca66f2cd6c7a1bfff86e996af4872459d76d010028ba9c39d290ddb32ef144d", "run_seconds": 0.920708179473877, "seed": 2051362474, "started_at": "2026-11-22T04:13:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.040293, 0.01363], "effect": -0.013332, "evaluated_at": "2026-11-22T04:17:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f48ee6903ccf", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.60952381, 0.372340426, 0.345744681, 0.37254902], "frozen": [0.180952381, 0.214539007, 0.106382979, 0.135620915], "full": [0.595238095, 0.358156028, 0.29964539, 0.39379085], "retrieval": [0.60952381, 0.338652482, 0.260638298, 0.390522876], "rules_nodetect": [0.595238095, 0.803191489, 0.29964539, 0.795751634]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "e0808f726f24219d16e39d2819bfda5542be182975ff0bae3e7bc2a598aeb24c", "primary_control": "facts", "result_hash": "9ca66f2cd6c7a1bfff86e996af4872459d76d010028ba9c39d290ddb32ef144d", "signature": "9df4ca29fb3c3d5fefa560c7aa26192d64947b84fec85bdffd732c8a44d653cfddf346094e59ebfdc4519e9ebc8a81c80bdb3f948a501b79134b153f28865e02", "state": "VERIFIED", "summary": {"facts": 0.425039, "frozen": 0.159374, "full": 0.411708, "retrieval": 0.399834, "rules_nodetect": 0.623457}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:17:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:00Z", "cpu_seconds": 0.9314093979999996, "created_at": "2026-11-22T04:13:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-00cec1169256", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "229f849bddabad45fd4c23fb90a44f963cf264ea13b39a17fd6450c4adef3ce2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-00cec1169256", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:13:47Z", "seed": 520976671, "treatment": "full"}, "priority": 0.2364818501656927, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "db3d38f341d4cb74666aea874c6404146e389374df4d51a10fc5a4307e163aaa", "run_seconds": 0.9457952976226807, "seed": 520976671, "started_at": "2026-11-22T04:13:59Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:17:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-00cec1169256", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.25, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "229f849bddabad45fd4c23fb90a44f963cf264ea13b39a17fd6450c4adef3ce2", "primary_control": "e141_best", "result_hash": "db3d38f341d4cb74666aea874c6404146e389374df4d51a10fc5a4307e163aaa", "signature": "88b4305534b46f894b1a6497321622c0b230b2a9fad81c4a589c93cb57cb61da6ec558f8f44a197d098238c2b9a20fc6d31c9d03f87258bc8890853bb070af01", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 0.9375, "random": 0.9375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:17:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:54Z", "cpu_seconds": 0.6084518079999981, "created_at": "2026-11-22T04:13:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3c4b3b95ee92", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "a2f49530efdc214111cf8aa728c01d1c6b688fd59404516d51200ec5ef961e14", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3c4b3b95ee92", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:13:47Z", "seed": 81529447, "treatment": "full"}, "priority": 0.24344131598774424, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8c24cb819beb17b933f26c3b9b915327a87445c5e1371aec57236e35b8f24849", "run_seconds": 0.617316484451294, "seed": 81529447, "started_at": "2026-11-22T04:13:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:17:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3c4b3b95ee92", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 1.0, 0.25], "fixed_policy": [0.5, 0.0, 0.25, 0.0], "full": [0.5, 0.5, 1.0, 0.25], "no_world_model": [0.25, 0.5, 1.0, 0.0], "random": [0.5, 0.5, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a2f49530efdc214111cf8aa728c01d1c6b688fd59404516d51200ec5ef961e14", "primary_control": "e141_best", "result_hash": "8c24cb819beb17b933f26c3b9b915327a87445c5e1371aec57236e35b8f24849", "signature": "45b465a8d80126f470fa10a20d70a2451879a611f1d9d5895ad540b1598b2fe9f3bdd753c42623cecbbddf696e6857ac31e9af8788e7910eea43a71546cab60c", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.1875, "full": 0.5625, "no_world_model": 0.4375, "random": 0.5}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:17:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:08Z", "cpu_seconds": 34.024414207999996, "created_at": "2026-11-22T04:13:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-56be5464838a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "f7b408f665f7414a2b06cbeabfe2ce43a0a975860f74dd227ae2f3b7c2f30cba", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-56be5464838a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:13:47Z", "seed": 1497379417, "treatment": "full"}, "priority": 0.23334699763404276, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "015658b30b7c7109bd47cba9d62ab7184419d849925457216d0b7f4bc4e71100", "run_seconds": 37.82258892059326, "seed": 1497379417, "started_at": "2026-11-22T04:14:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.191294, 0.441294], "effect": 0.125, "evaluated_at": "2026-11-22T04:18:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-56be5464838a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.0, 0.5, 0.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [0.25, 0.0, 0.75, 0.5], "no_world_model": [0.5, 0.0, 0.5, 0.75], "random": [0.75, 0.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f7b408f665f7414a2b06cbeabfe2ce43a0a975860f74dd227ae2f3b7c2f30cba", "primary_control": "e141_best", "result_hash": "015658b30b7c7109bd47cba9d62ab7184419d849925457216d0b7f4bc4e71100", "signature": "730357a1b48888724c58ec95b95d04dc32ee1299d8214ddfca5d7f4b4e43208a298664e2f23533a9d73041e781281a8d8838095cc33f9862119712eafbb75503", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 0.375, "no_world_model": 0.4375, "random": 0.1875}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:18:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:15Z", "cpu_seconds": 11.513389586999999, "created_at": "2026-11-22T04:13:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b8bb464a2311", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 8, "n": 4}, "prereg_hash": "a3ac4b0791494b49b37c9a55d1d4eb6802b5cfd7cd196aca7c3f21a46ac46043", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b8bb464a2311", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:13:47Z", "seed": 644963740, "treatment": "full"}, "priority": 0.23334998488115888, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9cf15f5028ab54dc142b4d21ea324c1e57f42081b7a3789545f833d236caf55b", "run_seconds": 12.221917629241943, "seed": 644963740, "started_at": "2026-11-22T04:14:02Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:17:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b8bb464a2311", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.8125], "fixed_policy": [1.0, 0.1875], "no_world_model": [1.0, 0.8125], "random": [1.0, 0.875]}, "prereg_hash": "a3ac4b0791494b49b37c9a55d1d4eb6802b5cfd7cd196aca7c3f21a46ac46043", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "9cf15f5028ab54dc142b4d21ea324c1e57f42081b7a3789545f833d236caf55b", "signature": "c02c18303c851047df16c1f5360a4332eb0970f7fad26724801bea8dff1fbbd4ec3c50896ca3c8608ec46bbfc9f24b1511131be96fbf4f64c923e47eb8bfcb0c", "state": "REJECTED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 0.8125, "random": 0.875}}, "verified_at": "2026-11-22T04:17:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:57Z", "cpu_seconds": 0.9306544959999989, "created_at": "2026-11-22T04:13:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f1c0380d20a3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "6b6938baba066107bd1ad1fadfe9d88e36605f7af8e32f3f22303f4ec70977eb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f1c0380d20a3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:13:47Z", "seed": 668329048, "treatment": "full"}, "priority": 0.2391961957484618, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f2a69eee7ca656eaf20645d6e28e273cae8beea86334d3d35781cc8de19ba24b", "run_seconds": 0.9458498954772949, "seed": 668329048, "started_at": "2026-11-22T04:13:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.151278, 0.776278], "effect": 0.3125, "evaluated_at": "2026-11-22T04:17:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f1c0380d20a3", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 0.0, 1.0], "fixed_policy": [0.25, 0.25, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.25, 1.0, 1.0, 1.0], "random": [1.0, 0.25, 0.75, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "6b6938baba066107bd1ad1fadfe9d88e36605f7af8e32f3f22303f4ec70977eb", "primary_control": "e141_best", "result_hash": "f2a69eee7ca656eaf20645d6e28e273cae8beea86334d3d35781cc8de19ba24b", "signature": "d832001201ce904e1074fcd1cdd30a9b3a42c3bd300f73c507da1853a7f6050c91eb7ff9caa3f6e9e92312276f403aca76d56b46bc630c5310eb0748fb64dc0b", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.125, "full": 1.0, "no_world_model": 0.8125, "random": 0.5}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:17:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:13:51Z", "cpu_seconds": 0.6784725749999971, "created_at": "2026-11-22T04:13:47Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-6a88195f49a6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "eaebd6797ba0d9b4d51a8a6c8e42b215a67d6ce581f0a49472d5be2d9ec262ea", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-6a88195f49a6", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:13:47Z", "seed": 32415249, "treatment": "gated"}, "priority": 0.24488471577850363, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6faacf291a13222cd2485b14aeb8b8c116f35c0d00c0fedf21d1aab44a73575c", "run_seconds": 0.6935522556304932, "seed": 32415249, "started_at": "2026-11-22T04:13:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.031587, -0.001092], "effect": -0.01634, "evaluated_at": "2026-11-22T04:17:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-6a88195f49a6", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.424836601, 0.418300654, 0.241830065, 0.307189542], "gated": [0.424836601, 0.418300654, 0.241830065, 0.307189542], "naive": [0.424836601, 0.418300654, 0.169934641, 0.274509804], "scratch": [0.45751634, 0.418300654, 0.248366013, 0.333333333], "ungated": [0.222222222, 0.274509804, 0.241830065, 0.307189542]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "eaebd6797ba0d9b4d51a8a6c8e42b215a67d6ce581f0a49472d5be2d9ec262ea", "primary_control": "scratch", "result_hash": "6faacf291a13222cd2485b14aeb8b8c116f35c0d00c0fedf21d1aab44a73575c", "signature": "15a3bfe935bceb4cba3c5831fa1a1fead18eef0e3b5f308fa9861735f464d33f11f48699f664dbf85515d9beb9b55ec39de89fa9c169c26c8e9194bf0602df03", "state": "VERIFIED", "summary": {"compressed": 0.348039, "gated": 0.348039, "naive": 0.321895, "scratch": 0.364379, "ungated": 0.261438}, "ties": 1, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:13Z", "cpu_seconds": 1.7264675310000008, "created_at": "2026-11-22T04:14:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0a31fe4ed3c6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "1237b96a9bc17df369c1074d290e27c3d95ba34a24929ead1664675a9925be53", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0a31fe4ed3c6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:14:04Z", "seed": 1922853795, "treatment": "full"}, "priority": 0.23964717846194902, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bcbb2872038f2ddb69f7cba717c60459943176bcb314fd13383503946791a73d", "run_seconds": 2.010443925857544, "seed": 1922853795, "started_at": "2026-11-22T04:14:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016412, 1.141412], "effect": 0.5625, "evaluated_at": "2026-11-22T04:17:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0a31fe4ed3c6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.5, 0.0], "fixed_policy": [0.25, 0.0, 0.5, 0.25], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.5, 0.25], "random": [0.75, 0.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "1237b96a9bc17df369c1074d290e27c3d95ba34a24929ead1664675a9925be53", "primary_control": "e141_best", "result_hash": "bcbb2872038f2ddb69f7cba717c60459943176bcb314fd13383503946791a73d", "signature": "4628283bc8d44e8f5aa33caafd39141c23a1329b5b2c6da20612fc8623c3cc57238c957b55acea7ec3fd746f45a8c11a47cacae7d1f7b2a50c98439ca7bbd60b", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.25, "full": 0.9375, "no_world_model": 0.6875, "random": 0.625}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:06Z", "cpu_seconds": 0.8539675189999976, "created_at": "2026-11-22T04:14:04Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8b2da78ca2e6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "de9970037be008fa462a5ce94c5a814d1a48e77aaa5f17b0c6b3431feaa82911", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8b2da78ca2e6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:14:04Z", "seed": 1055403698, "treatment": "full"}, "priority": 0.2396663413211794, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "37f1aaa35af64af110f91473b48f335b8542bb4e329377006cecc161e5181e18", "run_seconds": 1.0606622695922852, "seed": 1055403698, "started_at": "2026-11-22T04:14:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.064071, 0.097821], "effect": 0.016875, "evaluated_at": "2026-11-22T04:17:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8b2da78ca2e6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.613333333, 0.8, 0.9, 0.873333333], "frozen": [0.533333333, 0.6875, 0.55, 0.418333333], "full": [0.540833333, 0.925, 0.925, 0.863333333], "retrieval": [0.748333333, 0.925, 0.9625, 0.88], "rules_nodetect": [0.8, 0.925, 0.925, 0.863333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "de9970037be008fa462a5ce94c5a814d1a48e77aaa5f17b0c6b3431feaa82911", "primary_control": "facts", "result_hash": "37f1aaa35af64af110f91473b48f335b8542bb4e329377006cecc161e5181e18", "signature": "9d2c62a1e8b5ed8fb0b0131a08190202ac46b1ffad0b5e3f4b2f748b72e6f0e48677593140ed8696b70c5fda96a0edd923d63a80d455cbd082919b1c0e99bd07", "state": "VERIFIED", "summary": {"facts": 0.796667, "frozen": 0.547292, "full": 0.813542, "retrieval": 0.878958, "rules_nodetect": 0.878333}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:17:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:21Z", "cpu_seconds": 1.5051857879999986, "created_at": "2026-11-22T04:14:04Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d9e7fb584561", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "74f22a7088a572bd221b119158a838f33d7c814b2c5d7c01cf253d1f6f08a487", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d9e7fb584561", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:14:04Z", "seed": 1684997206, "treatment": "full"}, "priority": 0.23722280409382004, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0bb76597361e6804e1b98ca07424b25b94670c9403af5136dbedac9cc9468226", "run_seconds": 1.5271477699279785, "seed": 1684997206, "started_at": "2026-11-22T04:14:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001706, 0.218618], "effect": 0.110162, "evaluated_at": "2026-11-22T04:17:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d9e7fb584561", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.576923077, 0.145985401, 0.760416667, 0.07712766], "frozen": [0.038461538, 0.063868613, 0.0625, 0.01462766], "full": [0.548076923, 0.328467153, 0.833333333, 0.291223404], "retrieval": [0.557692308, 0.376824818, 0.833333333, 0.295212766], "rules_nodetect": [0.548076923, 0.328467153, 0.833333333, 0.29787234]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "74f22a7088a572bd221b119158a838f33d7c814b2c5d7c01cf253d1f6f08a487", "primary_control": "facts", "result_hash": "0bb76597361e6804e1b98ca07424b25b94670c9403af5136dbedac9cc9468226", "signature": "085391380d70913f8357875c4bc7fdf6a4f9949d87b4fbcb18ac13ec1a07d7c9d4b02b6b8da5bcb914fec7781e962db4b9f659567428e67be5d5215d8da39f0d", "state": "VERIFIED", "summary": {"facts": 0.390113, "frozen": 0.044864, "full": 0.500275, "retrieval": 0.515766, "rules_nodetect": 0.501937}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:17:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:19Z", "cpu_seconds": 0.8590816910000001, "created_at": "2026-11-22T04:14:04Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1b9c5f300aca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "e695d05a98766319ab1094202485517f4f7c4c75d0d5e123c488153b0b9d37ac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1b9c5f300aca", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:14:04Z", "seed": 88109088, "treatment": "gated"}, "priority": 0.2391323756867253, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "392eb218f112be2c19fb3c47102a2fb106e625eb8fd14d6e4309701627625763", "run_seconds": 0.8939332962036133, "seed": 88109088, "started_at": "2026-11-22T04:14:18Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:17:57Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1b9c5f300aca", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.379661017, 0.8625, 0.382291667, 0.714583333], "gated": [0.379661017, 0.8625, 0.382291667, 0.714583333], "naive": [0.400847458, 0.8, 0.39375, 0.582291667], "scratch": [0.379661017, 0.8625, 0.382291667, 0.714583333], "ungated": [0.534745763, 0.825, 0.471875, 0.714583333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e695d05a98766319ab1094202485517f4f7c4c75d0d5e123c488153b0b9d37ac", "primary_control": "scratch", "result_hash": "392eb218f112be2c19fb3c47102a2fb106e625eb8fd14d6e4309701627625763", "signature": "b7e3f996b6a9a66fc785313487d0f9827d7c4738b62e9bbe6b58fb70d6771b5c63a853a9226bfb01bddbf2d3e249851ae83ae3b2ebaf0b2fc3ecd1056adc2805", "state": "PARTIAL", "summary": {"compressed": 0.584759, "gated": 0.584759, "naive": 0.544222, "scratch": 0.584759, "ungated": 0.636551}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:57Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:25Z", "cpu_seconds": 0.6906311589999987, "created_at": "2026-11-22T04:14:04Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-c4cfc42decf9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "b24a1bae9fdd3407a8b91826e44716b528d8ed5a466d474f943626d82d8f9c6c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-c4cfc42decf9", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:14:04Z", "seed": 1750480445, "treatment": "gated"}, "priority": 0.2365610791973431, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "c79be8e0410bfac8b8290019d741a4fe0a00e95f0533b1a695c814af4c74418a", "run_seconds": 0.6913819313049316, "seed": 1750480445, "started_at": "2026-11-22T04:14:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024183, 0.007843], "effect": -0.00817, "evaluated_at": "2026-11-22T04:17:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-c4cfc42decf9", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.307189542, 0.411764706, 0.274509804, 0.411764706], "gated": [0.307189542, 0.411764706, 0.274509804, 0.411764706], "naive": [0.307189542, 0.267973856, 0.274509804, 0.411764706], "scratch": [0.307189542, 0.411764706, 0.274509804, 0.444444444], "ungated": [0.307189542, 0.411764706, 0.274509804, 0.418300654]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "b24a1bae9fdd3407a8b91826e44716b528d8ed5a466d474f943626d82d8f9c6c", "primary_control": "scratch", "result_hash": "c79be8e0410bfac8b8290019d741a4fe0a00e95f0533b1a695c814af4c74418a", "signature": "99568d0a060392c101cdb3a03054eb2d4ec44ddd17ef12843b907503ce745275586611d81175dc643d05dbe8b49ee922ab16261014569ec712fe662fb72b8700", "state": "VERIFIED", "summary": {"compressed": 0.351307, "gated": 0.351307, "naive": 0.315359, "scratch": 0.359477, "ungated": 0.352941}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:17:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:50Z", "cpu_seconds": 0.4552492880000045, "created_at": "2026-11-22T04:14:21Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-446e73f4df3d", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "7aa6175444d8304826584c61db395637f8e559e5df610bcc2d21471973f613c0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-446e73f4df3d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:14:21Z", "seed": 1713550845, "treatment": "frontier"}, "priority": 0.23112255222771433, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-725f52f61405", "result_hash": "2b65647a6a9424f0fe18564537b750c6a6843da63b7f89bfe7894cec4327f291", "run_seconds": 0.4651012420654297, "seed": 1713550845, "started_at": "2026-11-22T04:15:50Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.026003, 0.187366], "effect": 0.080682, "evaluated_at": "2026-11-22T04:22:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-446e73f4df3d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.981818182, 0.536363636, 0.472727273, 1.0], "frontier": [0.95, 0.527272727, 0.290909091, 1.0], "goal": [0.731818182, 0.509090909, 0.463636364, 1.0], "hypdisc": [0.645454545, 0.554545455, 0.490909091, 1.0], "infogain": [0.95, 0.327272727, 0.190909091, 1.0], "prederr": [0.95, 0.327272727, 0.190909091, 1.0], "random": [0.727272727, 0.418181818, 0.3, 1.0], "risk": [0.95, 0.527272727, 0.290909091, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7aa6175444d8304826584c61db395637f8e559e5df610bcc2d21471973f613c0", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-725f52f61405", "result_hash": "2b65647a6a9424f0fe18564537b750c6a6843da63b7f89bfe7894cec4327f291", "signature": "1f72c41e7312c56d4ba5600e8877ac035ee428a87cac5f26164e7c8565913899cb950c22253eaf5f619c71fecf3f7ae8b05bc9939f3175fc7a946adb33952f07", "state": "VERIFIED", "summary": {"count": 0.747727, "frontier": 0.692045, "goal": 0.676136, "hypdisc": 0.672727, "infogain": 0.617045, "prederr": 0.617045, "random": 0.611364, "risk": 0.692045}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:22:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:32Z", "cpu_seconds": 0.6377632619999929, "created_at": "2026-11-22T04:14:21Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-78aa4da6cf7c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "5a5a3b4c4f2d047a50749ec443376cb68453c073c72407a80ece18235f0d73ee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-78aa4da6cf7c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:14:21Z", "seed": 316342992, "treatment": "full"}, "priority": 0.22465850034251442, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-f286d0306f2a", "result_hash": "6ac589bfc61e5226a23e21867faca8b4f334387aa4916985fd872326f9f04ed3", "run_seconds": 0.654921293258667, "seed": 316342992, "started_at": "2026-11-22T04:16:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:24:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-78aa4da6cf7c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 0.5], "fixed_policy": [0.25, 0.5, 0.5, 0.5], "full": [0.75, 1.0, 1.0, 0.0], "no_world_model": [0.0, 1.0, 1.0, 0.0], "random": [0.75, 1.0, 1.0, 0.5]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5a5a3b4c4f2d047a50749ec443376cb68453c073c72407a80ece18235f0d73ee", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-f286d0306f2a", "result_hash": "6ac589bfc61e5226a23e21867faca8b4f334387aa4916985fd872326f9f04ed3", "signature": "9dacbd1d861a09ad45176a1ff5a73bbc720c20826514794c8fbedc49e5f69ec142613b9e7ea7b3475ca6b16c24738ca39314ba3b7f2a87202ea85ddb7791e903", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.4375, "full": 0.6875, "no_world_model": 0.5, "random": 0.8125}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:01Z", "cpu_seconds": 1.574481077999991, "created_at": "2026-11-22T04:14:21Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c870c1eef1aa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "a499251b650dc37f68550939ee3c4fe37cab4930f3718f1f74775bf37f1ffa0f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c870c1eef1aa", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:14:21Z", "seed": 1261108174, "treatment": "full"}, "priority": 0.22891011616109816, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e829b764135384fad45a6856be94ecc12da55e8c9dea706a10ceecc4fe903764", "run_seconds": 2.1092371940612793, "seed": 1261108174, "started_at": "2026-11-22T04:15:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024282, 0.040572], "effect": 0.008145, "evaluated_at": "2026-11-22T04:22:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c870c1eef1aa", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.910714286, 0.070866142, 0.911764706, 0.859375], "frozen": [0.0, 0.036089239, 0.088235294, 0.203125], "full": [0.928571429, 0.038713911, 0.911764706, 0.90625], "retrieval": [0.982142857, 0.072834646, 0.919117647, 0.859375], "rules_nodetect": [0.392857143, 0.038713911, 0.242647059, 0.203125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "a499251b650dc37f68550939ee3c4fe37cab4930f3718f1f74775bf37f1ffa0f", "primary_control": "facts", "result_hash": "e829b764135384fad45a6856be94ecc12da55e8c9dea706a10ceecc4fe903764", "signature": "277c24f14b8533c584e510b299bb626a5cb86e3ba3b7aac49905eb2c7dc10df0ae478d28ffea812caeb833b1994a5c3154c95bd0e9d41777851bf0890b458403", "state": "VERIFIED", "summary": {"facts": 0.68818, "frozen": 0.081862, "full": 0.696325, "retrieval": 0.708368, "rules_nodetect": 0.219336}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:22:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:27Z", "cpu_seconds": 1.1016943879999985, "created_at": "2026-11-22T04:14:21Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-88c480ed97c7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "prereg_hash": "479f3450f471826b4a3c3d00755049acd07fe5a5a917175a4535746216f6c5ac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-88c480ed97c7", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:14:21Z", "seed": 1600022637, "treatment": "gated"}, "priority": 0.23447862677464107, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "2762ce659b63ac15f9239050398bfc4bc27a47d659424732f18574b858fc8823", "run_seconds": 1.127903938293457, "seed": 1600022637, "started_at": "2026-11-22T04:14:26Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:17:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-88c480ed97c7", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.320402299, 0.5625, 0.269117647, 0.278289474], "gated": [0.320402299, 0.5625, 0.269117647, 0.278289474], "naive": [0.050287356, 0.354166667, 0.032352941, 0.259210526], "scratch": [0.320402299, 0.5625, 0.269117647, 0.278289474], "ungated": [0.327586207, 0.354166667, 0.025, 0.258552632]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "479f3450f471826b4a3c3d00755049acd07fe5a5a917175a4535746216f6c5ac", "primary_control": "scratch", "result_hash": "2762ce659b63ac15f9239050398bfc4bc27a47d659424732f18574b858fc8823", "signature": "e5fab0cc618ce69867e400873064dbeff4f593256dae4aebd6067d3923c5396f8aa08e7b5cb3dc66a43424191b6364500e012bb099d6adbe611d8b97c35cb806", "state": "PARTIAL", "summary": {"compressed": 0.357577, "gated": 0.357577, "naive": 0.174004, "scratch": 0.357577, "ungated": 0.241326}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:18:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:51Z", "cpu_seconds": 1.2181669080000006, "created_at": "2026-11-22T04:14:37Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-97523731c10d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "bd4290e7fba1839818726a4fbdba9c883e48f9bec944f38a3d27ec13b8896126", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-97523731c10d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:14:37Z", "seed": 563165643, "treatment": "full"}, "priority": 0.23744179709198715, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6116e24c1cb0a2f30a2acd8bf7ac7188b1837a40416aedba25d0f331f33b904b", "run_seconds": 1.3057637214660645, "seed": 563165643, "started_at": "2026-11-22T04:14:50Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:18:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-97523731c10d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "bd4290e7fba1839818726a4fbdba9c883e48f9bec944f38a3d27ec13b8896126", "primary_control": "e141_best", "result_hash": "6116e24c1cb0a2f30a2acd8bf7ac7188b1837a40416aedba25d0f331f33b904b", "signature": "f4e4d84281a9955cd498a5f85b2d0e3202b246f6ebb57baecc6ef13d0cf7b1b781ef34e1c24813286fb6a6fd5c0a47861662f62274021c2bd04cbfe8ff1b0006", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:18:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:10Z", "cpu_seconds": 0.6308251190000007, "created_at": "2026-11-22T04:14:37Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0840775e2823", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "12533c8f6ac3245d55d39c207d221120f172d7b104847fef3650968818d36be6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0840775e2823", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:14:37Z", "seed": 953645044, "treatment": "full"}, "priority": 0.23364530185147064, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c75f69ee80a83bdbee38fb632fa0fce5f2004b711aceb0c2586ac9c757ad39bc", "run_seconds": 0.6470046043395996, "seed": 953645044, "started_at": "2026-11-22T04:15:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.032615, 0.026504], "effect": -0.003056, "evaluated_at": "2026-11-22T04:18:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0840775e2823", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.857142857, 0.644444444, 0.907407407, 0.515555556], "frozen": [0.657142857, 0.311111111, 0.407407407, 0.417777778], "full": [0.857142857, 0.6, 0.935185185, 0.52], "retrieval": [0.857142857, 0.6, 0.898148148, 0.586666667], "rules_nodetect": [0.771428571, 0.6, 0.675925926, 0.351111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "12533c8f6ac3245d55d39c207d221120f172d7b104847fef3650968818d36be6", "primary_control": "facts", "result_hash": "c75f69ee80a83bdbee38fb632fa0fce5f2004b711aceb0c2586ac9c757ad39bc", "signature": "77c427eae66425e1aaef5e5472e25ce5357a7f1205c76ad92f7b0978069b82ea3fb3d4e5e60d413fc73e87f00f5d11cde2531e7df9edc6d80c14200236e47f0e", "state": "VERIFIED", "summary": {"facts": 0.731138, "frozen": 0.44836, "full": 0.728082, "retrieval": 0.735489, "rules_nodetect": 0.599616}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:18:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:04Z", "cpu_seconds": 1.5090734720000043, "created_at": "2026-11-22T04:14:37Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1a0997ec2f3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "f0ea1fefd5a0f1c3eb21ccec244d8561bfd0f360021e4005f3163b664de7d62d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1a0997ec2f3b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:14:37Z", "seed": 803054596, "treatment": "full"}, "priority": 0.23660343178905335, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5a33a2b728c86623cdb10282bf04abd792c9246c0a8bdc0f777f7040ce8663b0", "run_seconds": 1.9382333755493164, "seed": 803054596, "started_at": "2026-11-22T04:15:02Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.085569, 0.287578], "effect": 0.101004, "evaluated_at": "2026-11-22T04:18:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1a0997ec2f3b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.796875, 0.854166667, 0.616071429, 0.223958333], "frozen": [0.046875, 0.0, 0.089285714, 0.03125], "full": [0.8125, 0.875, 0.598214286, 0.609375], "retrieval": [0.8046875, 0.979166667, 0.607142857, 0.666666667], "rules_nodetect": [0.8125, 0.875, 0.598214286, 0.609375]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "f0ea1fefd5a0f1c3eb21ccec244d8561bfd0f360021e4005f3163b664de7d62d", "primary_control": "facts", "result_hash": "5a33a2b728c86623cdb10282bf04abd792c9246c0a8bdc0f777f7040ce8663b0", "signature": "76f49f72db8192aa7f59e324c06db08e930d995b0e5551a23d592a595a6c14b1ea29ef3381fc3675dc4f3625e467537b46af577d39278bb14e6699b726225e02", "state": "VERIFIED", "summary": {"facts": 0.622768, "frozen": 0.041853, "full": 0.723772, "retrieval": 0.764416, "rules_nodetect": 0.723772}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:18:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:45Z", "cpu_seconds": 4.973327447000003, "created_at": "2026-11-22T04:14:37Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-4585ad7ccd55", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "c69ee51ba451da8d006e78772be66a1c94f0b02495543155664cc45d2d85445c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-4585ad7ccd55", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:14:37Z", "seed": 1514249038, "treatment": "gated"}, "priority": 0.2506039700261754, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-58969546a769", "result_hash": "5cf5f401a5b1aa795a99b65f0b1dc494c12af22ec9df1377eb8074b966c59974", "run_seconds": 5.589319467544556, "seed": 1514249038, "started_at": "2026-11-22T04:14:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.096774, -0.002657], "effect": -0.049716, "evaluated_at": "2026-11-22T04:18:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-4585ad7ccd55", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.56, 0.528846154, 0.62037037, 0.543478261], "gated": [0.56, 0.528846154, 0.62037037, 0.543478261], "naive": [0.72, 0.653846154, 0.694444444, 0.565217391], "scratch": [0.6, 0.644230769, 0.62037037, 0.586956522], "ungated": [0.63, 0.548076923, 0.62037037, 0.543478261]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c69ee51ba451da8d006e78772be66a1c94f0b02495543155664cc45d2d85445c", "primary_control": "scratch", "replicated": false, "replication_of": "E141-ow_transfer-58969546a769", "result_hash": "5cf5f401a5b1aa795a99b65f0b1dc494c12af22ec9df1377eb8074b966c59974", "signature": "413a3817dd4035b5a27ffc52e349d7d521847c13a6f987cab5fdf69ad8e2742ed1ef1dc1592c6b2fba7e71099ad35c28a8d029396242bbfffe40a10d81a81a09", "state": "VERIFIED", "summary": {"compressed": 0.563174, "gated": 0.563174, "naive": 0.658377, "scratch": 0.612889, "ungated": 0.585481}, "ties": 1, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:18:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:40Z", "cpu_seconds": 0.8301359059999953, "created_at": "2026-11-22T04:14:54Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0c634c932220", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "f8e3c93176d8a8a770e0db13b97866e323bc551e36f6a1f6beb06a072691ccf0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0c634c932220", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:14:54Z", "seed": 1406371020, "treatment": "frontier"}, "priority": 0.23151382513339352, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f5f8729198890d53059cb169b938878315bf951ae9cc885b4e3a58bf87f1915a", "run_seconds": 1.2180616855621338, "seed": 1406371020, "started_at": "2026-11-22T04:15:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004505, 0.062839], "effect": 0.029167, "evaluated_at": "2026-11-22T04:18:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0c634c932220", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.916666667, 0.966666667, 0.983333333, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.95, 1.0, 0.933333333, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "f8e3c93176d8a8a770e0db13b97866e323bc551e36f6a1f6beb06a072691ccf0", "primary_control": "random", "result_hash": "f5f8729198890d53059cb169b938878315bf951ae9cc885b4e3a58bf87f1915a", "signature": "4acac5c15cb54e4916c4774fd59991b177f33350397ec98e11bec0ea2a8ed19e326e592e61c6a68a890ac6d13d3a5565524bae76a6a88e609f314fc6f357300d", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 1.0, "goal": 0.966667, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 0.970833, "risk": 1.0}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:18:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:14:57Z", "cpu_seconds": 0.7019777260000026, "created_at": "2026-11-22T04:14:54Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f4bb59a2f533", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "03eeaa9adf2eb72d91e332d8ceee2bc0507d7e7e390661329a1a9e00a3e27f27", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f4bb59a2f533", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:14:54Z", "seed": 1326198278, "treatment": "frontier"}, "priority": 0.23716496111883298, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6aaa7002d1521804a4b94b9a9e3f384e7202a5c3b2d16b0f5c34056bdbe13074", "run_seconds": 0.9546725749969482, "seed": 1326198278, "started_at": "2026-11-22T04:14:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.013, 0.062], "effect": 0.0375, "evaluated_at": "2026-11-22T04:18:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f4bb59a2f533", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.983333333, 1.0], "frontier": [1.0, 1.0, 0.966666667, 1.0], "goal": [0.95, 0.95, 0.966666667, 0.983333333], "hypdisc": [0.983333333, 0.983333333, 1.0, 1.0], "infogain": [1.0, 1.0, 0.966666667, 1.0], "prederr": [1.0, 1.0, 0.966666667, 1.0], "random": [0.95, 0.933333333, 0.95, 0.983333333], "risk": [1.0, 1.0, 0.966666667, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "03eeaa9adf2eb72d91e332d8ceee2bc0507d7e7e390661329a1a9e00a3e27f27", "primary_control": "random", "result_hash": "6aaa7002d1521804a4b94b9a9e3f384e7202a5c3b2d16b0f5c34056bdbe13074", "signature": "0f171c5e2b42b21c8cf21656f0d1f0dfe9db51ac6e5f34c4c03f9193553fa3df7090670fa191c9b50130a3f4bd13aac36a353795cc9d27dc4699b43fe041f800", "state": "VERIFIED", "summary": {"count": 0.995833, "frontier": 0.991667, "goal": 0.9625, "hypdisc": 0.991667, "infogain": 0.991667, "prederr": 0.991667, "random": 0.954167, "risk": 0.991667}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:18:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:17Z", "cpu_seconds": 1.1460718889999981, "created_at": "2026-11-22T04:15:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3d79681c3966", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "8c715c8f27a2da84949f6430e3a5fcb542a3b3fee196762243b35b6cfff9257d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3d79681c3966", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:15:11Z", "seed": 1167257710, "treatment": "frontier"}, "priority": 0.2380208037662559, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "86ac4fca05da1e523e9684700924070d41386bc456a24054970c068b988946d4", "run_seconds": 1.4697215557098389, "seed": 1167257710, "started_at": "2026-11-22T04:15:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.103853, 0.269905], "effect": 0.083026, "evaluated_at": "2026-11-22T04:18:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-3d79681c3966", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.565656566, 0.797979798, 0.717948718, 0.490740741], "frontier": [0.797979798, 0.797979798, 0.769230769, 0.5], "goal": [0.797979798, 0.797979798, 0.666666667, 0.611111111], "hypdisc": [0.777777778, 0.797979798, 0.794871795, 0.694444444], "infogain": [0.797979798, 0.797979798, 0.743589744, 0.657407407], "prederr": [0.717171717, 0.797979798, 0.794871795, 0.731481481], "random": [0.757575758, 0.434343434, 0.794871795, 0.546296296], "risk": [0.797979798, 0.797979798, 0.769230769, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "8c715c8f27a2da84949f6430e3a5fcb542a3b3fee196762243b35b6cfff9257d", "primary_control": "random", "result_hash": "86ac4fca05da1e523e9684700924070d41386bc456a24054970c068b988946d4", "signature": "056ec11bf67988613ff2b4fcd0c66478d70ce0d1a904d5e3dc4ad572d32dbf5a5b86528860f9f0c5f24f93cdee93bbeddee5824ab11ddd63b4c6ff545af6ad0a", "state": "VERIFIED", "summary": {"count": 0.643081, "frontier": 0.716298, "goal": 0.718434, "hypdisc": 0.766268, "infogain": 0.749239, "prederr": 0.760376, "random": 0.633272, "risk": 0.716298}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:18:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:18Z", "cpu_seconds": 3.1562642949999997, "created_at": "2026-11-22T04:15:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4561437620fd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "830f7e6d6972bb1d55d53d8bd3d3a6a6b07aac3640b392ab1b435135e9ac57e3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4561437620fd", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:15:11Z", "seed": 2011530531, "treatment": "frontier"}, "priority": 0.23825724931964093, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "77d943e79d04672b6878a499746990d96a9d9a6ebe7e16db4562e8cb108368a7", "run_seconds": 4.852089881896973, "seed": 2011530531, "started_at": "2026-11-22T04:15:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.031791, 0.261893], "effect": 0.115051, "evaluated_at": "2026-11-22T04:18:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4561437620fd", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.527777778, 0.621052632, 0.421052632, 0.794642857], "frontier": [0.561111111, 0.794736842, 0.339473684, 0.883928571], "goal": [0.538888889, 0.794736842, 0.339473684, 0.758928571], "hypdisc": [0.488888889, 0.678947368, 0.3, 0.767857143], "infogain": [0.516666667, 0.794736842, 0.418421053, 0.883928571], "prederr": [0.338888889, 0.794736842, 0.210526316, 0.883928571], "random": [0.433333333, 0.494736842, 0.405263158, 0.785714286], "risk": [0.561111111, 0.794736842, 0.339473684, 0.875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "830f7e6d6972bb1d55d53d8bd3d3a6a6b07aac3640b392ab1b435135e9ac57e3", "primary_control": "random", "result_hash": "77d943e79d04672b6878a499746990d96a9d9a6ebe7e16db4562e8cb108368a7", "signature": "bebd9992ce4df1a931cfa4bde1964f4d0ff3a6e0beee67dea8a9575918b2bb2fb5bc3e4956d62966b9353e8b88a0e7bcec7879f8f2dd9fd562cbe2140fdbc00c", "state": "VERIFIED", "summary": {"count": 0.591131, "frontier": 0.644813, "goal": 0.608007, "hypdisc": 0.558923, "infogain": 0.653438, "prederr": 0.55702, "random": 0.529762, "risk": 0.64258}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:18:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:23Z", "cpu_seconds": 0.9383488719999988, "created_at": "2026-11-22T04:15:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-550e39560381", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 8, "n": 4}, "prereg_hash": "5819be964ec2b7aa8f60dcf69b3b4e7c60dec3a6f7f9d802db0cd3a9ffb6eafa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-550e39560381", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:15:11Z", "seed": 1922119840, "treatment": "full"}, "priority": 0.23479442141281773, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e5e5cad9a2ef42f22f09bcf9727f81a9ca99066ee31abb5cb247bed7df904682", "run_seconds": 1.1059696674346924, "seed": 1922119840, "started_at": "2026-11-22T04:15:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0526, 0.172634], "effect": 0.060017, "evaluated_at": "2026-11-22T04:18:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-550e39560381", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.534117647, 0.472330097, 0.39375, 0.5015], "frozen": [0.418235294, 0.384951456, 0.315625, 0.421], "full": [0.465294118, 0.47038835, 0.527083333, 0.679], "retrieval": [0.821764706, 0.787378641, 0.776041667, 0.8145], "rules_nodetect": [0.465294118, 0.47038835, 0.527083333, 0.679]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "5819be964ec2b7aa8f60dcf69b3b4e7c60dec3a6f7f9d802db0cd3a9ffb6eafa", "primary_control": "facts", "result_hash": "e5e5cad9a2ef42f22f09bcf9727f81a9ca99066ee31abb5cb247bed7df904682", "signature": "bd74bf4b66d3262e5c3692edbbdd53c81771f47b8f7c4da615d0ac9b1486aaebbf2d7213bc58310bc17511229a0dde4f4b712c92ffc97bc3e40a1df6dd4c8009", "state": "VERIFIED", "summary": {"facts": 0.475424, "frozen": 0.384953, "full": 0.535441, "retrieval": 0.799921, "rules_nodetect": 0.535441}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:18:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:27Z", "cpu_seconds": 5.553392680000002, "created_at": "2026-11-22T04:15:26Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-057d05ad88ae", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 6, "n": 4}, "prereg_hash": "d0a6130a49baadb7358957ec52b7af951385ba733865578e02e52ddd7e598ea2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-057d05ad88ae", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:15:26Z", "seed": 1763318698, "treatment": "frontier"}, "priority": 0.22799828500959263, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-d8e52a86380e", "result_hash": "9d1d25894b47b35b089e43bdb892afcd99a05995cbd035d6e7abd1ab129a3323", "run_seconds": 5.886439085006714, "seed": 1763318698, "started_at": "2026-11-22T04:16:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.004582, 0.147085], "effect": 0.075833, "evaluated_at": "2026-11-22T04:23:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-057d05ad88ae", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.946666667, 0.295, 0.991666667, 0.908333333], "frontier": [0.973333333, 0.465, 1.0, 0.983333333], "goal": [0.92, 0.33, 0.95, 0.941666667], "hypdisc": [0.986666667, 0.3475, 0.983333333, 0.991666667], "infogain": [0.973333333, 0.465, 1.0, 0.983333333], "prederr": [0.586666667, 0.39, 0.891666667, 0.983333333], "random": [1.0, 0.335, 0.925, 0.858333333], "risk": [0.973333333, 0.465, 1.0, 0.983333333]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "d0a6130a49baadb7358957ec52b7af951385ba733865578e02e52ddd7e598ea2", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-d8e52a86380e", "result_hash": "9d1d25894b47b35b089e43bdb892afcd99a05995cbd035d6e7abd1ab129a3323", "signature": "a249a5376f34576a4ae6879c54d05a36752cba98e93a5ba8af5d19c15d5a38be209a5ff026216c6cbb901e678027b8d2eed02bdb131025caf19ab30113d6cc08", "state": "VERIFIED", "summary": {"count": 0.785417, "frontier": 0.855417, "goal": 0.785417, "hypdisc": 0.827292, "infogain": 0.855417, "prederr": 0.712917, "random": 0.779583, "risk": 0.855417}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:23:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:55Z", "cpu_seconds": 1.0700445209999998, "created_at": "2026-11-22T04:15:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-6c02324cee30", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "d68b2ac4b464220a55b196ab4304885f2843970bc125eb0119e8ac237436a56b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-6c02324cee30", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:15:26Z", "seed": 1442008595, "treatment": "full"}, "priority": 0.2299486959105329, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "91b7e55e084c1d0e67c576b57cf70303b64c1137f4c466ddc19b0cc00f0bd7ee", "run_seconds": 1.1087405681610107, "seed": 1442008595, "started_at": "2026-11-22T04:15:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.065, 0.31], "effect": 0.1875, "evaluated_at": "2026-11-22T04:22:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-6c02324cee30", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.75, 0.75, 0.75], "fixed_policy": [0.0, 0.5, 0.0, 0.25], "full": [0.75, 1.0, 1.0, 0.75], "no_world_model": [0.75, 0.75, 1.0, 0.75], "random": [0.5, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "d68b2ac4b464220a55b196ab4304885f2843970bc125eb0119e8ac237436a56b", "primary_control": "e141_best", "result_hash": "91b7e55e084c1d0e67c576b57cf70303b64c1137f4c466ddc19b0cc00f0bd7ee", "signature": "75dc0bfc5f69f72ccc0b04c8f3fd357e63cc9287a081bbc65f8bfa7e65f2bd20c6a4427b28a6648bfd038b19e392bf5e789acc00eb567b49addf3b3ed87cd30a", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.1875, "full": 0.875, "no_world_model": 0.8125, "random": 0.8125}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:22:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:31Z", "cpu_seconds": 3.2821521199999992, "created_at": "2026-11-22T04:15:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a630d0b2b35d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "658ef64329101359c397c6896573bfdc417444a3fbd8e048657259df18e9d372", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a630d0b2b35d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:15:26Z", "seed": 1573731382, "treatment": "full"}, "priority": 0.29284282862033384, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2d6fb6862e865af7dbaa4af884bfe54406d2af1fecde1559f0852aaeedcd7b8a", "run_seconds": 3.6471078395843506, "seed": 1573731382, "started_at": "2026-11-22T04:15:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.013885, 0.232079], "effect": 0.122982, "evaluated_at": "2026-11-22T04:18:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a630d0b2b35d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.85952381, 0.986111111, 0.31840796, 0.715686275], "frozen": [0.269047619, 0.0, 0.199004975, 0.31372549], "full": [0.923809524, 0.986111111, 0.564676617, 0.897058824], "retrieval": [0.973809524, 1.0, 0.592039801, 0.906862745], "rules_nodetect": [0.923809524, 0.986111111, 0.564676617, 0.897058824]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "658ef64329101359c397c6896573bfdc417444a3fbd8e048657259df18e9d372", "primary_control": "facts", "result_hash": "2d6fb6862e865af7dbaa4af884bfe54406d2af1fecde1559f0852aaeedcd7b8a", "signature": "933e3c1ace7733894c5721f5d0452b1e627e9449ec8b959405bef4ff8f9552751dffd2e5333aaa6cdcb4cb77693fdacbfd075b777c503a9de636efd0f2cdaa07", "state": "VERIFIED", "summary": {"facts": 0.719932, "frozen": 0.195445, "full": 0.842914, "retrieval": 0.868178, "rules_nodetect": 0.842914}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:18:47Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:13Z", "cpu_seconds": 201.38043232400003, "created_at": "2026-11-22T04:15:26Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-682c90ae15b2", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "a7402aee49a48e05ede3a50925ed641d4203d180bddfdffdbc65d6c84598db0a", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-682c90ae15b2", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:15:26Z", "seed": 814403811, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "39c8fcd254bebd9063901aca390ef02abd5c4a592f36a1fe680ebd839cc7070d", "run_seconds": 225.23802876472473, "seed": 814403811, "started_at": "2026-11-22T04:15:27Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"compressed": 0, "inherit_mem": 0, "inherit_nomem": 0, "random_search": 0, "reset_mem": 0, "retrieval": 0, "rootcause": 0, "xfer_mem": 2}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:26:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-682c90ae15b2", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0, "risk": 0.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.8}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "compressed": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "inherit_mem": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "inherit_nomem": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "random_search": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "reset_mem": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "retrieval": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "rootcause": [0.786111111, 1.0, 0.78125, 0.687056738, 0.655263158, 0.754403131, 0.901515152, 0.123529412, 0.898989899, 0.941666667], "xfer_mem": [0.823148148, 1.0, 0.802083333, 0.911347518, 0.79122807, 0.254403131, 0.96969697, 0.103529412, 0.898989899, 0.995833333]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "per_generation_sealed": [0.752979, 0.752979, 0.752979, 0.752979], "prereg_hash": "a7402aee49a48e05ede3a50925ed641d4203d180bddfdffdbc65d6c84598db0a", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "rootcause", "xfer_mem"], "result_hash": "39c8fcd254bebd9063901aca390ef02abd5c4a592f36a1fe680ebd839cc7070d", "signature": "3f8b4599049f5659b7b3682cb1000e5527f20cf14a6bbc5fde182eeaa864188b17630beb21a55c57111069af5d4252c5b4c547f7e82792547bed457458128d0b", "state": "PARTIAL", "summary": {"G0": 0.752979, "compressed": 0.752979, "inherit_mem": 0.752979, "inherit_nomem": 0.752979, "random_search": 0.752979, "reset_mem": 0.752979, "retrieval": 0.752979, "rootcause": 0.752979, "xfer_mem": 0.755026}, "ties": 10, "treatment": "inherit_mem", "wins": 0}, "verified_at": "2026-11-22T04:29:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:37Z", "cpu_seconds": 0.9249496130000026, "created_at": "2026-11-22T04:15:26Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-0bd506cbb728", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "2ddedfbd852a6e2f7d09b2b853d6e0962dcc36b1fe568a861f5d09d279b46fb2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-0bd506cbb728", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:15:26Z", "seed": 45452694, "treatment": "gated"}, "priority": 0.2319025564000804, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "3b7730dd1dd13eedf622c7843f94e90f32847c3b04dc07006b7091ddab363630", "run_seconds": 1.0124642848968506, "seed": 45452694, "started_at": "2026-11-22T04:15:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.038971, 0.001233], "effect": -0.018869, "evaluated_at": "2026-11-22T04:18:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-0bd506cbb728", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.491071429, 0.44047619, 0.326923077, 0.43], "gated": [0.491071429, 0.44047619, 0.326923077, 0.43], "naive": [0.580357143, 0.5, 0.423076923, 0.57], "scratch": [0.508928571, 0.488095238, 0.326923077, 0.44], "ungated": [0.580357143, 0.44047619, 0.346153846, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2ddedfbd852a6e2f7d09b2b853d6e0962dcc36b1fe568a861f5d09d279b46fb2", "primary_control": "scratch", "result_hash": "3b7730dd1dd13eedf622c7843f94e90f32847c3b04dc07006b7091ddab363630", "signature": "f723b7219a04d0457844cc88ec27462820de29fca9482e02bc5b79376002ff2935a743ee6e0e51f1beb2cbc993412bfbe21a35d73c746f5db175d5f54c873203", "state": "VERIFIED", "summary": {"compressed": 0.422118, "gated": 0.422118, "naive": 0.518359, "scratch": 0.440987, "ungated": 0.466747}, "ties": 1, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:18:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:47Z", "cpu_seconds": 1.2596306869999978, "created_at": "2026-11-22T04:15:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-25b3bdb656a4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "f1e04de084a06a38db0021bed88968ab448627ff29eb705021b1a6f9be256918", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-25b3bdb656a4", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:15:42Z", "seed": 106675097, "treatment": "frontier"}, "priority": 0.23161805364488713, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c57b036e738b5a2150b414060e37c61f213e9583e6effcb998378a7fc6838b78", "run_seconds": 1.8471276760101318, "seed": 106675097, "started_at": "2026-11-22T04:15:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022256, 0.09792], "effect": 0.037832, "evaluated_at": "2026-11-22T04:18:50Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-25b3bdb656a4", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.766081871, 0.882352941, 0.738562092, 0.6], "frontier": [0.684210526, 0.882352941, 0.790849673, 0.761904762], "goal": [0.684210526, 0.882352941, 0.790849673, 0.771428571], "hypdisc": [0.684210526, 0.882352941, 0.790849673, 0.504761905], "infogain": [0.742690058, 0.843137255, 0.790849673, 0.695238095], "prederr": [0.766081871, 0.843137255, 0.790849673, 0.761904762], "random": [0.695906433, 0.882352941, 0.751633987, 0.638095238], "risk": [0.684210526, 0.882352941, 0.790849673, 0.761904762]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f1e04de084a06a38db0021bed88968ab448627ff29eb705021b1a6f9be256918", "primary_control": "random", "result_hash": "c57b036e738b5a2150b414060e37c61f213e9583e6effcb998378a7fc6838b78", "signature": "e0dbef36de01e6032a37b25eae8932ef7942dd349cd7d10d475796fd074be3fbdebd8c1462a259d2e8ae0d499582807571849536fbea62f695a624581c924a06", "state": "VERIFIED", "summary": {"count": 0.746749, "frontier": 0.779829, "goal": 0.78221, "hypdisc": 0.715544, "infogain": 0.767979, "prederr": 0.790493, "random": 0.741997, "risk": 0.779829}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:18:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:54Z", "cpu_seconds": 1.6667981060000017, "created_at": "2026-11-22T04:15:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8548e9e3ffde", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "7eca058c4e214272d8a1395088f2e025918c39e178164b09479c612d1d9bd15f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8548e9e3ffde", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:15:42Z", "seed": 1965500444, "treatment": "frontier"}, "priority": 0.23031880939492663, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "30142408ff2960f68000bea976e8d53809dec385156fa88e4ce8670eb09a4b00", "run_seconds": 1.747849464416504, "seed": 1965500444, "started_at": "2026-11-22T04:15:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.010868, 0.114217], "effect": 0.062542, "evaluated_at": "2026-11-22T04:22:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-8548e9e3ffde", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.757352941, 0.964285714, 0.727272727, 0.861702128], "frontier": [0.786764706, 0.946428571, 0.863636364, 0.670212766], "goal": [0.786764706, 0.946428571, 0.803030303, 0.670212766], "hypdisc": [0.882352941, 0.964285714, 0.893939394, 0.64893617], "infogain": [0.867647059, 0.964285714, 0.863636364, 0.824468085], "prederr": [0.808823529, 0.607142857, 0.257575758, 0.819148936], "random": [0.676470588, 0.928571429, 0.757575758, 0.654255319], "risk": [0.786764706, 0.946428571, 0.863636364, 0.670212766]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "7eca058c4e214272d8a1395088f2e025918c39e178164b09479c612d1d9bd15f", "primary_control": "random", "result_hash": "30142408ff2960f68000bea976e8d53809dec385156fa88e4ce8670eb09a4b00", "signature": "e002837b449e03f50fcd33b8f84a4b4107062021d70d09f777dd248ccf2514bb5e7d16142a1f37466b52371c9c4c181439855e63056e427cbb5768ca12571006", "state": "VERIFIED", "summary": {"count": 0.827653, "frontier": 0.816761, "goal": 0.801609, "hypdisc": 0.847379, "infogain": 0.880009, "prederr": 0.623173, "random": 0.754218, "risk": 0.816761}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:22:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:09Z", "cpu_seconds": 12.668876226000002, "created_at": "2026-11-22T04:15:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f24287d8d4f7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 10, "n": 4}, "prereg_hash": "7c04ebfbf7dc329ab7bcffef40e1ce7bb202a9481af8f223e1f86c3ddeff393e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f24287d8d4f7", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:15:42Z", "seed": 1963806903, "treatment": "full"}, "priority": 0.22947975204057602, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "db2305a294bb7bcaf9bd4d1c42f02590556aed9bbae8fb3eee0704c3cee94cb3", "run_seconds": 13.741589546203613, "seed": 1963806903, "started_at": "2026-11-22T04:15:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019083, 0.855917], "effect": 0.4375, "evaluated_at": "2026-11-22T04:22:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f24287d8d4f7", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 0.25, 0.25], "fixed_policy": [1.0, 0.0, 0.0, 0.0], "full": [1.0, 0.5, 0.25, 0.75], "no_world_model": [0.5, 0.25, 0.0, 0.0], "random": [0.5, 0.5, 0.5, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "7c04ebfbf7dc329ab7bcffef40e1ce7bb202a9481af8f223e1f86c3ddeff393e", "primary_control": "e141_best", "result_hash": "db2305a294bb7bcaf9bd4d1c42f02590556aed9bbae8fb3eee0704c3cee94cb3", "signature": "14edd98ca8554eca6882fe9155f7ce7eac9e885ee9bacfa52a25cbe6c0c97aa232a757b804160b6a9d6725ca99be75bf4df87e4c881d0d56adef45f0232f1e0d", "state": "VERIFIED", "summary": {"e141_best": 0.1875, "fixed_policy": 0.25, "full": 0.625, "no_world_model": 0.1875, "random": 0.5}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:23:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:15:45Z", "cpu_seconds": 1.7130692340000024, "created_at": "2026-11-22T04:15:42Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-f8a8422284c7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "prereg_hash": "224301c30fc43c12730dd13002ee36edb1b1ce68ca25b4da4d60d450974eaeb5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-f8a8422284c7", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:15:42Z", "seed": 1615034969, "treatment": "gated"}, "priority": 0.23225611456453846, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "befc96a67516cdba4277193ffe00efec9605cba6578cba5be8e7286a0d0a8689", "run_seconds": 2.1296467781066895, "seed": 1615034969, "started_at": "2026-11-22T04:15:42Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:18:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-f8a8422284c7", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.531496063, 0.258806262, 0.30332681, 0.266144814], "gated": [0.531496063, 0.258806262, 0.30332681, 0.266144814], "naive": [0.545275591, 0.260763209, 0.163894325, 0.012720157], "scratch": [0.531496063, 0.258806262, 0.30332681, 0.266144814], "ungated": [0.531496063, 0.261252446, 0.272015656, 0.025440313]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "224301c30fc43c12730dd13002ee36edb1b1ce68ca25b4da4d60d450974eaeb5", "primary_control": "scratch", "result_hash": "befc96a67516cdba4277193ffe00efec9605cba6578cba5be8e7286a0d0a8689", "signature": "82718ab84132bbc965ae6ea50eddf4cc4aae91c0c453f5bfce90207547fb5f79f54c58d593cc9bcea263369784c0b18d5131805946054eb7249534a9ba49f50e", "state": "PARTIAL", "summary": {"compressed": 0.339943, "gated": 0.339943, "naive": 0.245663, "scratch": 0.339943, "ungated": 0.272551}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:18:50Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:15Z", "cpu_seconds": 1.071247115999995, "created_at": "2026-11-22T04:15:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0d14a9af9e69", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "cd6558357bd518dbb78a1dc0833e2284bc349d1b02204194d93150051e51b178", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0d14a9af9e69", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:15:59Z", "seed": 1170044338, "treatment": "frontier"}, "priority": 0.22817832294214507, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "787f46fd4cb32ef9495a0ba3692f4c6bcc4107d90d0697e85d8043a234c8235b", "run_seconds": 1.2050437927246094, "seed": 1170044338, "started_at": "2026-11-22T04:16:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.150417, 0.34398], "effect": 0.096781, "evaluated_at": "2026-11-22T04:23:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0d14a9af9e69", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.454545455, 0.74691358, 0.925925926, 0.709219858], "frontier": [0.793939394, 0.530864198, 0.888888889, 0.638297872], "goal": [0.793939394, 0.401234568, 0.851851852, 0.70212766], "hypdisc": [0.751515152, 0.75308642, 0.888888889, 0.730496454], "infogain": [0.793939394, 0.543209877, 0.888888889, 0.716312057], "prederr": [0.733333333, 0.604938272, 0.777777778, 0.113475177], "random": [0.478787879, 0.765432099, 0.851851852, 0.368794326], "risk": [0.793939394, 0.530864198, 0.888888889, 0.638297872]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "cd6558357bd518dbb78a1dc0833e2284bc349d1b02204194d93150051e51b178", "primary_control": "random", "result_hash": "787f46fd4cb32ef9495a0ba3692f4c6bcc4107d90d0697e85d8043a234c8235b", "signature": "dbbc58e951ea36794d6ea7cccc8938243cbfc2005e978318ff31b33943cf8d4df2529a70d16a0266452bd5d72e693cace80fc82282165d54285308cdf7b25508", "state": "VERIFIED", "summary": {"count": 0.709151, "frontier": 0.712998, "goal": 0.687288, "hypdisc": 0.780997, "infogain": 0.735588, "prederr": 0.557381, "random": 0.616217, "risk": 0.712998}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:23:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:28Z", "cpu_seconds": 2.0803290780000054, "created_at": "2026-11-22T04:15:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0dc0d4a95f02", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "dd3110b6c54a971d57bb63896e1e5d9e2669ea5f9ce813070a93c42e02c3f2ee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0dc0d4a95f02", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:15:59Z", "seed": 291857901, "treatment": "frontier"}, "priority": 0.22688755080820283, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d5a1daef9ce8449812a26349abfe070285d744bbcde32c4f255d9113d6c9d810", "run_seconds": 2.532748222351074, "seed": 291857901, "started_at": "2026-11-22T04:16:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.007697, 0.343494], "effect": 0.175595, "evaluated_at": "2026-11-22T04:23:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0dc0d4a95f02", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.875, 0.841666667, 0.910714286], "frontier": [0.925, 1.0, 1.0, 1.0], "goal": [0.9, 1.0, 1.0, 0.910714286], "hypdisc": [0.95, 0.866071429, 0.983333333, 1.0], "infogain": [0.925, 1.0, 1.0, 1.0], "prederr": [0.925, 1.0, 1.0, 1.0], "random": [0.975, 0.857142857, 0.658333333, 0.732142857], "risk": [0.925, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "dd3110b6c54a971d57bb63896e1e5d9e2669ea5f9ce813070a93c42e02c3f2ee", "primary_control": "random", "result_hash": "d5a1daef9ce8449812a26349abfe070285d744bbcde32c4f255d9113d6c9d810", "signature": "ce8af7c52717c06e36b5cc38f415b7b6bfac7c6d76ac15e746ee76bfacde77b43c1b768cabe5f392772a314e78f48b6a6d19ddd026c5f79b4adb228cf089860b", "state": "VERIFIED", "summary": {"count": 0.906845, "frontier": 0.98125, "goal": 0.952679, "hypdisc": 0.949851, "infogain": 0.98125, "prederr": 0.98125, "random": 0.805655, "risk": 0.98125}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:23:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:16Z", "cpu_seconds": 1.8043182690000066, "created_at": "2026-11-22T04:15:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-32b4cb0bbf32", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "62876a55fb839a389238d738d6bda33987bd5b1c87664785ea1d2bfc33a17ef5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-32b4cb0bbf32", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:15:59Z", "seed": 1799629074, "treatment": "full"}, "priority": 0.2287799402428888, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bb4176e914710ce4e0ef4da7d20be98d9c7167ee285f7dea1c1ee1974c1a813f", "run_seconds": 1.9024007320404053, "seed": 1799629074, "started_at": "2026-11-22T04:16:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T04:23:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-32b4cb0bbf32", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.5], "fixed_policy": [0.75, 0.5, 1.0, 1.0], "full": [0.75, 1.0, 1.0, 0.5], "no_world_model": [1.0, 1.0, 1.0, 0.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "62876a55fb839a389238d738d6bda33987bd5b1c87664785ea1d2bfc33a17ef5", "primary_control": "e141_best", "result_hash": "bb4176e914710ce4e0ef4da7d20be98d9c7167ee285f7dea1c1ee1974c1a813f", "signature": "0f641df1f13dfd36f921a37809d45472f8e9f75561b92594186a1f4e5fa9168849e5790ee1fe2f2d0ad305324f9215229de9ef8b1de8dba871e8d00d1726c104", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.8125, "full": 0.8125, "no_world_model": 0.75, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:23:11Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:02Z", "cpu_seconds": 2.094805124000004, "created_at": "2026-11-22T04:15:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-51d2cf6e800a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "0e3a9529586e20a3371e43f4d53c1610a7fadaeacbb5a39e4ed6b4397c032a0d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-51d2cf6e800a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:15:59Z", "seed": 899350251, "treatment": "full"}, "priority": 0.24045258700690733, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e88f40a39680ff275b6f9f276dfb2c99b0dc5bbceee4daf41fbb71c5b8b19530", "run_seconds": 2.5777716636657715, "seed": 899350251, "started_at": "2026-11-22T04:16:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T04:22:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-51d2cf6e800a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 0.75, 1.0], "fixed_policy": [0.0, 0.0, 1.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "0e3a9529586e20a3371e43f4d53c1610a7fadaeacbb5a39e4ed6b4397c032a0d", "primary_control": "e141_best", "result_hash": "e88f40a39680ff275b6f9f276dfb2c99b0dc5bbceee4daf41fbb71c5b8b19530", "signature": "0788f32c9808aee7727fd76621f14bedebc4f5d738429e77ad995cab418a1e7fc6605bae3bda3dac7edf6fd6951fc13cd46b83c15731a8621b564b4c12742701", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:22:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:09Z", "cpu_seconds": 1.5341695930000014, "created_at": "2026-11-22T04:15:59Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b69c0c6d57ec", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "9f9e6f237f84a80fc8eb8e3bad94fce5c7338eb741e20c9390d70fca9cb4cf1b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b69c0c6d57ec", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:15:59Z", "seed": 1183167892, "treatment": "full"}, "priority": 0.2312434879640893, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-c2419fed8634", "result_hash": "4de48d1bffcc90fec28a099334efb2d5e258ecfb494eda8e036be249c25732d0", "run_seconds": 1.7995479106903076, "seed": 1183167892, "started_at": "2026-11-22T04:16:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.006749, 0.120721], "effect": 0.063735, "evaluated_at": "2026-11-22T04:23:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b69c0c6d57ec", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.266666667, 0.583333333, 0.658823529, 0.406593407], "frozen": [0.0, 0.0, 0.087394958, 0.0], "full": [0.341666667, 0.583333333, 0.796638655, 0.448717949], "retrieval": [0.78, 0.826666667, 0.830252101, 0.84981685], "rules_nodetect": [0.291666667, 0.583333333, 0.796638655, 0.448717949]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "9f9e6f237f84a80fc8eb8e3bad94fce5c7338eb741e20c9390d70fca9cb4cf1b", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-c2419fed8634", "result_hash": "4de48d1bffcc90fec28a099334efb2d5e258ecfb494eda8e036be249c25732d0", "signature": "a51b53ada2403ee05871ab9b74bc8cc0cb8ed7e035ed66ac2dd7daab586241affd5180ee53bbf74d8d6a0c74c63b1ae8f274c52ae050c94b48c406ae85b22d03", "state": "VERIFIED", "summary": {"facts": 0.478854, "frozen": 0.021849, "full": 0.542589, "retrieval": 0.821684, "rules_nodetect": 0.530089}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:23:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:13Z", "cpu_seconds": 1.144856414000003, "created_at": "2026-11-22T04:15:59Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c96cbd7a560e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "5357af5dd8b6cee36c61f121cbb36482e00adf32b3e874bce6a03800d582ba9b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c96cbd7a560e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:15:59Z", "seed": 1170290640, "treatment": "full"}, "priority": 0.2305622795859118, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7356ffc52d9d558d18c02934dd9a3c99b8d97c4ce6cf9a6758bb37b505f4bd97", "run_seconds": 1.1882357597351074, "seed": 1170290640, "started_at": "2026-11-22T04:16:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.039187, 0.003048], "effect": -0.01807, "evaluated_at": "2026-11-22T04:23:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c96cbd7a560e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.489361702, 0.712765957, 0.562857143, 0.468085106], "frozen": [0.127659574, 0.127659574, 0.12, 0.127659574], "full": [0.463829787, 0.7, 0.571428571, 0.425531915], "retrieval": [0.489361702, 0.719148936, 0.557142857, 0.45106383], "rules_nodetect": [0.463829787, 0.7, 0.571428571, 0.657446809]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "5357af5dd8b6cee36c61f121cbb36482e00adf32b3e874bce6a03800d582ba9b", "primary_control": "facts", "result_hash": "7356ffc52d9d558d18c02934dd9a3c99b8d97c4ce6cf9a6758bb37b505f4bd97", "signature": "9f4e1503887b108f15ac2550ebdce21588de2ffbbf38be5bb89268c05b038500454155044dd91ae2e7d6d00c03a22140ebce04dc4a9ffb458492bd1d72074c05", "state": "VERIFIED", "summary": {"facts": 0.558267, "frozen": 0.125745, "full": 0.540198, "retrieval": 0.554179, "rules_nodetect": 0.598176}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:23:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:07Z", "cpu_seconds": 0.9112109349999997, "created_at": "2026-11-22T04:15:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1ce8f6045ef3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "f6981f9c5e89bb7b6eb43f11ac831a315548e9ce0e9c592e81643bc3475a21da", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1ce8f6045ef3", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:15:59Z", "seed": 1123727682, "treatment": "gated"}, "priority": 0.23610518874895176, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "3617105fdc30528af722565e2b93223ae064c397c525ddcd80bbb12e5cef394e", "run_seconds": 0.946972131729126, "seed": 1123727682, "started_at": "2026-11-22T04:16:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.042806, 0.060049], "effect": 0.008621, "evaluated_at": "2026-11-22T04:22:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-1ce8f6045ef3", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.19047619, 0.377777778, 0.219047619, 0.205479452], "gated": [0.19047619, 0.377777778, 0.219047619, 0.205479452], "naive": [0.061904762, 0.355555556, 0.085714286, 0.063013699], "scratch": [0.10952381, 0.422222222, 0.223809524, 0.202739726], "ungated": [0.123809524, 0.311111111, 0.09047619, 0.084931507]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "f6981f9c5e89bb7b6eb43f11ac831a315548e9ce0e9c592e81643bc3475a21da", "primary_control": "scratch", "result_hash": "3617105fdc30528af722565e2b93223ae064c397c525ddcd80bbb12e5cef394e", "signature": "96e4259712c921f74ac095af74de35657783ac7464afd876c574eab3231bb2ec9e7eb4aad8bd493d49738dee9d7e5f9e589e2f10ad6babe01b5d3f1499294805", "state": "VERIFIED", "summary": {"compressed": 0.248195, "gated": 0.248195, "naive": 0.141547, "scratch": 0.239574, "ungated": 0.152582}, "ties": 0, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:22:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:53Z", "cpu_seconds": 18.593857909999997, "created_at": "2026-11-22T04:16:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3c19bdbbc740", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 4, "n": 4}, "prereg_hash": "a3bc73b532beb995c4cd5d21ddf74bf2381116f02d5e33c30f83656dd449f8d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3c19bdbbc740", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:16:16Z", "seed": 1444331034, "treatment": "frontier"}, "priority": 0.21656948005505094, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "22ac3bb182f894a4e3333f403510a473bf55151696beab4ffa6d52ffb7f3c95b", "run_seconds": 20.85839080810547, "seed": 1444331034, "started_at": "2026-11-22T04:16:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035553, 0.156829], "effect": 0.060638, "evaluated_at": "2026-11-22T04:24:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-3c19bdbbc740", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.7078125, 0.471666667, 0.64765625, 0.690625], "frontier": [0.765625, 0.6525, 0.54296875, 0.78125], "goal": [0.7875, 0.6525, 0.54296875, 0.7078125], "hypdisc": [0.7578125, 0.584166667, 0.64609375, 0.7796875], "infogain": [0.765625, 0.6525, 0.54296875, 0.78125], "prederr": [0.765625, 0.6525, 0.54296875, 0.78125], "random": [0.684375, 0.534166667, 0.6265625, 0.6546875], "risk": [0.765625, 0.6525, 0.54296875, 0.78125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "a3bc73b532beb995c4cd5d21ddf74bf2381116f02d5e33c30f83656dd449f8d2", "primary_control": "random", "result_hash": "22ac3bb182f894a4e3333f403510a473bf55151696beab4ffa6d52ffb7f3c95b", "signature": "720d129b94fb3dab20ae89f1bb808069edd2e30518f40cfe889743e93b509e8633561a54f5fbfbbd57fdbc29a728607c34de95af04a49406706b8e573ef0ef0b", "state": "VERIFIED", "summary": {"count": 0.62944, "frontier": 0.685586, "goal": 0.672695, "hypdisc": 0.69194, "infogain": 0.685586, "prederr": 0.685586, "random": 0.624948, "risk": 0.685586}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:24:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:20Z", "cpu_seconds": 1.5401993509999983, "created_at": "2026-11-22T04:16:16Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-af2e815a441a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "cdf5f0ff5755b6bf54b143baae7f8ab09fa71a74ff834796de298b8cb1f421fe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-af2e815a441a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:16:16Z", "seed": 1672381218, "treatment": "frontier"}, "priority": 0.2387789899973178, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "46a8d7a14881fb90f295cd9fa3ca39e8321abbe30f42f62b4830744793e6717a", "run_seconds": 1.571887731552124, "seed": 1672381218, "started_at": "2026-11-22T04:16:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.077531, 0.193711], "effect": 0.135621, "evaluated_at": "2026-11-22T04:23:11Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-af2e815a441a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.751633987, 0.764705882, 0.562091503, 0.764705882], "frontier": [0.77124183, 0.888888889, 0.562091503, 0.849673203], "goal": [0.764705882, 0.816993464, 0.588235294, 0.633986928], "hypdisc": [0.679738562, 0.823529412, 0.555555556, 0.666666667], "infogain": [0.77124183, 0.888888889, 0.562091503, 0.849673203], "prederr": [0.77124183, 0.888888889, 0.562091503, 0.849673203], "random": [0.647058824, 0.679738562, 0.496732026, 0.705882353], "risk": [0.77124183, 0.888888889, 0.562091503, 0.849673203]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "cdf5f0ff5755b6bf54b143baae7f8ab09fa71a74ff834796de298b8cb1f421fe", "primary_control": "random", "result_hash": "46a8d7a14881fb90f295cd9fa3ca39e8321abbe30f42f62b4830744793e6717a", "signature": "253525e412e00006567cb738902393022b18b88f7a49b8bc78bc58f1e2cd7addeed33256bc3372cc1ce5e6feece97fadee3aa30cbc998720c48e4206b9e3020a", "state": "VERIFIED", "summary": {"count": 0.710784, "frontier": 0.767974, "goal": 0.70098, "hypdisc": 0.681373, "infogain": 0.767974, "prederr": 0.767974, "random": 0.632353, "risk": 0.767974}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:23:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:40Z", "cpu_seconds": 18.610938628999975, "created_at": "2026-11-22T04:16:16Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0921f66c17ea", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 8, "n": 4}, "prereg_hash": "1234205218c73041ebd49393ce541402e4480d3f358b1ab30a370cb818439764", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0921f66c17ea", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:16:16Z", "seed": 901863972, "treatment": "full"}, "priority": 0.21516036535397365, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bfd32058b90e02d43957b0993333ec05d9faed4d2c5b23eb2662c06d1f3f0155", "run_seconds": 24.540624856948853, "seed": 901863972, "started_at": "2026-11-22T04:17:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.19, 0.435], "effect": 0.3125, "evaluated_at": "2026-11-22T04:24:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0921f66c17ea", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.5, 0.75, 0.75], "fixed_policy": [0.25, 0.5, 0.5, 0.0], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [0.75, 0.5, 0.5, 0.75], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "1234205218c73041ebd49393ce541402e4480d3f358b1ab30a370cb818439764", "primary_control": "e141_best", "result_hash": "bfd32058b90e02d43957b0993333ec05d9faed4d2c5b23eb2662c06d1f3f0155", "signature": "b091f5e855516d72b759e8a9b34bd9b17c4a9d8bc7f620550bb1c7df4014369174c0f8e3f6edce1f61df1d301acc66c544767716f34ffe759bd7d8bcc552f501", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.3125, "full": 0.9375, "no_world_model": 0.625, "random": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:24:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:30Z", "cpu_seconds": 2.9843763820000504, "created_at": "2026-11-22T04:16:16Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-50c7a709ddab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "c33e2bb27d6936f02624e4c99ee4e71b1ea05679343e58c8653dc172061c246b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-50c7a709ddab", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:16:16Z", "seed": 867292260, "treatment": "full"}, "priority": 0.2251548344899328, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b0be104f5661ce98918303d888035c5bfb9412ff315132a2bbcd726ce29f36e2", "run_seconds": 3.27020263671875, "seed": 867292260, "started_at": "2026-11-22T04:16:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:23:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-50c7a709ddab", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 0.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 0.0, 1.0, 0.0], "no_world_model": [1.0, 0.0, 1.0, 0.25], "random": [1.0, 0.0, 1.0, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c33e2bb27d6936f02624e4c99ee4e71b1ea05679343e58c8653dc172061c246b", "primary_control": "e141_best", "result_hash": "b0be104f5661ce98918303d888035c5bfb9412ff315132a2bbcd726ce29f36e2", "signature": "df17c1cbf608977f0f134853387bc76861e2708fd90d986c0233c2cc71664186926bc8f954fb8e6eb64841ce67ff45d177829d6cf29a08e818b7f1ce6648d20b", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.0625, "full": 0.5, "no_world_model": 0.5625, "random": 0.5625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:22Z", "cpu_seconds": 1.44201823100002, "created_at": "2026-11-22T04:16:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-87ad21c98236", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 6, "n": 4}, "prereg_hash": "108cd8615b3a60626128bcc72c24ae9587f06876473dd85a57bca8de79c62024", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-87ad21c98236", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:16:16Z", "seed": 1731687150, "treatment": "full"}, "priority": 0.22995126856023057, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "005098006d91171876cc8ef9e271dc746fc2f4f7e7ccc7279d2ec4aa3e280178", "run_seconds": 1.5202717781066895, "seed": 1731687150, "started_at": "2026-11-22T04:16:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.039686, 0.289552], "effect": 0.164619, "evaluated_at": "2026-11-22T04:23:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-87ad21c98236", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.975, 0.417391304, 0.424390244, 0.733333333], "frozen": [0.358333333, 0.095652174, 0.107317073, 0.346666667], "full": [0.958333333, 0.698550725, 0.631707317, 0.92], "retrieval": [0.966666667, 0.707246377, 0.670731707, 0.96], "rules_nodetect": [0.5, 0.342028986, 0.241463415, 0.373333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "108cd8615b3a60626128bcc72c24ae9587f06876473dd85a57bca8de79c62024", "primary_control": "facts", "result_hash": "005098006d91171876cc8ef9e271dc746fc2f4f7e7ccc7279d2ec4aa3e280178", "signature": "e091c45453f3c062cdde8dd1d8b9dfd0dc658fa6790d827be6ba08901d116bfb4f00b3c21ecfad2a3b8a2fafafc640af05c13948ddcee06d20103ba81d11ef0b", "state": "VERIFIED", "summary": {"facts": 0.637529, "frozen": 0.226992, "full": 0.802148, "retrieval": 0.826161, "rules_nodetect": 0.364206}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:23:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:25Z", "cpu_seconds": 1.1172962170000034, "created_at": "2026-11-22T04:16:16Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a143b7b8b479", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "e7cb995ac3117cfdf7f1e9c2ee5a9d035127d49434a2460bdc3a720a54691e54", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a143b7b8b479", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:16:16Z", "seed": 208012905, "treatment": "full"}, "priority": 0.21373677690877832, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "dc08563835b11f0529ce13a7b69d73facbbf6953b82c5bf5dbece1e672441ecd", "run_seconds": 2.720369577407837, "seed": 208012905, "started_at": "2026-11-22T04:17:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.036208, 0.066151], "effect": 0.014972, "evaluated_at": "2026-11-22T04:24:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a143b7b8b479", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.55862069, 0.768421053, 0.523529412, 0.271428571], "frozen": [0.103448276, 0.115789474, 0.035294118, 0.123809524], "full": [0.589655172, 0.768421053, 0.6, 0.223809524], "retrieval": [0.6, 0.768421053, 0.558823529, 0.233333333], "rules_nodetect": [0.589655172, 0.768421053, 0.6, 0.223809524]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "e7cb995ac3117cfdf7f1e9c2ee5a9d035127d49434a2460bdc3a720a54691e54", "primary_control": "facts", "result_hash": "dc08563835b11f0529ce13a7b69d73facbbf6953b82c5bf5dbece1e672441ecd", "signature": "d7dd499a1315bdca6b7b652a2dc91c05d149f4e67482736c0c3a1a95f6be327aa3ce5c2426f3397a782c757c701e29e70e268650dbc425706927c06ce0d7350b", "state": "VERIFIED", "summary": {"facts": 0.5305, "frozen": 0.094585, "full": 0.545471, "retrieval": 0.540144, "rules_nodetect": 0.545471}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:24:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:08Z", "cpu_seconds": 2.08742466999999, "created_at": "2026-11-22T04:16:33Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ce0419be9d47", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "0ddb05bddbd0b5833179c82b91182cf5f1dd4a33ab270b788db28fdc4e2dc094", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ce0419be9d47", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:16:33Z", "seed": 1539842184, "treatment": "frontier"}, "priority": 0.21701448315610672, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5b099237dff2e1b622160b987434d39ed3712e933598bdc1d7b6aaa5f46e62f8", "run_seconds": 2.610585927963257, "seed": 1539842184, "started_at": "2026-11-22T04:17:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.07507, 0.189874], "effect": 0.057402, "evaluated_at": "2026-11-22T04:24:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ce0419be9d47", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.62, 0.958333333, 0.980769231, 0.886363636], "frontier": [0.52, 1.0, 1.0, 0.988636364], "goal": [0.52, 1.0, 0.932692308, 0.977272727], "hypdisc": [0.52, 1.0, 1.0, 1.0], "infogain": [0.52, 1.0, 1.0, 0.988636364], "prederr": [0.52, 1.0, 1.0, 0.988636364], "random": [0.66, 0.854166667, 0.855769231, 0.909090909], "risk": [0.52, 1.0, 1.0, 0.988636364]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "0ddb05bddbd0b5833179c82b91182cf5f1dd4a33ab270b788db28fdc4e2dc094", "primary_control": "random", "result_hash": "5b099237dff2e1b622160b987434d39ed3712e933598bdc1d7b6aaa5f46e62f8", "signature": "249a8e80ba2123aa480e41cc7d50e5bda831a8ce93a25ef1ea7c1b6af2e24b3dec9a368a36c70448b09cdd429e0f4bbb9994dcfe1527200a51619a6bbe28da0f", "state": "VERIFIED", "summary": {"count": 0.861367, "frontier": 0.877159, "goal": 0.857491, "hypdisc": 0.88, "infogain": 0.877159, "prederr": 0.877159, "random": 0.819757, "risk": 0.877159}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:24:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:49Z", "cpu_seconds": 1.555905475000003, "created_at": "2026-11-22T04:16:33Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d49148c64473", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "e390f5ad4cd6946f5684898de231e7268f55cfe743d12d10fc085fed2ad4275c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d49148c64473", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:16:33Z", "seed": 2076288732, "treatment": "frontier"}, "priority": 0.22459511618557587, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4b153be8012e4197c482ab58067c418bd1e62ea4b825dbc636c5219d37b06d18", "run_seconds": 2.1135268211364746, "seed": 2076288732, "started_at": "2026-11-22T04:16:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.048607, -0.009726], "effect": -0.029167, "evaluated_at": "2026-11-22T04:24:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d49148c64473", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.916666667, 1.0, 0.958333333, 0.925], "goal": [0.875, 1.0, 0.966666667, 0.841666667], "hypdisc": [0.941666667, 1.0, 0.958333333, 0.933333333], "infogain": [0.916666667, 1.0, 0.958333333, 0.925], "prederr": [0.916666667, 1.0, 0.958333333, 0.925], "random": [0.95, 1.0, 1.0, 0.966666667], "risk": [0.916666667, 1.0, 0.958333333, 0.925]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e390f5ad4cd6946f5684898de231e7268f55cfe743d12d10fc085fed2ad4275c", "primary_control": "random", "result_hash": "4b153be8012e4197c482ab58067c418bd1e62ea4b825dbc636c5219d37b06d18", "signature": "08adeb9043f59ccd3f9c50661ac95f1aafca2f0d0bad9dab718239af32d5d4df02ea6086e7ccaed0961c70e6ec45d32a8278887a7704956e61aba9f097002508", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.95, "goal": 0.920833, "hypdisc": 0.958333, "infogain": 0.95, "prederr": 0.95, "random": 0.979167, "risk": 0.95}, "ties": 1, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:24:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:42Z", "cpu_seconds": 4.877070914000001, "created_at": "2026-11-22T04:16:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-be297dc849b5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "5f656c26a8ebde50d401a9c83f38eaeed9799c5b9537057f57faaf665f1739c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-be297dc849b5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:16:33Z", "seed": 1067359711, "treatment": "full"}, "priority": 0.22852150511993624, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0d964fe99d49d9e211b751097cccd5143aa65289cce14320d4da418de1a3d062", "run_seconds": 6.055575847625732, "seed": 1067359711, "started_at": "2026-11-22T04:16:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.04707, 0.42207], "effect": 0.1875, "evaluated_at": "2026-11-22T04:24:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-be297dc849b5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.5, 0.0, 0.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [1.0, 0.75, 0.0, 0.0], "no_world_model": [0.75, 0.5, 0.25, 0.0], "random": [0.25, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "5f656c26a8ebde50d401a9c83f38eaeed9799c5b9537057f57faaf665f1739c3", "primary_control": "e141_best", "result_hash": "0d964fe99d49d9e211b751097cccd5143aa65289cce14320d4da418de1a3d062", "signature": "df552e14efe060ac27c68910f905e21792844b62406ce1dc7272d1e0cdb6243081402c8f0bac2f6a38028821a1de5a28efc4148d9254b6f5c0d80874daf3ac09", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 0.4375, "no_world_model": 0.375, "random": 0.3125}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:24:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:04Z", "cpu_seconds": 0.1701080609999508, "created_at": "2026-11-22T04:16:33Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ebe631d71edf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "f1f298fd4495c2b3582cff91cdb83065449a0d642604a62087b3ddf7b63f4e5e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ebe631d71edf", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:16:33Z", "seed": 1205579934, "treatment": "full"}, "priority": 0.21998025073706173, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5bdb0142cae007c937683b41d61f3f9f4568dcf363c58c1cb0cb8ae86088bdc5", "run_seconds": 0.280519962310791, "seed": 1205579934, "started_at": "2026-11-22T04:17:04Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:24:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ebe631d71edf", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.75, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f1f298fd4495c2b3582cff91cdb83065449a0d642604a62087b3ddf7b63f4e5e", "primary_control": "e141_best", "result_hash": "5bdb0142cae007c937683b41d61f3f9f4568dcf363c58c1cb0cb8ae86088bdc5", "signature": "8104846995643a911b2959ed58b534cf4e2ca50c2d50f95d9d56395b4b238ce95d7782325f4f74ee9519f6cc564c3f0c4d9ca1a45bea401e2422f688ebef1901", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:49Z", "cpu_seconds": 1.7688696339999979, "created_at": "2026-11-22T04:16:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-257f07574a07", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "c70878810733b2c88430e9ddfa5b388b461048bb232208e270a8468f44620c2f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-257f07574a07", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:16:33Z", "seed": 1181986179, "treatment": "full"}, "priority": 0.2225494144078936, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "eb8adde26371f0f08b023cfc865bb003d2a546c8a8e90f260b3edaaccc34f950", "run_seconds": 2.7427308559417725, "seed": 1181986179, "started_at": "2026-11-22T04:16:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.019144, 0.109475], "effect": 0.045165, "evaluated_at": "2026-11-22T04:24:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-257f07574a07", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.958333333, 0.493333333, 0.96, 0.516455696], "frozen": [0.358333333, 0.0, 0.33, 0.108860759], "full": [0.958333333, 0.502222222, 0.99, 0.658227848], "retrieval": [0.975, 0.855555556, 0.99, 0.73164557], "rules_nodetect": [0.958333333, 0.502222222, 0.99, 0.658227848]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c70878810733b2c88430e9ddfa5b388b461048bb232208e270a8468f44620c2f", "primary_control": "facts", "result_hash": "eb8adde26371f0f08b023cfc865bb003d2a546c8a8e90f260b3edaaccc34f950", "signature": "b4f1551d20ae562cd07e7a741ebc2aefb5d47160ef917e2096040fbb31d5747f013fdb5b29e3446a0854ca53c6941bbc9463a5814d1d8819d087ada55adaa905", "state": "VERIFIED", "summary": {"facts": 0.732031, "frozen": 0.199299, "full": 0.777196, "retrieval": 0.88805, "rules_nodetect": 0.777196}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:24:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:42Z", "cpu_seconds": 3.380228794000004, "created_at": "2026-11-22T04:16:33Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a8856eb0810f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 4, "n": 4}, "prereg_hash": "5aa27076a87ca849930a1281b79d835e94341fa79205a53148ab56c5015ae44a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a8856eb0810f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:16:33Z", "seed": 1714504508, "treatment": "full"}, "priority": 0.22710252379670137, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "82b96b84c882d2756b1dd1ef12d98220722b6a9d0d8e14534f43fd219ada34ae", "run_seconds": 4.122394323348999, "seed": 1714504508, "started_at": "2026-11-22T04:16:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006331, 0.235498], "effect": 0.114583, "evaluated_at": "2026-11-22T04:24:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a8856eb0810f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.722222222, 1.0, 0.861111111, 0.958333333], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.930555556, 1.0, 0.965277778, 0.986111111], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "5aa27076a87ca849930a1281b79d835e94341fa79205a53148ab56c5015ae44a", "primary_control": "facts", "result_hash": "82b96b84c882d2756b1dd1ef12d98220722b6a9d0d8e14534f43fd219ada34ae", "signature": "d72e5fd182c94dfcce97e2a766b96d3409885341c244ff7a49fa54e1f4cdd945ed677fb896b6fa5559da041e2bb234b312ca26210aadd1a31144a66f3ca2eb0e", "state": "VERIFIED", "summary": {"facts": 0.885417, "frozen": 0.0, "full": 1.0, "retrieval": 0.970486, "rules_nodetect": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:24:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:06Z", "cpu_seconds": 1.118426765999999, "created_at": "2026-11-22T04:16:33Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-96e0330d7510", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "prereg_hash": "e6797ef1a285dd972580f1f018423ccc8f62e45f1cdbec287a9d541928ace384", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-96e0330d7510", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:16:33Z", "seed": 219672988, "treatment": "gated"}, "priority": 0.21807015612491465, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "e362d5f823c87c92fa8472369173f53ef6317fc9aba157d6a41b6a7aa324324b", "run_seconds": 1.3447065353393555, "seed": 219672988, "started_at": "2026-11-22T04:17:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007273, 0.022424], "effect": 0.007576, "evaluated_at": "2026-11-22T04:24:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-96e0330d7510", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.147474747, 0.195402299, 0.487356322, 0.18989899], "gated": [0.177777778, 0.195402299, 0.487356322, 0.18989899], "naive": [0.15959596, 0.144827586, 0.222988506, 0.163636364], "scratch": [0.147474747, 0.195402299, 0.487356322, 0.18989899], "ungated": [0.15959596, 0.117241379, 0.222988506, 0.115151515]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "e6797ef1a285dd972580f1f018423ccc8f62e45f1cdbec287a9d541928ace384", "primary_control": "scratch", "result_hash": "e362d5f823c87c92fa8472369173f53ef6317fc9aba157d6a41b6a7aa324324b", "signature": "fe7d29fe838ae9e129fb1ca380fc1c45bc21db6c7bafb63fee3367debb33950b3fa1f7f117e7ebe63132ab59dba20e3d1767ce6700e970c3a254743ddff22d03", "state": "VERIFIED", "summary": {"compressed": 0.255033, "gated": 0.262609, "naive": 0.172762, "scratch": 0.255033, "ungated": 0.153744}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:24:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:59Z", "cpu_seconds": 3.444997381999997, "created_at": "2026-11-22T04:16:50Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-04f76abf4bf9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "569ad7f38e79dae9313c483f43a86a6bb663ab41d615fff5ee7a4396a00d58d9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-04f76abf4bf9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:16:50Z", "seed": 1415485139, "treatment": "frontier"}, "priority": 0.22738867800983867, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4773b8a965e235c6e80a0e877c7b69f263c47c89b00937a222bd724a8c8d6da7", "run_seconds": 3.894681453704834, "seed": 1415485139, "started_at": "2026-11-22T04:16:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.088864, 0.027322], "effect": -0.030771, "evaluated_at": "2026-11-22T04:24:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-04f76abf4bf9", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.341711957, 0.864583333, 0.592447917, 0.613095238], "frontier": [0.300271739, 0.6875, 0.595052083, 0.598958333], "goal": [0.286005435, 0.697916667, 0.5703125, 0.602678571], "hypdisc": [0.31861413, 0.708333333, 0.574869792, 0.595238095], "infogain": [0.300271739, 0.6875, 0.595052083, 0.598958333], "prederr": [0.300271739, 0.6875, 0.595052083, 0.598958333], "random": [0.330163043, 0.802083333, 0.578125, 0.594494048], "risk": [0.300271739, 0.6875, 0.595052083, 0.598958333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "569ad7f38e79dae9313c483f43a86a6bb663ab41d615fff5ee7a4396a00d58d9", "primary_control": "random", "result_hash": "4773b8a965e235c6e80a0e877c7b69f263c47c89b00937a222bd724a8c8d6da7", "signature": "556fd5001c2ef1afb6b01c1a6c304a75635f2770706de5aa50e8a1b8f4fec8a37a01b1981fdb8a6b7be5e54b674ae1772e3e36d87bccb234660a041d4cb0ae0b", "state": "VERIFIED", "summary": {"count": 0.60296, "frontier": 0.545446, "goal": 0.539228, "hypdisc": 0.549264, "infogain": 0.545446, "prederr": 0.545446, "random": 0.576216, "risk": 0.545446}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:24:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:00Z", "cpu_seconds": 0.5315552949999898, "created_at": "2026-11-22T04:16:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b4530dfd23e8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "58b6958c54298138c9beac57260bd1a5341536f40f7ab9398d0f2479255e1ffb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b4530dfd23e8", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:16:50Z", "seed": 479597425, "treatment": "full"}, "priority": 0.22095092131360622, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a33a82dc1c91c19483ad52e785c737039037a2e0933df5af5506f2695742e5c8", "run_seconds": 0.5471808910369873, "seed": 479597425, "started_at": "2026-11-22T04:17:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.596482, 0.096482], "effect": -0.25, "evaluated_at": "2026-11-22T04:24:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b4530dfd23e8", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 0.75, 1.0], "fixed_policy": [0.25, 0.0, 0.25, 0.0], "full": [0.25, 1.0, 0.0, 1.0], "no_world_model": [0.0, 1.0, 0.75, 1.0], "random": [0.5, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "58b6958c54298138c9beac57260bd1a5341536f40f7ab9398d0f2479255e1ffb", "primary_control": "e141_best", "result_hash": "a33a82dc1c91c19483ad52e785c737039037a2e0933df5af5506f2695742e5c8", "signature": "1182b86671e2bc78f699441409e378ae3a63fc0c0cfa957c1db064e1a04f4e352f3bfa3853e250d4c23bd5dd2d9f66b9f70c47dfc544a667bdbb3bdc45378e09", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.125, "full": 0.5625, "no_world_model": 0.6875, "random": 0.625}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:16:55Z", "cpu_seconds": 1.1169338699999969, "created_at": "2026-11-22T04:16:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5f788d3aace0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "a0920684c243ee7b2d8300441b9297d08a368f8d63222c4f010a56ef4d459025", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5f788d3aace0", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:16:50Z", "seed": 310887896, "treatment": "full"}, "priority": 0.23367051247320883, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "8dcd91ae2eff35207801a42e1119cd8762037d0fea2a79370f521c6c42c39ca6", "run_seconds": 1.2242252826690674, "seed": 310887896, "started_at": "2026-11-22T04:16:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.074336, -0.021652], "effect": -0.047994, "evaluated_at": "2026-11-22T04:24:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5f788d3aace0", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.47172619, 0.723214286, 0.200892857, 0.384057971], "frozen": [0.180059524, 0.166666667, 0.180059524, 0.175362319], "full": [0.436011905, 0.703869048, 0.119047619, 0.328985507], "retrieval": [0.485119048, 0.729166667, 0.095238095, 0.37826087], "rules_nodetect": [0.751488095, 0.703869048, 0.742559524, 0.328985507]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "a0920684c243ee7b2d8300441b9297d08a368f8d63222c4f010a56ef4d459025", "primary_control": "facts", "result_hash": "8dcd91ae2eff35207801a42e1119cd8762037d0fea2a79370f521c6c42c39ca6", "signature": "374d2367a5c95651661fd6c3a5956d21b8959e16ede81f298494326f7f5cee320634489b1c12c7c846d48aee74e09f82adb6958abcf1d1c52f8d7916b8d6330c", "state": "VERIFIED", "summary": {"facts": 0.444973, "frozen": 0.175537, "full": 0.396979, "retrieval": 0.421946, "rules_nodetect": 0.631726}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:00Z", "cpu_seconds": 1.4978537029999899, "created_at": "2026-11-22T04:16:50Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-63c7f8cf5c08", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "prereg_hash": "52f2336df2d2303d0ac5dc9c02310e07e65aad86b5111954447f16f385b31e94", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-63c7f8cf5c08", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:16:50Z", "seed": 1617812312, "treatment": "gated"}, "priority": 0.2245257189129838, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f05960281e766211c51e9aefa669a27b7174cac5939d183e063af880ae88fa8e", "run_seconds": 1.5192458629608154, "seed": 1617812312, "started_at": "2026-11-22T04:16:59Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:24:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-63c7f8cf5c08", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.333333333, 0.473214286, 0.388392857, 0.33277027], "gated": [0.333333333, 0.473214286, 0.388392857, 0.33277027], "naive": [0.322916667, 0.473214286, 0.370535714, 0.366554054], "scratch": [0.333333333, 0.473214286, 0.388392857, 0.33277027], "ungated": [0.322916667, 0.464285714, 0.379464286, 0.342905405]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "52f2336df2d2303d0ac5dc9c02310e07e65aad86b5111954447f16f385b31e94", "primary_control": "scratch", "result_hash": "f05960281e766211c51e9aefa669a27b7174cac5939d183e063af880ae88fa8e", "signature": "fa4270de186b0a9583db925204cc1c611f248c3f3af32357cc099bdda9c37e5701f7465d6ad4a40234452a1ca197e22aa2b1515716d0f942406b5a151ba0c002", "state": "PARTIAL", "summary": {"compressed": 0.381928, "gated": 0.381928, "naive": 0.383305, "scratch": 0.381928, "ungated": 0.377393}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:24:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:16Z", "cpu_seconds": 4.424604234, "created_at": "2026-11-22T04:17:07Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a2300ee0960c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "e884b1d982ce2494cdbc4c3ddcb9f2c5a8acc8605265f117d20a90aefc5156a4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a2300ee0960c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:17:07Z", "seed": 1189271001, "treatment": "frontier"}, "priority": 0.2201139553039579, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "117111aad59465a01e17ee452654c65f61cf65128881cd5df16d7ae8c7a8e77b", "run_seconds": 4.8086347579956055, "seed": 1189271001, "started_at": "2026-11-22T04:17:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.180714, 0.110111], "effect": -0.035302, "evaluated_at": "2026-11-22T04:24:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a2300ee0960c", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.555555556, 0.503875969, 0.633986928, 0.627906977], "frontier": [0.274509804, 0.364341085, 0.699346405, 0.418604651], "goal": [0.274509804, 0.457364341, 0.647058824, 0.658914729], "hypdisc": [0.45751634, 0.457364341, 0.614379085, 0.627906977], "infogain": [0.274509804, 0.333333333, 0.692810458, 0.372093023], "prederr": [0.274509804, 0.186046512, 0.254901961, 0.341085271], "random": [0.470588235, 0.418604651, 0.535947712, 0.472868217], "risk": [0.274509804, 0.364341085, 0.699346405, 0.418604651]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "e884b1d982ce2494cdbc4c3ddcb9f2c5a8acc8605265f117d20a90aefc5156a4", "primary_control": "random", "result_hash": "117111aad59465a01e17ee452654c65f61cf65128881cd5df16d7ae8c7a8e77b", "signature": "fe6e5a4e5dfa14b554ec5e37c663f0217cbfe430e1aafcb584bbac5a68641403101b6a7dd65b2a9911142fb5e8003450b8746ac9ebe37526e207088e6afb5f01", "state": "VERIFIED", "summary": {"count": 0.580331, "frontier": 0.4392, "goal": 0.509462, "hypdisc": 0.539292, "infogain": 0.418187, "prederr": 0.264136, "random": 0.474502, "risk": 0.4392}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:24:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:24Z", "cpu_seconds": 2.7119276909999996, "created_at": "2026-11-22T04:17:07Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-b5c68fee109e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "89ddbb47efcfc8f683f4708182ee66b770c24046e9105a1a2a0880637ac75e67", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-b5c68fee109e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:17:07Z", "seed": 1767308808, "treatment": "frontier"}, "priority": 0.2143232506519746, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3e6709d6ae31ce4bf5dfde1e29579f1bd66f08f84c4de1f2b87b7ba402e41a03", "run_seconds": 3.891352415084839, "seed": 1767308808, "started_at": "2026-11-22T04:17:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.156823, 0.114847], "effect": -0.020988, "evaluated_at": "2026-11-22T04:24:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-b5c68fee109e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.485245902, 0.597368421, 0.463157895, 0.7], "frontier": [0.518032787, 0.397368421, 0.431578947, 0.605263158], "goal": [0.406557377, 0.468421053, 0.368421053, 0.607894737], "hypdisc": [0.514754098, 0.457894737, 0.421052632, 0.605263158], "infogain": [0.459016393, 0.397368421, 0.547368421, 0.676315789], "prederr": [0.554098361, 0.397368421, 0.494736842, 0.676315789], "random": [0.396721311, 0.592105263, 0.378947368, 0.668421053], "risk": [0.518032787, 0.397368421, 0.431578947, 0.605263158]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "89ddbb47efcfc8f683f4708182ee66b770c24046e9105a1a2a0880637ac75e67", "primary_control": "random", "result_hash": "3e6709d6ae31ce4bf5dfde1e29579f1bd66f08f84c4de1f2b87b7ba402e41a03", "signature": "3750c7393c5105ba71f8cc3ce430e2e07d5ba9c695c542ea849cffa5f3990d20494c7de6dbe574cf51b001af5e1ab4085d2e65f7a533678a09385fbfbdc8c306", "state": "VERIFIED", "summary": {"count": 0.561443, "frontier": 0.488061, "goal": 0.462824, "hypdisc": 0.499741, "infogain": 0.520017, "prederr": 0.53063, "random": 0.509049, "risk": 0.488061}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:24:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:33Z", "cpu_seconds": 1.6562193499999998, "created_at": "2026-11-22T04:17:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2dc6ff1d34f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "75a529e774dd946bcf8af0c3ad3048298ab6c3a154f9ffe163356a96f928f893", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2dc6ff1d34f0", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:17:07Z", "seed": 775900729, "treatment": "full"}, "priority": 0.21357699012457046, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ebb776a1fe5970d91c439b33149bfc451395399741611dbd8e9beb132ecbcb66", "run_seconds": 2.4407031536102295, "seed": 775900729, "started_at": "2026-11-22T04:18:31Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:24:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2dc6ff1d34f0", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "75a529e774dd946bcf8af0c3ad3048298ab6c3a154f9ffe163356a96f928f893", "primary_control": "e141_best", "result_hash": "ebb776a1fe5970d91c439b33149bfc451395399741611dbd8e9beb132ecbcb66", "signature": "01648fbcbb7f0ce1a3d180fab3736d8119a2eba8f386b1d92e4ca691f04ec48c49be623faf4a948f590474f6a5314725cc75ed105a59cddd50557a36bee6630b", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:11Z", "cpu_seconds": 1.762790946999985, "created_at": "2026-11-22T04:17:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-cf7beff3a07e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "baa1f87a1fe43f5227b5bede53228d01665ec9b76b93cbbd02f612274832d8b7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-cf7beff3a07e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:17:07Z", "seed": 120365097, "treatment": "full"}, "priority": 0.2237769637980642, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8dba83b167f688260ab64890b6c1a18b3e005027b1e028625d6a715e410acc8e", "run_seconds": 1.8826115131378174, "seed": 120365097, "started_at": "2026-11-22T04:17:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.151278, 0.776278], "effect": 0.3125, "evaluated_at": "2026-11-22T04:24:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-cf7beff3a07e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.75, 1.0, 0.75, 1.0], "no_world_model": [0.5, 1.0, 1.0, 1.0], "random": [0.75, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "baa1f87a1fe43f5227b5bede53228d01665ec9b76b93cbbd02f612274832d8b7", "primary_control": "e141_best", "result_hash": "8dba83b167f688260ab64890b6c1a18b3e005027b1e028625d6a715e410acc8e", "signature": "979f12e00204a816a1d9250338f76eabd33915e462f728cb1e5c8d661a2fc3d3ed534895e9f489214e92715787e03645ecd9722909e2760e0645d1a31cbd3701", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.0, "full": 0.875, "no_world_model": 0.875, "random": 0.6875}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:24:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:01Z", "cpu_seconds": 1.1712854989999997, "created_at": "2026-11-22T04:17:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-97b4c3523474", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "ec308baa95fffd5834e5693480eb6f611624342262a3d32e06e01782e986a1ee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-97b4c3523474", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:17:07Z", "seed": 858214714, "treatment": "full"}, "priority": 0.21212917215744576, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "98951e8a8b86112df81c79dc5eec2d1d38c37476244e26028e374dcd736d3fae", "run_seconds": 1.2802369594573975, "seed": 858214714, "started_at": "2026-11-22T04:18:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007265, 0.034073], "effect": 0.013404, "evaluated_at": "2026-11-22T04:26:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-97b4c3523474", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.779411765, 0.549382716, 0.9, 0.673387097], "frozen": [0.220588235, 0.265432099, 0.4, 0.362903226], "full": [0.779411765, 0.558641975, 0.9, 0.717741935], "retrieval": [0.779411765, 0.614197531, 0.883333333, 0.782258065], "rules_nodetect": [0.102941176, 0.37654321, 0.416666667, 0.85483871]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "ec308baa95fffd5834e5693480eb6f611624342262a3d32e06e01782e986a1ee", "primary_control": "facts", "result_hash": "98951e8a8b86112df81c79dc5eec2d1d38c37476244e26028e374dcd736d3fae", "signature": "686196633faf1fa59cb055c706b92706d03b4027ca611f9b4d8288ff371f2a56c88697133cb5301fdc816c96dadf60ce19f6c2173d335392fc488ccfec7d910a", "state": "VERIFIED", "summary": {"facts": 0.725545, "frozen": 0.312231, "full": 0.738949, "retrieval": 0.7648, "rules_nodetect": 0.437747}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:26:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:16Z", "cpu_seconds": 2.3238153599999976, "created_at": "2026-11-22T04:17:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b79bf0e069c5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 8, "n": 4}, "prereg_hash": "2faa1b209cae9a3e9e072d61b3eb88992b526163a012f39353fffe9a3471fe3b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b79bf0e069c5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:17:07Z", "seed": 2081749927, "treatment": "full"}, "priority": 0.21742163207386622, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4ec2db04d19b3e5ad62181d2991bf47220553103444c569b419c17ab25513590", "run_seconds": 2.4648196697235107, "seed": 2081749927, "started_at": "2026-11-22T04:17:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.017357, 0.198374], "effect": 0.107866, "evaluated_at": "2026-11-22T04:24:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b79bf0e069c5", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.955, 0.527118644, 0.571764706, 0.893333333], "frozen": [0.33, 0.108474576, 0.0, 0.413333333], "full": [0.985, 0.76779661, 0.665882353, 0.96], "retrieval": [1.0, 0.794915254, 0.903529412, 0.933333333], "rules_nodetect": [0.985, 0.76779661, 0.665882353, 0.96]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2faa1b209cae9a3e9e072d61b3eb88992b526163a012f39353fffe9a3471fe3b", "primary_control": "facts", "result_hash": "4ec2db04d19b3e5ad62181d2991bf47220553103444c569b419c17ab25513590", "signature": "4a1e886a356671f7b09b1814dccd7da50dc99a0a2727216358a7a30a561e40a6c3b4818632831dc786a0bd051346bd0f598973a7e70400aefd9a2c14aa1a280f", "state": "VERIFIED", "summary": {"facts": 0.736804, "frozen": 0.212952, "full": 0.84467, "retrieval": 0.907944, "rules_nodetect": 0.84467}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:24:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:49Z", "cpu_seconds": 1.8825039559999999, "created_at": "2026-11-22T04:17:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d1e19b78c7b2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "3c820c61c173a57ce1d9980da6e24a3dbba810abcdff87bab53ddcfd574651ab", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d1e19b78c7b2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:17:07Z", "seed": 1355841509, "treatment": "full"}, "priority": 0.2124471889878295, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ba54c18b0e3009fdb5f50d935d2bd2920aa7d88678285c32de27919cd7c58f81", "run_seconds": 2.216157913208008, "seed": 1355841509, "started_at": "2026-11-22T04:18:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.164427, 0.418559], "effect": 0.291493, "evaluated_at": "2026-11-22T04:24:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d1e19b78c7b2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.71875, 0.754166667, 0.833333333, 0.527777778], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.958333333, 0.970833333, 0.979166667, 0.861111111], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "3c820c61c173a57ce1d9980da6e24a3dbba810abcdff87bab53ddcfd574651ab", "primary_control": "facts", "result_hash": "ba54c18b0e3009fdb5f50d935d2bd2920aa7d88678285c32de27919cd7c58f81", "signature": "105fa299b28a5c2e2c8fa437311b95232ce24bd556e346ced639c1aac53edb99d0a255bb780716a33356f213f038666b86fde808387a3b58197a1d89aa35dd02", "state": "VERIFIED", "summary": {"facts": 0.708507, "frozen": 0.0, "full": 1.0, "retrieval": 0.942361, "rules_nodetect": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:24:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:54Z", "cpu_seconds": 0.331389778, "created_at": "2026-11-22T04:17:24Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-490dd7ae6de2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "f15b4f9a33c1c6923c6eab4ccfe588d140756c8d6dc4d522f942375e437e21e9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-490dd7ae6de2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:17:24Z", "seed": 53464479, "treatment": "full"}, "priority": 0.2203308282834082, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e6d7d80663ca732b347e91050829a0c3cad4d86eca590e45952324ad47a398dd", "run_seconds": 0.3388195037841797, "seed": 53464479, "started_at": "2026-11-22T04:17:54Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:24:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-490dd7ae6de2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f15b4f9a33c1c6923c6eab4ccfe588d140756c8d6dc4d522f942375e437e21e9", "primary_control": "e141_best", "result_hash": "e6d7d80663ca732b347e91050829a0c3cad4d86eca590e45952324ad47a398dd", "signature": "bebe98dedbca70dfff8f27fcfc00c71e4a253b0c6c16d337662165da6e32bbc32bab8f1555213a74827375a8f79271760eef672958bd3e0287b75b1a3ac2b00e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.75, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:19Z", "cpu_seconds": 0.489743742, "created_at": "2026-11-22T04:17:24Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a03f1659bf23", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "9f6074e8be27932382d29549f910c88f34857ef33e810928403596c1aaa50776", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a03f1659bf23", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:17:24Z", "seed": 2088900160, "treatment": "full"}, "priority": 0.2176419417614473, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "46862fc848ec285cc2620da27be423b426ea707d809d084f8ec7c92bbedbae94", "run_seconds": 0.5959839820861816, "seed": 2088900160, "started_at": "2026-11-22T04:18:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.68, 0.055], "effect": -0.3125, "evaluated_at": "2026-11-22T04:24:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a03f1659bf23", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 0.0, 0.5], "fixed_policy": [0.5, 0.5, 0.0, 0.25], "full": [0.0, 0.75, 0.0, 0.0], "no_world_model": [0.0, 0.75, 0.0, 0.0], "random": [0.75, 0.75, 0.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "9f6074e8be27932382d29549f910c88f34857ef33e810928403596c1aaa50776", "primary_control": "e141_best", "result_hash": "46862fc848ec285cc2620da27be423b426ea707d809d084f8ec7c92bbedbae94", "signature": "90013b18c84809b26849d5dca38e2d47fb84bda3ec1c92f53599f35a70234d9aa007dc249aecb7f4df487404d2f2b5647c28acb017d12fa828071d0fe9cedc08", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.3125, "full": 0.1875, "no_world_model": 0.1875, "random": 0.5}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:15Z", "cpu_seconds": 1.8578467490000001, "created_at": "2026-11-22T04:17:24Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1d63d5ea13e3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "a2ea1a248c814a150cff608b8a5c21ac239e8aa97a4bb08b25b3f58969750814", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1d63d5ea13e3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:17:24Z", "seed": 538328082, "treatment": "full"}, "priority": 0.21157708899561567, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "eb755a805018f718c90cc2060fd0c2a7b15d2922063ce9c59dff5a58a2decf73", "run_seconds": 1.9409525394439697, "seed": 538328082, "started_at": "2026-11-22T04:19:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014379, 0.490074], "effect": 0.237847, "evaluated_at": "2026-11-22T04:29:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1d63d5ea13e3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.8125, 0.875, 0.388888889, 0.972222222], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.979166667, 0.930555556, 0.857638889, 0.986111111], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a2ea1a248c814a150cff608b8a5c21ac239e8aa97a4bb08b25b3f58969750814", "primary_control": "facts", "result_hash": "eb755a805018f718c90cc2060fd0c2a7b15d2922063ce9c59dff5a58a2decf73", "signature": "906b940ace41ed4cb791cca33dbf1e88e6b63e025f0f6811cec73560665e3ecf8f85a82350f3b0ebd81e957ae9837c258466afd5a887fdc737bb0a8dbd970f05", "state": "VERIFIED", "summary": {"facts": 0.762153, "frozen": 0.0, "full": 1.0, "retrieval": 0.938368, "rules_nodetect": 1.0}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:29:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:17:58Z", "cpu_seconds": 3.439699572, "created_at": "2026-11-22T04:17:24Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-27dc1970dfd4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 2, "n": 4}, "prereg_hash": "805579bf234b5dbb76580a2f4a75e559656705e443df9fc5d9607699ea745221", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-27dc1970dfd4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:17:24Z", "seed": 1175094979, "treatment": "full"}, "priority": 0.22062617765827652, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "00692ac7804a672e39e120b21d6eaa4ba3d124c8e47eb44a4b3f523e1dd808b5", "run_seconds": 3.9796640872955322, "seed": 1175094979, "started_at": "2026-11-22T04:17:54Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003871, 0.011935], "effect": 0.004032, "evaluated_at": "2026-11-22T04:24:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-27dc1970dfd4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.983870968, 1.0, 1.0], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 0.994623656, 1.0, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "805579bf234b5dbb76580a2f4a75e559656705e443df9fc5d9607699ea745221", "primary_control": "facts", "result_hash": "00692ac7804a672e39e120b21d6eaa4ba3d124c8e47eb44a4b3f523e1dd808b5", "signature": "ee424e35f4265b39ddd53f93554c0771c305d0ff125815bb17ced2f9e9b12983e5472d45c70ddb7b3dc6ba7f1afd894fac99346cbed4f926bffe95345f5fe30f", "state": "VERIFIED", "summary": {"facts": 0.995968, "frozen": 0.0, "full": 1.0, "retrieval": 0.998656, "rules_nodetect": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:24:39Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:27Z", "cpu_seconds": 1.98784707, "created_at": "2026-11-22T04:17:24Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-93b1ded184a6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "e8a916d1dbd9ca7c4e981cbbca471731997003373c887feed890962776a6438e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-93b1ded184a6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:17:24Z", "seed": 1458093100, "treatment": "full"}, "priority": 0.21567046105471488, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "090e3dc7091a49ad1b3018b1c474cc8b0ebfc6435d1f39be3b588b4eab459f8f", "run_seconds": 2.0606141090393066, "seed": 1458093100, "started_at": "2026-11-22T04:18:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.000457, 0.47973], "effect": 0.239636, "evaluated_at": "2026-11-22T04:24:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-93b1ded184a6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.571428571, 1.0, 0.941964286, 0.528061224], "frozen": [0.0, 0.285714286, 0.142857143, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [0.851190476, 1.0, 0.955357143, 0.87244898], "rules_nodetect": [1.0, 0.392857143, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "e8a916d1dbd9ca7c4e981cbbca471731997003373c887feed890962776a6438e", "primary_control": "facts", "result_hash": "090e3dc7091a49ad1b3018b1c474cc8b0ebfc6435d1f39be3b588b4eab459f8f", "signature": "5eead425f70226df5f6f53da2a7ff6bddcd795873f9f5b446870ca6421415346e4b968d58e6cc9dd49cae039456a96e71642202d7981e67df3df1b8de6f80300", "state": "VERIFIED", "summary": {"facts": 0.760364, "frozen": 0.107143, "full": 1.0, "retrieval": 0.919749, "rules_nodetect": 0.848214}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:24:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:48Z", "cpu_seconds": 5.0194226539999995, "created_at": "2026-11-22T04:17:24Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8a41d984aa97", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "194702eb7fe21aac0032076f20ced9ca6476a503b593628b2c2261d49161d0b9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8a41d984aa97", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:17:24Z", "seed": 774778394, "treatment": "gated"}, "priority": 0.21251621412082444, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6577cb5451b4275f046463ea021d43d7867c692d8dedbc7e5bbf652e8346d72e", "run_seconds": 5.737076044082642, "seed": 774778394, "started_at": "2026-11-22T04:18:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002, 0.006167], "effect": 0.002083, "evaluated_at": "2026-11-22T04:24:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-8a41d984aa97", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.783333333, 0.866666667, 0.197979798, 0.9], "gated": [0.783333333, 0.866666667, 0.197979798, 0.9], "naive": [0.658333333, 0.866666667, 0.119191919, 0.6], "scratch": [0.783333333, 0.866666667, 0.197979798, 0.891666667], "ungated": [0.658333333, 0.866666667, 0.119191919, 0.6]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "194702eb7fe21aac0032076f20ced9ca6476a503b593628b2c2261d49161d0b9", "primary_control": "scratch", "result_hash": "6577cb5451b4275f046463ea021d43d7867c692d8dedbc7e5bbf652e8346d72e", "signature": "11a03da5232fe62d2846503c84a70980bb1b5408b9af7a7fde8d8945316d83d0d6464ac1476aa5ea5b7cd9ec1fc80d7ed1b03f7e29bf88bd1fc754200122530a", "state": "VERIFIED", "summary": {"compressed": 0.686995, "gated": 0.686995, "naive": 0.561048, "scratch": 0.684912, "ungated": 0.561048}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:24:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:54Z", "cpu_seconds": 52.934009788, "created_at": "2026-11-22T04:17:41Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-2ac40ef850d1", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "79aed09d6ea9eb03acba5690ddb7fa4d9559f9b0e3dff329624d7882f43f34cc", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-2ac40ef850d1", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:17:41Z", "seed": 279342929, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "51d8366b1698f4d4361707032a7da43de2e82ee4e7cd959fb4a761c839e621ea", "run_seconds": 60.1457462310791, "seed": 279342929, "started_at": "2026-11-22T04:17:54Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"compressed": 0, "inherit_mem": 0, "inherit_nomem": 1, "random_search": 1, "reset_mem": 0, "retrieval": 0, "rootcause": 1, "xfer_mem": 0}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:24:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-2ac40ef850d1", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"hypdisc": 0.298, "random": 1.0, "risk": 0.283}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0, "risk": 0.6}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.907608696, 1.0, 0.947916667, 0.772058824, 0.645, 0.163992172, 0.977272727, 0.243617021, 0.907216495, 0.75890411], "compressed": [0.907608696, 1.0, 0.947916667, 0.772058824, 0.645, 0.163992172, 0.977272727, 0.243617021, 0.907216495, 0.75890411], "inherit_mem": [0.907608696, 1.0, 0.947916667, 0.772058824, 0.645, 0.163992172, 0.977272727, 0.243617021, 0.907216495, 0.75890411], "inherit_nomem": [0.983695652, 1.0, 0.989583333, 0.818627451, 0.8175, 0.306017613, 0.977272727, 0.277659574, 0.912371134, 0.794520548], "random_search": [0.967391304, 1.0, 0.989583333, 0.801470588, 0.68375, 0.176712329, 0.96969697, 0.553191489, 0.917525773, 0.75890411], "reset_mem": [0.907608696, 1.0, 0.947916667, 0.772058824, 0.645, 0.163992172, 0.977272727, 0.243617021, 0.907216495, 0.75890411], "retrieval": [0.907608696, 1.0, 0.947916667, 0.772058824, 0.645, 0.163992172, 0.977272727, 0.243617021, 0.907216495, 0.75890411], "rootcause": [0.983695652, 1.0, 0.989583333, 0.803921569, 0.8225, 0.180039139, 1.0, 0.296808511, 0.912371134, 0.794520548], "xfer_mem": [0.907608696, 1.0, 0.947916667, 0.772058824, 0.645, 0.163992172, 0.977272727, 0.243617021, 0.907216495, 0.75890411]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "per_generation_sealed": [0.732359, 0.732359, 0.732359, 0.732359], "prereg_hash": "79aed09d6ea9eb03acba5690ddb7fa4d9559f9b0e3dff329624d7882f43f34cc", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "compressed", "random_search"], "result_hash": "51d8366b1698f4d4361707032a7da43de2e82ee4e7cd959fb4a761c839e621ea", "signature": "97a4b97148ed081462d227822f99174ec8424f2392e1266f7f98c87a815571a688f8943f39f1bf579fc7b6e7c9228d78fed5febae19bb181686926129b323d0b", "state": "PARTIAL", "summary": {"G0": 0.732359, "compressed": 0.732359, "inherit_mem": 0.732359, "inherit_nomem": 0.787725, "random_search": 0.781823, "reset_mem": 0.732359, "retrieval": 0.732359, "rootcause": 0.778344, "xfer_mem": 0.732359}, "ties": 10, "treatment": "inherit_mem", "wins": 0}, "verified_at": "2026-11-22T04:26:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:26Z", "cpu_seconds": 0.52193774, "created_at": "2026-11-22T04:17:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8ae85b149a60", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "16132dbadc8dd7d5909e8407705e6d99cb32c425e3ef86509dd3527d6aa86aa8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8ae85b149a60", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:17:58Z", "seed": 1267838232, "treatment": "full"}, "priority": 0.2136452243435527, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f3a515afcd5a7a675bde53fe16fe473d3267e84bfe491f38fa62ce075427bc6d", "run_seconds": 0.5416343212127686, "seed": 1267838232, "started_at": "2026-11-22T04:18:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.003232, 0.094748], "effect": 0.04899, "evaluated_at": "2026-11-22T04:24:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-8ae85b149a60", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.75177305, 0.168604651, 0.89122807, 0.851851852], "frozen": [0.095744681, 0.081395349, 0.094736842, 0.092592593], "full": [0.765957447, 0.284883721, 0.912280702, 0.896296296], "retrieval": [0.79787234, 0.313953488, 0.898245614, 0.892592593], "rules_nodetect": [0.765957447, 0.238372093, 0.912280702, 0.896296296]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "16132dbadc8dd7d5909e8407705e6d99cb32c425e3ef86509dd3527d6aa86aa8", "primary_control": "facts", "result_hash": "f3a515afcd5a7a675bde53fe16fe473d3267e84bfe491f38fa62ce075427bc6d", "signature": "725c8bd5680d641861ae47187714655fc86d72f98ff621390ceef6319593a29468d786f9f704570c73dec204d5b9efad5ed31921d8c1a4fac71f92c8d6e81a0a", "state": "VERIFIED", "summary": {"facts": 0.665864, "frozen": 0.091117, "full": 0.714855, "retrieval": 0.725666, "rules_nodetect": 0.703227}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:24:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:01Z", "cpu_seconds": 1.181141112, "created_at": "2026-11-22T04:17:58Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-042fc62413f5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "12db6f910e2da13ae91af45f9aaaa6104b3a17b718cbe30dd9b8a7e92d2bbabf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-042fc62413f5", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:17:58Z", "seed": 38531996, "treatment": "gated"}, "priority": 0.3762179155976272, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "437b9bc986d732b11945f3af0c3229802d0a9860f5298135da07d0dad751f66e", "run_seconds": 1.5351860523223877, "seed": 38531996, "started_at": "2026-11-22T04:17:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.059213, 0.004638], "effect": -0.027288, "evaluated_at": "2026-11-22T04:24:39Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-042fc62413f5", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.488888889, 0.294117647, 0.394117647, 0.2], "gated": [0.488888889, 0.294117647, 0.394117647, 0.2], "naive": [0.444444444, 0.270588235, 0.382352941, 0.247619048], "scratch": [0.533333333, 0.358823529, 0.394117647, 0.2], "ungated": [0.422222222, 0.364705882, 0.382352941, 0.257142857]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "12db6f910e2da13ae91af45f9aaaa6104b3a17b718cbe30dd9b8a7e92d2bbabf", "primary_control": "scratch", "result_hash": "437b9bc986d732b11945f3af0c3229802d0a9860f5298135da07d0dad751f66e", "signature": "5fe164e4154dd4159fd1ec585eaded8408148cf35947584dc5f94ae0d359ca35925c95aacc1ea237026d66ed4fc7733e226401ee29e64f18df429a392d4e100c", "state": "VERIFIED", "summary": {"compressed": 0.344281, "gated": 0.344281, "naive": 0.336251, "scratch": 0.371569, "ungated": 0.356606}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:24:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:20Z", "cpu_seconds": 1.409891858, "created_at": "2026-11-22T04:18:14Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7b92e9d008d1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "eb1ab1a52e64877c865e7963bdcff94be52853008214dea8a5a27454aad9e2bb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7b92e9d008d1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:18:14Z", "seed": 1332172397, "treatment": "full"}, "priority": 0.3652854899635702, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "542f082ecd24780a22543f1638978fb52e130ab695a3ba7aa3558e21afb80a98", "run_seconds": 1.5393779277801514, "seed": 1332172397, "started_at": "2026-11-22T04:18:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:24:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7b92e9d008d1", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.5, 0.5], "fixed_policy": [0.25, 0.25, 0.0, 0.0], "full": [1.0, 1.0, 0.0, 0.5], "no_world_model": [1.0, 1.0, 0.25, 0.5], "random": [1.0, 1.0, 0.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "eb1ab1a52e64877c865e7963bdcff94be52853008214dea8a5a27454aad9e2bb", "primary_control": "e141_best", "result_hash": "542f082ecd24780a22543f1638978fb52e130ab695a3ba7aa3558e21afb80a98", "signature": "4fa71afb72846160f4f320733e71b08d3c79056d02a39761c78f8bbc8d1935c371526504bc90031e9449951570268cc1cd875712856e6e06da40765b2c4f2302", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.125, "full": 0.625, "no_world_model": 0.6875, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:24:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:42Z", "cpu_seconds": 2.7157854969999997, "created_at": "2026-11-22T04:18:31Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-344271a04971", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "8e4c62044c4a0e2ead56487b07b04b2789fc98b9480d8bf31e3b9e85f30bb6a8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-344271a04971", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:18:31Z", "seed": 555913891, "treatment": "frontier"}, "priority": 0.2248907321607084, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "07cbbd7afd9acb0a6436d0bae9e7182d7a38d01bd8c5991862738fe663d07904", "run_seconds": 3.3551878929138184, "seed": 555913891, "started_at": "2026-11-22T04:18:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.032814, 0.377877], "effect": 0.172532, "evaluated_at": "2026-11-22T04:24:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-344271a04971", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.813953488, 1.0, 0.892857143, 1.0], "frontier": [0.947674419, 1.0, 0.910714286, 1.0], "goal": [0.674418605, 1.0, 0.910714286, 0.826923077], "hypdisc": [0.906976744, 1.0, 1.0, 1.0], "infogain": [0.947674419, 1.0, 0.910714286, 1.0], "prederr": [0.947674419, 1.0, 0.910714286, 1.0], "random": [0.494186047, 0.863636364, 0.964285714, 0.846153846], "risk": [0.947674419, 1.0, 0.910714286, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "8e4c62044c4a0e2ead56487b07b04b2789fc98b9480d8bf31e3b9e85f30bb6a8", "primary_control": "random", "result_hash": "07cbbd7afd9acb0a6436d0bae9e7182d7a38d01bd8c5991862738fe663d07904", "signature": "56f8ee9e07e47f98dd01dc0c2bebce732021d9049563e38730bfa33772238e1552df38d2284309bfea901c0f056a3aca0664e59562bb3f9a1c899d40cfb0ad0c", "state": "VERIFIED", "summary": {"count": 0.926703, "frontier": 0.964597, "goal": 0.853014, "hypdisc": 0.976744, "infogain": 0.964597, "prederr": 0.964597, "random": 0.792065, "risk": 0.964597}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:24:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:02Z", "cpu_seconds": 0.5464963950000001, "created_at": "2026-11-22T04:18:31Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8c1d74267fa8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "b4314128481186f30c223b9048d30b0e1cdd7f66cccac46a5ff064390a3fc814", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8c1d74267fa8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:18:31Z", "seed": 2030937938, "treatment": "frontier"}, "priority": 0.21208581178966524, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b6226f8e27fa3b80a78cf869120a2388b2edd9a58864be501d1192a3f20a1011", "run_seconds": 0.5922703742980957, "seed": 2030937938, "started_at": "2026-11-22T04:19:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.001615, 0.024531], "effect": 0.011458, "evaluated_at": "2026-11-22T04:26:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-8c1d74267fa8", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.916666667, 0.95, 1.0, 0.916666667], "frontier": [0.895833333, 0.95, 1.0, 0.895833333], "goal": [0.916666667, 0.875, 1.0, 0.875], "hypdisc": [0.916666667, 0.95, 1.0, 0.854166667], "infogain": [0.875, 0.95, 1.0, 0.916666667], "prederr": [0.791666667, 0.95, 1.0, 0.916666667], "random": [0.895833333, 0.925, 1.0, 0.875], "risk": [0.895833333, 0.95, 1.0, 0.895833333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "b4314128481186f30c223b9048d30b0e1cdd7f66cccac46a5ff064390a3fc814", "primary_control": "random", "result_hash": "b6226f8e27fa3b80a78cf869120a2388b2edd9a58864be501d1192a3f20a1011", "signature": "ecde96892e9f9ae7805088a5332327eb3802321cee3407be95c97d0d365799b081c9ef4376a87c36aeccbf5b600b819224452ba7e986ee7e8f4629060d38ac03", "state": "VERIFIED", "summary": {"count": 0.945833, "frontier": 0.935417, "goal": 0.916667, "hypdisc": 0.930208, "infogain": 0.935417, "prederr": 0.914583, "random": 0.923958, "risk": 0.935417}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:26:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:41Z", "cpu_seconds": 22.410015574, "created_at": "2026-11-22T04:18:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5000d5a8ac0d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "0aeefcd13369d1bff444d79a0829f907a9b2c881c053dc88c788c880a4e5eb98", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5000d5a8ac0d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:18:31Z", "seed": 92173590, "treatment": "full"}, "priority": 0.21126878764332346, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "613a572e11f075c3f59caa0373c1adf3b91b7029f38a67eae1fcb1fe2411e0df", "run_seconds": 23.81079912185669, "seed": 92173590, "started_at": "2026-11-22T04:19:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.094139, 0.844139], "effect": 0.375, "evaluated_at": "2026-11-22T04:29:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5000d5a8ac0d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 0.0], "fixed_policy": [0.0, 1.0, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 0.25, 1.0, 0.5], "random": [1.0, 0.5, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "0aeefcd13369d1bff444d79a0829f907a9b2c881c053dc88c788c880a4e5eb98", "primary_control": "e141_best", "result_hash": "613a572e11f075c3f59caa0373c1adf3b91b7029f38a67eae1fcb1fe2411e0df", "signature": "6214573089502071902186b6937fa83a46e8e661ef33a22f71e12bfbbef74cd9357c84a592b38bacdb811f36eb4a71fcc539486bbb8b70ecb31791f4134a4909", "state": "VERIFIED", "summary": {"e141_best": 0.625, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 0.6875, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:29:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:19Z", "cpu_seconds": 0.7782163199999985, "created_at": "2026-11-22T04:18:31Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-30eb28c83412", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "9fa84c1f214c8de6bee0fdfc2d882e49168ec900e2c3300d7ae20c13dba9745f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-30eb28c83412", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:18:31Z", "seed": 22779148, "treatment": "gated"}, "priority": 0.2108872326543138, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "48a25d0533f7cc52836c6eb34f2a4be2064832b8081686e535102cc97958185f", "run_seconds": 0.8680760860443115, "seed": 22779148, "started_at": "2026-11-22T04:19:18Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.078144, 0.015995], "effect": -0.031074, "evaluated_at": "2026-11-22T04:29:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-30eb28c83412", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.628205128, 0.695652174, 0.666666667, 0.522222222], "gated": [0.628205128, 0.695652174, 0.666666667, 0.522222222], "naive": [0.730769231, 0.782608696, 0.73015873, 0.7], "scratch": [0.679487179, 0.695652174, 0.650793651, 0.611111111], "ungated": [0.730769231, 0.695652174, 0.428571429, 0.333333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "9fa84c1f214c8de6bee0fdfc2d882e49168ec900e2c3300d7ae20c13dba9745f", "primary_control": "scratch", "result_hash": "48a25d0533f7cc52836c6eb34f2a4be2064832b8081686e535102cc97958185f", "signature": "ed11c21f55a7ddb6a7f68962d5919591f7fd702c8dcbe262f5c3bdc0d0062ced400b3ae70c828fd0e530e439fe416acde5e9299f2f35cae9139244518f803009", "state": "VERIFIED", "summary": {"compressed": 0.628187, "gated": 0.628187, "naive": 0.735884, "scratch": 0.659261, "ungated": 0.547082}, "ties": 1, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:29:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:37Z", "cpu_seconds": 3.7204100369999997, "created_at": "2026-11-22T04:18:31Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-aad1c8a2fae5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "00d3c514b2763d66af083c06c09f2bf2514028ca9ccc9e3e3b4c01095919d143", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-aad1c8a2fae5", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:18:31Z", "seed": 39973605, "treatment": "gated"}, "priority": 0.2744826023408501, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "442c9afccba4db8df33ea2206af2409f374bb099bb127375dbf6c9a700423022", "run_seconds": 5.861194372177124, "seed": 39973605, "started_at": "2026-11-22T04:18:32Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017619, 0.005714], "effect": -0.005952, "evaluated_at": "2026-11-22T04:24:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-aad1c8a2fae5", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.71875, 0.611111111, 0.672413793, 0.511904762], "gated": [0.71875, 0.611111111, 0.672413793, 0.511904762], "naive": [0.8125, 0.680555556, 0.74137931, 0.583333333], "scratch": [0.71875, 0.611111111, 0.672413793, 0.535714286], "ungated": [0.84375, 0.611111111, 0.646551724, 0.583333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "00d3c514b2763d66af083c06c09f2bf2514028ca9ccc9e3e3b4c01095919d143", "primary_control": "scratch", "result_hash": "442c9afccba4db8df33ea2206af2409f374bb099bb127375dbf6c9a700423022", "signature": "04e9258e37b8e46340eecdb65a70508387afc1af37f1e6c1960b2ee8936b844aa71adcc3581412bea3b46efacca40c2fd0e8d70d057f22ce001dd2d7b74d2f05", "state": "VERIFIED", "summary": {"compressed": 0.628545, "gated": 0.628545, "naive": 0.704442, "scratch": 0.634497, "ungated": 0.671187}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:24:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:21Z", "cpu_seconds": 1.2851951859999993, "created_at": "2026-11-22T04:18:48Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-974b6bcbe22a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "34d165dcebe7e2cfdeae1ba86e0950831418d60e9b3c667fe79ce28a944d3641", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-974b6bcbe22a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:18:48Z", "seed": 1080184511, "treatment": "frontier"}, "priority": 0.20935029409653846, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "635b07a8f553c2b0d3dba7405e4be9b81a1ee7b14cc8d149b2ef786f870ef8eb", "run_seconds": 1.2965059280395508, "seed": 1080184511, "started_at": "2026-11-22T04:19:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.012923, 0.293103], "effect": 0.14009, "evaluated_at": "2026-11-22T04:29:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-974b6bcbe22a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.517857143, 0.793103448, 0.822916667, 0.826923077], "frontier": [0.4375, 0.896551724, 0.958333333, 0.942307692], "goal": [0.544642857, 0.724137931, 0.791666667, 0.701923077], "hypdisc": [0.535714286, 0.74137931, 0.822916667, 0.894230769], "infogain": [0.4375, 0.913793103, 0.96875, 0.884615385], "prederr": [0.410714286, 0.698275862, 0.083333333, 0.5], "random": [0.526785714, 0.637931034, 0.75, 0.759615385], "risk": [0.598214286, 0.896551724, 0.958333333, 0.942307692]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "34d165dcebe7e2cfdeae1ba86e0950831418d60e9b3c667fe79ce28a944d3641", "primary_control": "random", "result_hash": "635b07a8f553c2b0d3dba7405e4be9b81a1ee7b14cc8d149b2ef786f870ef8eb", "signature": "083e2d87206a0ba6b59af583e7ddbf90fa747b9aebddd51ac8f15fd8dc200e318f898d35443e0c326751cd74d1faf0c20d6223dd1118d9afb1f8efbd56aa270f", "state": "VERIFIED", "summary": {"count": 0.7402, "frontier": 0.808673, "goal": 0.690593, "hypdisc": 0.74856, "infogain": 0.801165, "prederr": 0.423081, "random": 0.668583, "risk": 0.848852}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:29:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:56Z", "cpu_seconds": 1.8382040999999987, "created_at": "2026-11-22T04:18:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3828b03a3ec1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 3, "n": 4}, "prereg_hash": "cc4f003626973c7c3ced795a5afab62269e621ba91fbbf145972a6f1e16994d7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3828b03a3ec1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:18:48Z", "seed": 1804589158, "treatment": "full"}, "priority": 0.21415585516784913, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0c303a130b488fd85d0bdcc45159b5535242f31b60a803f9e687e9502f5134a2", "run_seconds": 1.8506789207458496, "seed": 1804589158, "started_at": "2026-11-22T04:18:54Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:26:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3828b03a3ec1", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.75, 0.5, 0.0, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "cc4f003626973c7c3ced795a5afab62269e621ba91fbbf145972a6f1e16994d7", "primary_control": "e141_best", "result_hash": "0c303a130b488fd85d0bdcc45159b5535242f31b60a803f9e687e9502f5134a2", "signature": "24eb838d7246520167ff5b989044d5c1a629481000415c237e0f8686297eaf077314f0f46ae3ef34fcdffb71c1eda75a0fa5094ed1836fcc000b13b0ab4d7201", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:26:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:02Z", "cpu_seconds": 2.4055891060000008, "created_at": "2026-11-22T04:18:48Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-35dab12cb516", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lights|lights|1|9"}, "prereg_hash": "d20d3e80938aa8ab923b868f858b32b02bd0f5d5c60aeafd25fbb069dcb65913", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-35dab12cb516", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lights|lights|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:18:48Z", "seed": 1882791824, "treatment": "gated"}, "priority": 0.2122931954023234, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "cbb23063cf3b6b8ccad0e7d33a28459029ed3a349723405c124c56fb0ee6f154", "run_seconds": 2.6531782150268555, "seed": 1882791824, "started_at": "2026-11-22T04:18:59Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:26:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-35dab12cb516", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.555555556, 0.25, 0.409722222, 0.534722222], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.555555556, 0.75, 0.694444444, 0.694444444]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d20d3e80938aa8ab923b868f858b32b02bd0f5d5c60aeafd25fbb069dcb65913", "primary_control": "scratch", "result_hash": "cbb23063cf3b6b8ccad0e7d33a28459029ed3a349723405c124c56fb0ee6f154", "signature": "2a33911e4ac573d8beca901c2ed9bcf403238dc5bfdcb37d0accb6d62cd9f736cdf909169ed2a85109c7b78d667b13a474f529acef5c613a868d74ce9e373203", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.4375, "scratch": 1.0, "ungated": 0.673611}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:26:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:18:54Z", "cpu_seconds": 1.1581533400000001, "created_at": "2026-11-22T04:18:48Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-b859ea3b81eb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "prereg_hash": "8d763a4082c3d91a02273b7ab39d5f189486806efd53c9734d0037187590f4e1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-b859ea3b81eb", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:18:48Z", "seed": 685378484, "treatment": "gated"}, "priority": 0.2265543068037201, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "54de2d71074c1947fcfd8993a5cc3fe8239ace9ceb769636a7aa9adc5b8ecf28", "run_seconds": 1.1753718852996826, "seed": 685378484, "started_at": "2026-11-22T04:18:53Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:24:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-b859ea3b81eb", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.796875, 0.66, 0.489583333, 0.666666667], "gated": [0.796875, 0.66, 0.489583333, 0.666666667], "naive": [0.65625, 0.54, 0.489583333, 0.694444444], "scratch": [0.796875, 0.66, 0.489583333, 0.666666667], "ungated": [0.65625, 0.59, 0.541666667, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8d763a4082c3d91a02273b7ab39d5f189486806efd53c9734d0037187590f4e1", "primary_control": "scratch", "result_hash": "54de2d71074c1947fcfd8993a5cc3fe8239ace9ceb769636a7aa9adc5b8ecf28", "signature": "5570fb7db913fed0ccc3691c42fbffb1d29c23f0bb9a8a80fc1f05f8593e62627528516ce8cbbfefc9bb723fa2a7581c2827a2e8ef4fe9d3b1ad0025e6b7120c", "state": "PARTIAL", "summary": {"compressed": 0.653281, "gated": 0.653281, "naive": 0.595069, "scratch": 0.653281, "ungated": 0.634479}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:24:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:08Z", "cpu_seconds": 1.3836599019999998, "created_at": "2026-11-22T04:19:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6e6c2fc574c5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "2c990873a0b09db024c9436fdc60ade7a945733cc7d75adb10700ef15728ea90", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6e6c2fc574c5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:05Z", "seed": 336445707, "treatment": "frontier"}, "priority": 0.2237994005683042, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "de4bf80ab17cac3dd866dff3dc1a5945f624760d08575e43f20f4319d68e1824", "run_seconds": 1.5926623344421387, "seed": 336445707, "started_at": "2026-11-22T04:19:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.008618, 0.161506], "effect": 0.085062, "evaluated_at": "2026-11-22T04:26:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6e6c2fc574c5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.676767677, 0.789473684, 0.864583333, 0.715686275], "frontier": [0.707070707, 0.824561404, 0.822916667, 0.803921569], "goal": [0.606060606, 0.824561404, 0.770833333, 0.803921569], "hypdisc": [0.737373737, 0.824561404, 0.822916667, 0.803921569], "infogain": [0.686868687, 0.824561404, 0.822916667, 0.803921569], "prederr": [0.757575758, 0.824561404, 0.760416667, 0.803921569], "random": [0.686868687, 0.649122807, 0.697916667, 0.784313725], "risk": [0.707070707, 0.824561404, 0.822916667, 0.803921569]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2c990873a0b09db024c9436fdc60ade7a945733cc7d75adb10700ef15728ea90", "primary_control": "random", "result_hash": "de4bf80ab17cac3dd866dff3dc1a5945f624760d08575e43f20f4319d68e1824", "signature": "df3b7da9d5d197d6f5f516eeb5530233de5d1262405accb315a7aa8fe23230a0b3cf3022916090d831782f5028668981c153b1f3e9bd6e6f609b7c59c9821101", "state": "VERIFIED", "summary": {"count": 0.761628, "frontier": 0.789618, "goal": 0.751344, "hypdisc": 0.797193, "infogain": 0.784567, "prederr": 0.786619, "random": 0.704555, "risk": 0.789618}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:26:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:08Z", "cpu_seconds": 1.1187107089999984, "created_at": "2026-11-22T04:19:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f4422d8c875d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "208df86a8d5f5da27eb148118606ad9df608ddde60f34de84c84060165eca28f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f4422d8c875d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:05Z", "seed": 1082661558, "treatment": "frontier"}, "priority": 0.21729660766311987, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c143fba38a517b6cfd44b44b2294bb4c1e5f9fbcc89eae4a392eb849f355674d", "run_seconds": 1.2813396453857422, "seed": 1082661558, "started_at": "2026-11-22T04:19:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.117428, 0.215755], "effect": 0.049164, "evaluated_at": "2026-11-22T04:26:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f4422d8c875d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.858585859, 0.888888889, 0.854166667, 0.916666667], "frontier": [0.696969697, 0.708333333, 0.854166667, 0.916666667], "goal": [0.696969697, 0.680555556, 0.854166667, 0.916666667], "hypdisc": [0.565656566, 0.888888889, 0.854166667, 0.916666667], "infogain": [0.696969697, 0.638888889, 0.854166667, 0.916666667], "prederr": [0.696969697, 0.708333333, 0.854166667, 0.888888889], "random": [0.656565657, 0.833333333, 0.572916667, 0.916666667], "risk": [0.696969697, 0.708333333, 0.854166667, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "208df86a8d5f5da27eb148118606ad9df608ddde60f34de84c84060165eca28f", "primary_control": "random", "result_hash": "c143fba38a517b6cfd44b44b2294bb4c1e5f9fbcc89eae4a392eb849f355674d", "signature": "31d3f59d518033263ce459f609ede0d756bf65abc784ad16c8a2d1932805bd983fdd6367604b1f6387ec4603dacd1c16f4a7a256801c73fcb771dcdd4ad49501", "state": "VERIFIED", "summary": {"count": 0.879577, "frontier": 0.794034, "goal": 0.78709, "hypdisc": 0.806345, "infogain": 0.776673, "prederr": 0.78709, "random": 0.744871, "risk": 0.794034}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:26:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:12Z", "cpu_seconds": 0.4289993039999995, "created_at": "2026-11-22T04:19:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d5dbbd48ac8d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "071e33499251b29463798776fa1c7552f7ef0b6ac30ff56c467c3d1607f9f859", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d5dbbd48ac8d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:19:05Z", "seed": 1139445304, "treatment": "full"}, "priority": 0.21593750545298207, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "60a648950b0150dee7a3fe72218c14a723ee1750dc50cb17f0540da60f4030c2", "run_seconds": 0.46238064765930176, "seed": 1139445304, "started_at": "2026-11-22T04:19:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:26:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d5dbbd48ac8d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 0.5, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.0, 0.5, 0.5, 0.75], "no_world_model": [0.0, 0.0, 0.25, 0.75], "random": [0.0, 0.5, 0.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "071e33499251b29463798776fa1c7552f7ef0b6ac30ff56c467c3d1607f9f859", "primary_control": "e141_best", "result_hash": "60a648950b0150dee7a3fe72218c14a723ee1750dc50cb17f0540da60f4030c2", "signature": "07d0607dce7d003263bcb0eab12cb0c7524a2b2279785fd9355cab04e7e00129c7a9c390313992fc4f485cc87836734e7610bf9bd912950b43c887a5c0c8b607", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.0, "full": 0.4375, "no_world_model": 0.25, "random": 0.3125}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:26:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:13Z", "cpu_seconds": 0.47576190599999535, "created_at": "2026-11-22T04:19:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f6a4bb7e4a7a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "b22e1527bd0f633ee94dbf9724eb3c8635ca3b173802538220ac2349d540039f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f6a4bb7e4a7a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:19:05Z", "seed": 1135301883, "treatment": "full"}, "priority": 0.21495969690581784, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "bf06b573e6119c58f97dd13cb9478a3d0eb4910d21df73bef8921224411060f8", "run_seconds": 0.485107421875, "seed": 1135301883, "started_at": "2026-11-22T04:19:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T04:26:25Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f6a4bb7e4a7a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.5, 1.0, 0.5], "fixed_policy": [0.0, 0.5, 0.0, 0.0], "full": [0.0, 0.5, 1.0, 0.5], "no_world_model": [0.0, 0.0, 0.75, 0.0], "random": [0.0, 0.0, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "b22e1527bd0f633ee94dbf9724eb3c8635ca3b173802538220ac2349d540039f", "primary_control": "e141_best", "result_hash": "bf06b573e6119c58f97dd13cb9478a3d0eb4910d21df73bef8921224411060f8", "signature": "368b1d4f3c4f4a3246ebd44c02cdbd2f4f9cb2e7553c0b848d2d20d127bcb3e96a376cef734a3339456c80b623192fa5390807d68e263d5bcd9a0e5cccef1901", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.125, "full": 0.5, "no_world_model": 0.1875, "random": 0.375}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:26:25Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:07Z", "cpu_seconds": 1.0530287510000012, "created_at": "2026-11-22T04:19:05Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-0e57df5b9897", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "8b23cddbd087e7b9f79da9664cc9ea9c6dcef9a8c58bb5466c7aa01534d68b03", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-0e57df5b9897", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:19:05Z", "seed": 1542878198, "treatment": "gated"}, "priority": 0.39625380768374735, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "03623b6088b6ae3cf395d06ac98e0d4dcf4ca9aaaad8934d14fbe7ab0d574119", "run_seconds": 1.2203655242919922, "seed": 1542878198, "started_at": "2026-11-22T04:19:06Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:26:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-0e57df5b9897", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.523809524, 0.423076923, 0.40625, 0.31], "gated": [0.523809524, 0.423076923, 0.40625, 0.31], "naive": [0.571428571, 0.567307692, 0.541666667, 0.47], "scratch": [0.523809524, 0.423076923, 0.40625, 0.31], "ungated": [0.523809524, 0.567307692, 0.447916667, 0.4]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8b23cddbd087e7b9f79da9664cc9ea9c6dcef9a8c58bb5466c7aa01534d68b03", "primary_control": "scratch", "result_hash": "03623b6088b6ae3cf395d06ac98e0d4dcf4ca9aaaad8934d14fbe7ab0d574119", "signature": "8a08c0a816b4ec213f6130888e919d6e40f678fd51ba76796a2344ae7f6e6b83d743db8a662b4238672ea5bd90f923a0a20b79f3c503eeeb8cad5b95e1adb302", "state": "PARTIAL", "summary": {"compressed": 0.415784, "gated": 0.415784, "naive": 0.537601, "scratch": 0.415784, "ungated": 0.484758}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:26:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:10Z", "cpu_seconds": 3.8421389830000123, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3085df5ddcda", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "213285e7c2822e7571fd246c0b40fa82b2d30fae3a2ce620bd4491506a76702b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3085df5ddcda", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:21Z", "seed": 1622293731, "treatment": "frontier"}, "priority": 0.20719759653722067, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "795f0bffe87ca0db3451022ec6a625d8b743000f0161d23339c961383d8bead5", "run_seconds": 4.260834217071533, "seed": 1622293731, "started_at": "2026-11-22T04:20:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.0825, 0.146581], "effect": 0.032041, "evaluated_at": "2026-11-22T04:30:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-3085df5ddcda", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.571794872, 0.467961165, 0.576923077, 0.629411765], "frontier": [0.430769231, 0.512621359, 0.666666667, 0.611764706], "goal": [0.320512821, 0.512621359, 0.046153846, 0.541176471], "hypdisc": [0.284615385, 0.469902913, 0.569230769, 0.647058824], "infogain": [0.633333333, 0.233009709, 0.561538462, 0.611764706], "prederr": [0.566666667, 0.246601942, 0.548717949, 0.611764706], "random": [0.569230769, 0.462135922, 0.556410256, 0.505882353], "risk": [0.430769231, 0.211650485, 0.666666667, 0.611764706]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "213285e7c2822e7571fd246c0b40fa82b2d30fae3a2ce620bd4491506a76702b", "primary_control": "random", "result_hash": "795f0bffe87ca0db3451022ec6a625d8b743000f0161d23339c961383d8bead5", "signature": "bd109216ea5925410bb2c55d30738cbf37125dd97d4ad3b89473f3cd562bc7a96e91e110a4619a00a31ea4acff7b295b1bd155b9c71d00f12379f15b56baef0d", "state": "VERIFIED", "summary": {"count": 0.561523, "frontier": 0.555455, "goal": 0.355116, "hypdisc": 0.492702, "infogain": 0.509912, "prederr": 0.493438, "random": 0.523415, "risk": 0.480213}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:30:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:40Z", "cpu_seconds": 3.752283549000005, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6ac2694fb9e1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "dd3e0d6acc57e8b91bc1a09df69bd1a888b30270bc551739fae76031eb313dd4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6ac2694fb9e1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:21Z", "seed": 1025291484, "treatment": "frontier"}, "priority": 0.20899867209523207, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5caa5aa5edaba747a5c8c2e988d662605f96566c0993dc1b29d024e1eaa842e2", "run_seconds": 4.863676071166992, "seed": 1025291484, "started_at": "2026-11-22T04:19:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.206371, -0.00667], "effect": -0.10652, "evaluated_at": "2026-11-22T04:29:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6ac2694fb9e1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.794117647, 0.896551724, 0.360687023, 0.736111111], "frontier": [0.713235294, 0.655172414, 0.369751908, 0.430555556], "goal": [0.669117647, 0.517241379, 0.327290076, 0.576388889], "hypdisc": [0.753676471, 0.827586207, 0.344942748, 0.631944444], "infogain": [0.713235294, 0.655172414, 0.369751908, 0.430555556], "prederr": [0.713235294, 0.655172414, 0.369751908, 0.430555556], "random": [0.786764706, 0.801724138, 0.353530534, 0.652777778], "risk": [0.713235294, 0.655172414, 0.369751908, 0.430555556]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "dd3e0d6acc57e8b91bc1a09df69bd1a888b30270bc551739fae76031eb313dd4", "primary_control": "random", "result_hash": "5caa5aa5edaba747a5c8c2e988d662605f96566c0993dc1b29d024e1eaa842e2", "signature": "2416214375cbd1bbe9421245590df4272e929b5096046308af65a22ae0f46ac417684abf61a0bb3daff4030bfce49a3b6dd9b62723bca143f57cda4b1d55d40c", "state": "VERIFIED", "summary": {"count": 0.696867, "frontier": 0.542179, "goal": 0.522509, "hypdisc": 0.639537, "infogain": 0.542179, "prederr": 0.542179, "random": 0.648699, "risk": 0.542179}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:29:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:11Z", "cpu_seconds": 58.571107055999995, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7eb2c7fee6ad", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 6, "n": 4}, "prereg_hash": "b686ac4bca486aca5dfdbc5d186b016f9cab70a365955d957f82b5a2ad730075", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7eb2c7fee6ad", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:21Z", "seed": 270392823, "treatment": "frontier"}, "priority": 0.20578193587480106, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "66d20eded319342ffe43dc4856f081fbf250c213e134b67ecb506a36e862ac9e", "run_seconds": 61.30038571357727, "seed": 270392823, "started_at": "2026-11-22T04:20:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00207, 0.165927], "effect": 0.081928, "evaluated_at": "2026-11-22T04:30:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7eb2c7fee6ad", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.581578947, 0.553571429, 0.498333333, 0.4375], "frontier": [0.455263158, 0.616071429, 0.636666667, 0.590833333], "goal": [0.536842105, 0.514285714, 0.5325, 0.471666667], "hypdisc": [0.619298246, 0.454464286, 0.515833333, 0.548333333], "infogain": [0.538596491, 0.5875, 0.635, 0.6125], "prederr": [0.521052632, 0.202678571, 0.6175, 0.134166667], "random": [0.460526316, 0.546428571, 0.436666667, 0.5275], "risk": [0.455263158, 0.616071429, 0.636666667, 0.590833333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "b686ac4bca486aca5dfdbc5d186b016f9cab70a365955d957f82b5a2ad730075", "primary_control": "random", "result_hash": "66d20eded319342ffe43dc4856f081fbf250c213e134b67ecb506a36e862ac9e", "signature": "4415c677d2f94ab7af28e933d9876c07a9c7ed580562bcd47245e264cd797bffadd44745c46c6ba1c0c6456631c13ae40059d109acb513f97fcd9a5d0c521a01", "state": "VERIFIED", "summary": {"count": 0.517746, "frontier": 0.574709, "goal": 0.513824, "hypdisc": 0.534482, "infogain": 0.593399, "prederr": 0.368849, "random": 0.49278, "risk": 0.574709}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:30:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:25Z", "cpu_seconds": 0.6181798350000065, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-cd1241d154c9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "0c5fd2a6f5f8b4f7247c6c1fb40b24d9d640e2bd235439f57713e4ddbb74af19", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-cd1241d154c9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:21Z", "seed": 308414399, "treatment": "frontier"}, "priority": 0.21095357584710134, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "38aaf5c624408263cc1aed2b4ed30dccac9fcec6ef585f969d5326944b435230", "run_seconds": 0.6217842102050781, "seed": 308414399, "started_at": "2026-11-22T04:19:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.23543, 0.098347], "effect": -0.068542, "evaluated_at": "2026-11-22T04:29:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-cd1241d154c9", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.95, 0.979166667, 0.916666667, 0.875], "frontier": [0.97, 0.604166667, 0.916666667, 0.708333333], "goal": [0.79, 0.875, 0.883333333, 0.770833333], "hypdisc": [0.93, 0.979166667, 0.933333333, 0.833333333], "infogain": [0.97, 0.9375, 0.9, 0.875], "prederr": [0.97, 0.333333333, 0.916666667, 0.770833333], "random": [0.89, 0.916666667, 0.916666667, 0.75], "risk": [0.97, 0.604166667, 0.916666667, 0.708333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "0c5fd2a6f5f8b4f7247c6c1fb40b24d9d640e2bd235439f57713e4ddbb74af19", "primary_control": "random", "result_hash": "38aaf5c624408263cc1aed2b4ed30dccac9fcec6ef585f969d5326944b435230", "signature": "c3561073b2a0856d9963222e6afda56d9bd3bea777ed5de58fa775885eea4ddcb981ef10908c7184987d44c2e85e549d4be252cc4c226f62ec73139a1ce9f005", "state": "VERIFIED", "summary": {"count": 0.930208, "frontier": 0.799792, "goal": 0.829792, "hypdisc": 0.918958, "infogain": 0.920625, "prederr": 0.747708, "random": 0.868333, "risk": 0.799792}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:29:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:35Z", "cpu_seconds": 2.2466222349999967, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-99c591fadd12", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 2, "n": 4}, "prereg_hash": "b4ab098413a646356f5a8d78539de3241d07de2c7198c413e07f386ccb659b54", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-99c591fadd12", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:19:21Z", "seed": 1678932495, "treatment": "full"}, "priority": 0.20952020287309914, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a6ac30a99d59011963bfb74acb7aabc70ba74bb3b7cc7a15fc7c6df78b2c846a", "run_seconds": 2.681873083114624, "seed": 1678932495, "started_at": "2026-11-22T04:19:32Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:29:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-99c591fadd12", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.0, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "b4ab098413a646356f5a8d78539de3241d07de2c7198c413e07f386ccb659b54", "primary_control": "e141_best", "result_hash": "a6ac30a99d59011963bfb74acb7aabc70ba74bb3b7cc7a15fc7c6df78b2c846a", "signature": "3fe39a4d44ea11500373a13e67ee747d87c748f803631dfab94be224b82fb27de0aa0db759a379b8b0126a4b84a22af2f207f2edfd8b5511e24b304ddefa0300", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:29:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:27Z", "cpu_seconds": 0.5911480830000002, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f1c485525fc4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "ea34efa5a78f585e736197df939b2ccd2dd6a693364924b51487d1eb464bacff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f1c485525fc4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:19:21Z", "seed": 1414242520, "treatment": "full"}, "priority": 0.21009346854204708, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4e8b32fe34cb413ac3730ee3c61a3f89dbdfdfee74e47aafecd62ca0b6131aa9", "run_seconds": 0.5924124717712402, "seed": 1414242520, "started_at": "2026-11-22T04:19:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T04:29:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-f1c485525fc4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.5, 0.75, 1.0, 1.0], "random": [0.75, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "ea34efa5a78f585e736197df939b2ccd2dd6a693364924b51487d1eb464bacff", "primary_control": "e141_best", "result_hash": "4e8b32fe34cb413ac3730ee3c61a3f89dbdfdfee74e47aafecd62ca0b6131aa9", "signature": "c1148722b4b5ff76b41416dc749c9f0ddbc6e93da1e9b5f2724ae45dd5fcf78b0520c7e6c9d8f46bb55001215bcc42221a5f773b3716a619985902344f7d240e", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 0.8125, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:29:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:09Z", "cpu_seconds": 1.6325989720000003, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-42838f76da2d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "a3ea49de22706b393d542f4d945ac5eede78d0233d4a1d134c260be4d04cd93b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-42838f76da2d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:19:21Z", "seed": 411573815, "treatment": "full"}, "priority": 0.20624268163781542, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "776264ea5c76d58a2a2ace6638bf6b36fc822c7bae31c60bfdabc12b47c6d238", "run_seconds": 1.863447904586792, "seed": 411573815, "started_at": "2026-11-22T04:20:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.151495, 0.395672], "effect": 0.273583, "evaluated_at": "2026-11-22T04:30:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-42838f76da2d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.509868421, 0.213235294, 0.093427835, 0.169131238], "frozen": [0.026315789, 0.073529412, 0.037371134, 0.041127542], "full": [0.759868421, 0.669117647, 0.297036082, 0.353974122], "retrieval": [0.769736842, 0.573529412, 0.332474227, 0.386321627], "rules_nodetect": [0.759868421, 0.669117647, 0.297036082, 0.353974122]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a3ea49de22706b393d542f4d945ac5eede78d0233d4a1d134c260be4d04cd93b", "primary_control": "facts", "result_hash": "776264ea5c76d58a2a2ace6638bf6b36fc822c7bae31c60bfdabc12b47c6d238", "signature": "1c4f595ed4eb2b5206a4db4fc66c37d2db1adc3a90b58134ba072de21d8e85d0d508574d85e6e2de3fc4325ea968a437c0f3aab40bd08f33e4ffb7bc8465240f", "state": "VERIFIED", "summary": {"facts": 0.246416, "frozen": 0.044586, "full": 0.519999, "retrieval": 0.515516, "rules_nodetect": 0.519999}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:30:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:30Z", "cpu_seconds": 0.45030242899999706, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7eed2143d4c3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "37596c3c9a59b026ff2aa43a1e87c2ec5e4eee0d3907a20ac00ff00ebaf07c6c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7eed2143d4c3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:19:21Z", "seed": 1369567754, "treatment": "full"}, "priority": 0.20961603836850307, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "04e73e41b3bcf9bcff18f82b894ea1bdadfddc74c27b912d6af9fa80d964f6dc", "run_seconds": 0.4629223346710205, "seed": 1369567754, "started_at": "2026-11-22T04:19:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.070713, 0.124672], "effect": 0.02698, "evaluated_at": "2026-11-22T04:29:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7eed2143d4c3", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.426356589, 0.791666667, 0.555555556, 0.318181818], "frozen": [0.426356589, 0.395833333, 0.176470588, 0.28030303], "full": [0.364341085, 0.791666667, 0.725490196, 0.318181818], "retrieval": [0.426356589, 0.791666667, 0.633986928, 0.318181818], "rules_nodetect": [0.426356589, 0.791666667, 0.908496732, 0.310606061]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "37596c3c9a59b026ff2aa43a1e87c2ec5e4eee0d3907a20ac00ff00ebaf07c6c", "primary_control": "facts", "result_hash": "04e73e41b3bcf9bcff18f82b894ea1bdadfddc74c27b912d6af9fa80d964f6dc", "signature": "fd45ff20ee5531f2225d2c78ecafcb873e9783789a8c41ea5d54da35a06be8d5613b3c0a2241fdcc0ab7931863fa90f55ac4ea49ae7cdd63ff3a7b44da967a07", "state": "VERIFIED", "summary": {"facts": 0.52294, "frozen": 0.319741, "full": 0.54992, "retrieval": 0.542548, "rules_nodetect": 0.609282}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:29:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:52Z", "cpu_seconds": 0.524358055999997, "created_at": "2026-11-22T04:19:21Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-96d278890c3d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "b85b4f9ea127361a82ffd8a50b232c4e547614dcfeb1eb111b14df563a42fbb0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-96d278890c3d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:19:21Z", "seed": 731672947, "treatment": "full"}, "priority": 0.2078120359536248, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "853b98c77319d6578e319d998f9452a96e7b5267547b7060834be9d1f762fc94", "run_seconds": 0.5258781909942627, "seed": 731672947, "started_at": "2026-11-22T04:19:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.07741, 0.069015], "effect": -0.004197, "evaluated_at": "2026-11-22T04:29:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-96d278890c3d", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.468085106, 0.784313725, 0.623188406, 0.63963964], "frozen": [0.34751773, 0.333333333, 0.391304348, 0.342342342], "full": [0.368794326, 0.77124183, 0.637681159, 0.720720721], "retrieval": [0.489361702, 0.77124183, 0.666666667, 0.72972973], "rules_nodetect": [0.780141844, 0.77124183, 0.768115942, 0.81981982]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "b85b4f9ea127361a82ffd8a50b232c4e547614dcfeb1eb111b14df563a42fbb0", "primary_control": "facts", "result_hash": "853b98c77319d6578e319d998f9452a96e7b5267547b7060834be9d1f762fc94", "signature": "a07fbf3a7309f842639ee99a874542b688db4a497a52b4c45b90c9485a0358b7c905ce03da6129a1a13a82cc593804e2ab5894b05da162711f3f5dc845883309", "state": "VERIFIED", "summary": {"facts": 0.628807, "frozen": 0.353624, "full": 0.62461, "retrieval": 0.66425, "rules_nodetect": 0.78483}, "ties": 0, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:29:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:43Z", "cpu_seconds": 3.225950989000001, "created_at": "2026-11-22T04:19:38Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0bdaf47a53d2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "80008d7269120e45e841f233597fc4fa1f6b16b24bfce769e836c47458d6bf11", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0bdaf47a53d2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:38Z", "seed": 827720335, "treatment": "frontier"}, "priority": 0.21957914434895212, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b6ce41c15849c0628ede0e341dc57e42e36148b1006d785d07469fc4f809d989", "run_seconds": 3.303971529006958, "seed": 827720335, "started_at": "2026-11-22T04:19:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.242092, 0.084221], "effect": -0.078935, "evaluated_at": "2026-11-22T04:29:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-0bdaf47a53d2", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.71875, 0.95, 0.865384615], "frontier": [0.970588235, 0.645833333, 0.9, 0.509615385], "goal": [0.970588235, 0.606770833, 0.85, 0.538461538], "hypdisc": [1.0, 0.6484375, 0.85, 0.826923077], "infogain": [0.970588235, 0.645833333, 0.9, 0.509615385], "prederr": [0.970588235, 0.645833333, 0.9, 0.509615385], "random": [0.941176471, 0.6640625, 0.9, 0.836538462], "risk": [0.970588235, 0.645833333, 0.9, 0.509615385]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "80008d7269120e45e841f233597fc4fa1f6b16b24bfce769e836c47458d6bf11", "primary_control": "random", "result_hash": "b6ce41c15849c0628ede0e341dc57e42e36148b1006d785d07469fc4f809d989", "signature": "b1e319d78a3b2667fca30a5e122a25d77075ff39b93f73a061b4bf8942bfa8953061b751894c663c74756d346574ced8ee79bc22a12f67537c267021e327a40a", "state": "VERIFIED", "summary": {"count": 0.883534, "frontier": 0.756509, "goal": 0.741455, "hypdisc": 0.83134, "infogain": 0.756509, "prederr": 0.756509, "random": 0.835444, "risk": 0.756509}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:29:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:50Z", "cpu_seconds": 1.9032254369999997, "created_at": "2026-11-22T04:19:38Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1a467d9dfda5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "2ef65ecfdfb44f96f1a3e1104ee1c3845580eda563fec51fefda8ddd39565557", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1a467d9dfda5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:19:38Z", "seed": 1095422015, "treatment": "full"}, "priority": 0.2103549210940141, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1eb2cc14b434279f922f51ed698174375969aa761fa78ec13fbb5cf3120ca6cf", "run_seconds": 1.939192295074463, "seed": 1095422015, "started_at": "2026-11-22T04:19:48Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:29:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1a467d9dfda5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.0, 0.75, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.5, 1.0], "random": [1.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "2ef65ecfdfb44f96f1a3e1104ee1c3845580eda563fec51fefda8ddd39565557", "primary_control": "e141_best", "result_hash": "1eb2cc14b434279f922f51ed698174375969aa761fa78ec13fbb5cf3120ca6cf", "signature": "b1d44a8da04e299bb3e463dcc77428a79321388bf0cd08f0bcba2b478f58a49c7f6cf2cf56c527215b25abe7d778668f8fd67e77a77eaad4e192953d5394bd0a", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 0.875, "random": 0.9375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:29:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:01Z", "cpu_seconds": 9.561794659999997, "created_at": "2026-11-22T04:19:38Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bf4be351b65d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "aad1a687944c17c39473f3cd2e9ab69ce982e89dcae737d4c353f9f68b079a3c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bf4be351b65d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:19:38Z", "seed": 102842930, "treatment": "full"}, "priority": 0.21012917718717605, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "675e948bfa5233a728702fc8e73564b1ea8a4ee1653aec7411e06a8312492621", "run_seconds": 9.618157386779785, "seed": 102842930, "started_at": "2026-11-22T04:19:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.42207, 0.04707], "effect": -0.1875, "evaluated_at": "2026-11-22T04:29:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-bf4be351b65d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.0, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.25, 0.0, 0.25, 0.0], "no_world_model": [0.75, 0.0, 0.5, 0.0], "random": [0.75, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "aad1a687944c17c39473f3cd2e9ab69ce982e89dcae737d4c353f9f68b079a3c", "primary_control": "e141_best", "result_hash": "675e948bfa5233a728702fc8e73564b1ea8a4ee1653aec7411e06a8312492621", "signature": "1c457b34d67f25b61d10e7d06c714edef33aa3863c6d7fa6df1ae2f16933ae50dda0832e667177b536dfbf89d20e24000fe9b1e62719a311d77375c68ec34107", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.0, "full": 0.125, "no_world_model": 0.3125, "random": 0.4375}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:30:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:46Z", "cpu_seconds": 0.6003523500000014, "created_at": "2026-11-22T04:19:38Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-4c51d7aec2a2", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "c0171c1bb6319e7520b7aa0fb73d61345900f3290dae654e2d1f605d6c9c3805", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-4c51d7aec2a2", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:19:38Z", "seed": 1058203467, "treatment": "gated"}, "priority": 0.21095058134291875, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-81a45483ab66", "result_hash": "d7ce3caabb623d2c73b87a05e5e8c95ab0de76e01bef0e725d580cba7a796335", "run_seconds": 0.637357234954834, "seed": 1058203467, "started_at": "2026-11-22T04:19:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.038693, 0.012549], "effect": -0.013072, "evaluated_at": "2026-11-22T04:29:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-4c51d7aec2a2", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.320261438, 0.458333333, 0.310606061, 0.418300654], "gated": [0.320261438, 0.458333333, 0.310606061, 0.418300654], "naive": [0.320261438, 0.458333333, 0.212121212, 0.418300654], "scratch": [0.320261438, 0.458333333, 0.310606061, 0.470588235], "ungated": [0.339869281, 0.458333333, 0.212121212, 0.359477124]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c0171c1bb6319e7520b7aa0fb73d61345900f3290dae654e2d1f605d6c9c3805", "primary_control": "scratch", "replicated": false, "replication_of": "E141-ow_transfer-81a45483ab66", "result_hash": "d7ce3caabb623d2c73b87a05e5e8c95ab0de76e01bef0e725d580cba7a796335", "signature": "e8438c0281ce0004f4de9fe9bd5c1b43dedc0ec719eeb438b7b49bdaa738586bac3683b9441978c3ae7188f6bc3dcd54b674f0a29b2e856c099f556959003204", "state": "VERIFIED", "summary": {"compressed": 0.376875, "gated": 0.376875, "naive": 0.352254, "scratch": 0.389947, "ungated": 0.34245}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:29:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:47Z", "cpu_seconds": 0.5775968420000055, "created_at": "2026-11-22T04:19:38Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-de8cb8e44331", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "d6064b5fd105e0e2756db52319badfce5987d5870ed300fe06bde81e44e8b9d6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-de8cb8e44331", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:19:38Z", "seed": 628811897, "treatment": "gated"}, "priority": 0.210477444758958, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "25a5f20ce51521f7276b66537e85747ab77fb1f822d65bd841ec36db11132190", "run_seconds": 0.578716516494751, "seed": 628811897, "started_at": "2026-11-22T04:19:46Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:29:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-de8cb8e44331", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.375, 0.437908497, 0.496124031, 0.307189542], "gated": [0.375, 0.437908497, 0.511627907, 0.307189542], "naive": [0.28125, 0.261437908, 0.255813953, 0.189542484], "scratch": [0.375, 0.437908497, 0.511627907, 0.307189542], "ungated": [0.385416667, 0.437908497, 0.496124031, 0.307189542]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d6064b5fd105e0e2756db52319badfce5987d5870ed300fe06bde81e44e8b9d6", "primary_control": "scratch", "result_hash": "25a5f20ce51521f7276b66537e85747ab77fb1f822d65bd841ec36db11132190", "signature": "39852791b7441f1b881ef0d45c7fb011ce232327f9deefe1ffb0c91d60ba0bbd26c87284aa9336897f0f53ab3c7ad4ab2b4f2438a2e75c04fb88156db1264406", "state": "PARTIAL", "summary": {"compressed": 0.404056, "gated": 0.407931, "naive": 0.247011, "scratch": 0.407931, "ungated": 0.40666}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:29:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:56Z", "cpu_seconds": 0.588724360999997, "created_at": "2026-11-22T04:19:55Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e2fa45f92c2f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "93720a4ea87737ab112ab0c99317cec5f738df3a2b06fa09a211a960e7169645", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e2fa45f92c2f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:19:55Z", "seed": 779279431, "treatment": "frontier"}, "priority": 0.21532937843108296, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "05eaec87f4ae5be224d871cba6ec0a04c4bf5dcbf77cadf98f1dd0581c5661ef", "run_seconds": 0.5896661281585693, "seed": 779279431, "started_at": "2026-11-22T04:19:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022224, 0.036808], "effect": 0.007292, "evaluated_at": "2026-11-22T04:29:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e2fa45f92c2f", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.979166667, 0.925, 1.0, 0.958333333], "goal": [1.0, 0.875, 1.0, 0.9375], "hypdisc": [1.0, 0.925, 1.0, 0.9375], "infogain": [0.979166667, 0.925, 1.0, 0.958333333], "prederr": [0.979166667, 0.925, 1.0, 0.958333333], "random": [0.979166667, 0.875, 1.0, 0.979166667], "risk": [0.979166667, 0.925, 1.0, 0.958333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "93720a4ea87737ab112ab0c99317cec5f738df3a2b06fa09a211a960e7169645", "primary_control": "random", "result_hash": "05eaec87f4ae5be224d871cba6ec0a04c4bf5dcbf77cadf98f1dd0581c5661ef", "signature": "3db5c5d685c0df075c6a006a4b0913fe0b12e562bfa90706c7f524a5b6de3f2de6e3a235a48aed16d2b42da8d03db3dee9d77756791f65ca0acb51d4ea60c60f", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.965625, "goal": 0.953125, "hypdisc": 0.965625, "infogain": 0.965625, "prederr": 0.965625, "random": 0.958333, "risk": 0.965625}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:29:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:19:58Z", "cpu_seconds": 0.45897918000000004, "created_at": "2026-11-22T04:19:55Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b7324489aff2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "cf9f4c9083e33d1874b67d135a9443abd9c88a3076b6f6e68071ff7493854aa1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b7324489aff2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:19:55Z", "seed": 850041417, "treatment": "full"}, "priority": 0.2130960453572943, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e6fed78927050167dd062c30d591033a654083b3f2887440ae366740bbbdd3de", "run_seconds": 0.4597609043121338, "seed": 850041417, "started_at": "2026-11-22T04:19:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.24, 0.74], "effect": 0.25, "evaluated_at": "2026-11-22T04:29:48Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b7324489aff2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.25, 1.0, 0.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [0.0, 0.25, 1.0, 1.0], "no_world_model": [0.0, 0.0, 1.0, 1.0], "random": [0.0, 0.25, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "cf9f4c9083e33d1874b67d135a9443abd9c88a3076b6f6e68071ff7493854aa1", "primary_control": "e141_best", "result_hash": "e6fed78927050167dd062c30d591033a654083b3f2887440ae366740bbbdd3de", "signature": "8597d59b42488a4116b4c09b01896e6869cebdc369f40a47fc8403f9afe5b3469fc168f8441e0e78358b52abfb7714c33292047b245d1bb5decf11ac948e630c", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.0625, "full": 0.5625, "no_world_model": 0.5, "random": 0.5625}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:29:48Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:12Z", "cpu_seconds": 0.5584155709999976, "created_at": "2026-11-22T04:19:55Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-19afb8d73e3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "574215351a78706e39f2ad551679957d1927495cf0120e895273dbccd202a033", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-19afb8d73e3b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:19:55Z", "seed": 1530572940, "treatment": "full"}, "priority": 0.20376655778843275, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d9388405e4fe440b85122e0627b53bbfc278f06999f4b594a041ed5f618c783c", "run_seconds": 0.56777024269104, "seed": 1530572940, "started_at": "2026-11-22T04:21:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.000295, 0.034212], "effect": 0.017253, "evaluated_at": "2026-11-22T04:30:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-19afb8d73e3b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.989583333, 0.311111111, 0.795454545, 0.732026144], "frozen": [0.0625, 0.081481481, 0.075757576, 0.091503268], "full": [0.989583333, 0.333333333, 0.803030303, 0.77124183], "retrieval": [0.989583333, 0.488888889, 0.825757576, 0.777777778], "rules_nodetect": [0.989583333, 0.333333333, 0.803030303, 0.77124183]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "574215351a78706e39f2ad551679957d1927495cf0120e895273dbccd202a033", "primary_control": "facts", "result_hash": "d9388405e4fe440b85122e0627b53bbfc278f06999f4b594a041ed5f618c783c", "signature": "078e680df8e8746172ca66e87b7840f207d4d58e94732ddb33037e6cd31a2e1759cd2104bae2b523ee350764a8ccded40d08c53d088bc3c1667eff7d5978a30b", "state": "VERIFIED", "summary": {"facts": 0.707044, "frozen": 0.077811, "full": 0.724297, "retrieval": 0.770502, "rules_nodetect": 0.724297}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:30:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:04Z", "cpu_seconds": 1.2986884060000037, "created_at": "2026-11-22T04:19:55Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-86ac5e5efc9c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "2d833b8431394bd7a356b8fbe4d769f2ac68f23b3e86b3234122f9c21aac6e32", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-86ac5e5efc9c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:19:55Z", "seed": 970486181, "treatment": "full"}, "priority": 0.2074029131699425, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-6c22899d2520", "result_hash": "ac0b74794ddc7fc6e9db71e5c7a384433b6e5911e61b42ec38fabdd37b0bc73c", "run_seconds": 1.5628011226654053, "seed": 970486181, "started_at": "2026-11-22T04:20:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.045423, -0.014101], "effect": -0.029762, "evaluated_at": "2026-11-22T04:30:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-86ac5e5efc9c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.685714286, 0.544326241, 0.343971631, 0.755319149], "frozen": [0.180952381, 0.203900709, 0.108156028, 0.106382979], "full": [0.65, 0.526595745, 0.294326241, 0.739361702], "retrieval": [0.642857143, 0.492907801, 0.322695035, 0.739361702], "rules_nodetect": [0.65, 0.526595745, 0.294326241, 0.739361702]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "2d833b8431394bd7a356b8fbe4d769f2ac68f23b3e86b3234122f9c21aac6e32", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-6c22899d2520", "result_hash": "ac0b74794ddc7fc6e9db71e5c7a384433b6e5911e61b42ec38fabdd37b0bc73c", "signature": "47c6ef2939bd73aa39122915779ccbdd64621040901b70adcd4014da3cf0200a7eca8fff040ef637f8782f7dcf45083d5cb6c866955285d19296c0b05a03e008", "state": "VERIFIED", "summary": {"facts": 0.582333, "frozen": 0.149848, "full": 0.552571, "retrieval": 0.549455, "rules_nodetect": 0.552571}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:30:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:43Z", "cpu_seconds": 0.9810721930000028, "created_at": "2026-11-22T04:19:55Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f1bbbe4878e2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "35a3ee0298c8009712b85cd190c7da1641af2f93c844019ef37400630f652a6a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f1bbbe4878e2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:19:55Z", "seed": 1311220684, "treatment": "full"}, "priority": 0.2052990653449269, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a337a904569c708064212a34e6fc71973cc23210a58d223696cdc96b7668f6ed", "run_seconds": 1.1990420818328857, "seed": 1311220684, "started_at": "2026-11-22T04:20:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.029091, -0.005688], "effect": -0.017389, "evaluated_at": "2026-11-22T04:30:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f1bbbe4878e2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.55372807, 0.195281782, 0.399449036, 0.223404255], "frozen": [0.184210526, 0.152031455, 0.239669421, 0.20212766], "full": [0.542763158, 0.165137615, 0.395316804, 0.199088146], "retrieval": [0.552631579, 0.149410223, 0.338842975, 0.205167173], "rules_nodetect": [0.213815789, 0.560943644, 0.316804408, 0.386018237]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "35a3ee0298c8009712b85cd190c7da1641af2f93c844019ef37400630f652a6a", "primary_control": "facts", "result_hash": "a337a904569c708064212a34e6fc71973cc23210a58d223696cdc96b7668f6ed", "signature": "4513eedbc292fa3134e7610c1ee2f1c341dc98472092ba049f1058a7ac47b18a153df1b52dd2e4ec8ede8fb5ce159f898dcf12ec6fa81a3b43b54c2aaca39d02", "state": "VERIFIED", "summary": {"facts": 0.342966, "frozen": 0.19451, "full": 0.325576, "retrieval": 0.311513, "rules_nodetect": 0.369396}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:30:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:02Z", "cpu_seconds": 1.6224948400000017, "created_at": "2026-11-22T04:19:55Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-4a4b05f4423a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "prereg_hash": "46c2c25f86b00892d8d2695699eebbdb5c6eb1360ed42dba37ee47ea0498e50d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-4a4b05f4423a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:19:55Z", "seed": 950041047, "treatment": "gated"}, "priority": 0.21193132177762058, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "85b1bf6840c90dfa7483c7c43101d2a261fb514a09323f84ce8425a0436f1c41", "run_seconds": 1.664602518081665, "seed": 950041047, "started_at": "2026-11-22T04:20:01Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-4a4b05f4423a", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.616666667, 0.8, 0.633333333, 0.8], "gated": [0.616666667, 0.8, 0.633333333, 0.8], "naive": [0.333333333, 0.383333333, 0.216666667, 0.4], "scratch": [0.616666667, 0.8, 0.633333333, 0.8], "ungated": [0.666666667, 0.683333333, 0.483333333, 0.716666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "46c2c25f86b00892d8d2695699eebbdb5c6eb1360ed42dba37ee47ea0498e50d", "primary_control": "scratch", "result_hash": "85b1bf6840c90dfa7483c7c43101d2a261fb514a09323f84ce8425a0436f1c41", "signature": "d98680f04d3ae59d1d0a9dfda497b15dfdc622500ef67ef5cc34256fab1e2f6ece61cc1ee3421380315f2b4eda5301568d136d46f368d509b9e56f924cc31809", "state": "PARTIAL", "summary": {"compressed": 0.7125, "gated": 0.7125, "naive": 0.333333, "scratch": 0.7125, "ungated": 0.6375}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:22Z", "cpu_seconds": 1.4467714079999894, "created_at": "2026-11-22T04:20:12Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-afb3a319a502", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "d369280ce0e84f312326748493ef32bfe0f6fce3daa2fde6934bf91602ad0182", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-afb3a319a502", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:20:12Z", "seed": 1719857160, "treatment": "frontier"}, "priority": 0.21291600944610375, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a445bde797295bc1c95a8205f6d0b11b5eb5bbdebae57558783440a7c3a2d6b8", "run_seconds": 1.4791860580444336, "seed": 1719857160, "started_at": "2026-11-22T04:20:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.027753, 0.117165], "effect": 0.072459, "evaluated_at": "2026-11-22T04:30:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-afb3a319a502", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.932692308, 0.982142857, 0.803571429], "frontier": [1.0, 0.980769231, 1.0, 0.803571429], "goal": [1.0, 0.884615385, 0.839285714, 0.785714286], "hypdisc": [0.825, 0.932692308, 1.0, 0.892857143], "infogain": [1.0, 0.980769231, 1.0, 0.803571429], "prederr": [1.0, 0.980769231, 1.0, 0.803571429], "random": [0.875, 0.923076923, 0.982142857, 0.714285714], "risk": [1.0, 0.980769231, 1.0, 0.803571429]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "d369280ce0e84f312326748493ef32bfe0f6fce3daa2fde6934bf91602ad0182", "primary_control": "random", "result_hash": "a445bde797295bc1c95a8205f6d0b11b5eb5bbdebae57558783440a7c3a2d6b8", "signature": "3e5b971576d34729ea8d55b24ba43e5b7a2a6d1be5b3d042dd7dbdd2c4d8437186af77a20e25a42386b8ee91aef41dfcd7aebf902151a08cadc8a3d12c9a9103", "state": "VERIFIED", "summary": {"count": 0.929602, "frontier": 0.946085, "goal": 0.877404, "hypdisc": 0.912637, "infogain": 0.946085, "prederr": 0.946085, "random": 0.873626, "risk": 0.946085}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:30:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:28Z", "cpu_seconds": 0.7395053800000113, "created_at": "2026-11-22T04:20:12Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c1107f9bf68e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "35de536f0289c4e0a5d8e554b7969e403b7872eabad4ae8a675bf576b0cfbb44", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c1107f9bf68e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:20:12Z", "seed": 7837521, "treatment": "frontier"}, "priority": 0.20878673929604616, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3040478baa608c97c98ec224d29eb70d63134d5cea0049f795caa00835160aaf", "run_seconds": 0.7440226078033447, "seed": 7837521, "started_at": "2026-11-22T04:20:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.104454, 0.04253], "effect": -0.030962, "evaluated_at": "2026-11-22T04:30:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c1107f9bf68e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.925925926, 0.504065041, 0.98989899, 0.956521739], "frontier": [0.814814815, 0.674796748, 0.919191919, 0.927536232], "goal": [0.777777778, 0.731707317, 0.919191919, 0.869565217], "hypdisc": [0.740740741, 0.707317073, 0.919191919, 0.927536232], "infogain": [0.777777778, 0.731707317, 0.919191919, 0.652173913], "prederr": [0.777777778, 0.731707317, 0.676767677, 0.884057971], "random": [0.925925926, 0.674796748, 0.98989899, 0.869565217], "risk": [0.814814815, 0.674796748, 0.919191919, 0.927536232]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "35de536f0289c4e0a5d8e554b7969e403b7872eabad4ae8a675bf576b0cfbb44", "primary_control": "random", "result_hash": "3040478baa608c97c98ec224d29eb70d63134d5cea0049f795caa00835160aaf", "signature": "0b8ffeafd1894acdfd4916682f00be9824437314b34ddf4f4bf5f61edd0310ea4d9e882315d485013d801093c709660ff08da51a69d7b57c168b76d24d70b500", "state": "VERIFIED", "summary": {"count": 0.844103, "frontier": 0.834085, "goal": 0.824561, "hypdisc": 0.823696, "infogain": 0.770213, "prederr": 0.767578, "random": 0.865047, "risk": 0.834085}, "ties": 1, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:30:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:30Z", "cpu_seconds": 2.1133244839999996, "created_at": "2026-11-22T04:20:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c3fb255fcde2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "6fd2678b39a0978a92fa9e2fa01f7e8f60773f974d0add22f8452bf7a3920441", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c3fb255fcde2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:20:12Z", "seed": 384600373, "treatment": "full"}, "priority": 0.2082060601353278, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0892c0aa7c12dc1985a5783ae4fd48417dc7e18275e3cc67e2d53900bce70517", "run_seconds": 2.16342830657959, "seed": 384600373, "started_at": "2026-11-22T04:20:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.021935, 0.067634], "effect": 0.022849, "evaluated_at": "2026-11-22T04:30:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c3fb255fcde2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 1.0, 1.0, 0.908602151], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 1.0, 0.97311828], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6fd2678b39a0978a92fa9e2fa01f7e8f60773f974d0add22f8452bf7a3920441", "primary_control": "facts", "result_hash": "0892c0aa7c12dc1985a5783ae4fd48417dc7e18275e3cc67e2d53900bce70517", "signature": "ab34a4c09d0b2188b2eaf9c1786f87305b13e522f2187187973ce15cdc08e45581454ed89341cf46f1f427488953f8096c85c285a5447d56766df23134731f08", "state": "VERIFIED", "summary": {"facts": 0.977151, "frozen": 0.0, "full": 1.0, "retrieval": 0.99328, "rules_nodetect": 1.0}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:30:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:23Z", "cpu_seconds": 0.6932488779999986, "created_at": "2026-11-22T04:20:12Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2929ad37b378", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "41c649b8e9e3f8151c120915754bbbe7732e1517ba70ab3bfd56837d7f89bb9a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2929ad37b378", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:20:12Z", "seed": 1522941834, "treatment": "gated"}, "priority": 0.20989859647970216, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "97fe10db65202f483a4500ad71f9d1ac686429764a8c7c81fe7c22b0748c2eda", "run_seconds": 0.7464468479156494, "seed": 1522941834, "started_at": "2026-11-22T04:20:22Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2929ad37b378", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.588392857, 0.490625, 0.594166667, 0.95], "gated": [0.588392857, 0.490625, 0.594166667, 0.95], "naive": [0.4125, 0.334375, 0.415, 0.95], "scratch": [0.588392857, 0.490625, 0.594166667, 0.95], "ungated": [0.4125, 0.490625, 0.585833333, 0.95]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "41c649b8e9e3f8151c120915754bbbe7732e1517ba70ab3bfd56837d7f89bb9a", "primary_control": "scratch", "result_hash": "97fe10db65202f483a4500ad71f9d1ac686429764a8c7c81fe7c22b0748c2eda", "signature": "b0946050c97a11256d04f8f5a01cf45030001a03d0d376a0943f7eefac710becdecc0abedc24ae58e08caf06cdb18dbbf6806dc4e8e20a3809cd4c6847a0dd01", "state": "PARTIAL", "summary": {"compressed": 0.655796, "gated": 0.655796, "naive": 0.527969, "scratch": 0.655796, "ungated": 0.60974}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:42Z", "cpu_seconds": 1.6869108310000058, "created_at": "2026-11-22T04:20:12Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3a244ab6eb78", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lights|lights|1|1"}, "prereg_hash": "c29465a133c49ad08afe1675d733a4f60b61193e802234414fc246a2e6d1e95f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3a244ab6eb78", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lights|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:20:12Z", "seed": 184251341, "treatment": "gated"}, "priority": 0.20819962309877713, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "62effbb4e4cac14ad4fc075c9d5e060953c56231bec17e63f5cf8be3cdce1b1e", "run_seconds": 1.838975429534912, "seed": 184251341, "started_at": "2026-11-22T04:20:40Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-3a244ab6eb78", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.444444444, 0.344444444, 0.322222222, 0.392473118], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.577777778, 0.744444444, 0.588888889, 0.844086022]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c29465a133c49ad08afe1675d733a4f60b61193e802234414fc246a2e6d1e95f", "primary_control": "scratch", "result_hash": "62effbb4e4cac14ad4fc075c9d5e060953c56231bec17e63f5cf8be3cdce1b1e", "signature": "98e42b556e26ebe9dcc7779f63c06908762508ca312782c370f978d3be1427c46d8e5ea3a18895286dde23fedb3c9a6eb51c02076629281de3e1ed7b6b482e07", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.375896, "scratch": 1.0, "ungated": 0.688799}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:16Z", "cpu_seconds": 0.6388299910000086, "created_at": "2026-11-22T04:20:12Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-cacce566fb15", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "efc9754380859b225fbbb88dee69778e2a3c1ef2ce5dc8b85b5b4acbd7e47e8d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-cacce566fb15", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:20:12Z", "seed": 1478246060, "treatment": "gated"}, "priority": 0.2141758184632383, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "829691ee630bf9d559317e416bceacf38f1c6987eeb4b8d14bc7afb87d1cf746", "run_seconds": 0.6726489067077637, "seed": 1478246060, "started_at": "2026-11-22T04:20:15Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-cacce566fb15", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.666666667, 0.599033816, 0.743589744, 0.833333333], "gated": [0.666666667, 0.599033816, 0.743589744, 0.833333333], "naive": [0.608695652, 0.381642512, 0.58974359, 0.833333333], "scratch": [0.666666667, 0.599033816, 0.743589744, 0.833333333], "ungated": [0.652173913, 0.400966184, 0.64957265, 0.833333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "efc9754380859b225fbbb88dee69778e2a3c1ef2ce5dc8b85b5b4acbd7e47e8d", "primary_control": "scratch", "result_hash": "829691ee630bf9d559317e416bceacf38f1c6987eeb4b8d14bc7afb87d1cf746", "signature": "2b8e1abddad5d4b3ad833b8fc11c2c7cfcad7098d2bfe38edf474f25e87ad6dd14eb2583c23247c4ce9b63bca21374ed9b58e290eb508695c5f458f7b5126405", "state": "PARTIAL", "summary": {"compressed": 0.710656, "gated": 0.710656, "naive": 0.603354, "scratch": 0.710656, "ungated": 0.634012}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:17Z", "cpu_seconds": 3.0521828850000006, "created_at": "2026-11-22T04:20:12Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e2f6fe8992c0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lights|lights|1|1"}, "prereg_hash": "151b280463480e881872db8fb5d81b272ea2453e3b431290d8abd9352fa71f53", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e2f6fe8992c0", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lights|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:20:12Z", "seed": 1062275264, "treatment": "gated"}, "priority": 0.21991045535114565, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "59ef7e7ba8de645ae83ac52bb7a61401662c4fe2f80952f24ce770bd6c17527e", "run_seconds": 3.148148536682129, "seed": 1062275264, "started_at": "2026-11-22T04:20:14Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-e2f6fe8992c0", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [1.0, 0.85483871, 1.0, 1.0], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "151b280463480e881872db8fb5d81b272ea2453e3b431290d8abd9352fa71f53", "primary_control": "scratch", "result_hash": "59ef7e7ba8de645ae83ac52bb7a61401662c4fe2f80952f24ce770bd6c17527e", "signature": "74e0d6473a31db838d6a3a1d4370f5a8a515179564c53fcf2a599e11660fcd14061eb5534a0132c29f158c46b14e1ceaa33e9d6f82de5555ced0a01d9455090f", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.96371, "scratch": 1.0, "ungated": 1.0}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:35Z", "cpu_seconds": 1.8297206069999987, "created_at": "2026-11-22T04:20:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a02eee93ec52", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "2fe5affdda7338b61122d9ee7a5d40170b103dc894e25655af9eb0b37a9cec47", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a02eee93ec52", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:20:28Z", "seed": 1654908251, "treatment": "frontier"}, "priority": 0.21727204413048495, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c37998b2e74cc4c1531e44cf87e7a8e2cecb595ac5b9d76351e5a8f5893c4eaf", "run_seconds": 2.1328125, "seed": 1654908251, "started_at": "2026-11-22T04:20:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.079755, 0.310626], "effect": 0.115435, "evaluated_at": "2026-11-22T04:30:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a02eee93ec52", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.884615385, 0.564814815, 0.708333333, 1.0], "frontier": [0.990384615, 0.916666667, 1.0, 0.821428571], "goal": [0.932692308, 0.861111111, 1.0, 0.785714286], "hypdisc": [1.0, 0.759259259, 1.0, 0.964285714], "infogain": [0.990384615, 0.916666667, 1.0, 0.821428571], "prederr": [0.990384615, 0.916666667, 1.0, 0.821428571], "random": [0.826923077, 0.675925926, 0.763888889, 1.0], "risk": [0.990384615, 0.916666667, 1.0, 0.821428571]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "2fe5affdda7338b61122d9ee7a5d40170b103dc894e25655af9eb0b37a9cec47", "primary_control": "random", "result_hash": "c37998b2e74cc4c1531e44cf87e7a8e2cecb595ac5b9d76351e5a8f5893c4eaf", "signature": "242fcac1d8fdcbf1c3f4d895850497309d9b772238272849b9fda34bcd671dc5cdd528fb71b3405fc3c5f6967fadbffb4b03b3c03c08bb94eca317241fa8390d", "state": "VERIFIED", "summary": {"count": 0.789441, "frontier": 0.93212, "goal": 0.894879, "hypdisc": 0.930886, "infogain": 0.93212, "prederr": 0.93212, "random": 0.816684, "risk": 0.93212}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:30:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:20Z", "cpu_seconds": 2.5534692730000046, "created_at": "2026-11-22T04:20:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d0e11f95aa21", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "4109d06528f1f837ac64758f218586bf2d0be8f4a29d84e2b8cea08a03726455", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d0e11f95aa21", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:20:28Z", "seed": 794628842, "treatment": "frontier"}, "priority": 0.2031044550888744, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f1cc60da9a8680fe22cb2b59a01c4f457202a74a6149f89cb00e403f2d7320b2", "run_seconds": 2.6818807125091553, "seed": 794628842, "started_at": "2026-11-22T04:21:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.315056, -0.043089], "effect": -0.179073, "evaluated_at": "2026-11-22T04:30:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d0e11f95aa21", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.921875, 0.338315217, 0.970588235, 0.53997093], "frontier": [0.734375, 0.029211957, 0.588235294, 0.520348837], "goal": [0.765625, 0.033967391, 0.75, 0.542877907], "hypdisc": [0.890625, 0.305706522, 0.941176471, 0.545784884], "infogain": [0.796875, 0.029211957, 0.985294118, 0.533430233], "prederr": [0.53125, 0.029211957, 0.779411765, 0.528343023], "random": [0.828125, 0.313858696, 0.897058824, 0.549418605], "risk": [0.734375, 0.029211957, 0.588235294, 0.520348837]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "4109d06528f1f837ac64758f218586bf2d0be8f4a29d84e2b8cea08a03726455", "primary_control": "random", "result_hash": "f1cc60da9a8680fe22cb2b59a01c4f457202a74a6149f89cb00e403f2d7320b2", "signature": "29924c2b1ab6f222f244126bbea87c66b13270723e13992a68fefc7254c2458a550c3b8283ce89d7cef348129037e5a3806b30791611a41650a767bc7fb74501", "state": "VERIFIED", "summary": {"count": 0.692687, "frontier": 0.468043, "goal": 0.523118, "hypdisc": 0.670823, "infogain": 0.586203, "prederr": 0.467054, "random": 0.647115, "risk": 0.468043}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:30:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:59Z", "cpu_seconds": 1.0364423470000048, "created_at": "2026-11-22T04:20:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e275167a8251", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "9b983718e56990c4e100e4b1e91a90e7c7636f93bd8f28cbd5d54b34f079c3b9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e275167a8251", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:20:28Z", "seed": 1576270530, "treatment": "frontier"}, "priority": 0.20444114934282132, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "626fee0084782f91d8824c359d015af9ac611554b6f8aa9fc5f2bfa1be84f4d5", "run_seconds": 1.0654850006103516, "seed": 1576270530, "started_at": "2026-11-22T04:20:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.149047, 0.329651], "effect": 0.090302, "evaluated_at": "2026-11-22T04:30:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e275167a8251", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.424836601, 0.534883721, 0.846153846, 0.717948718], "frontier": [0.764705882, 0.441860465, 0.846153846, 0.846153846], "goal": [0.790849673, 0.689922481, 0.820512821, 0.846153846], "hypdisc": [0.790849673, 0.705426357, 0.846153846, 0.846153846], "infogain": [0.758169935, 0.76744186, 0.846153846, 0.846153846], "prederr": [0.784313725, 0.751937984, 0.846153846, 0.846153846], "random": [0.555555556, 0.674418605, 0.794871795, 0.512820513], "risk": [0.764705882, 0.441860465, 0.846153846, 0.846153846]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "9b983718e56990c4e100e4b1e91a90e7c7636f93bd8f28cbd5d54b34f079c3b9", "primary_control": "random", "result_hash": "626fee0084782f91d8824c359d015af9ac611554b6f8aa9fc5f2bfa1be84f4d5", "signature": "6af5bb89ab4d79bb44e7bdafd5a11c4c67c3723bcbf8ae0628b2dbccf5f4911b5ad69b4510ad88a9f6b419b386f6bb7506f606068086d582e2a5a04b40b9f204", "state": "VERIFIED", "summary": {"count": 0.630956, "frontier": 0.724719, "goal": 0.78686, "hypdisc": 0.797146, "infogain": 0.80448, "prederr": 0.80714, "random": 0.634417, "risk": 0.724719}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:30:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:36Z", "cpu_seconds": 0.820023122000002, "created_at": "2026-11-22T04:20:28Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7bf5f7ef026e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "e33fe70e31ea17a79cafef51557c953627554e3907fe23ab1ccd32cb6b774e63", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7bf5f7ef026e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:20:28Z", "seed": 2129059496, "treatment": "full"}, "priority": 0.214414117853322, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "eff8312cf71c86336ffb296a9860331cd5742e7f38b571e3b94ad7cba5f7247a", "run_seconds": 0.9171948432922363, "seed": 2129059496, "started_at": "2026-11-22T04:20:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06758, 0.94258], "effect": 0.4375, "evaluated_at": "2026-11-22T04:30:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7bf5f7ef026e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.0, 0.25], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [0.75, 1.0, 0.75, 1.0], "random": [0.0, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "e33fe70e31ea17a79cafef51557c953627554e3907fe23ab1ccd32cb6b774e63", "primary_control": "e141_best", "result_hash": "eff8312cf71c86336ffb296a9860331cd5742e7f38b571e3b94ad7cba5f7247a", "signature": "301eb647c5616037b91f55ce43d54daa5180fd7d36de9f581d05854f5541101051241fa844b4f2c81238fe6a411768dc864d82159c95bcb024d0680494f90106", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.125, "full": 0.9375, "no_world_model": 0.875, "random": 0.5}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:30:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:53Z", "cpu_seconds": 0.52046043, "created_at": "2026-11-22T04:20:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-9198f008b59c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "3cd904365903d6faa2bc7f7bfa0378f7f99a4865adf5d1ae5e4622e562b78ff5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-9198f008b59c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:20:28Z", "seed": 435267958, "treatment": "gated"}, "priority": 0.20459988073215774, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7b52368fc0a566499f60f998cc91d56cf93c56da8544e660052d78b0f7126ebb", "run_seconds": 0.5234427452087402, "seed": 435267958, "started_at": "2026-11-22T04:20:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.017209, 0.005581], "effect": -0.005814, "evaluated_at": "2026-11-22T04:30:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-9198f008b59c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.510416667, 0.174242424, 0.4, 0.240310078], "gated": [0.510416667, 0.174242424, 0.4, 0.240310078], "naive": [0.208333333, 0.136363636, 0.14, 0.224806202], "scratch": [0.510416667, 0.174242424, 0.4, 0.263565891], "ungated": [0.510416667, 0.204545455, 0.4, 0.310077519]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "3cd904365903d6faa2bc7f7bfa0378f7f99a4865adf5d1ae5e4622e562b78ff5", "primary_control": "scratch", "result_hash": "7b52368fc0a566499f60f998cc91d56cf93c56da8544e660052d78b0f7126ebb", "signature": "0ba0b55b2f3898ad8046b4dc16e0ed06a8c330ea5b1cf44025cad47695826b4252da3621301e3571a7028fc9c0eed0f1c19a929fc649631d3f6d79891f1e3605", "state": "VERIFIED", "summary": {"compressed": 0.331242, "gated": 0.331242, "naive": 0.177376, "scratch": 0.337056, "ungated": 0.35626}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:16Z", "cpu_seconds": 0.5628234719999909, "created_at": "2026-11-22T04:20:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-d28fb853b0da", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "f45f0e59c5f00168e1476425fae350ca9d4013739d7d9dbdbe3623844df165a5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d28fb853b0da", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:20:28Z", "seed": 453635251, "treatment": "gated"}, "priority": 0.20362991624440022, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "dae7be32d9621e2f4837dab52c1910be480852ca83f9dc81f0c466c8343691e5", "run_seconds": 0.578294038772583, "seed": 453635251, "started_at": "2026-11-22T04:21:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.029804, 0.091895], "effect": 0.031046, "evaluated_at": "2026-11-22T04:30:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-d28fb853b0da", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.385620915, 0.241830065, 0.189542484, 0.405228758], "gated": [0.385620915, 0.241830065, 0.189542484, 0.405228758], "naive": [0.248366013, 0.248366013, 0.281045752, 0.254901961], "scratch": [0.385620915, 0.241830065, 0.189542484, 0.281045752], "ungated": [0.274509804, 0.241830065, 0.14379085, 0.346405229]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f45f0e59c5f00168e1476425fae350ca9d4013739d7d9dbdbe3623844df165a5", "primary_control": "scratch", "result_hash": "dae7be32d9621e2f4837dab52c1910be480852ca83f9dc81f0c466c8343691e5", "signature": "f638216f563a1e393e49e09d95edcedac911b05def804ddc13eb2f5b239124430865d784c5f56bc3656f590f09a4da8194dc16e362345a7ea296c7c60a950500", "state": "VERIFIED", "summary": {"compressed": 0.305556, "gated": 0.305556, "naive": 0.25817, "scratch": 0.27451, "ungated": 0.251634}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:30:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:48Z", "cpu_seconds": 0.6868974960000003, "created_at": "2026-11-22T04:20:45Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-79fe01111c00", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "e6f6eee04c0c3cbf45b63908a0402a715f4674bd32c88b84f88c7e8b9628abec", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-79fe01111c00", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:20:45Z", "seed": 320869470, "treatment": "frontier"}, "priority": 0.21191634092241446, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-f4bb59a2f533", "result_hash": "e16ba2b0b740ccff336d659edc7e238d112567a247115c4c337bcaa5af88e41d", "run_seconds": 0.7292349338531494, "seed": 320869470, "started_at": "2026-11-22T04:20:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.007828, 0.092172], "effect": 0.05, "evaluated_at": "2026-11-22T04:30:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-79fe01111c00", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 0.983333333], "frontier": [1.0, 0.966666667, 1.0, 1.0], "goal": [0.933333333, 0.9, 1.0, 0.983333333], "hypdisc": [1.0, 1.0, 1.0, 0.983333333], "infogain": [1.0, 0.966666667, 1.0, 1.0], "prederr": [1.0, 0.966666667, 1.0, 1.0], "random": [0.9, 0.966666667, 0.966666667, 0.933333333], "risk": [1.0, 0.966666667, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "e6f6eee04c0c3cbf45b63908a0402a715f4674bd32c88b84f88c7e8b9628abec", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-f4bb59a2f533", "result_hash": "e16ba2b0b740ccff336d659edc7e238d112567a247115c4c337bcaa5af88e41d", "signature": "48705e8533d91ca668d2b0b77c4d6a288311960949f391421291f6a1355ad501fd4beb790bda0357962f9b22da08525b282bc149d89d451ff185b84ec93c120b", "state": "VERIFIED", "summary": {"count": 0.995833, "frontier": 0.991667, "goal": 0.954167, "hypdisc": 0.995833, "infogain": 0.991667, "prederr": 0.991667, "random": 0.941667, "risk": 0.991667}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:30:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:56Z", "cpu_seconds": 5.139990264000005, "created_at": "2026-11-22T04:20:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-24e28f40c1af", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 6, "n": 4}, "prereg_hash": "d3f4d5e9c2b989196af8daa5ecca17c72c898b71b763b4797ac3188e10b53b75", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-24e28f40c1af", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:20:45Z", "seed": 1085576098, "treatment": "full"}, "priority": 0.20070859621757958, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d2ada197d12da415c7c0e0429d79339fe4881c100efec4894f39c34b9dd10a14", "run_seconds": 5.202829837799072, "seed": 1085576098, "started_at": "2026-11-22T04:21:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.096482, 0.596482], "effect": 0.25, "evaluated_at": "2026-11-22T04:31:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-24e28f40c1af", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 1.0, 0.25, 0.5], "fixed_policy": [0.0, 1.0, 0.0, 1.0], "full": [0.25, 1.0, 1.0, 0.75], "no_world_model": [0.25, 1.0, 0.75, 0.5], "random": [1.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "d3f4d5e9c2b989196af8daa5ecca17c72c898b71b763b4797ac3188e10b53b75", "primary_control": "e141_best", "result_hash": "d2ada197d12da415c7c0e0429d79339fe4881c100efec4894f39c34b9dd10a14", "signature": "ea83987e676c4f03b85acab3a5beee3f7f7e9e967c06ebbeb8beecc2ad5db5ab4eae0391f10a674fc6df04ad22755ef88617c94301c12ce368e1dd993f93010a", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.5, "full": 0.75, "no_world_model": 0.625, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:31:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:59Z", "cpu_seconds": 3.4507692119999973, "created_at": "2026-11-22T04:20:45Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8db88962c254", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "980e283686ceb9c5f0a3a6a28288b75d4fa15ce80b484138bba9d68b6d3debad", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8db88962c254", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:20:45Z", "seed": 1174188208, "treatment": "full"}, "priority": 0.20451264972759722, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "127176cd3bb1c4bf10a23a8954bcf384b5a1aed41d8e07f8cbabc4e1b3bbf30a", "run_seconds": 3.469808578491211, "seed": 1174188208, "started_at": "2026-11-22T04:20:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.12, 0.37], "effect": 0.125, "evaluated_at": "2026-11-22T04:30:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8db88962c254", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.5], "fixed_policy": [0.0, 0.75, 0.25, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 0.75, 1.0, 0.75], "random": [0.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "980e283686ceb9c5f0a3a6a28288b75d4fa15ce80b484138bba9d68b6d3debad", "primary_control": "e141_best", "result_hash": "127176cd3bb1c4bf10a23a8954bcf384b5a1aed41d8e07f8cbabc4e1b3bbf30a", "signature": "536c19644dd1926c34c88b5ac7e130e5c90e554f89b4ef608ea0205db046d7683090487e1a661cb5918ccbc1ccb93cfb2675f8463230fc7bc10e91173096d70c", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 0.875, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:30:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:20:51Z", "cpu_seconds": 2.719114189999999, "created_at": "2026-11-22T04:20:45Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-fccc7d98d56d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lights|lights|1|9"}, "prereg_hash": "e0872d3bfa63e0cf5262a75d9988be882dabf302c79357b077a98b4a52759cdf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-fccc7d98d56d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lights|lights|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:20:45Z", "seed": 1357683464, "treatment": "gated"}, "priority": 0.20955227450492664, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "5078576c8b4bf7aab4412907fd9d8f2232d85f742c2ed77b63702ee4e76fffaa", "run_seconds": 2.9767866134643555, "seed": 1357683464, "started_at": "2026-11-22T04:20:48Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-fccc7d98d56d", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [0.763888889, 1.0, 1.0, 0.847222222], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [0.888888889, 1.0, 1.0, 0.847222222]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "e0872d3bfa63e0cf5262a75d9988be882dabf302c79357b077a98b4a52759cdf", "primary_control": "scratch", "result_hash": "5078576c8b4bf7aab4412907fd9d8f2232d85f742c2ed77b63702ee4e76fffaa", "signature": "2b39a2ec1db4f4d053d9b8da25f6db6773439646e4bf4d12651567a498cf1b7270384d4a94c31b46a3d8d3d614cf9dada90b5842fd72169d1d2530d2ba3e520b", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.902778, "scratch": 1.0, "ungated": 0.934028}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:49Z", "cpu_seconds": 146.206526016, "created_at": "2026-11-22T04:21:02Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5c7925b116c8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "7e042dd8508c4955a1e067f96298c0cfd597b3495084586d7b75a535cbeb9153", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5c7925b116c8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:21:02Z", "seed": 1756817542, "treatment": "frontier"}, "priority": 0.20381084575444353, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "727e64217648b663950dcbe5c36b0a2404acb695940b20ad27018a4002d7bbd3", "run_seconds": 159.29395198822021, "seed": 1756817542, "started_at": "2026-11-22T04:21:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023245, 0.08835], "effect": 0.032553, "evaluated_at": "2026-11-22T04:33:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5c7925b116c8", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.592156863, 0.56627451, 0.48, 0.591338583], "frontier": [0.57254902, 0.588235294, 0.598431373, 0.57480315], "goal": [0.57254902, 0.67372549, 0.641568627, 0.57480315], "hypdisc": [0.635294118, 0.580392157, 0.643921569, 0.636220472], "infogain": [0.57254902, 0.588235294, 0.598431373, 0.57480315], "prederr": [0.57254902, 0.588235294, 0.598431373, 0.57480315], "random": [0.584313725, 0.565490196, 0.483137255, 0.570866142], "risk": [0.57254902, 0.588235294, 0.598431373, 0.57480315]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "7e042dd8508c4955a1e067f96298c0cfd597b3495084586d7b75a535cbeb9153", "primary_control": "random", "result_hash": "727e64217648b663950dcbe5c36b0a2404acb695940b20ad27018a4002d7bbd3", "signature": "ce1db526bc9ebacb7bff19f3f79b3e379af9dca11691ff73c5a86f406252a346c0de800f68fcdef49a6ab2c96153bc59209fc97cbec76416a77f1af3b271010e", "state": "VERIFIED", "summary": {"count": 0.557442, "frontier": 0.583505, "goal": 0.615662, "hypdisc": 0.623957, "infogain": 0.583505, "prederr": 0.583505, "random": 0.550952, "risk": 0.583505}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:33:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:58Z", "cpu_seconds": 2.086533086000003, "created_at": "2026-11-22T04:21:02Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c442ec3b877f", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "806c02401290fb89e0e57ea07d9c0ef53860b74ede0db33264e4e1075d980776", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c442ec3b877f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:21:02Z", "seed": 453052090, "treatment": "frontier"}, "priority": 0.20060874944504384, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f85467059bb444c31c9bf3cfeec0e91107cfb94eb3589e5d192dfd08e5953b1e", "run_seconds": 2.0928092002868652, "seed": 453052090, "started_at": "2026-11-22T04:21:56Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:31:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c442ec3b877f", "family": "ow_explore", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"count": [1.0, 0.776962], "frontier": [1.0, 0.813714], "goal": [1.0, 0.678955], "hypdisc": [1.0, 0.729236], "infogain": [1.0, 0.813714], "prederr": [1.0, 0.813714], "random": [1.0, 0.660545], "risk": [1.0, 0.813714]}, "prereg_hash": "806c02401290fb89e0e57ea07d9c0ef53860b74ede0db33264e4e1075d980776", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "f85467059bb444c31c9bf3cfeec0e91107cfb94eb3589e5d192dfd08e5953b1e", "signature": "3087b7b08f750728f76ddc5c04e10bd764503ff0f0fb9c4b1e5a39ac57be586069be8232a6534b4fb27fef300332722a1ba78e1516537f5601208b910c347604", "state": "REJECTED", "summary": {"count": 0.776962, "frontier": 0.813714, "goal": 0.678955, "hypdisc": 0.729236, "infogain": 0.813714, "prederr": 0.813714, "random": 0.660545, "risk": 0.813714}}, "verified_at": "2026-11-22T04:31:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:32Z", "cpu_seconds": 6.156630766000006, "created_at": "2026-11-22T04:21:02Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-127479a02001", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 3, "n": 4}, "prereg_hash": "236acbbf5a6241ba768ac3807159122204d908ad292b4e30b9508a187df467d3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-127479a02001", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:21:02Z", "seed": 1270549711, "treatment": "full"}, "priority": 0.20242875138527255, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6f66479ef01f1eeebd098b94a2ee4a2126d57678f14e494287fcac58e454bd4b", "run_seconds": 6.689152717590332, "seed": 1270549711, "started_at": "2026-11-22T04:21:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:30:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-127479a02001", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.25, 0.75], "full": [1.0, 0.0, 1.0, 1.0], "no_world_model": [0.75, 0.75, 1.0, 1.0], "random": [0.5, 0.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "236acbbf5a6241ba768ac3807159122204d908ad292b4e30b9508a187df467d3", "primary_control": "e141_best", "result_hash": "6f66479ef01f1eeebd098b94a2ee4a2126d57678f14e494287fcac58e454bd4b", "signature": "03041cdacd7273043386eae520529152e90161ecbe11e84006f648f7bb4d8fb0915a5a56314dd0206f563116c719ed4c7f84c3cc5fc9bd80affe62bbf59c7b09", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.25, "full": 0.75, "no_world_model": 0.875, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:31:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:05Z", "cpu_seconds": 0.6706892059999987, "created_at": "2026-11-22T04:21:02Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d9b94f0dbb7b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "a6817374ffa6e261442a7c296909cfa07d62b78d0bf1ce9cdbffb9212f7bb43d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d9b94f0dbb7b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:21:02Z", "seed": 217015080, "treatment": "full"}, "priority": 0.2099507243620149, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "49b5bfb4fa81a770ec320243105ed249b0da8accc87ce890142c456b69f1cf91", "run_seconds": 0.7265622615814209, "seed": 217015080, "started_at": "2026-11-22T04:21:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T04:30:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d9b94f0dbb7b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.25, 0.0], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [1.0, 0.75, 1.0, 1.0], "random": [1.0, 0.75, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "a6817374ffa6e261442a7c296909cfa07d62b78d0bf1ce9cdbffb9212f7bb43d", "primary_control": "e141_best", "result_hash": "49b5bfb4fa81a770ec320243105ed249b0da8accc87ce890142c456b69f1cf91", "signature": "f44897cfc51a3207bce37f9a598759ac203bf0f9ad037b4026513f23fed60918b16a18fea9d3254c450f41aaca03eb35774bcf34cdff865236b1f622ecf3b403", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.125, "full": 0.9375, "no_world_model": 0.9375, "random": 0.9375}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:30:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:07Z", "cpu_seconds": 2.124400470999987, "created_at": "2026-11-22T04:21:02Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-684ef4e05ea4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "prereg_hash": "13ac4951893ef9dd5729140830d0c0945378dae0c24f5657353b4f47831d564b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-684ef4e05ea4", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:21:02Z", "seed": 1730152046, "treatment": "gated"}, "priority": 0.20409719465488046, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d29cfb9bd4e3cb4f92a59b5ce68ae12f55de72f8f40332f06b1251b61bbc52df", "run_seconds": 2.674952507019043, "seed": 1730152046, "started_at": "2026-11-22T04:21:04Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-684ef4e05ea4", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.444444444, 0.611702128, 0.566929134, 0.424603175], "gated": [0.444444444, 0.611702128, 0.566929134, 0.424603175], "naive": [0.404761905, 0.606382979, 0.476377953, 0.44047619], "scratch": [0.444444444, 0.611702128, 0.566929134, 0.424603175], "ungated": [0.44047619, 0.611702128, 0.480314961, 0.392857143]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "13ac4951893ef9dd5729140830d0c0945378dae0c24f5657353b4f47831d564b", "primary_control": "scratch", "result_hash": "d29cfb9bd4e3cb4f92a59b5ce68ae12f55de72f8f40332f06b1251b61bbc52df", "signature": "dd502872c87516d8a948c68f53c92115cd2f3ba671991f8df2f1a2786ab7835f5ab33513c5a14bf3dba043ae956d7400badd4fde340d2ad6c14a0b3256652f08", "state": "PARTIAL", "summary": {"compressed": 0.51192, "gated": 0.51192, "naive": 0.482, "scratch": 0.51192, "ungated": 0.481338}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:30:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:43Z", "cpu_seconds": 12.995914517999992, "created_at": "2026-11-22T04:21:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f76c4fb0c2c5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "5e4c247cdd212747af61ee44ca296f81e4160bdcb37eab2d2d55c085a036b488", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f76c4fb0c2c5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:21:19Z", "seed": 1903776687, "treatment": "frontier"}, "priority": 0.2009482893429207, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "660f588a6c315cb60e45d3bb9595aaf625e0fcecaa194339776b3b8f865590a8", "run_seconds": 14.273767232894897, "seed": 1903776687, "started_at": "2026-11-22T04:21:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.011693, 0.117719], "effect": 0.064706, "evaluated_at": "2026-11-22T04:31:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f76c4fb0c2c5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.530718954, 0.369934641, 0.387581699, 0.381045752], "frontier": [0.57124183, 0.601960784, 0.483006536, 0.513071895], "goal": [0.57124183, 0.601960784, 0.51372549, 0.565359477], "hypdisc": [0.470588235, 0.524183007, 0.531372549, 0.558823529], "infogain": [0.57124183, 0.601960784, 0.483006536, 0.513071895], "prederr": [0.57124183, 0.601960784, 0.483006536, 0.513071895], "random": [0.558823529, 0.462091503, 0.437908497, 0.451633987], "risk": [0.57124183, 0.601960784, 0.483006536, 0.513071895]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "5e4c247cdd212747af61ee44ca296f81e4160bdcb37eab2d2d55c085a036b488", "primary_control": "random", "result_hash": "660f588a6c315cb60e45d3bb9595aaf625e0fcecaa194339776b3b8f865590a8", "signature": "ebd26bfd4a70e7a564b98425aac80212e03b60db68bd86af59689775674a7ed9ae6b18202fc2f24af7597e090a8b0b1eccbc781e66a0e6d09a56563e23cd0206", "state": "VERIFIED", "summary": {"count": 0.41732, "frontier": 0.54232, "goal": 0.563072, "hypdisc": 0.521242, "infogain": 0.54232, "prederr": 0.54232, "random": 0.477614, "risk": 0.54232}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:31:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:48Z", "cpu_seconds": 0.2309003910000058, "created_at": "2026-11-22T04:21:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0ec4195f6013", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "93773c75b8503f3cb6e4409830bd3f20900902887dbb9c229db2e866b573c6bc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0ec4195f6013", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:21:19Z", "seed": 1464184127, "treatment": "full"}, "priority": 0.1928437276742882, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e7957b075e50411290951c4f3050d1eac082b1192cfc7d9b12bdcb3e00af46c5", "run_seconds": 0.24593710899353027, "seed": 1464184127, "started_at": "2026-11-22T04:22:48Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:31:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0ec4195f6013", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.0, 0.5, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "93773c75b8503f3cb6e4409830bd3f20900902887dbb9c229db2e866b573c6bc", "primary_control": "e141_best", "result_hash": "e7957b075e50411290951c4f3050d1eac082b1192cfc7d9b12bdcb3e00af46c5", "signature": "25d4c60b051a51b43004b45e97f916d69bef277c7f1cc3facc5073a6bd8ad65702f88108918bbbc5a952bbcb979f91bf562bd405efb0f8fd154dabe540645c00", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:31:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:24Z", "cpu_seconds": 2.308926504999988, "created_at": "2026-11-22T04:21:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-be5f4a71aa76", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "f3bad2ede59a1d005778463d143b853a1396df52e0c704669abd96e95b293194", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-be5f4a71aa76", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:21:19Z", "seed": 385596859, "treatment": "full"}, "priority": 0.20580607014205826, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "958aa92376307b8d247fd4a0a2b0f471c385f9baabb247124c91af93cb4b95d1", "run_seconds": 2.3331515789031982, "seed": 385596859, "started_at": "2026-11-22T04:21:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T04:30:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-be5f4a71aa76", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.0, 0.0, 0.75], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [0.75, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "f3bad2ede59a1d005778463d143b853a1396df52e0c704669abd96e95b293194", "primary_control": "e141_best", "result_hash": "958aa92376307b8d247fd4a0a2b0f471c385f9baabb247124c91af93cb4b95d1", "signature": "c1fdf1833608d0497c439340eb9fc923bb5c96b4720287b9df4f1e449378fa734a7b58747c4623ba9d6531ea19e633fd0e29c7e6282c3e9745963f897da56c0a", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.3125, "full": 0.9375, "no_world_model": 0.9375, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:30:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:43Z", "cpu_seconds": 0.22591937299999643, "created_at": "2026-11-22T04:21:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a1add2fbe2c8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "baea3343e7862b430db219a245559fda668df475b9314221acaa60e3b7865e5a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a1add2fbe2c8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:21:19Z", "seed": 1321678121, "treatment": "full"}, "priority": 0.193304409201414, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b7af631798dc221042b8e6acb81ee15542c57f89e8c39f465541ffc56212599d", "run_seconds": 0.281383752822876, "seed": 1321678121, "started_at": "2026-11-22T04:22:43Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:31:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-a1add2fbe2c8", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.939393939, 1.0, 1.0, 1.0], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.015151515], "full": [0.939393939, 1.0, 1.0, 1.0], "retrieval": [0.939393939, 1.0, 1.0, 1.0], "rules_nodetect": [0.939393939, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "baea3343e7862b430db219a245559fda668df475b9314221acaa60e3b7865e5a", "primary_control": "facts", "result_hash": "b7af631798dc221042b8e6acb81ee15542c57f89e8c39f465541ffc56212599d", "signature": "cc5cd5ca12cbcccf4be84c9a4c7f06ee0a513dc960a5d9ba8326d241e082e3313afa7fdd892537fb76c293e7d693e270fed4981ec0baf69aa2d7058f77ed5f04", "state": "PARTIAL", "summary": {"facts": 0.984848, "frozen": 0.026515, "full": 0.984848, "retrieval": 0.984848, "rules_nodetect": 0.984848}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:31:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:08Z", "cpu_seconds": 1.341452824000001, "created_at": "2026-11-22T04:21:19Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-32c4f1ea805c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "prereg_hash": "7df86c3d18f6bace6f9f712b2393fa07c9426718ff811dabf4e74f74b540ea40", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-32c4f1ea805c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:21:19Z", "seed": 964298194, "treatment": "gated"}, "priority": 0.1959359312776572, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d25126141f05059fd5b47bb3ce83d7285124d133862e604eb96d6fee7fbf72cf", "run_seconds": 1.455833911895752, "seed": 964298194, "started_at": "2026-11-22T04:22:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:31:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-32c4f1ea805c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.75, 1.0, 0.75, 0.566666667], "gated": [0.75, 1.0, 0.75, 0.816666667], "naive": [0.1, 0.1, 0.083333333, 0.35], "scratch": [0.75, 1.0, 0.75, 0.566666667], "ungated": [0.316666667, 0.1, 0.3, 0.466666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "7df86c3d18f6bace6f9f712b2393fa07c9426718ff811dabf4e74f74b540ea40", "primary_control": "scratch", "result_hash": "d25126141f05059fd5b47bb3ce83d7285124d133862e604eb96d6fee7fbf72cf", "signature": "edd0bc7024bf6f18f5c3e4fd504c5d4d68cc470d7ec30cc36a4926a5bf35d02ba0c50e778a55b71f9628b89305d16c42cdae200d966dcb0b526a38a09d4b7005", "state": "VERIFIED", "summary": {"compressed": 0.766667, "gated": 0.829167, "naive": 0.158333, "scratch": 0.766667, "ungated": 0.295833}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:31:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:38Z", "cpu_seconds": 0.23524504099999888, "created_at": "2026-11-22T04:21:19Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-82285bf12046", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "prereg_hash": "a8a9a3242f378cc377b02b79d64aa84e68b24afff8ca34956c412d4a76389170", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-82285bf12046", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:21:19Z", "seed": 1367221709, "treatment": "gated"}, "priority": 0.19392192135025332, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d54fb0e7c288d564eb57e46a9c2956a83da34d4a17ac66af362326bb6b6b3b23", "run_seconds": 0.25729990005493164, "seed": 1367221709, "started_at": "2026-11-22T04:22:38Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:31:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-82285bf12046", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.409090909, 0.212121212, 0.409090909, 0.317460317], "gated": [0.409090909, 0.212121212, 0.409090909, 0.317460317], "naive": [0.196969697, 0.212121212, 0.196969697, 0.317460317], "scratch": [0.409090909, 0.212121212, 0.409090909, 0.317460317], "ungated": [0.196969697, 0.212121212, 0.196969697, 0.317460317]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a8a9a3242f378cc377b02b79d64aa84e68b24afff8ca34956c412d4a76389170", "primary_control": "scratch", "result_hash": "d54fb0e7c288d564eb57e46a9c2956a83da34d4a17ac66af362326bb6b6b3b23", "signature": "e83b031e949492d28bf61909864cb1571cde05a9acb9cea66e4b6056c283a73cea5f39d3442e76a1a87102f5d886d529cc9f4b732d08821068d284722ee53d03", "state": "PARTIAL", "summary": {"compressed": 0.336941, "gated": 0.336941, "naive": 0.23088, "scratch": 0.336941, "ungated": 0.23088}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:31:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:51Z", "cpu_seconds": 3.084902378999999, "created_at": "2026-11-22T04:21:34Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-357050a34480", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "9ddc2df704d892d455441eee0afa8467b4930dba7513e91435ea72599711755c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-357050a34480", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:21:34Z", "seed": 1284558213, "treatment": "full"}, "priority": 0.20236268961683945, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2b8496cfe6c0835ca67baf96b625bc09f16d32ffdfb96d3310524da5d8200298", "run_seconds": 3.3040404319763184, "seed": 1284558213, "started_at": "2026-11-22T04:21:48Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:31:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-357050a34480", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.77592], "frozen": [1.0, 0.208036], "full": [1.0, 0.834551], "retrieval": [1.0, 0.882882], "rules_nodetect": [1.0, 0.834551]}, "prereg_hash": "9ddc2df704d892d455441eee0afa8467b4930dba7513e91435ea72599711755c", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "2b8496cfe6c0835ca67baf96b625bc09f16d32ffdfb96d3310524da5d8200298", "signature": "a10a5c037590581c2c1b42a8b7acbbc41c1d06cf8c3c62bc26431a4b36402d237edd2cf7812cb41669f59b62def14f0c3f932ffc7230e7ad11207924362b1b07", "state": "REJECTED", "summary": {"facts": 0.77592, "frozen": 0.208036, "full": 0.834551, "retrieval": 0.882882, "rules_nodetect": 0.834551}}, "verified_at": "2026-11-22T04:31:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:46Z", "cpu_seconds": 2.720555173000008, "created_at": "2026-11-22T04:21:34Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-dea119b7a5ac", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 3, "n": 4}, "prereg_hash": "852834bfc723840935cb731a8a2baed33e5ad625beb4e3a6c59994241e119e6e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-dea119b7a5ac", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:21:34Z", "seed": 1402485877, "treatment": "full"}, "priority": 0.20585971730339822, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2f1941ed9568ac288afcabacee0f9043c0a604dbe86a6da6b438985a22288941", "run_seconds": 2.8632733821868896, "seed": 1402485877, "started_at": "2026-11-22T04:21:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.033277, 0.136451], "effect": 0.051587, "evaluated_at": "2026-11-22T04:31:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-dea119b7a5ac", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.951388889, 0.307539683, 1.0, 0.671717172], "frozen": [0.3125, 0.099206349, 0.3125, 0.0], "full": [0.986111111, 0.486111111, 0.993055556, 0.671717172], "retrieval": [0.986111111, 0.60515873, 1.0, 0.919191919], "rules_nodetect": [0.986111111, 0.486111111, 0.993055556, 0.671717172]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "852834bfc723840935cb731a8a2baed33e5ad625beb4e3a6c59994241e119e6e", "primary_control": "facts", "result_hash": "2f1941ed9568ac288afcabacee0f9043c0a604dbe86a6da6b438985a22288941", "signature": "b8b1e312d11937c29648d8dbc5171801c2f511357233218319503a6188c77067b607916931afd7139013bdf61e3dab43f0b95c7c0b1794edf32997d9430e340a", "state": "VERIFIED", "summary": {"facts": 0.732661, "frozen": 0.181052, "full": 0.784249, "retrieval": 0.877615, "rules_nodetect": 0.784249}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:31:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:21:38Z", "cpu_seconds": 0.6247851570000051, "created_at": "2026-11-22T04:21:34Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-a3d3b9debfca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "9e0b20e15f66a411f99f5d24213e91525ec96b42413c3cc47347eaf6a17a66dd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-a3d3b9debfca", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:21:34Z", "seed": 936737800, "treatment": "gated"}, "priority": 0.21260665975041976, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "90a0f50bda03e69d12c5bbd54e1e2ad683e274544666160f6b17e9251f958c70", "run_seconds": 0.8856494426727295, "seed": 936737800, "started_at": "2026-11-22T04:21:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.033856, 0.01098], "effect": -0.011438, "evaluated_at": "2026-11-22T04:31:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-a3d3b9debfca", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.424836601, 0.294117647, 0.411764706, 0.405228758], "gated": [0.424836601, 0.294117647, 0.411764706, 0.405228758], "naive": [0.424836601, 0.281045752, 0.39869281, 0.405228758], "scratch": [0.424836601, 0.294117647, 0.411764706, 0.450980392], "ungated": [0.385620915, 0.320261438, 0.444444444, 0.346405229]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "9e0b20e15f66a411f99f5d24213e91525ec96b42413c3cc47347eaf6a17a66dd", "primary_control": "scratch", "result_hash": "90a0f50bda03e69d12c5bbd54e1e2ad683e274544666160f6b17e9251f958c70", "signature": "572fb635a886797332620780205661088f0f951935cf02d99a283167510337befc6f2be3120a9297e0d6a9527f06aa6c670d328eb4b61ad2fd9d469b6578ae0e", "state": "VERIFIED", "summary": {"compressed": 0.383987, "gated": 0.383987, "naive": 0.377451, "scratch": 0.395425, "ungated": 0.374183}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:31:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:02Z", "cpu_seconds": 0.9510023529999927, "created_at": "2026-11-22T04:21:51Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6e9af831ef69", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "db63ec930f8a3510726303af6a7c234aa22825490fe929431b82cae3fb28b748", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6e9af831ef69", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:21:51Z", "seed": 1989126280, "treatment": "frontier"}, "priority": 0.1992516300341672, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c6530fbce414258f7a11645b5aa41f8c315f7148ea2c207704f18809a8adaff3", "run_seconds": 0.9609208106994629, "seed": 1989126280, "started_at": "2026-11-22T04:22:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.137264, 0.153173], "effect": 0.007955, "evaluated_at": "2026-11-22T04:31:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6e9af831ef69", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.813636364, 0.613636364, 0.787878788, 0.916666667], "frontier": [0.545454545, 0.8, 0.734848485, 0.916666667], "goal": [0.595454545, 0.822727273, 0.537878788, 0.916666667], "hypdisc": [0.354545455, 0.786363636, 0.553030303, 0.916666667], "infogain": [0.754545455, 0.740909091, 0.545454545, 0.916666667], "prederr": [0.681818182, 0.822727273, 0.598484848, 0.611111111], "random": [0.736363636, 0.759090909, 0.71969697, 0.75], "risk": [0.545454545, 0.8, 0.734848485, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "db63ec930f8a3510726303af6a7c234aa22825490fe929431b82cae3fb28b748", "primary_control": "random", "result_hash": "c6530fbce414258f7a11645b5aa41f8c315f7148ea2c207704f18809a8adaff3", "signature": "0ef83b9eb9a1aa138a895303c23917761c2d83db6683265542aa3c08361602da39a429f697904de956c2b3f298247a9079077af706702148087a272cdf344601", "state": "VERIFIED", "summary": {"count": 0.782955, "frontier": 0.749242, "goal": 0.718182, "hypdisc": 0.652652, "infogain": 0.739394, "prederr": 0.678535, "random": 0.741288, "risk": 0.749242}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:31:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:33Z", "cpu_seconds": 1.46300909, "created_at": "2026-11-22T04:21:51Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ef6d7695bd78", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "74841d6c658b2fc55f123770fa3113b9824db1a692bf0e83e425d7ac29d52b6f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ef6d7695bd78", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:21:51Z", "seed": 603609301, "treatment": "frontier"}, "priority": 0.1947781264751457, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5b7bde5a914c8a7f2a1da06ff5f2f928725e1a0a3b7a854b851ff23be096fa05", "run_seconds": 1.5794124603271484, "seed": 603609301, "started_at": "2026-11-22T04:22:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.000664, 0.156271], "effect": 0.078467, "evaluated_at": "2026-11-22T04:31:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ef6d7695bd78", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.441509434, 0.363157895, 0.547058824, 0.327659574], "frontier": [0.350943396, 0.418421053, 0.594117647, 0.54893617], "goal": [0.286792453, 0.281578947, 0.529411765, 0.129787234], "hypdisc": [0.373584906, 0.294736842, 0.505882353, 0.387234043], "infogain": [0.350943396, 0.4, 0.594117647, 0.565957447], "prederr": [0.294339623, 0.131578947, 0.205882353, 0.055319149], "random": [0.366037736, 0.376315789, 0.464705882, 0.391489362], "risk": [0.350943396, 0.418421053, 0.594117647, 0.578723404]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "74841d6c658b2fc55f123770fa3113b9824db1a692bf0e83e425d7ac29d52b6f", "primary_control": "random", "result_hash": "5b7bde5a914c8a7f2a1da06ff5f2f928725e1a0a3b7a854b851ff23be096fa05", "signature": "38625f09bc3f413a9ac482b4a8fafe9cd69a7db513853d41be66ec3897ab8d91a3255e1fa696bee24028bd70aee21970fb31d9db1914fc788b9c068a615d3a08", "state": "VERIFIED", "summary": {"count": 0.419846, "frontier": 0.478105, "goal": 0.306893, "hypdisc": 0.39036, "infogain": 0.477755, "prederr": 0.17178, "random": 0.399637, "risk": 0.485551}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:31:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:05Z", "cpu_seconds": 0.9606675619999976, "created_at": "2026-11-22T04:21:51Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-576d77d1045b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "6c946144bdfd871b758a63818bfe505df2940f70047a541ad0de6cd38e76db02", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-576d77d1045b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:21:51Z", "seed": 217811342, "treatment": "full"}, "priority": 0.1967786546487626, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f02f9725bc3cf2cbcd5807bf27bb72f9fbd432b50b3aca6061d914177df73fbd", "run_seconds": 1.07454252243042, "seed": 217811342, "started_at": "2026-11-22T04:22:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005588, 0.051619], "effect": 0.023015, "evaluated_at": "2026-11-22T04:31:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-576d77d1045b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.9375, 1.0, 0.890625, 0.898648649], "frozen": [0.40625, 0.425, 0.265625, 0.236486486], "full": [0.9375, 1.0, 0.921875, 0.959459459], "retrieval": [0.9375, 1.0, 0.953125, 0.939189189], "rules_nodetect": [0.9375, 1.0, 0.921875, 0.959459459]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "6c946144bdfd871b758a63818bfe505df2940f70047a541ad0de6cd38e76db02", "primary_control": "facts", "result_hash": "f02f9725bc3cf2cbcd5807bf27bb72f9fbd432b50b3aca6061d914177df73fbd", "signature": "b4fa11e3daf31ba16b67161844546498670597f4394fb74baaad7842bc3655395f0e59dac09037039ceb9411806a694b7098165e6aaa7d64cc00ed96d6f76806", "state": "VERIFIED", "summary": {"facts": 0.931693, "frozen": 0.33334, "full": 0.954709, "retrieval": 0.957454, "rules_nodetect": 0.954709}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:31:31Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:15Z", "cpu_seconds": 1.0699779059999912, "created_at": "2026-11-22T04:22:07Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f788124cde37", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "d8e447beb0f5efe3e60cc08360350329bebb5c8eae784838c826f95c0ba8ec6b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f788124cde37", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:22:07Z", "seed": 566100887, "treatment": "frontier"}, "priority": 0.2097709940907117, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "67bd81d7a66bb088b08f526aef330d56150614625832c93f3b973ea876802860", "run_seconds": 1.196486473083496, "seed": 566100887, "started_at": "2026-11-22T04:22:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.047912, 0.190259], "effect": 0.071173, "evaluated_at": "2026-11-22T04:31:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f788124cde37", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.797385621, 0.875816993, 0.76744186, 0.635658915], "frontier": [0.758169935, 0.91503268, 0.759689922, 0.589147287], "goal": [0.751633987, 0.882352941, 0.80620155, 0.705426357], "hypdisc": [0.758169935, 0.758169935, 0.76744186, 0.782945736], "infogain": [0.758169935, 0.91503268, 0.759689922, 0.589147287], "prederr": [0.758169935, 0.91503268, 0.759689922, 0.589147287], "random": [0.745098039, 0.666666667, 0.782945736, 0.542635659], "risk": [0.758169935, 0.921568627, 0.759689922, 0.589147287]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "d8e447beb0f5efe3e60cc08360350329bebb5c8eae784838c826f95c0ba8ec6b", "primary_control": "random", "result_hash": "67bd81d7a66bb088b08f526aef330d56150614625832c93f3b973ea876802860", "signature": "40eb0d4dd43292941cfc424931f34ab1c9dd125c977ff4b3a8b701fff20b5b7c8e1571e1a693e3866c900adb37826161e03436c7da1924a1593bcbfb4d134f07", "state": "VERIFIED", "summary": {"count": 0.769076, "frontier": 0.75551, "goal": 0.786404, "hypdisc": 0.766682, "infogain": 0.75551, "prederr": 0.75551, "random": 0.684337, "risk": 0.757144}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:31:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:14Z", "cpu_seconds": 3.5768460400000066, "created_at": "2026-11-22T04:22:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-3eb7264a6996", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 10, "n": 4}, "prereg_hash": "35e07e4cfe7eff90ca88711d31268490d7a5937c9c459a6b8f4afb7dab0cb8a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-3eb7264a6996", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:07Z", "seed": 822260823, "treatment": "full"}, "priority": 0.2126575193220458, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "23236230ed0a1ebb5335aa93ff58b98d1d13cdd53af9eda1a2f7c74b57d3fe32", "run_seconds": 4.108224153518677, "seed": 822260823, "started_at": "2026-11-22T04:22:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058706, 0.691294], "effect": 0.375, "evaluated_at": "2026-11-22T04:31:31Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-3eb7264a6996", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 0.25, 0.25], "fixed_policy": [1.0, 1.0, 0.75, 0.5], "full": [1.0, 0.5, 0.75, 0.5], "no_world_model": [0.0, 0.5, 0.25, 0.0], "random": [0.75, 0.5, 0.75, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "35e07e4cfe7eff90ca88711d31268490d7a5937c9c459a6b8f4afb7dab0cb8a2", "primary_control": "e141_best", "result_hash": "23236230ed0a1ebb5335aa93ff58b98d1d13cdd53af9eda1a2f7c74b57d3fe32", "signature": "82954d7de4d30af0b9db8e02f64927329b383da18c9c5dcaf4c16f3ac1a81f6c1ff2273ee9c3462d762e58b3a9f04159fe2f671c324a0ddae79f23e871d92e04", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.8125, "full": 0.6875, "no_world_model": 0.1875, "random": 0.5625}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:31:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:20Z", "cpu_seconds": 0.6666947089999979, "created_at": "2026-11-22T04:22:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-649a5d64658c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "04d39a0b2d2eaa1b94bce2b97ab60b23bbe867d85ba4d2e64fd68f7d873cc611", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-649a5d64658c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:07Z", "seed": 1589076707, "treatment": "full"}, "priority": 0.19675608249429577, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c400eb7458308ecb8b789f96bf2f67c9e050a2315986abfe5b942eec3f4584b1", "run_seconds": 0.674705982208252, "seed": 1589076707, "started_at": "2026-11-22T04:22:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.44258, 0.56758], "effect": 0.0625, "evaluated_at": "2026-11-22T04:31:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-649a5d64658c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.75, 1.0, 0.0], "fixed_policy": [0.0, 0.75, 0.0, 0.25], "full": [1.0, 0.25, 1.0, 0.0], "no_world_model": [0.5, 1.0, 1.0, 0.0], "random": [1.0, 0.25, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "04d39a0b2d2eaa1b94bce2b97ab60b23bbe867d85ba4d2e64fd68f7d873cc611", "primary_control": "e141_best", "result_hash": "c400eb7458308ecb8b789f96bf2f67c9e050a2315986abfe5b942eec3f4584b1", "signature": "317946248404e1a55a203fcea4858830769739179d5e6ff831b6266eff72664dfda4296223531b041ae0b2748b4ad99447a78bf9161f710f13cff89db97bbe0a", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.25, "full": 0.5625, "no_world_model": 0.625, "random": 0.8125}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:31:38Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:19Z", "cpu_seconds": 0.5442678529999938, "created_at": "2026-11-22T04:22:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-99f3f3dd4ba6", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "933dc880323b2677ef6a92d0554bc9c2fcc5a8a9a5686843f0208c415a07aca8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-99f3f3dd4ba6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:07Z", "seed": 752176548, "treatment": "full"}, "priority": 0.2043725789709106, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-0f4f8c44f978", "result_hash": "5ac7ebda927b4a55d4176263b2a9710f33f5f1d984755ec9fa1606feb9acbef3", "run_seconds": 0.6021947860717773, "seed": 752176548, "started_at": "2026-11-22T04:22:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.058337, 0.129244], "effect": 0.035454, "evaluated_at": "2026-11-22T04:31:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-99f3f3dd4ba6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.522875817, 0.699346405, 0.325581395, 0.267973856], "frozen": [0.176470588, 0.176470588, 0.294573643, 0.281045752], "full": [0.692810458, 0.725490196, 0.271317829, 0.267973856], "retrieval": [0.647058824, 0.751633987, 0.310077519, 0.241830065], "rules_nodetect": [0.908496732, 0.908496732, 0.333333333, 0.307189542]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "933dc880323b2677ef6a92d0554bc9c2fcc5a8a9a5686843f0208c415a07aca8", "primary_control": "facts", "replicated": false, "replication_of": "E141-ow_model-0f4f8c44f978", "result_hash": "5ac7ebda927b4a55d4176263b2a9710f33f5f1d984755ec9fa1606feb9acbef3", "signature": "574d178c094e4f6da4e54de8c032c53c735a507dca6f54ced06d5d78875818ea0da87d7f774671d0f141cdbf760fd2800660188c28fa0af10a2e0732229df60f", "state": "VERIFIED", "summary": {"facts": 0.453944, "frozen": 0.23214, "full": 0.489398, "retrieval": 0.48765, "rules_nodetect": 0.614379}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:31:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:31Z", "cpu_seconds": 0.8983214430000004, "created_at": "2026-11-22T04:22:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9f70a5f68183", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "a577aaf63218ff3ae9967b3c2af6801ab77c34c0c3326917e7f1bcf427c3146b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9f70a5f68183", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:07Z", "seed": 197634839, "treatment": "full"}, "priority": 0.19558459986545312, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "95c93e2729b7c290e4aa6d05385f368d9469b96d52eab67a52ae3d1cca174586", "run_seconds": 0.989598274230957, "seed": 197634839, "started_at": "2026-11-22T04:22:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.018225, 0.06987], "effect": 0.044047, "evaluated_at": "2026-11-22T04:31:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-9f70a5f68183", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.552941176, 0.755555556, 0.709090909, 0.547540984], "frozen": [0.035294118, 0.155555556, 0.109090909, 0.108196721], "full": [0.629411765, 0.777777778, 0.763636364, 0.570491803], "retrieval": [0.582352941, 0.822222222, 0.709090909, 0.547540984], "rules_nodetect": [0.629411765, 0.777777778, 0.763636364, 0.570491803]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a577aaf63218ff3ae9967b3c2af6801ab77c34c0c3326917e7f1bcf427c3146b", "primary_control": "facts", "result_hash": "95c93e2729b7c290e4aa6d05385f368d9469b96d52eab67a52ae3d1cca174586", "signature": "65aa5061f62d7348a245ec81896fa4388d83b30ac097ddbb81b134e616845eef4d57556bade411c0d84ac56ae2251b9960febca048153eac8d924febe6916006", "state": "VERIFIED", "summary": {"facts": 0.641282, "frozen": 0.102034, "full": 0.685329, "retrieval": 0.665302, "rules_nodetect": 0.685329}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:31:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:26Z", "cpu_seconds": 0.5832890110000051, "created_at": "2026-11-22T04:22:24Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-089583c31fc5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "1ebba7cefb20fafc861b1712e9e865d7b6b8f2522c4280eac873724396acb913", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-089583c31fc5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:24Z", "seed": 628745220, "treatment": "full"}, "priority": 0.200597338993319, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5ce48c69e9025624ce3dd109b6845a7684a253f55f3425e05e6ef0e9c99db4d0", "run_seconds": 0.6013574600219727, "seed": 628745220, "started_at": "2026-11-22T04:22:25Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:31:40Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-089583c31fc5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.25, 1.0, 0.5], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 0.25, 1.0, 0.5], "no_world_model": [1.0, 0.0, 1.0, 0.0], "random": [1.0, 0.25, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1ebba7cefb20fafc861b1712e9e865d7b6b8f2522c4280eac873724396acb913", "primary_control": "e141_best", "result_hash": "5ce48c69e9025624ce3dd109b6845a7684a253f55f3425e05e6ef0e9c99db4d0", "signature": "043e75e9ec36f3e9dedb240156d6cee90a1362715a587b7e7221a057a25f8e56ebc9e9643d33a5a3c24a6c08596781789c8bd011d982f2baf7b050879f27aa04", "state": "PARTIAL", "summary": {"e141_best": 0.6875, "fixed_policy": 0.0625, "full": 0.6875, "no_world_model": 0.5, "random": 0.6875}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:31:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:25Z", "cpu_seconds": 0.8332945690000031, "created_at": "2026-11-22T04:22:24Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8508269275ae", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "bae27dae7702f01f676f0e1f103ef87b6056e78b9ff5d6cdf46e1a2aa80ec2f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8508269275ae", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:24Z", "seed": 1532215827, "treatment": "full"}, "priority": 0.20747971518117053, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ab6f20ca5f5f689babf6d5b8837bae9c10831e24902cfe6f8178bb2eda7fe6c7", "run_seconds": 0.8359363079071045, "seed": 1532215827, "started_at": "2026-11-22T04:22:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.266451, 0.016451], "effect": -0.125, "evaluated_at": "2026-11-22T04:31:38Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8508269275ae", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 1.0, 0.75], "fixed_policy": [0.25, 1.0, 0.0, 0.5], "full": [1.0, 0.5, 0.75, 0.75], "no_world_model": [0.5, 0.5, 1.0, 0.75], "random": [1.0, 0.5, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "bae27dae7702f01f676f0e1f103ef87b6056e78b9ff5d6cdf46e1a2aa80ec2f4", "primary_control": "e141_best", "result_hash": "ab6f20ca5f5f689babf6d5b8837bae9c10831e24902cfe6f8178bb2eda7fe6c7", "signature": "4c160a89247f9e61d6e60f08822675a2751a9adf06c30fc48da308c06652746d1490e5f0037d6d4b50add349b405e4ac74f77dc2c4cd6bc948f94c1583148902", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.4375, "full": 0.75, "no_world_model": 0.6875, "random": 0.8125}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:31:40Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:43Z", "cpu_seconds": 0.20240029400000026, "created_at": "2026-11-22T04:22:24Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ef31bb367d57", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "8cf158eca7f6de08fe4450d4158ceb493bcc60ebc238a91638442774d3c90937", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ef31bb367d57", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:24Z", "seed": 891690486, "treatment": "full"}, "priority": 0.19334829872219142, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9f749bacfbd7bfc5aebaf2c85a2c3a1fdc072d55d7a25706b41bf9a69900eeda", "run_seconds": 0.22705435752868652, "seed": 891690486, "started_at": "2026-11-22T04:22:42Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:31:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ef31bb367d57", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.0, 0.25, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8cf158eca7f6de08fe4450d4158ceb493bcc60ebc238a91638442774d3c90937", "primary_control": "e141_best", "result_hash": "9f749bacfbd7bfc5aebaf2c85a2c3a1fdc072d55d7a25706b41bf9a69900eeda", "signature": "69cff46783e04d1b8d2a1ef120441cba51106cf3afad699fa7c00e61146721c064f05fb7cf70dfc582811b94ae69629110bdb3c594a6b0e73ee45488ee3cfb0e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:31:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:37Z", "cpu_seconds": 1.1213481739999906, "created_at": "2026-11-22T04:22:24Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c85429dfa6fb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "cdc1280a9f1b5527708830b7957dd02953001c726c7ec7286b5843d3d1420352", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c85429dfa6fb", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:24Z", "seed": 1195658608, "treatment": "full"}, "priority": 0.19457104760875166, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f3c3ff73bf028750613188cd60270bd793ec3ecebcdbbbe515bb88b986b5b706", "run_seconds": 1.1711468696594238, "seed": 1195658608, "started_at": "2026-11-22T04:22:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.020571, 0.063429], "effect": 0.021429, "evaluated_at": "2026-11-22T04:31:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c85429dfa6fb", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.825, 0.957894737, 1.0, 0.771428571], "frozen": [0.475, 0.378947368, 0.452631579, 0.485714286], "full": [0.825, 0.957894737, 1.0, 0.857142857], "retrieval": [0.8, 0.978947368, 1.0, 0.885714286], "rules_nodetect": [0.45, 0.515789474, 1.0, 0.8]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "cdc1280a9f1b5527708830b7957dd02953001c726c7ec7286b5843d3d1420352", "primary_control": "facts", "result_hash": "f3c3ff73bf028750613188cd60270bd793ec3ecebcdbbbe515bb88b986b5b706", "signature": "71d95fd5abae91a80b043af3f8ba77bd3397f4c7a5b6f97b9412981f68381eac6bd795322a0a7292a1651a012d2f4ac27c5f170214929aeefb94e04857c5d10b", "state": "VERIFIED", "summary": {"facts": 0.888581, "frozen": 0.448073, "full": 0.910009, "retrieval": 0.916165, "rules_nodetect": 0.691447}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:31:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:59Z", "cpu_seconds": 5.9479081140000005, "created_at": "2026-11-22T04:22:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4da7e201d5cb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "0abb0320238feaee712d1532c08d45646724406f8c97425a7e5acb08365e024d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4da7e201d5cb", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:22:41Z", "seed": 689687448, "treatment": "frontier"}, "priority": 0.1918454771706984, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "89780fcc126eeb6fb241e887450d37b6356b94843ab6ec04824f460c0cba556a", "run_seconds": 6.111531019210815, "seed": 689687448, "started_at": "2026-11-22T04:22:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003333, 0.010278], "effect": 0.003472, "evaluated_at": "2026-11-22T04:31:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-4da7e201d5cb", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 0.991071429, 1.0], "goal": [1.0, 1.0, 0.991071429, 0.993055556], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 0.991071429, 1.0], "prederr": [1.0, 1.0, 0.982142857, 1.0], "random": [1.0, 1.0, 0.991071429, 0.986111111], "risk": [1.0, 1.0, 0.991071429, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "0abb0320238feaee712d1532c08d45646724406f8c97425a7e5acb08365e024d", "primary_control": "random", "result_hash": "89780fcc126eeb6fb241e887450d37b6356b94843ab6ec04824f460c0cba556a", "signature": "c7e01331a5a2b1eeee002ac15cee9634c24621aa7e1d0fdc05d7a0b8fc47deebd4159c57ba4a62bfc974e5f0a70956dba3913a9360c4bb1367864889f68ecb0f", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.997768, "goal": 0.996032, "hypdisc": 1.0, "infogain": 0.997768, "prederr": 0.995536, "random": 0.994296, "risk": 0.997768}, "ties": 3, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:31:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:55Z", "cpu_seconds": 0.2450592269999987, "created_at": "2026-11-22T04:22:41Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-63b0041aff54", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "47c0e6d7c9b8d8a27cebbdb8f8d31cca77709d7ccd55a5d2a61535c6db22c692", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-63b0041aff54", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:22:41Z", "seed": 1656602323, "treatment": "frontier"}, "priority": 0.19140067631198954, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "359acb423d7679c11b4161c7deb1dfdb957a201a90b82ec643a5351c7e503426", "run_seconds": 0.2529604434967041, "seed": 1656602323, "started_at": "2026-11-22T04:22:55Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:31:42Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-63b0041aff54", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "47c0e6d7c9b8d8a27cebbdb8f8d31cca77709d7ccd55a5d2a61535c6db22c692", "primary_control": "random", "result_hash": "359acb423d7679c11b4161c7deb1dfdb957a201a90b82ec643a5351c7e503426", "signature": "29a468dab79d4b37d4fab622bb975fce8df889796490c49dc358a0f8d5dbfc8c4d2fe02e84bf47a97b3260aea7387ace1dfadda062edd32e6022cc9ca7ae6a0e", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:31:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:22:50Z", "cpu_seconds": 1.6355707780000017, "created_at": "2026-11-22T04:22:41Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-7b4c02f41ab2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "f12a486a01e79f123679a68669ffb4720f8239d9f26252b42030ba81b890d982", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-7b4c02f41ab2", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:41Z", "seed": 1733179630, "treatment": "full"}, "priority": 0.1920191699483141, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1b46c03538e58bc6ed6ab2e1505fb6f80af40643f56d482f25c917f425e32063", "run_seconds": 1.7632458209991455, "seed": 1733179630, "started_at": "2026-11-22T04:22:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.191294, 0.441294], "effect": 0.125, "evaluated_at": "2026-11-22T04:31:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-7b4c02f41ab2", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 0.5, 1.0], "fixed_policy": [0.5, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 0.75, 0.75], "no_world_model": [1.0, 0.25, 0.5, 0.75], "random": [1.0, 1.0, 0.5, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "f12a486a01e79f123679a68669ffb4720f8239d9f26252b42030ba81b890d982", "primary_control": "e141_best", "result_hash": "1b46c03538e58bc6ed6ab2e1505fb6f80af40643f56d482f25c917f425e32063", "signature": "0269e9bbd25bd836436872e9506cbbbbdde7ec022fb3bae809f1c96a920fe6df08e3ec97857e3bf200c47abd1a72b9e67ce349634aa5d0997e1bd7182ba84b05", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.875, "full": 0.875, "no_world_model": 0.625, "random": 0.8125}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:31:42Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:02Z", "cpu_seconds": 2.2366320129999906, "created_at": "2026-11-22T04:22:41Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-85366efa1312", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "b8028ae5f35d6a676363fa2aea6e6a2dbe6bc4cdfb6bc130256557f293d5d43d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-85366efa1312", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:41Z", "seed": 601072803, "treatment": "full"}, "priority": 0.1913668252124343, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8b1c27b2a40d70fa7f9aca714682c69d88e8120700b5142c194cf16710352835", "run_seconds": 2.248847723007202, "seed": 601072803, "started_at": "2026-11-22T04:24:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.049958, 0.450042], "effect": 0.25, "evaluated_at": "2026-11-22T04:34:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-85366efa1312", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 1.0, 0.5], "fixed_policy": [0.5, 0.5, 1.0, 1.0], "full": [0.5, 1.0, 1.0, 0.75], "no_world_model": [0.5, 0.5, 0.75, 0.5], "random": [0.5, 0.75, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b8028ae5f35d6a676363fa2aea6e6a2dbe6bc4cdfb6bc130256557f293d5d43d", "primary_control": "e141_best", "result_hash": "8b1c27b2a40d70fa7f9aca714682c69d88e8120700b5142c194cf16710352835", "signature": "82d1769a706d6c214c81f89a750b6a87cf7fc18ff8a24e5699b41381220d6d014115857d22863bf554ae2df4f24d917690cb0fe06f8f2667fde9b935c2330e02", "state": "VERIFIED", "summary": {"e141_best": 0.5625, "fixed_policy": 0.75, "full": 0.8125, "no_world_model": 0.5625, "random": 0.8125}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:34:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:22Z", "cpu_seconds": 2.0601624699999945, "created_at": "2026-11-22T04:22:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5786d1bc0486", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "53537bbb349777c62bf2a143dc829f4c25857c2ea6d3967d5d66836d28e87521", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5786d1bc0486", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:41Z", "seed": 584050327, "treatment": "full"}, "priority": 0.18979653022061557, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5258d180d9b7230b188837b6ed05c0e07473bdfa9dc4335d3299425a9d0a44f1", "run_seconds": 2.0852272510528564, "seed": 584050327, "started_at": "2026-11-22T04:24:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.036493, 0.034255], "effect": -0.001119, "evaluated_at": "2026-11-22T04:34:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5786d1bc0486", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.234042553, 0.431684335, 0.191489362, 0.483471074], "frozen": [0.156028369, 0.15606596, 0.156028369, 0.067493113], "full": [0.221631206, 0.399293286, 0.180851064, 0.534435262], "retrieval": [0.29964539, 0.444640754, 0.163120567, 0.550964187], "rules_nodetect": [0.221631206, 0.447585395, 0.180851064, 0.534435262]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "53537bbb349777c62bf2a143dc829f4c25857c2ea6d3967d5d66836d28e87521", "primary_control": "facts", "result_hash": "5258d180d9b7230b188837b6ed05c0e07473bdfa9dc4335d3299425a9d0a44f1", "signature": "ba1fb15be387903c21479c1841b3416585c1818deac4023209e260711238cf3d79c9fef570d1f553d0ee9f0bc3d7f8ba4cf6f8ff1ba5577b10c219d05db0f90c", "state": "VERIFIED", "summary": {"facts": 0.335172, "frozen": 0.133904, "full": 0.334053, "retrieval": 0.364593, "rules_nodetect": 0.346126}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:34:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:06Z", "cpu_seconds": 0.39251711, "created_at": "2026-11-22T04:22:41Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-bf8bc67daa00", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "31c919ef8499e509ebfea029436f598a6a022079669f98dc92c7f0484ad6c14a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bf8bc67daa00", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:41Z", "seed": 816682100, "treatment": "full"}, "priority": 0.19047851003644556, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "1c286a96f9f446f2647697b5eb22591f4ba3545891a42397ac8b9434e7da0739", "run_seconds": 0.4037313461303711, "seed": 816682100, "started_at": "2026-11-22T04:24:05Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:34:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-bf8bc67daa00", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.97826087, 0.994565217, 0.956521739, 1.0], "frozen": [0.016304348, 0.016304348, 0.016304348, 0.016304348], "full": [0.97826087, 0.994565217, 0.956521739, 1.0], "retrieval": [0.972826087, 0.994565217, 0.961956522, 1.0], "rules_nodetect": [0.97826087, 0.994565217, 0.956521739, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "31c919ef8499e509ebfea029436f598a6a022079669f98dc92c7f0484ad6c14a", "primary_control": "facts", "result_hash": "1c286a96f9f446f2647697b5eb22591f4ba3545891a42397ac8b9434e7da0739", "signature": "6082c28fbfa9a52f2fd5a4cd29975e6b54257944a7f1e78894b27ed8f07cd9cb22c9c011399c8d7754c3f284538f90dee0af618bc905fc75d949e70e7abb470a", "state": "PARTIAL", "summary": {"facts": 0.982337, "frozen": 0.016304, "full": 0.982337, "retrieval": 0.982337, "rules_nodetect": 0.982337}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:34:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:08Z", "cpu_seconds": 3.6915159139999787, "created_at": "2026-11-22T04:22:58Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-22590930abcf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "f5175cbf127b7b1a4efd5d55764567fe16ef68c690f7636962c4216f7dbe3d54", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-22590930abcf", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:22:58Z", "seed": 1248417883, "treatment": "frontier"}, "priority": 0.19107917019978043, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "21cbc7109d3bb4d942db841dd3813286d4fcb078931be9cc011e4f21dadaaee8", "run_seconds": 3.7903711795806885, "seed": 1248417883, "started_at": "2026-11-22T04:24:05Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.22732, 0.122953], "effect": -0.052183, "evaluated_at": "2026-11-22T04:34:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-22590930abcf", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.5515625, 1.0, 0.354817708, 0.961538462], "frontier": [0.6, 0.8515625, 0.297526042, 0.480769231], "goal": [0.5234375, 0.828125, 0.350911458, 0.355769231], "hypdisc": [0.484375, 0.90625, 0.332682292, 0.615384615], "infogain": [0.6, 0.8515625, 0.315104167, 0.480769231], "prederr": [0.6, 0.8515625, 0.322265625, 0.480769231], "random": [0.490625, 0.859375, 0.300130208, 0.788461538], "risk": [0.6, 0.8515625, 0.297526042, 0.480769231]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "f5175cbf127b7b1a4efd5d55764567fe16ef68c690f7636962c4216f7dbe3d54", "primary_control": "random", "result_hash": "21cbc7109d3bb4d942db841dd3813286d4fcb078931be9cc011e4f21dadaaee8", "signature": "1cff27a82ce20593e045098c6e2dd5b059a3ea6bad177cca7300de2d2ea66228f55abe7c06632784ebe2fe8230d01b21ab9cacba6f36885af440fe5ee042c106", "state": "VERIFIED", "summary": {"count": 0.71698, "frontier": 0.557464, "goal": 0.514561, "hypdisc": 0.584673, "infogain": 0.561859, "prederr": 0.563649, "random": 0.609648, "risk": 0.557464}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:34:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:05Z", "cpu_seconds": 0.9623399689999985, "created_at": "2026-11-22T04:22:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2fc708d366bc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "4da093a152b91a95ed5878d7d1b7fe74da13a1cfdbd37567fd30f31c4ac484cd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2fc708d366bc", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:58Z", "seed": 1970484936, "treatment": "full"}, "priority": 0.20313849634951062, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6e3598155cc6298e11adf45714e4cdc901d87763b05c380236b576fc28969252", "run_seconds": 1.1142597198486328, "seed": 1970484936, "started_at": "2026-11-22T04:23:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:31:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2fc708d366bc", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 0.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "4da093a152b91a95ed5878d7d1b7fe74da13a1cfdbd37567fd30f31c4ac484cd", "primary_control": "e141_best", "result_hash": "6e3598155cc6298e11adf45714e4cdc901d87763b05c380236b576fc28969252", "signature": "3c78cb33df238ea03ae0b7cb7a695a42838084e2d0f0e66bf8c1dac76ba2461f63576f135c50d77837f5517ac4ef07c526ac3cf0fdb708075e150242ba248106", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 1.0, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:31:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:55Z", "cpu_seconds": 0.27783348500000216, "created_at": "2026-11-22T04:22:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a74fdde7ae7c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "c53e918c087365c50385e2c4225e9b69afbb2f210dfae0050e6302745b14acde", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a74fdde7ae7c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:22:58Z", "seed": 679353268, "treatment": "full"}, "priority": 0.1935707627507637, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9b4c5b23ea347d809a797db74aa420663722a02f31ed211086e7d7f3a3af6bac", "run_seconds": 0.29055285453796387, "seed": 679353268, "started_at": "2026-11-22T04:23:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.13, 0.62], "effect": 0.375, "evaluated_at": "2026-11-22T04:33:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a74fdde7ae7c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.0, 1.0, 0.5], "fixed_policy": [0.75, 1.0, 1.0, 1.0], "full": [0.75, 0.5, 1.0, 1.0], "no_world_model": [0.75, 0.5, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "c53e918c087365c50385e2c4225e9b69afbb2f210dfae0050e6302745b14acde", "primary_control": "e141_best", "result_hash": "9b4c5b23ea347d809a797db74aa420663722a02f31ed211086e7d7f3a3af6bac", "signature": "f5e6dceefc6789c3e92e8c3ecb67e634d1714cebe26daf748afe75f6958061cf19b4937ecf34147fefc8d67d214b023b782312fd51e428356e40c01dcd6cc409", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.9375, "full": 0.8125, "no_world_model": 0.8125, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:33:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:00Z", "cpu_seconds": 0.8365538360000073, "created_at": "2026-11-22T04:22:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-193636b11d46", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "dab93bca092ef0bcffb9e6704220a066ec2c155149edbc02637c3c6c080eb567", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-193636b11d46", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:58Z", "seed": 1519198205, "treatment": "full"}, "priority": 0.19164129389287354, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-1ab4ec458c3e", "result_hash": "f70cd71b2141de69595a5d0de29ec73d97cc4c6adbb4df39c53da50f27745bc3", "run_seconds": 0.8430678844451904, "seed": 1519198205, "started_at": "2026-11-22T04:23:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01, 0.030833], "effect": 0.010417, "evaluated_at": "2026-11-22T04:33:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-193636b11d46", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.871794872, 0.725490196, 0.847222222, 0.833333333], "frozen": [0.269230769, 0.323529412, 0.243055556, 0.229166667], "full": [0.871794872, 0.725490196, 0.847222222, 0.875], "retrieval": [0.871794872, 0.803921569, 0.847222222, 0.75], "rules_nodetect": [0.871794872, 0.764705882, 0.847222222, 0.875]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "dab93bca092ef0bcffb9e6704220a066ec2c155149edbc02637c3c6c080eb567", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-1ab4ec458c3e", "result_hash": "f70cd71b2141de69595a5d0de29ec73d97cc4c6adbb4df39c53da50f27745bc3", "signature": "8a879d4a5856e7366e9a7c17f5ef097acc11d3e7d38dea1de799c5e2cd0970e69fe33c1505a590803416413c9fe1ec730114854f00864224aa1d688766294907", "state": "VERIFIED", "summary": {"facts": 0.81946, "frozen": 0.266246, "full": 0.829877, "retrieval": 0.818235, "rules_nodetect": 0.839681}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:33:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:11Z", "cpu_seconds": 0.6365695679999988, "created_at": "2026-11-22T04:22:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-3963e710c7f4", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "033a95b6be2d97a1c8ee76356928f79504660425954c51f34755bab77570950b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-3963e710c7f4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:58Z", "seed": 272905041, "treatment": "full"}, "priority": 0.1945203830450175, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-3c9a93570695", "result_hash": "2cd801aaaec4c4bf7d9de8c088b9a19c29bc430ef71d3befb6dfde3ac14f282d", "run_seconds": 0.9887361526489258, "seed": 272905041, "started_at": "2026-11-22T04:23:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.014337, 0.029186], "effect": 0.021762, "evaluated_at": "2026-11-22T04:31:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-3963e710c7f4", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.68627451, 0.85620915, 0.816993464, 0.765151515], "frozen": [0.091503268, 0.091503268, 0.091503268, 0.098484848], "full": [0.718954248, 0.875816993, 0.836601307, 0.78030303], "retrieval": [0.869281046, 0.882352941, 0.908496732, 0.818181818], "rules_nodetect": [0.718954248, 0.875816993, 0.836601307, 0.78030303]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "033a95b6be2d97a1c8ee76356928f79504660425954c51f34755bab77570950b", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-3c9a93570695", "result_hash": "2cd801aaaec4c4bf7d9de8c088b9a19c29bc430ef71d3befb6dfde3ac14f282d", "signature": "360d2420da6c4162643d4c1f1f66fbb2c0009a45d7d9dd7a37c423fedf4ef5967192271978f71d909e6e86be1b1364ed97a611c758f774dbf9c4b8af217eb60d", "state": "VERIFIED", "summary": {"facts": 0.781157, "frozen": 0.093249, "full": 0.802919, "retrieval": 0.869578, "rules_nodetect": 0.802919}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:31:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:22Z", "cpu_seconds": 0.13058184399997685, "created_at": "2026-11-22T04:22:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7ad5cadef91a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "66d29fd5152e9074c985aff3771f281359ce1a808a23ab2a218abdc4558ffc89", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7ad5cadef91a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:22:58Z", "seed": 1254190304, "treatment": "full"}, "priority": 0.18881407280835732, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "88a21f0a2e9ca03ea1c0c064933811d42fe9d6ebfd850c4f7acef44495fc428f", "run_seconds": 0.13164901733398438, "seed": 1254190304, "started_at": "2026-11-22T04:24:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.01861, 0.148056], "effect": 0.083333, "evaluated_at": "2026-11-22T04:34:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7ad5cadef91a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.666666667, 0.681818182, 0.696969697, 0.560606061], "frozen": [0.015151515, 0.03030303, 0.03030303, 0.03030303], "full": [0.803030303, 0.742424242, 0.696969697, 0.696969697], "retrieval": [0.787878788, 0.712121212, 0.681818182, 0.666666667], "rules_nodetect": [0.363636364, 0.393939394, 0.409090909, 0.606060606]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "66d29fd5152e9074c985aff3771f281359ce1a808a23ab2a218abdc4558ffc89", "primary_control": "facts", "result_hash": "88a21f0a2e9ca03ea1c0c064933811d42fe9d6ebfd850c4f7acef44495fc428f", "signature": "06d01707d44a4b6b3bd4778819c73ea0ccaad5508087650b1e6e524b3026eb0e4e1b2b7278b26b428566e75d37ddc3ef7b3673e087e7cf705266d48601c5d90b", "state": "VERIFIED", "summary": {"facts": 0.651515, "frozen": 0.026515, "full": 0.734848, "retrieval": 0.712121, "rules_nodetect": 0.443182}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:34:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:00Z", "cpu_seconds": 51.84153263200001, "created_at": "2026-11-22T04:22:58Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-f703f157d1d1", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "3f64fcd4f376390819f0649cfdbd123158bf75aa3f8ecd788211447633cf4ba5", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-f703f157d1d1", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:22:58Z", "seed": 203429722, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "5e610a340bf89538a492c2ca5a38b343dc0fd238fe778fd6ba667341a2b3b9b7", "run_seconds": 59.28906011581421, "seed": 203429722, "started_at": "2026-11-22T04:23:00Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 2, "inherit_mem": 2, "inherit_nomem": 3, "random_search": 1, "reset_mem": 2, "retrieval": 0, "rootcause": 2, "xfer_mem": 2}, "better": "higher", "ci95": [0.047563, 0.240253], "effect": 0.143908, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:33:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-f703f157d1d1", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 0.2}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"count": 0.8, "frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 1.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.7}, "random_search": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 1.015, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.97}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"count": 0.6, "frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 0.2}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.65, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"hypdisc": 0.6, "prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": true, "instance_scores": {"G0": [0.35326087, 1.0, 0.96875, 0.677325581, 0.347488038, 0.892361111, 0.909090909, 0.23271028, 0.770833333, 0.728282828], "compressed": [0.97826087, 1.0, 0.9921875, 0.941860465, 0.50645933, 0.684027778, 1.0, 0.606074766, 0.520833333, 0.86969697], "inherit_mem": [0.788043478, 1.0, 0.984375, 0.914728682, 0.617105263, 0.899305556, 0.977272727, 0.53317757, 0.770833333, 0.834343434], "inherit_nomem": [0.972826087, 1.0, 0.9921875, 0.945736434, 0.86937799, 0.798611111, 0.984848485, 0.809345794, 0.520833333, 0.845454545], "random_search": [0.956521739, 1.0, 0.96875, 0.934108527, 0.623803828, 0.871527778, 0.954545455, 0.425, 0.520833333, 0.690151515], "reset_mem": [0.945652174, 1.0, 1.0, 0.895348837, 0.494976077, 0.902777778, 0.992424242, 0.660046729, 0.770833333, 0.852525253], "retrieval": [0.35326087, 1.0, 0.96875, 0.677325581, 0.347488038, 0.892361111, 0.909090909, 0.23271028, 0.770833333, 0.728282828], "rootcause": [0.97826087, 1.0, 0.9921875, 0.941860465, 0.50645933, 0.684027778, 1.0, 0.606074766, 0.520833333, 0.86969697], "xfer_mem": [0.967391304, 1.0, 0.9921875, 0.903100775, 0.620454545, 0.84375, 0.939393939, 0.643224299, 0.520833333, 0.783838384]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -2.4082, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "SUPPORTED", "p_sign_display": "0.00391", "per_generation_sealed": [0.68801, 0.761766, 0.831919, 0.831919], "prereg_hash": "3f64fcd4f376390819f0649cfdbd123158bf75aa3f8ecd788211447633cf4ba5", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "inherit_nomem", "reset_mem"], "result_hash": "5e610a340bf89538a492c2ca5a38b343dc0fd238fe778fd6ba667341a2b3b9b7", "signature": "2d139d6dcc5185d8678623bf67ab845b733acacbe60e09e148b255780a85dfc0e80433edc874883f8dff7d909b6595606edf560c22289c2ad27ea3601e4d7f0c", "state": "VERIFIED", "summary": {"G0": 0.68801, "compressed": 0.80994, "inherit_mem": 0.831919, "inherit_nomem": 0.873922, "random_search": 0.794524, "reset_mem": 0.851458, "retrieval": 0.68801, "rootcause": 0.80994, "xfer_mem": 0.821417}, "ties": 2, "treatment": "inherit_mem", "wins": 8}, "verified_at": "2026-11-22T04:34:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:31Z", "cpu_seconds": 5.631399090000002, "created_at": "2026-11-22T04:23:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4e069f20157d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 2, "n": 4}, "prereg_hash": "d3105af18fcb4955762dee36b328e531d6c2537a73a33cf28a35530d78a56d57", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4e069f20157d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:23:15Z", "seed": 149011455, "treatment": "full"}, "priority": 0.2072910957675834, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6373d10f5b7d5926a14fbff272455369f6c6686d58e98c3c6cbd51c4faa1d132", "run_seconds": 6.029330730438232, "seed": 149011455, "started_at": "2026-11-22T04:23:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.151278, 0.776278], "effect": 0.3125, "evaluated_at": "2026-11-22T04:31:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4e069f20157d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 0.0, 0.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [0.25, 1.0, 0.0, 1.0], "no_world_model": [0.75, 1.0, 0.0, 0.5], "random": [0.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "d3105af18fcb4955762dee36b328e531d6c2537a73a33cf28a35530d78a56d57", "primary_control": "e141_best", "result_hash": "6373d10f5b7d5926a14fbff272455369f6c6686d58e98c3c6cbd51c4faa1d132", "signature": "ed39cec769ea74929dd62881637d85796f814744f8fa094595be8199bd1b4f3cbdd662cd354a2f6dc97508838a53a026a6f1eb308bfbbb756b47467cda44bf0e", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 0.5625, "no_world_model": 0.5625, "random": 0.25}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:33:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:19Z", "cpu_seconds": 2.319200238999997, "created_at": "2026-11-22T04:23:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f1b933233239", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "a3bae79198440192fead38436f9e813835f2ade8f5a73ce36544260d4fe622fb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f1b933233239", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:23:15Z", "seed": 1259347831, "treatment": "full"}, "priority": 0.28680403014427114, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "bb41e1c04c6dbd434807f3394e07d025f12c2566dfa49f0b39588085589d495f", "run_seconds": 2.9719395637512207, "seed": 1259347831, "started_at": "2026-11-22T04:23:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.031168, 0.142959], "effect": 0.087063, "evaluated_at": "2026-11-22T04:31:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f1b933233239", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.893333333, 0.847619048, 0.451388889, 0.751111111], "frozen": [0.5, 0.433333333, 0.319444444, 0.296666667], "full": [0.96, 0.938095238, 0.479166667, 0.914444444], "retrieval": [0.96, 0.966666667, 0.493055556, 0.962222222], "rules_nodetect": [0.96, 0.40952381, 0.479166667, 0.312222222]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a3bae79198440192fead38436f9e813835f2ade8f5a73ce36544260d4fe622fb", "primary_control": "facts", "result_hash": "bb41e1c04c6dbd434807f3394e07d025f12c2566dfa49f0b39588085589d495f", "signature": "65a903c6516d1f706507b456fcebad9702f4a6ead14e35df36a7c9be4f95b0a056ef9124acd316dce6ec07969e0fad25bc1e63192e060c40b49e06e4c7296108", "state": "VERIFIED", "summary": {"facts": 0.735863, "frozen": 0.387361, "full": 0.822927, "retrieval": 0.845486, "rules_nodetect": 0.540228}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:31:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:45Z", "cpu_seconds": 1.0692461160000164, "created_at": "2026-11-22T04:23:32Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-dbd7f765828a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "512e0698c323ba61346ed2866bcaa0c5627e658ddc07b7cbc1b3a242259ce73c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-dbd7f765828a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:23:32Z", "seed": 1726269439, "treatment": "frontier"}, "priority": 0.20302515703059176, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "555c25fe64656b536e2391d3c7c1434c7bb210617833b5ca26d5f4d85210578b", "run_seconds": 1.6705379486083984, "seed": 1726269439, "started_at": "2026-11-22T04:23:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019656, 0.235241], "effect": 0.127448, "evaluated_at": "2026-11-22T04:33:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-dbd7f765828a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.6875, 0.769230769, 0.846153846, 0.925925926], "frontier": [0.8125, 0.807692308, 0.846153846, 0.888888889], "goal": [0.8125, 0.807692308, 0.846153846, 0.851851852], "hypdisc": [0.8125, 0.807692308, 0.846153846, 0.888888889], "infogain": [0.8125, 0.807692308, 0.846153846, 0.814814815], "prederr": [0.458333333, 0.756410256, 0.512820513, 0.481481481], "random": [0.75, 0.679487179, 0.564102564, 0.851851852], "risk": [0.8125, 0.807692308, 0.846153846, 0.888888889]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "512e0698c323ba61346ed2866bcaa0c5627e658ddc07b7cbc1b3a242259ce73c", "primary_control": "random", "result_hash": "555c25fe64656b536e2391d3c7c1434c7bb210617833b5ca26d5f4d85210578b", "signature": "b67104a34b20eefe1ed5af3b8bd341c1200ebe3f8d0c615c2b2af4f32d238a512b9978134f864251233d8775592facdbefa15dd70d446cea4700b9e6bf7f1906", "state": "VERIFIED", "summary": {"count": 0.807203, "frontier": 0.838809, "goal": 0.82955, "hypdisc": 0.838809, "infogain": 0.82029, "prederr": 0.552261, "random": 0.71136, "risk": 0.838809}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:33:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:38Z", "cpu_seconds": 1.7959545630000093, "created_at": "2026-11-22T04:23:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1be777398131", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "5463c0a10fd76dc62e19614d950f3d8d8f348df0a4f69fea46ffecc9423c89db", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1be777398131", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:23:32Z", "seed": 1278828674, "treatment": "full"}, "priority": 0.20733012731027964, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "cebf973bbf3cae117ca472a2f89f6319416301c1140907784186bb73ad4a2bd1", "run_seconds": 2.3228025436401367, "seed": 1278828674, "started_at": "2026-11-22T04:23:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T04:33:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1be777398131", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.75, 0.25], "fixed_policy": [1.0, 0.0, 0.5, 0.25], "full": [1.0, 0.0, 0.5, 0.25], "no_world_model": [1.0, 0.0, 0.25, 0.0], "random": [1.0, 0.0, 0.25, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "5463c0a10fd76dc62e19614d950f3d8d8f348df0a4f69fea46ffecc9423c89db", "primary_control": "e141_best", "result_hash": "cebf973bbf3cae117ca472a2f89f6319416301c1140907784186bb73ad4a2bd1", "signature": "701f24d045909bd24854719783477d04c8358d1d695e0431b6dd5d116cf62c5bf2cef9be0f226cf34f430d1e86a55799f52f975acf67d552a2ef7f0b6a63de00", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.4375, "full": 0.4375, "no_world_model": 0.3125, "random": 0.375}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:33:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:50Z", "cpu_seconds": 0.09743983199999207, "created_at": "2026-11-22T04:23:49Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-33acd8091371", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "a624f2a8170e22b26111777b966a5e4d6c328b22c2865227ec38c1a14819d8b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-33acd8091371", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:23:49Z", "seed": 147539138, "treatment": "full"}, "priority": 0.19723654023105922, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f12839608390029c68d152daa5625da30bb75d6d33bf68b45d07f69fb2bb4d99", "run_seconds": 0.11800670623779297, "seed": 147539138, "started_at": "2026-11-22T04:23:50Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:33:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-33acd8091371", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.25, 0.5, 0.5], "fixed_policy": [0.5, 0.5, 0.5, 0.75], "full": [0.75, 0.25, 0.5, 0.5], "no_world_model": [0.75, 0.25, 0.5, 0.5], "random": [0.75, 0.25, 0.5, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a624f2a8170e22b26111777b966a5e4d6c328b22c2865227ec38c1a14819d8b6", "primary_control": "e141_best", "result_hash": "f12839608390029c68d152daa5625da30bb75d6d33bf68b45d07f69fb2bb4d99", "signature": "6dfd9b71eba14a6526107a966c215715b91d7b20ff69362791f0c5d380e3aa9a595e0a4da7ca801e1d3c95d2fa1962c657202ae267ac0fa1ed6f693de50cd40f", "state": "PARTIAL", "summary": {"e141_best": 0.5, "fixed_policy": 0.5625, "full": 0.5, "no_world_model": 0.5, "random": 0.5}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:33:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:23:54Z", "cpu_seconds": 0.21066734299998302, "created_at": "2026-11-22T04:23:49Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-be848067c98a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "prereg_hash": "49e7ff834d12ce243597fb21de16f7fa62da2f95ce18a192cafb267aceac72b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-be848067c98a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:23:49Z", "seed": 472865422, "treatment": "gated"}, "priority": 0.19416432705064107, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "ff8cdfaad0de4d52fe555915f9554ebc0339ce46333a4dd95d730b06f3d9db71", "run_seconds": 0.2413804531097412, "seed": 472865422, "started_at": "2026-11-22T04:23:54Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:33:54Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-be848067c98a", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.212121212, 0.151515152, 0.318181818, 0.272727273], "gated": [0.212121212, 0.151515152, 0.318181818, 0.272727273], "naive": [0.212121212, 0.090909091, 0.333333333, 0.106060606], "scratch": [0.212121212, 0.151515152, 0.318181818, 0.272727273], "ungated": [0.212121212, 0.090909091, 0.333333333, 0.106060606]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "49e7ff834d12ce243597fb21de16f7fa62da2f95ce18a192cafb267aceac72b6", "primary_control": "scratch", "result_hash": "ff8cdfaad0de4d52fe555915f9554ebc0339ce46333a4dd95d730b06f3d9db71", "signature": "ef13458ca9c94d956d3301ccad9a2424dca72c864fae701992b6a3bc3ae609c7d7d109c7e2890bf36c7656e33271216d574c08c814987ae4826e796f25f25f0f", "state": "PARTIAL", "summary": {"compressed": 0.238636, "gated": 0.238636, "naive": 0.185606, "scratch": 0.238636, "ungated": 0.185606}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:33:54Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:10Z", "cpu_seconds": 2.5426492480000036, "created_at": "2026-11-22T04:24:05Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d27ef5bf7ca0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "45c5a7dc2abcaca81ffc083058c14a12b309c7857826354418a85f0ac8da8226", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d27ef5bf7ca0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:05Z", "seed": 723219566, "treatment": "frontier"}, "priority": 0.19552200652383317, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5f1f4bd79b76d95c668f834d4428ff7e8360100e75029849f85a3a6094cd8dd2", "run_seconds": 2.636491060256958, "seed": 723219566, "started_at": "2026-11-22T04:24:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.009887, 0.013359], "effect": 0.001736, "evaluated_at": "2026-11-22T04:34:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d27ef5bf7ca0", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.986111111, 1.0, 1.0, 1.0], "frontier": [0.986111111, 1.0, 0.996527778, 1.0], "goal": [0.986111111, 1.0, 0.996527778, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [0.986111111, 1.0, 0.996527778, 1.0], "prederr": [0.986111111, 1.0, 0.996527778, 1.0], "random": [1.0, 0.986111111, 0.996527778, 0.993055556], "risk": [0.986111111, 1.0, 0.996527778, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "45c5a7dc2abcaca81ffc083058c14a12b309c7857826354418a85f0ac8da8226", "primary_control": "random", "result_hash": "5f1f4bd79b76d95c668f834d4428ff7e8360100e75029849f85a3a6094cd8dd2", "signature": "ab2604c11e8acefc559080c3e38614bd6052db0287f041f66977e570de0b2093421b35edf40ab0ce04f4d4fe1bd852bae0341ab0d4348ac398b9fbbd91fae102", "state": "VERIFIED", "summary": {"count": 0.996528, "frontier": 0.99566, "goal": 0.99566, "hypdisc": 1.0, "infogain": 0.99566, "prederr": 0.99566, "random": 0.993924, "risk": 0.99566}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:34:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:11Z", "cpu_seconds": 0.4291936109999881, "created_at": "2026-11-22T04:24:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-339847f351b7", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "2126b27e33efb5cf0e5e2f2ab10bfa80e263cf82411ad2a14aef99b53ffa6c6f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-339847f351b7", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:24:05Z", "seed": 1147202058, "treatment": "full"}, "priority": 0.19511791245506158, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-4340ef3eb815", "result_hash": "1ab1ddaa5b2448b7f107b1c4be55ecd3c2e31c2ba213f0465dd4ea0cdb47c5fc", "run_seconds": 0.446727991104126, "seed": 1147202058, "started_at": "2026-11-22T04:24:11Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.56758, 0.44258], "effect": -0.0625, "evaluated_at": "2026-11-22T04:34:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-339847f351b7", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.0, 0.75], "fixed_policy": [0.25, 0.0, 0.25, 0.25], "full": [0.0, 0.0, 0.5, 0.0], "no_world_model": [0.5, 0.0, 0.0, 0.25], "random": [0.0, 0.0, 0.5, 0.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "2126b27e33efb5cf0e5e2f2ab10bfa80e263cf82411ad2a14aef99b53ffa6c6f", "primary_control": "e141_best", "replicated": false, "replication_of": "E141-ow_goal-4340ef3eb815", "result_hash": "1ab1ddaa5b2448b7f107b1c4be55ecd3c2e31c2ba213f0465dd4ea0cdb47c5fc", "signature": "17070038ef8a48ef68a936a6d87f4ad46e665a47d75c70c140f14c97836a42ad76d58dbfd21e0a9ee24856552387c6e1dae3dd168eb6ee1a17f015bf906bda0e", "state": "VERIFIED", "summary": {"e141_best": 0.1875, "fixed_policy": 0.1875, "full": 0.125, "no_world_model": 0.1875, "random": 0.125}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:34:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:17Z", "cpu_seconds": 0.3114648730000056, "created_at": "2026-11-22T04:24:05Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-cb5d977ae48f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "c2e07b3ce538d0d839a4a60dc482f6e770fa91ee06082272f777a59a056b5928", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-cb5d977ae48f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:24:05Z", "seed": 1226830796, "treatment": "full"}, "priority": 0.19368313524768133, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9cbb1d792dd1617621e94c43b48752d44a52099dff7e1c95fb75b241d469dded", "run_seconds": 0.39415884017944336, "seed": 1226830796, "started_at": "2026-11-22T04:24:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T04:34:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-cb5d977ae48f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 0.75], "fixed_policy": [0.0, 0.25, 0.25, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [0.0, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "c2e07b3ce538d0d839a4a60dc482f6e770fa91ee06082272f777a59a056b5928", "primary_control": "e141_best", "result_hash": "9cbb1d792dd1617621e94c43b48752d44a52099dff7e1c95fb75b241d469dded", "signature": "302ecb344c95ce9ba8fbf4c53f77f5578e322ffd0ffc65b9806e5cd21ba64464f3dd052d043f0936cdc4af142d5ab4eb921064affb221e6113d36c5c8e07bb0b", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.1875, "full": 1.0, "no_world_model": 0.9375, "random": 0.6875}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:34:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:14Z", "cpu_seconds": 0.13607075400000213, "created_at": "2026-11-22T04:24:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-457ccb6f0b24", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "45e2abde4fafa9a0a1a05a7c1917c6a2f9ad107c3e68b3d99a0906e9cfec3553", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-457ccb6f0b24", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:05Z", "seed": 749088794, "treatment": "full"}, "priority": 0.1949292543947321, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2bed672a6ac53b4e00165e6d03e33e99938f7f383d7eaf5e07ba09f4c068ee84", "run_seconds": 0.14249491691589355, "seed": 749088794, "started_at": "2026-11-22T04:24:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002727, 0.008409], "effect": 0.002841, "evaluated_at": "2026-11-22T04:34:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-457ccb6f0b24", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.965909091, 0.863636364, 1.0, 1.0], "frozen": [0.022727273, 0.034090909, 0.022727273, 0.011363636], "full": [0.977272727, 0.863636364, 1.0, 1.0], "retrieval": [0.977272727, 0.840909091, 1.0, 1.0], "rules_nodetect": [0.977272727, 0.863636364, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "45e2abde4fafa9a0a1a05a7c1917c6a2f9ad107c3e68b3d99a0906e9cfec3553", "primary_control": "facts", "result_hash": "2bed672a6ac53b4e00165e6d03e33e99938f7f383d7eaf5e07ba09f4c068ee84", "signature": "0054e7d020b61c1b949f5f45624d22fe1dbf9f1bb34ca4449273ebf9597794f626ce6e279ab4925f2efe867f94ae50013f847a8b49dd6b30ea1526ee8e39340b", "state": "VERIFIED", "summary": {"facts": 0.957386, "frozen": 0.022727, "full": 0.960227, "retrieval": 0.954545, "rules_nodetect": 0.960227}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:34:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:15Z", "cpu_seconds": 0.12747653400001013, "created_at": "2026-11-22T04:24:05Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-755efcb8e439", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "fb8a8e2a88c090ac0a64096be3e6cf5bcf3e87edc7c3542a8ba77676bf164969", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-755efcb8e439", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:05Z", "seed": 1831258026, "treatment": "full"}, "priority": 0.19450524772095296, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9f83c567d17f285dd04ecdbb4aefb862cc4e49bd8ea7622c52beff69a2826aae", "run_seconds": 0.12782955169677734, "seed": 1831258026, "started_at": "2026-11-22T04:24:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.015265, 0.076349], "effect": 0.045807, "evaluated_at": "2026-11-22T04:34:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-755efcb8e439", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.825757576, 0.75, 0.795454545, 0.852713178], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.023255814], "full": [0.893939394, 0.75, 0.848484848, 0.914728682], "retrieval": [0.893939394, 0.833333333, 0.939393939, 0.891472868], "rules_nodetect": [0.893939394, 0.75, 0.848484848, 0.914728682]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "fb8a8e2a88c090ac0a64096be3e6cf5bcf3e87edc7c3542a8ba77676bf164969", "primary_control": "facts", "result_hash": "9f83c567d17f285dd04ecdbb4aefb862cc4e49bd8ea7622c52beff69a2826aae", "signature": "b97acf61429bdaba21cc22b81d1af9e079fe9ec2e83d8b3c86371e2d59d18e542ec0e3661f9f35beab2b7d594f05a43930083e93d5b81a55b870c9b33d56b90e", "state": "VERIFIED", "summary": {"facts": 0.805981, "frozen": 0.028541, "full": 0.851788, "retrieval": 0.889535, "rules_nodetect": 0.851788}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:34:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:19Z", "cpu_seconds": 0.43234225499998047, "created_at": "2026-11-22T04:24:05Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-16572821213e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "prereg_hash": "a17887416f8dc080ac9cb90b13b6f0d7820f1d84bfd94277fb1419eafc6a2a8c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-16572821213e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:24:05Z", "seed": 740517891, "treatment": "gated"}, "priority": 0.1914564000104262, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "0acb48a49a37fe66a58c1d80858fbcb29ba38d446ae967e30a5364b8e4f9df7d", "run_seconds": 0.47394537925720215, "seed": 740517891, "started_at": "2026-11-22T04:24:19Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:34:36Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-16572821213e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.766666667, 0.833333333, 0.766666667, 0.6], "gated": [0.766666667, 0.833333333, 0.766666667, 0.6], "naive": [0.766666667, 0.716666667, 0.666666667, 0.65], "scratch": [0.766666667, 0.833333333, 0.766666667, 0.6], "ungated": [0.766666667, 0.716666667, 0.666666667, 0.65]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "a17887416f8dc080ac9cb90b13b6f0d7820f1d84bfd94277fb1419eafc6a2a8c", "primary_control": "scratch", "result_hash": "0acb48a49a37fe66a58c1d80858fbcb29ba38d446ae967e30a5364b8e4f9df7d", "signature": "f0f58775279e85da3edc9a565f6a68d40bc66f8747d1e4d4ea30753e4ebc9da9ed3cc69fdbd9ff943a5aa12e1ae9fd6435180f7910afe48670619a764ba75701", "state": "PARTIAL", "summary": {"compressed": 0.741667, "gated": 0.741667, "naive": 0.7, "scratch": 0.741667, "ungated": 0.7}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:34:36Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:52Z", "cpu_seconds": 0.27571329799999944, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d89a9fc1c895", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "f0b372570ae32c4b35b228e547f8a00d0ab7fe0ad289e6717bd47a1c82b44e5c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d89a9fc1c895", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:22Z", "seed": 1879046856, "treatment": "frontier"}, "priority": 0.18958427740054945, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5549b3f74dea77db48464767302b4ea432f1ee898380cce98bb2942f418fc44e", "run_seconds": 0.294696569442749, "seed": 1879046856, "started_at": "2026-11-22T04:24:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.018118, 0.215087], "effect": 0.098485, "evaluated_at": "2026-11-22T04:36:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d89a9fc1c895", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.666666667, 0.393939394, 0.454545455, 0.590909091], "frontier": [0.772727273, 0.666666667, 0.878787879, 0.696969697], "goal": [0.515151515, 0.651515152, 0.696969697, 0.742424242], "hypdisc": [0.621212121, 0.545454545, 0.621212121, 0.696969697], "infogain": [0.727272727, 0.772727273, 0.848484848, 0.696969697], "prederr": [0.060606061, 0.772727273, 0.833333333, 0.696969697], "random": [0.696969697, 0.636363636, 0.606060606, 0.681818182], "risk": [0.772727273, 0.666666667, 0.878787879, 0.696969697]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f0b372570ae32c4b35b228e547f8a00d0ab7fe0ad289e6717bd47a1c82b44e5c", "primary_control": "random", "result_hash": "5549b3f74dea77db48464767302b4ea432f1ee898380cce98bb2942f418fc44e", "signature": "88d0982791e03f343f70911e8443d367bf6d36591746c5ef67bcb03b844b632d600ea8ca9173cd8def334f3e8eb8fcf1b8737d69665eb8f787472b96fdefd302", "state": "VERIFIED", "summary": {"count": 0.526515, "frontier": 0.753788, "goal": 0.651515, "hypdisc": 0.621212, "infogain": 0.761364, "prederr": 0.590909, "random": 0.655303, "risk": 0.753788}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:36:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:38Z", "cpu_seconds": 5.007978551000008, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-db709aa9ff8d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 10, "n": 4}, "prereg_hash": "4b4182c9be764ea39ec63ee59edbc498cdc1caf3ca8e613ae3326181b364eb36", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-db709aa9ff8d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:22Z", "seed": 1487402363, "treatment": "frontier"}, "priority": 0.19259889179382952, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "772a7180738cb24cd519a8fa807afa02baae05f49294a21e27b10ce3d706b468", "run_seconds": 5.052960395812988, "seed": 1487402363, "started_at": "2026-11-22T04:24:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.037412, 0.120853], "effect": 0.041721, "evaluated_at": "2026-11-22T04:36:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-db709aa9ff8d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.75, 0.696778711, 0.597470238, 0.601694915], "frontier": [0.752604167, 0.595238095, 0.522321429, 0.574576271], "goal": [0.611979167, 0.572829132, 0.522321429, 0.490677966], "hypdisc": [0.778645833, 0.667366947, 0.514136905, 0.634745763], "infogain": [0.7890625, 0.595238095, 0.524553571, 0.647457627], "prederr": [0.721354167, 0.595238095, 0.09375, 0.403389831], "random": [0.606770833, 0.614145658, 0.547619048, 0.509322034], "risk": [0.752604167, 0.595238095, 0.522321429, 0.574576271]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "4b4182c9be764ea39ec63ee59edbc498cdc1caf3ca8e613ae3326181b364eb36", "primary_control": "random", "result_hash": "772a7180738cb24cd519a8fa807afa02baae05f49294a21e27b10ce3d706b468", "signature": "cd41d0b118ecb6ffd551383490c60cad0ee759d1da0a55551c61e91e29529444da3da8744ae45b4563adbddc698724b2f8853c9c18ca3094554bb6b032d23408", "state": "VERIFIED", "summary": {"count": 0.661486, "frontier": 0.611185, "goal": 0.549452, "hypdisc": 0.648724, "infogain": 0.639078, "prederr": 0.453433, "random": 0.569464, "risk": 0.611185}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:36:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:31Z", "cpu_seconds": 4.5088899579999975, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f6f914437679", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "77e23e0b3e92eebd4a7c9aee88cc4df443decb2e60064580e0f2678d82265e30", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f6f914437679", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:22Z", "seed": 743176408, "treatment": "frontier"}, "priority": 0.1992389237062868, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "59a06eeafe695f414e1d4773b66830d6d09140ebe94576a3a35f2afc99ba3eda", "run_seconds": 4.594331979751587, "seed": 743176408, "started_at": "2026-11-22T04:24:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.034565, 0.140006], "effect": 0.052721, "evaluated_at": "2026-11-22T04:36:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f6f914437679", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.407246377, 0.427536232, 0.508090615, 0.389880952], "frontier": [0.414492754, 0.286956522, 0.608414239, 0.412202381], "goal": [0.363768116, 0.320289855, 0.22815534, 0.401785714], "hypdisc": [0.385507246, 0.334782609, 0.504854369, 0.430059524], "infogain": [0.414492754, 0.286956522, 0.655339806, 0.412202381], "prederr": [0.414492754, 0.286956522, 0.655339806, 0.412202381], "random": [0.433333333, 0.302898551, 0.44012945, 0.334821429], "risk": [0.414492754, 0.286956522, 0.608414239, 0.412202381]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "77e23e0b3e92eebd4a7c9aee88cc4df443decb2e60064580e0f2678d82265e30", "primary_control": "random", "result_hash": "59a06eeafe695f414e1d4773b66830d6d09140ebe94576a3a35f2afc99ba3eda", "signature": "13897b04107a2aa34a35340cb53d8d3679b9f8d6181d268e309a0ad517cc9627e0f5eecf645d8aa7032f55b028a5ac8426cc5ee21375cb3b9958533a4c0e4704", "state": "VERIFIED", "summary": {"count": 0.433189, "frontier": 0.430516, "goal": 0.3285, "hypdisc": 0.413801, "infogain": 0.442248, "prederr": 0.442248, "random": 0.377796, "risk": 0.430516}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:36:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:25Z", "cpu_seconds": 1.2786018920000117, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-874ff2fe86fc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "7305e180512d7fc62c27663e0b7501c1ed57d892a35b57588860867a663e7b62", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-874ff2fe86fc", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:24:22Z", "seed": 1505965773, "treatment": "full"}, "priority": 0.18843954625218584, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "326243e9b09cd9f0aef92d61cc4e677f4f3293a1a0d5888f6e66a1956406b969", "run_seconds": 1.3644969463348389, "seed": 1505965773, "started_at": "2026-11-22T04:25:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.096482, 0.596482], "effect": 0.25, "evaluated_at": "2026-11-22T04:36:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-874ff2fe86fc", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 0.25], "fixed_policy": [0.75, 0.75, 0.75, 1.0], "full": [1.0, 0.75, 1.0, 1.0], "no_world_model": [0.75, 0.25, 1.0, 0.25], "random": [1.0, 0.75, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "7305e180512d7fc62c27663e0b7501c1ed57d892a35b57588860867a663e7b62", "primary_control": "e141_best", "result_hash": "326243e9b09cd9f0aef92d61cc4e677f4f3293a1a0d5888f6e66a1956406b969", "signature": "84a14e396bea3cdc6c7fe8e6187972d41693b17234addeb7743d62ce24797f7d368e4f6267cdc72d1c2ed7dc69fcc848be99f9aef580567f97edb6adde74dd0f", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.8125, "full": 0.9375, "no_world_model": 0.5625, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:36:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:28Z", "cpu_seconds": 3.5998421510000185, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d4f1df6f275b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "e1027bd25ff65591808dae0dc59a3c5575dde72e9efd5fa68295683d5c40d12a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d4f1df6f275b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:24:22Z", "seed": 1284602788, "treatment": "full"}, "priority": 0.3682631736011205, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5934e210ea947f8f0aa83be7e0f2b24962b3086cdbb45025bde9e94fb40f4277", "run_seconds": 3.6231415271759033, "seed": 1284602788, "started_at": "2026-11-22T04:24:24Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.54707, -0.07793], "effect": -0.3125, "evaluated_at": "2026-11-22T04:34:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-d4f1df6f275b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.5, 0.25, 0.5], "fixed_policy": [0.25, 0.0, 0.25, 0.25], "full": [0.0, 0.0, 0.0, 0.0], "no_world_model": [0.0, 0.75, 0.0, 0.5], "random": [0.75, 0.25, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e1027bd25ff65591808dae0dc59a3c5575dde72e9efd5fa68295683d5c40d12a", "primary_control": "e141_best", "result_hash": "5934e210ea947f8f0aa83be7e0f2b24962b3086cdbb45025bde9e94fb40f4277", "signature": "70157fbf42b780cac9ec97ef37a86eca9b9688d7be0a03c7ac4ea66102a05a8a47d29eb639b6ee33b298a5d6ad7f6933f574ffb3b9065b2f912a0caba03fcd03", "state": "VERIFIED", "summary": {"e141_best": 0.3125, "fixed_policy": 0.1875, "full": 0.0, "no_world_model": 0.3125, "random": 0.25}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:34Z", "cpu_seconds": 1.0845241410000028, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-09e59ea4a679", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "a139653151a975dd803f7e1f20047dee649fd044cd6459371de2adf761b7b5b1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-09e59ea4a679", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:22Z", "seed": 1890788867, "treatment": "full"}, "priority": 0.19256533480750598, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c015b2b74c18d82a543206579bd67f4d411cf12e98052d23fffa1a6a3bd539e1", "run_seconds": 1.1204204559326172, "seed": 1890788867, "started_at": "2026-11-22T04:24:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.049675, -0.002013], "effect": -0.025844, "evaluated_at": "2026-11-22T04:36:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-09e59ea4a679", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.487234043, 0.482978723, 0.746808511, 0.230434783], "frozen": [0.127659574, 0.134042553, 0.127659574, 0.134782609], "full": [0.465957447, 0.459574468, 0.746808511, 0.17173913], "retrieval": [0.529787234, 0.49787234, 0.746808511, 0.219565217], "rules_nodetect": [0.465957447, 0.636170213, 0.746808511, 0.17173913]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "a139653151a975dd803f7e1f20047dee649fd044cd6459371de2adf761b7b5b1", "primary_control": "facts", "result_hash": "c015b2b74c18d82a543206579bd67f4d411cf12e98052d23fffa1a6a3bd539e1", "signature": "170cc090bb525f916c0a422a85b5df604c8246bdb04fbc9006bed4c37ac3ee6156b363145e300432ab65fcd471668052fd49aae436b02ba911c08ff97c95ec0c", "state": "VERIFIED", "summary": {"facts": 0.486864, "frozen": 0.131036, "full": 0.46102, "retrieval": 0.498508, "rules_nodetect": 0.505169}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:37Z", "cpu_seconds": 0.25094258299998273, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-97d063676d42", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "f6a38e6673614827883d5e999da28b1b55f6904d696ea09e17e6600543e80b81", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-97d063676d42", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:22Z", "seed": 69794600, "treatment": "full"}, "priority": 0.19246535948154594, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "caf3b57bdf497f3107d76ff6c42c3b2353b86257923665fa1824c017ca728923", "run_seconds": 0.25409865379333496, "seed": 69794600, "started_at": "2026-11-22T04:24:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.182371, 0.313462], "effect": 0.247917, "evaluated_at": "2026-11-22T04:36:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-97d063676d42", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.770833333, 0.583333333, 0.25, 0.541666667], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.9375, 0.8125, 0.575, 0.8125], "retrieval": [0.916666667, 0.833333333, 0.5, 0.895833333], "rules_nodetect": [0.333333333, 0.270833333, 0.375, 0.333333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "f6a38e6673614827883d5e999da28b1b55f6904d696ea09e17e6600543e80b81", "primary_control": "facts", "result_hash": "caf3b57bdf497f3107d76ff6c42c3b2353b86257923665fa1824c017ca728923", "signature": "232c6c7cbd31ceb583a1b2bc1c5efb4b82a0ec36dc84ea7fef9d60db3df03c1473202bb6bbec03c13d25a5098a729e299a1495a4d8a9063790139aed3883c209", "state": "VERIFIED", "summary": {"facts": 0.536458, "frozen": 0.0, "full": 0.784375, "retrieval": 0.786458, "rules_nodetect": 0.328125}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:28Z", "cpu_seconds": 0.11970204799999351, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d78d229c2a14", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "cd694a069dfb450af9818abc92cfbc9a5e38a70a78815261f468dad7ac8685be", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d78d229c2a14", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:22Z", "seed": 22959782, "treatment": "full"}, "priority": 0.19392776083994281, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "57c9b40315838dce66ac304cf0e837e9a2068589621fee2c402644020c268b3e", "run_seconds": 0.12959957122802734, "seed": 22959782, "started_at": "2026-11-22T04:24:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01191, 0.130017], "effect": 0.059054, "evaluated_at": "2026-11-22T04:34:37Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d78d229c2a14", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.238636364, 0.922727273, 0.130434783], "frozen": [1.0, 0.034090909, 0.036363636, 0.026086957], "full": [1.0, 0.392045455, 0.927272727, 0.208695652], "retrieval": [1.0, 0.369318182, 0.927272727, 0.165217391], "rules_nodetect": [1.0, 0.534090909, 0.927272727, 0.426086957]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "cd694a069dfb450af9818abc92cfbc9a5e38a70a78815261f468dad7ac8685be", "primary_control": "facts", "result_hash": "57c9b40315838dce66ac304cf0e837e9a2068589621fee2c402644020c268b3e", "signature": "d2b591a6fc2963da7624f4d03519e61e21a829d7a8e603189e70c105d9f565161cc6cc3beb67344bb37afba2f8a3722a5faf013ee5e615a40be90875ce8ff406", "state": "VERIFIED", "summary": {"facts": 0.57295, "frozen": 0.274135, "full": 0.632003, "retrieval": 0.615452, "rules_nodetect": 0.721863}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:34:37Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:54Z", "cpu_seconds": 0.22723243100000445, "created_at": "2026-11-22T04:24:22Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-76f054446c21", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "prereg_hash": "f00a44cae00e26ce8410c16a63e0b145838a34591791087aa6fc2c79494aec77", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-76f054446c21", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|math_oracle|math_oracle|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:24:22Z", "seed": 102769912, "treatment": "gated"}, "priority": 0.18855781660266216, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "4df7749e1497d235db40afb7a2afbcf8356e7b28a485cb590715308b3654f7df", "run_seconds": 0.22743463516235352, "seed": 102769912, "started_at": "2026-11-22T04:24:54Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-76f054446c21", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.454545455, 0.212121212, 0.378787879, 0.196969697], "gated": [0.454545455, 0.212121212, 0.378787879, 0.196969697], "naive": [0.454545455, 0.212121212, 0.378787879, 0.196969697], "scratch": [0.454545455, 0.212121212, 0.378787879, 0.196969697], "ungated": [0.454545455, 0.212121212, 0.378787879, 0.196969697]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f00a44cae00e26ce8410c16a63e0b145838a34591791087aa6fc2c79494aec77", "primary_control": "scratch", "result_hash": "4df7749e1497d235db40afb7a2afbcf8356e7b28a485cb590715308b3654f7df", "signature": "0e42411afa438c82a58096bbb1fe3289b8cd4f036010a34555955e28a68a494f235c9720a7e3786d5bb0c12b3d9110b02ba6f30af7ff4d3edc3db1b7071be103", "state": "PARTIAL", "summary": {"compressed": 0.310606, "gated": 0.310606, "naive": 0.310606, "scratch": 0.310606, "ungated": 0.310606}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:43Z", "cpu_seconds": 1.050474190000017, "created_at": "2026-11-22T04:24:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-105a89a32da1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "9242db168c63ba180431e6b74a87ba0a52b6ca9bdd11015fa4aa5d8a9624f678", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-105a89a32da1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:39Z", "seed": 1013091325, "treatment": "frontier"}, "priority": 0.3440436543815121, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "781fbab6be4eb2a9ec5c2bba7d713fa23147f6b318f3faa9f1f9c8330d540faf", "run_seconds": 1.1422498226165771, "seed": 1013091325, "started_at": "2026-11-22T04:24:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.085046, 0.280075], "effect": 0.097514, "evaluated_at": "2026-11-22T04:36:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-105a89a32da1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.366013072, 0.627906977, 0.565891473, 0.729166667], "frontier": [0.575163399, 0.674418605, 0.395348837, 0.677083333], "goal": [0.529411765, 0.666666667, 0.395348837, 0.697916667], "hypdisc": [0.535947712, 0.604651163, 0.511627907, 0.729166667], "infogain": [0.614379085, 0.736434109, 0.496124031, 0.677083333], "prederr": [0.568627451, 0.736434109, 0.372093023, 0.604166667], "random": [0.294117647, 0.480620155, 0.542635659, 0.614583333], "risk": [0.575163399, 0.674418605, 0.395348837, 0.677083333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "9242db168c63ba180431e6b74a87ba0a52b6ca9bdd11015fa4aa5d8a9624f678", "primary_control": "random", "result_hash": "781fbab6be4eb2a9ec5c2bba7d713fa23147f6b318f3faa9f1f9c8330d540faf", "signature": "245d970dd9860f7f72a3d2eb9953f2ed04d382f789840bbd399ed84a2408e5f5cd225d247c6ba92de8560bff9b85e4f41b46beaa9439fa02a486828f4ffb1006", "state": "VERIFIED", "summary": {"count": 0.572245, "frontier": 0.580504, "goal": 0.572336, "hypdisc": 0.595348, "infogain": 0.631005, "prederr": 0.57033, "random": 0.482989, "risk": 0.580504}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:36:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:41Z", "cpu_seconds": 1.9373766850000038, "created_at": "2026-11-22T04:24:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1b31f61cfc95", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "2411ccd80c64af448ad7e9b9613b3c4858d16b8c7d279906007c95a208005d39", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1b31f61cfc95", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:39Z", "seed": 1800967542, "treatment": "frontier"}, "priority": 0.38327978115758055, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "aeef746e54e11412292ffe5f27548b158db3f3fe35c435e437bb7ff383d8fbb5", "run_seconds": 1.9744348526000977, "seed": 1800967542, "started_at": "2026-11-22T04:24:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.004097, 0.014811], "effect": 0.005357, "evaluated_at": "2026-11-22T04:36:14Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1b31f61cfc95", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.983333333, 0.942857143, 0.991666667, 0.985714286], "goal": [0.95, 0.764285714, 0.95, 0.9], "hypdisc": [0.991666667, 0.871428571, 0.991666667, 0.985714286], "infogain": [0.983333333, 0.942857143, 0.991666667, 0.985714286], "prederr": [0.983333333, 0.942857143, 0.991666667, 0.985714286], "random": [0.991666667, 0.935714286, 0.983333333, 0.971428571], "risk": [0.983333333, 0.942857143, 0.991666667, 0.985714286]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "2411ccd80c64af448ad7e9b9613b3c4858d16b8c7d279906007c95a208005d39", "primary_control": "random", "result_hash": "aeef746e54e11412292ffe5f27548b158db3f3fe35c435e437bb7ff383d8fbb5", "signature": "fabb2d9100cd7deec9feb059af2920b456c8ebcbbb5d930166b81b619685f30abb9c5c59db065dd0df04f25821d3950d42c4f091c9b777b76616036f053da10e", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.975893, "goal": 0.891071, "hypdisc": 0.960119, "infogain": 0.975893, "prederr": 0.975893, "random": 0.970536, "risk": 0.975893}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:36:14Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:54Z", "cpu_seconds": 3.20396251599999, "created_at": "2026-11-22T04:24:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6486739d8a05", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "ad69d6afb3f45f5590ea789eaef2fdd29a128f3ebd5e35efac082a91f9e72839", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6486739d8a05", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:39Z", "seed": 683726563, "treatment": "frontier"}, "priority": 0.19049196313708486, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "96206975ed51e768aad815581b65f08b08d4a3a2b7fe3d10d1937c0485ff9f18", "run_seconds": 3.2942912578582764, "seed": 683726563, "started_at": "2026-11-22T04:24:50Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6486739d8a05", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ad69d6afb3f45f5590ea789eaef2fdd29a128f3ebd5e35efac082a91f9e72839", "primary_control": "random", "result_hash": "96206975ed51e768aad815581b65f08b08d4a3a2b7fe3d10d1937c0485ff9f18", "signature": "4ddd3275c8e0269f3c76844cba7114c842c3d2a390d2578632ceda093968de328b6934f42be6217917b504427241b59dc34ee0c1c89555590839f610bce79206", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:36:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:45Z", "cpu_seconds": 2.5592202499999814, "created_at": "2026-11-22T04:24:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-88c9ec5db890", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "801ef3b7c3cb0b51af83b8243302262e01c8725c0caa7f3518e9bb94db906ea6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-88c9ec5db890", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:39Z", "seed": 1162721037, "treatment": "frontier"}, "priority": 0.20000091692691935, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "293209c989aa4a57a1763688d1f807bb67b9b012c201146fcafbc64e98d57236", "run_seconds": 2.744436740875244, "seed": 1162721037, "started_at": "2026-11-22T04:24:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001985, 0.12559], "effect": 0.063788, "evaluated_at": "2026-11-22T04:36:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-88c9ec5db890", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.944444444, 0.75, 0.579487179, 0.610526316], "frontier": [0.916666667, 0.716666667, 0.676923077, 0.673684211], "goal": [0.916666667, 0.7, 0.579487179, 0.705263158], "hypdisc": [0.944444444, 0.666666667, 0.58974359, 0.610526316], "infogain": [0.916666667, 0.733333333, 0.620512821, 0.673684211], "prederr": [0.916666667, 0.7, 0.430769231, 0.673684211], "random": [0.944444444, 0.6, 0.594871795, 0.589473684], "risk": [0.916666667, 0.716666667, 0.676923077, 0.673684211]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "801ef3b7c3cb0b51af83b8243302262e01c8725c0caa7f3518e9bb94db906ea6", "primary_control": "random", "result_hash": "293209c989aa4a57a1763688d1f807bb67b9b012c201146fcafbc64e98d57236", "signature": "f83400f18e11f9687bc7af7857495965b596923be50fdd794ca004ae3a5f9a97d6c45ac49ff5bff56e65b8d55db1d537fb95eed9917916b4cce976fb7cb59d09", "state": "VERIFIED", "summary": {"count": 0.721114, "frontier": 0.745985, "goal": 0.725354, "hypdisc": 0.702845, "infogain": 0.736049, "prederr": 0.68028, "random": 0.682197, "risk": 0.745985}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:36:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:47Z", "cpu_seconds": 0.4526056339999798, "created_at": "2026-11-22T04:24:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4d3716566745", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "4952cde535e34fa59772263867dd719f8274acd8e3a1deea2a953690ce18e43d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4d3716566745", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:24:39Z", "seed": 1163235616, "treatment": "full"}, "priority": 0.1919588521080013, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-533c29f8a90c", "result_hash": "8bb56a9643ead02569e8f3f47753cfd44ca5ad48f12f838059f5824518aef716", "run_seconds": 0.4701108932495117, "seed": 1163235616, "started_at": "2026-11-22T04:24:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.805, -0.07], "effect": -0.4375, "evaluated_at": "2026-11-22T04:36:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4d3716566745", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.75, 0.75, 0.75], "fixed_policy": [0.25, 0.0, 0.5, 0.25], "full": [0.5, 0.0, 0.0, 0.5], "no_world_model": [0.0, 0.0, 0.0, 0.75], "random": [0.25, 0.0, 0.0, 0.75]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "4952cde535e34fa59772263867dd719f8274acd8e3a1deea2a953690ce18e43d", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-533c29f8a90c", "result_hash": "8bb56a9643ead02569e8f3f47753cfd44ca5ad48f12f838059f5824518aef716", "signature": "19fd378fa0a2a5aed08ac6a72d7d539d58274894c469ae29fbc3207a7cceeac0ce7ea802a7613e18a3b37690609cf83291b2abf1ade647828e573d455e6ff90e", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.25, "full": 0.25, "no_world_model": 0.1875, "random": 0.25}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:49Z", "cpu_seconds": 1.075815461000019, "created_at": "2026-11-22T04:24:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f32c836dba89", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "f57e72919aa978d6de116e73bc169f4772a7596ae3170a948cfcac139e6faf9e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f32c836dba89", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:39Z", "seed": 1881862730, "treatment": "full"}, "priority": 0.19137403356429517, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f4c885d854a4fb46cc94de2374958cbac40fb799ac33c44668627eea5c871a98", "run_seconds": 1.1938300132751465, "seed": 1881862730, "started_at": "2026-11-22T04:24:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.023134, 0.035885], "effect": 0.006375, "evaluated_at": "2026-11-22T04:36:16Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-f32c836dba89", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.590163934, 0.551724138, 0.822222222, 0.527586207], "frozen": [0.108196721, 0.103448276, 0.155555556, 0.103448276], "full": [0.593442623, 0.517241379, 0.844444444, 0.562068966], "retrieval": [0.613114754, 0.565517241, 0.844444444, 0.517241379], "rules_nodetect": [0.593442623, 0.517241379, 0.844444444, 0.562068966]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "f57e72919aa978d6de116e73bc169f4772a7596ae3170a948cfcac139e6faf9e", "primary_control": "facts", "result_hash": "f4c885d854a4fb46cc94de2374958cbac40fb799ac33c44668627eea5c871a98", "signature": "44e594f9f4ad2d5dd358ec0ecec11f439b9d8ba7fe358f8c1e45bef16331d78f806445222c241e2bf66050a16e48819b02b7a3f5471bd2b35d031fc726c1a009", "state": "VERIFIED", "summary": {"facts": 0.622924, "frozen": 0.117662, "full": 0.629299, "retrieval": 0.635079, "rules_nodetect": 0.629299}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:36:16Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:02Z", "cpu_seconds": 2.132032071999987, "created_at": "2026-11-22T04:24:54Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5984743c1fc9", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "a0c18708753ed1395963f9df506b1359c4ad7285a120b8c93cd51a11e722bfeb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5984743c1fc9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:24:54Z", "seed": 1546394397, "treatment": "frontier"}, "priority": 0.2765067146031379, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-e3763ea9e75a", "result_hash": "0c2e86045f2cf293eaaa31fecc572dc1edbed638e1ecc3bbc87edd4790480d5d", "run_seconds": 2.4948205947875977, "seed": 1546394397, "started_at": "2026-11-22T04:24:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.226101, 0.142095], "effect": -0.042003, "evaluated_at": "2026-11-22T04:36:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5984743c1fc9", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.375, 0.844444444, 0.673684211, 0.531578947], "frontier": [0.3725, 0.422222222, 0.368421053, 0.618421053], "goal": [0.3075, 0.466666667, 0.378947368, 0.223684211], "hypdisc": [0.3125, 0.466666667, 0.473684211, 0.363157895], "infogain": [0.4325, 0.422222222, 0.368421053, 0.181578947], "prederr": [0.3775, 0.422222222, 0.263157895, 0.210526316], "random": [0.3025, 0.644444444, 0.547368421, 0.455263158], "risk": [0.3725, 0.422222222, 0.368421053, 0.618421053]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "a0c18708753ed1395963f9df506b1359c4ad7285a120b8c93cd51a11e722bfeb", "primary_control": "random", "replicated": false, "replication_of": "E141-ow_explore-e3763ea9e75a", "result_hash": "0c2e86045f2cf293eaaa31fecc572dc1edbed638e1ecc3bbc87edd4790480d5d", "signature": "d42ac3d0c96010457be11ec5ad7b8b578e45611abdfffa16d91c7a20db9b1b9990bb95db504956bd4375640a97c4dd7cf6ef9c3f834f22c13646d78bb1176500", "state": "VERIFIED", "summary": {"count": 0.606177, "frontier": 0.445391, "goal": 0.3442, "hypdisc": 0.404002, "infogain": 0.351181, "prederr": 0.318352, "random": 0.487394, "risk": 0.445391}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:36:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:07Z", "cpu_seconds": 0.34555054000000496, "created_at": "2026-11-22T04:24:54Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6a0c33cc0bbd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "b3a3dbb28975ac771476ee0c4570ecee681d27a429cff9a94375de02cc244a60", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6a0c33cc0bbd", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:54Z", "seed": 1479813187, "treatment": "full"}, "priority": 0.19073427268216517, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "720df1acf7fbce4703679d9afc30d055642cdc916ef5b15145f717646c21c3cf", "run_seconds": 0.35125041007995605, "seed": 1479813187, "started_at": "2026-11-22T04:25:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.077531, 0.275594], "effect": 0.176562, "evaluated_at": "2026-11-22T04:36:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-6a0c33cc0bbd", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.9, 0.828125, 0.78125, 0.65625], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [0.95, 1.0, 0.96875, 0.953125], "retrieval": [0.9875, 0.953125, 0.96875, 0.921875], "rules_nodetect": [0.95, 1.0, 0.96875, 0.953125]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "b3a3dbb28975ac771476ee0c4570ecee681d27a429cff9a94375de02cc244a60", "primary_control": "facts", "result_hash": "720df1acf7fbce4703679d9afc30d055642cdc916ef5b15145f717646c21c3cf", "signature": "cf52456b1d71edb6dac736fadcf5e75ff066d0c529792baa485f0142762cd7d53dddc0b18045e1d9e58c1dc520ea11c249893f20dfe0166ffc40ff342ce0ea02", "state": "VERIFIED", "summary": {"facts": 0.791406, "frozen": 0.0, "full": 0.967969, "retrieval": 0.957812, "rules_nodetect": 0.967969}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:07Z", "cpu_seconds": 3.4875631099999964, "created_at": "2026-11-22T04:24:54Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-cc616a34aa39", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "fcb522cde3468eb39a8d21d1ad72e338dc4f7e63d423a6ed39c9bdbf3e658c5d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-cc616a34aa39", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:24:54Z", "seed": 1147371566, "treatment": "full"}, "priority": 0.19980677128157476, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "27039173a34c133032870ff54a8b00ca015a5fd3a083373a42117d3870bc8602", "run_seconds": 3.579265594482422, "seed": 1147371566, "started_at": "2026-11-22T04:25:03Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:36:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-cc616a34aa39", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.784195], "frozen": [1.0, 0.294612], "full": [1.0, 0.870905], "retrieval": [1.0, 0.886566], "rules_nodetect": [1.0, 0.870905]}, "prereg_hash": "fcb522cde3468eb39a8d21d1ad72e338dc4f7e63d423a6ed39c9bdbf3e658c5d", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "27039173a34c133032870ff54a8b00ca015a5fd3a083373a42117d3870bc8602", "signature": "0d9b4b8cdcbcfb93c74fec473106b03943750c48a6542ee59e0294ae6ee343814ff2eff761c2e0476f719546d8d86d3448dda04a859dad4af2408aef1c87c10d", "state": "REJECTED", "summary": {"facts": 0.784195, "frozen": 0.294612, "full": 0.870905, "retrieval": 0.886566, "rules_nodetect": 0.870905}}, "verified_at": "2026-11-22T04:36:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:02Z", "cpu_seconds": 2.4431546790000027, "created_at": "2026-11-22T04:24:54Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-18c7865455ea", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "65a693ebc2e2b58b52d4ff5d4a116846c34f9e51b2ad7f9da12c83ea46143d25", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-18c7865455ea", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:24:54Z", "seed": 1305074622, "treatment": "gated"}, "priority": 0.2807163003396242, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "5567398b38a33afefb87e81fc49ef137a446975c3b477dd6aec2ce7d8a10dc45", "run_seconds": 2.84511661529541, "seed": 1305074622, "started_at": "2026-11-22T04:24:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.01, 0.030833], "effect": 0.010417, "evaluated_at": "2026-11-22T04:36:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-18c7865455ea", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.683333333, 0.625, 0.733870968, 0.569444444], "gated": [0.683333333, 0.625, 0.733870968, 0.569444444], "naive": [0.716666667, 0.625, 0.733870968, 0.611111111], "scratch": [0.683333333, 0.625, 0.733870968, 0.527777778], "ungated": [0.7, 0.625, 0.733870968, 0.611111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "65a693ebc2e2b58b52d4ff5d4a116846c34f9e51b2ad7f9da12c83ea46143d25", "primary_control": "scratch", "result_hash": "5567398b38a33afefb87e81fc49ef137a446975c3b477dd6aec2ce7d8a10dc45", "signature": "13e75128f538597f8a3e39316331679dd11ccf165b7fa7718b7ce0dcbfc45e1501945357b9018b2e5d2d4d7712fc6725a4704e8f375dc0d1523e80883ac50f00", "state": "VERIFIED", "summary": {"compressed": 0.652912, "gated": 0.652912, "naive": 0.671662, "scratch": 0.642496, "ungated": 0.667496}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:36:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:08Z", "cpu_seconds": 1.1972363149999978, "created_at": "2026-11-22T04:24:54Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5aa6d814357b", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "prereg_hash": "2cd0921f322eb89b28d8948ce68f26503121d978bc75b00c3888492b9c5bf0ef", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5aa6d814357b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:24:54Z", "seed": 2101604375, "treatment": "gated"}, "priority": 0.189309615051602, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-6bc75240bf73", "result_hash": "8df0b2766a2d22e0aae55bc28e26a7680e34c3157ea260f101f5519c67d40eaa", "run_seconds": 1.2271549701690674, "seed": 2101604375, "started_at": "2026-11-22T04:25:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.014205, 0.007818], "effect": -0.003194, "evaluated_at": "2026-11-22T04:36:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-5aa6d814357b", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.151648352, 0.087912088, 0.095121951, 0.189361702], "gated": [0.151648352, 0.087912088, 0.095121951, 0.189361702], "naive": [0.164835165, 0.120879121, 0.126829268, 0.125531915], "scratch": [0.151648352, 0.107692308, 0.090243902, 0.187234043], "ungated": [0.182417582, 0.178021978, 0.126829268, 0.172340426]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "2cd0921f322eb89b28d8948ce68f26503121d978bc75b00c3888492b9c5bf0ef", "primary_control": "scratch", "replicated": true, "replication_of": "E141-ow_transfer-6bc75240bf73", "result_hash": "8df0b2766a2d22e0aae55bc28e26a7680e34c3157ea260f101f5519c67d40eaa", "signature": "159d35e5946a90a04e1142a43c8357c43deaf35fd972499de39c0d25a6c93e1a6b74feaf3d876427e72936ce22e251bf8bb41cfbb0b005c353a3d3250c66d109", "state": "VERIFIED", "summary": {"compressed": 0.131011, "gated": 0.131011, "naive": 0.134519, "scratch": 0.134205, "ungated": 0.164902}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:36:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:24:58Z", "cpu_seconds": 1.1867759099999944, "created_at": "2026-11-22T04:24:54Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8838d4ff099e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "c8c40c69963c8f7ff6c521fa27927c83310fc589ddfc718f85fc3a34db3c63c9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8838d4ff099e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:24:54Z", "seed": 1640378553, "treatment": "gated"}, "priority": 0.2991554529993148, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "740fe752c1c7917beff421bdc571ebfd92ed5799e58c0e8d28d25d072aa4b3c0", "run_seconds": 1.2801306247711182, "seed": 1640378553, "started_at": "2026-11-22T04:24:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005967, 0.00037], "effect": -0.002798, "evaluated_at": "2026-11-22T04:36:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-8838d4ff099e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.191428571, 0.167123288, 0.327586207, 0.213157895], "gated": [0.191428571, 0.167123288, 0.327586207, 0.213157895], "naive": [0.297142857, 0.265753425, 0.324137931, 0.213157895], "scratch": [0.197142857, 0.17260274, 0.327586207, 0.213157895], "ungated": [0.168571429, 0.167123288, 0.324137931, 0.234210526]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "c8c40c69963c8f7ff6c521fa27927c83310fc589ddfc718f85fc3a34db3c63c9", "primary_control": "scratch", "result_hash": "740fe752c1c7917beff421bdc571ebfd92ed5799e58c0e8d28d25d072aa4b3c0", "signature": "f6d7eb9d787d5a33b0a18fdbe977c65529db046e665bbe3bbe1b35713d99a46001865ecd415f54e0ac4b027e20b6d2d893551ca282b3a483c9b3137fdaf5ca0e", "state": "VERIFIED", "summary": {"compressed": 0.224824, "gated": 0.224824, "naive": 0.275048, "scratch": 0.227622, "ungated": 0.223511}, "ties": 2, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:23Z", "cpu_seconds": 0.48896070800000757, "created_at": "2026-11-22T04:24:54Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-faf69e466761", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "59a55fc07fde2cf1923b1a1318a7e7c66d10561682f832c1903c74864cd0df71", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-faf69e466761", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:24:54Z", "seed": 1545464990, "treatment": "gated"}, "priority": 0.18929776585640892, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-864e08f7fec4", "result_hash": "2a1f5db90c0c0710facb0feb9ff846092b2b2b5d48f5d9a4e91811098bd9f2e7", "run_seconds": 0.5534343719482422, "seed": 1545464990, "started_at": "2026-11-22T04:25:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.061667, 0.02], "effect": -0.020833, "evaluated_at": "2026-11-22T04:36:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-faf69e466761", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.679487179, 0.654320988, 0.619883041, 0.75], "gated": [0.679487179, 0.654320988, 0.619883041, 0.75], "naive": [0.679487179, 0.654320988, 0.619883041, 0.75], "scratch": [0.679487179, 0.654320988, 0.619883041, 0.833333333], "ungated": [0.564102564, 0.654320988, 0.619883041, 0.75]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "59a55fc07fde2cf1923b1a1318a7e7c66d10561682f832c1903c74864cd0df71", "primary_control": "scratch", "replicated": true, "replication_of": "E141-ow_transfer-864e08f7fec4", "result_hash": "2a1f5db90c0c0710facb0feb9ff846092b2b2b5d48f5d9a4e91811098bd9f2e7", "signature": "656c8a286c7c460d74934dddfbf9d1b51027c45bef69798089f3f6a96e4e79b6dca06e2d7c465096fb98210b9fdf5228897a709559f17a69408619f15b5b9505", "state": "VERIFIED", "summary": {"compressed": 0.675923, "gated": 0.675923, "naive": 0.675923, "scratch": 0.696756, "ungated": 0.647077}, "ties": 3, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:23Z", "cpu_seconds": 3.616818278000011, "created_at": "2026-11-22T04:25:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c56a144f58dd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "fb74c227ccc3adebb7466be475b81a027aa85db0676cfada7581af49906be4d7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c56a144f58dd", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:25:11Z", "seed": 2057053372, "treatment": "frontier"}, "priority": 0.18965785144055422, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9d21b9f654fe60d1ab0d20b4b0350217023c7ef666328d2462e3a133bae7b073", "run_seconds": 4.5892274379730225, "seed": 2057053372, "started_at": "2026-11-22T04:25:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.175398, 0.013089], "effect": -0.081155, "evaluated_at": "2026-11-22T04:36:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c56a144f58dd", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.440068493, 1.0, 0.374313187, 0.894230769], "frontier": [0.330479452, 0.964285714, 0.364010989, 0.548076923], "goal": [0.321917808, 0.857142857, 0.342032967, 0.519230769], "hypdisc": [0.378424658, 1.0, 0.360576923, 0.711538462], "infogain": [0.330479452, 0.964285714, 0.364010989, 0.548076923], "prederr": [0.330479452, 0.964285714, 0.364010989, 0.548076923], "random": [0.419520548, 1.0, 0.352335165, 0.759615385], "risk": [0.330479452, 0.964285714, 0.364010989, 0.548076923]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "fb74c227ccc3adebb7466be475b81a027aa85db0676cfada7581af49906be4d7", "primary_control": "random", "result_hash": "9d21b9f654fe60d1ab0d20b4b0350217023c7ef666328d2462e3a133bae7b073", "signature": "ad49b0bd06a32335db55d99fa6859ed36145c1eb8b65849ba9f9f6003b2a71fbd6b5eb7b0f2c94b5e850b77aebef73cdd1be9cdf5c9c440222f74ae52334ef0e", "state": "VERIFIED", "summary": {"count": 0.677153, "frontier": 0.551713, "goal": 0.510081, "hypdisc": 0.612635, "infogain": 0.551713, "prederr": 0.551713, "random": 0.632868, "risk": 0.551713}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:36:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:19Z", "cpu_seconds": 0.27149883300000965, "created_at": "2026-11-22T04:25:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2a300310a9a8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "60998e79d5f78d20e7f14d3f3d00590fee95df6293b896d79e0489259196325d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2a300310a9a8", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:11Z", "seed": 463754276, "treatment": "full"}, "priority": 0.19041791577139114, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b1ae9b9610335ec652e9364ad0e00896282c5e57ee6967d3f14666c16a6e40b5", "run_seconds": 0.3175849914550781, "seed": 463754276, "started_at": "2026-11-22T04:25:18Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2a300310a9a8", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "60998e79d5f78d20e7f14d3f3d00590fee95df6293b896d79e0489259196325d", "primary_control": "e141_best", "result_hash": "b1ae9b9610335ec652e9364ad0e00896282c5e57ee6967d3f14666c16a6e40b5", "signature": "e27e2433a8291355ab96bd0a9c03d7511799017ca5c7dd4f9f8c7b495d1dc2a4cc0c27fab7d47da1c805ab0f6a79289f313620a575a5956638f059e7e4fda00a", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:52Z", "cpu_seconds": 0.31061135500002024, "created_at": "2026-11-22T04:25:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-39e95b20fadd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "911b43df592e82bb10edab1bb4c666456d4711c7999e98b4029e3bc431a09d5d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-39e95b20fadd", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:11Z", "seed": 1036670312, "treatment": "full"}, "priority": 0.18732022159601386, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "72f1b6bf514c149ba8c3b8882e886a222a4bdd5d9c5f1803530dbfdb84d6d7b6", "run_seconds": 0.3210322856903076, "seed": 1036670312, "started_at": "2026-11-22T04:25:52Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-39e95b20fadd", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "911b43df592e82bb10edab1bb4c666456d4711c7999e98b4029e3bc431a09d5d", "primary_control": "e141_best", "result_hash": "72f1b6bf514c149ba8c3b8882e886a222a4bdd5d9c5f1803530dbfdb84d6d7b6", "signature": "316d23f7a88c2e0cdc853ad6cdd911db20975f501a89c7965e1845c9fde6401ce8dabb74d08e2f8fff9abb3d0ea30e5e7eab62654709a68d0a20d92d4253030c", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:12Z", "cpu_seconds": 0.2895135020000055, "created_at": "2026-11-22T04:25:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4937b3d2f992", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "9e1c50657c3df1106562bbf4f7cccdfd17eaa35b097213ce704e8e5263a036b2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4937b3d2f992", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:11Z", "seed": 1106446255, "treatment": "full"}, "priority": 0.19427927837461076, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b66401453d2f3d64447b2c29a28b5dbd0a482d81507827e4ded428b2275a2755", "run_seconds": 0.337343692779541, "seed": 1106446255, "started_at": "2026-11-22T04:25:12Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4937b3d2f992", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "9e1c50657c3df1106562bbf4f7cccdfd17eaa35b097213ce704e8e5263a036b2", "primary_control": "e141_best", "result_hash": "b66401453d2f3d64447b2c29a28b5dbd0a482d81507827e4ded428b2275a2755", "signature": "56dd57f4d0dd52a243aaeca083abbb40c5d3fd6735fe5ed8da8a0a0c489489318e7269b2fc4d4b3f083d5c224ad1b5982fac2eade6b4b95778242d01848cc30d", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:14Z", "cpu_seconds": 1.4630801480000173, "created_at": "2026-11-22T04:25:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9fcd9788e7a8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "e45d64ddaf2939f64652752bc4a3070d8428539f4b1bb40a41b6171a895ffb7e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9fcd9788e7a8", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:11Z", "seed": 1861551165, "treatment": "full"}, "priority": 0.3186430836815872, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4a7e95785b9f1b924dff62e887f6860baef91576f36f5b1c24dfdfe18055c84a", "run_seconds": 1.7789361476898193, "seed": 1861551165, "started_at": "2026-11-22T04:25:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.266451, 0.016451], "effect": -0.125, "evaluated_at": "2026-11-22T04:36:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9fcd9788e7a8", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.25, 0.75], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [1.0, 1.0, 0.0, 0.5], "no_world_model": [1.0, 1.0, 0.0, 0.0], "random": [0.75, 1.0, 0.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e45d64ddaf2939f64652752bc4a3070d8428539f4b1bb40a41b6171a895ffb7e", "primary_control": "e141_best", "result_hash": "4a7e95785b9f1b924dff62e887f6860baef91576f36f5b1c24dfdfe18055c84a", "signature": "bfd25ac75f566b5e0364a8bbea39f590219ba7d7ade8d700a133f3f7eeb9546744cbc2fa3e2f8ebbc39521b0d54898aa2a6ddf1724546f7edc31ea833bb4eb0a", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.0625, "full": 0.625, "no_world_model": 0.5, "random": 0.625}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:13Z", "cpu_seconds": 0.19050513299998784, "created_at": "2026-11-22T04:25:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a74cf96c56f5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "89a23a84377438429a5c662c0d5517fa2254b217380e642508ecd502b15448b7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a74cf96c56f5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:11Z", "seed": 38599346, "treatment": "full"}, "priority": 0.19336801008626178, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "9924d403343f41ae26b68a6763d3fbf96e05dffc0addc5aac6bca364dbde14ac", "run_seconds": 0.3763761520385742, "seed": 38599346, "started_at": "2026-11-22T04:25:13Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a74cf96c56f5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.5, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "89a23a84377438429a5c662c0d5517fa2254b217380e642508ecd502b15448b7", "primary_control": "e141_best", "result_hash": "9924d403343f41ae26b68a6763d3fbf96e05dffc0addc5aac6bca364dbde14ac", "signature": "23a05be899dbae03be98508c678e5d9422c021fcc8f7c63651849329800c96fb0eb0098f832c9ef734d11aec16cf0f27497ebc5d94eaba2d2b8aaead12eb7c0a", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:18Z", "cpu_seconds": 0.39511616400000094, "created_at": "2026-11-22T04:25:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-cea99045cbad", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "8d7f80e9ff381312cf49ad8a4796b3532a9af0db104111207f1a38dc1f902540", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-cea99045cbad", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:11Z", "seed": 1918509298, "treatment": "full"}, "priority": 0.1919825641756656, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b11759b3338d42ca0e74072a50d5367c2579752f561117329a63a5956ecdfe51", "run_seconds": 0.46503782272338867, "seed": 1918509298, "started_at": "2026-11-22T04:25:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.197307, 0.697307], "effect": 0.25, "evaluated_at": "2026-11-22T04:36:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-cea99045cbad", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.5, 0.0, 0.0], "fixed_policy": [0.25, 0.25, 0.0, 0.0], "full": [0.0, 0.25, 0.5, 0.75], "no_world_model": [0.0, 0.0, 0.5, 0.75], "random": [0.0, 0.25, 0.5, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "8d7f80e9ff381312cf49ad8a4796b3532a9af0db104111207f1a38dc1f902540", "primary_control": "e141_best", "result_hash": "b11759b3338d42ca0e74072a50d5367c2579752f561117329a63a5956ecdfe51", "signature": "69da80ee8dacae464544e718eed7b02ddc1278213efd13a78df397cd60e06bbe80606d68a7e58657a3ac5cb577de08aaa840a077c5c73dc0a99712e72e22190c", "state": "VERIFIED", "summary": {"e141_best": 0.125, "fixed_policy": 0.125, "full": 0.375, "no_world_model": 0.3125, "random": 0.1875}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:36:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:29Z", "cpu_seconds": 0.3087403149999943, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2d044dec71be", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "0a59c652110907dc3fba57ce883b794f785be02da8acd324cd08dd45eea7fbb5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2d044dec71be", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:26Z", "seed": 201739508, "treatment": "full"}, "priority": 0.19203491770527953, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0ce3b1044ebb683d8595fdba082f51934868f86aec86c967ee4a5f8ca09ff3d9", "run_seconds": 0.31627917289733887, "seed": 201739508, "started_at": "2026-11-22T04:25:28Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2d044dec71be", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.25, 0.25, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "0a59c652110907dc3fba57ce883b794f785be02da8acd324cd08dd45eea7fbb5", "primary_control": "e141_best", "result_hash": "0ce3b1044ebb683d8595fdba082f51934868f86aec86c967ee4a5f8ca09ff3d9", "signature": "77cc869214a9debccab634d7a08063c6b5965a5fdd56104afefc9ce33965d78d3de14eb7a064ed5a9016be372c5996fed05ae0e9b82d537d9e2f361dba0eb20f", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:37Z", "cpu_seconds": 3.072772981000014, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9ce81bf37028", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "42a1957caf3b36bd98990b0f7b913fad70be9bdf0060de8d0ea6e385bb8542b4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9ce81bf37028", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:26Z", "seed": 1009732055, "treatment": "full"}, "priority": 0.18891042756144022, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f043b4f84ba7cc925f850c25b38fab6a9fe2d2e4d633ef3ec80911917e338166", "run_seconds": 3.1963350772857666, "seed": 1009732055, "started_at": "2026-11-22T04:25:34Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.63, 1.12], "effect": 0.875, "evaluated_at": "2026-11-22T04:36:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9ce81bf37028", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.0, 0.0, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.25], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 0.75], "random": [0.75, 0.75, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "42a1957caf3b36bd98990b0f7b913fad70be9bdf0060de8d0ea6e385bb8542b4", "primary_control": "e141_best", "result_hash": "f043b4f84ba7cc925f850c25b38fab6a9fe2d2e4d633ef3ec80911917e338166", "signature": "21e03c6e3f86e27aff33387438c2c4775b4e61ae236d31ec4e01b8562f1ca0473a6e7c31e3bb16a7cb72e6c9b469f8a639f6469abc01c02efb50bcda10d7080a", "state": "VERIFIED", "summary": {"e141_best": 0.0625, "fixed_policy": 0.0625, "full": 0.9375, "no_world_model": 0.875, "random": 0.875}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:42Z", "cpu_seconds": 0.2028245429999913, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e326268ac3c8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "3a58bbcd2e79f779a148e3efc3b2e1c8cf6d3976bd514681a396d7e87774b24b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e326268ac3c8", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:26Z", "seed": 1453261403, "treatment": "full"}, "priority": 0.187406668476724, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "28a8b0ed519d0f66ab8822634a3330f37a7570e1cd807cfe3d4bad577b1ffd1d", "run_seconds": 0.230804443359375, "seed": 1453261403, "started_at": "2026-11-22T04:25:42Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e326268ac3c8", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.5, 0.5, 0.25, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "3a58bbcd2e79f779a148e3efc3b2e1c8cf6d3976bd514681a396d7e87774b24b", "primary_control": "e141_best", "result_hash": "28a8b0ed519d0f66ab8822634a3330f37a7570e1cd807cfe3d4bad577b1ffd1d", "signature": "728a517441126ee8f219304dca6a51a23ad91333651560728a81d7dd94dd1c7caa50b03387faca0b5e287486d418e52cdae0e04f144c7725e7d5b50c34bff309", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 0.9375}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:37Z", "cpu_seconds": 1.5251476580000087, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-02173c3c847f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "769720c6e8e65e2cb3c4255da85f7203dfd8b1b323e7db01824d7c4ee33cbfcc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-02173c3c847f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:26Z", "seed": 127659139, "treatment": "full"}, "priority": 0.18809152801878157, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "eceafe2537055ad470e1e02ffb23a5d348ee8ecf4781ef7431d1792f95e9ef1e", "run_seconds": 1.5719256401062012, "seed": 127659139, "started_at": "2026-11-22T04:25:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.12094, 0.116327], "effect": -0.002307, "evaluated_at": "2026-11-22T04:36:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-02173c3c847f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.75625, 1.0, 0.928571429, 0.527777778], "frozen": [0.40625, 0.333333333, 0.321428571, 0.305555556], "full": [0.6875, 1.0, 0.821428571, 0.694444444], "retrieval": [0.7625, 1.0, 0.821428571, 0.777777778], "rules_nodetect": [0.6875, 1.0, 0.821428571, 0.694444444]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "769720c6e8e65e2cb3c4255da85f7203dfd8b1b323e7db01824d7c4ee33cbfcc", "primary_control": "facts", "result_hash": "eceafe2537055ad470e1e02ffb23a5d348ee8ecf4781ef7431d1792f95e9ef1e", "signature": "92c368ee4a752dd0b14fe7a5b1144ee617e88043b61139fe44a360b7c1b775bed0f04a1fec787074bfb0e6ed03b04cd2621f06dce9cc7d6171f20cccda639200", "state": "VERIFIED", "summary": {"facts": 0.80315, "frozen": 0.341642, "full": 0.800843, "retrieval": 0.840427, "rules_nodetect": 0.800843}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:36:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:43Z", "cpu_seconds": 0.5570361919999982, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-86954d82ab3e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "a859bafa0d70f460add7aea10a7a02d858e3a2a06ff74b64312b75d677786a99", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-86954d82ab3e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:26Z", "seed": 1912360701, "treatment": "full"}, "priority": 0.187387287683, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ad97e5e13a7fb8cfe5ea95d8b63efc1b2cb2cd4a7d9b61b3c86334494ec923bf", "run_seconds": 0.6407666206359863, "seed": 1912360701, "started_at": "2026-11-22T04:25:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.033996, 0.061592], "effect": 0.047794, "evaluated_at": "2026-11-22T04:36:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-86954d82ab3e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.803921569, 0.730392157, 0.715686275, 0.519607843], "frozen": [0.073529412, 0.068627451, 0.073529412, 0.12745098], "full": [0.833333333, 0.779411765, 0.779411765, 0.568627451], "retrieval": [0.882352941, 0.852941176, 0.799019608, 0.705882353], "rules_nodetect": [0.833333333, 0.779411765, 0.779411765, 0.568627451]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a859bafa0d70f460add7aea10a7a02d858e3a2a06ff74b64312b75d677786a99", "primary_control": "facts", "result_hash": "ad97e5e13a7fb8cfe5ea95d8b63efc1b2cb2cd4a7d9b61b3c86334494ec923bf", "signature": "60a380fcd2e4005cfe586b24b11046fcc46a9b8691a2fb07770caa9834fe3a691ad5ec1f175c68f7ba7012c14357eae79f21a64fdaa00d879d434bd3b4b98007", "state": "VERIFIED", "summary": {"facts": 0.692402, "frozen": 0.085784, "full": 0.740196, "retrieval": 0.810049, "rules_nodetect": 0.740196}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:31Z", "cpu_seconds": 2.1236599990000116, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d76986d1b40c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 6, "n": 4}, "prereg_hash": "38f823029021430e19efebfd1d647b980081f98fbf16c6a04830f6ae9556a5f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d76986d1b40c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:26Z", "seed": 308565250, "treatment": "full"}, "priority": 0.20016048513925833, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-3ec8ee6c989a", "result_hash": "efd376c1d23a74ff3775665566aa9891cb4ed1751733dd9b1f564be0baea0f2a", "run_seconds": 2.158658504486084, "seed": 308565250, "started_at": "2026-11-22T04:25:28Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005829, 0.220744], "effect": 0.107457, "evaluated_at": "2026-11-22T04:36:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d76986d1b40c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.975, 0.415189873, 0.524137931, 0.126666667], "frozen": [0.35, 0.106329114, 0.089655172, 0.0], "full": [0.975, 0.632911392, 0.720689655, 0.142222222], "retrieval": [0.975, 0.698734177, 0.782758621, 0.275555556], "rules_nodetect": [0.325, 0.336708861, 0.413793103, 0.142222222]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "38f823029021430e19efebfd1d647b980081f98fbf16c6a04830f6ae9556a5f4", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-3ec8ee6c989a", "result_hash": "efd376c1d23a74ff3775665566aa9891cb4ed1751733dd9b1f564be0baea0f2a", "signature": "854024a27cdd09a29cfe910549b5146b0af85c55eb1c8fed11dc5732760029a9eb9b556d07f11a9068830fd3c5bcf6aebe08f45820df18bd2dd0060f7dc71808", "state": "VERIFIED", "summary": {"facts": 0.510249, "frozen": 0.136496, "full": 0.617706, "retrieval": 0.683012, "rules_nodetect": 0.304431}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:36:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:40Z", "cpu_seconds": 0.41113271899999404, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-231c3d3ea05c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "prereg_hash": "ac27241d14ba66e5151d97d3bdcdc6f06b7d9945fccdf75ad108900688200740", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-231c3d3ea05c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:25:26Z", "seed": 427729891, "treatment": "gated"}, "priority": 0.18752586775792052, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "ff41c65ab020b1500c0381b2464f2e961c86a7a3c57f36b9fee718970ef19a8f", "run_seconds": 0.43597412109375, "seed": 427729891, "started_at": "2026-11-22T04:25:40Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:23Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-231c3d3ea05c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.9, 0.895833333, 0.854166667, 0.895833333], "gated": [0.9, 0.895833333, 0.854166667, 0.895833333], "naive": [0.3, 0.6875, 0.229166667, 0.395833333], "scratch": [0.9, 0.895833333, 0.854166667, 0.895833333], "ungated": [0.5, 0.8125, 0.5, 0.5625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ac27241d14ba66e5151d97d3bdcdc6f06b7d9945fccdf75ad108900688200740", "primary_control": "scratch", "result_hash": "ff41c65ab020b1500c0381b2464f2e961c86a7a3c57f36b9fee718970ef19a8f", "signature": "860b3405157c0264c1d3431dbf180e5655e24ec6c34c784f19dafd706ef3305d725843a44312e705f3d4c57c9b9732c6133f05f20f6c38250efc735c5b65ca0e", "state": "PARTIAL", "summary": {"compressed": 0.886458, "gated": 0.886458, "naive": 0.403125, "scratch": 0.886458, "ungated": 0.59375}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:35Z", "cpu_seconds": 4.693350068000001, "created_at": "2026-11-22T04:25:26Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-b9706c2c90d1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "prereg_hash": "d322e33922603016430590bf65d638867926de5474d30c64f11a9e870b075230", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-b9706c2c90d1", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:25:26Z", "seed": 1430866765, "treatment": "gated"}, "priority": 0.18914287357736517, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "3a83e5a84ea45a5ac9960ed8bd776e117007dac9d44a958a764ff17181bd3ba8", "run_seconds": 4.886087894439697, "seed": 1430866765, "started_at": "2026-11-22T04:25:30Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-b9706c2c90d1", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.425974026, 0.280555556, 0.2, 0.141666667], "gated": [0.425974026, 0.280555556, 0.2, 0.141666667], "naive": [0.163636364, 0.177777778, 0.163414634, 0.08125], "scratch": [0.425974026, 0.280555556, 0.2, 0.141666667], "ungated": [0.168831169, 0.147222222, 0.141463415, 0.116666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "d322e33922603016430590bf65d638867926de5474d30c64f11a9e870b075230", "primary_control": "scratch", "result_hash": "3a83e5a84ea45a5ac9960ed8bd776e117007dac9d44a958a764ff17181bd3ba8", "signature": "c0ec93401e16db60032e6008dc8c19c3e7eee5b6d5d11420829dac9038e0d9d2237d2748cd0766abd4e38a404f0d175a563ead84a7bf4c1fc7a33878b5505609", "state": "PARTIAL", "summary": {"compressed": 0.262049, "gated": 0.262049, "naive": 0.14652, "scratch": 0.262049, "ungated": 0.143546}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:23Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:56Z", "cpu_seconds": 0.41107947100002207, "created_at": "2026-11-22T04:25:42Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-45bbafd1302e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "a6c89a5085d4bbe03140f406863ceca69551d0113f9781b7e6a0bdd1d736ef9c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-45bbafd1302e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:25:42Z", "seed": 2024571707, "treatment": "frontier"}, "priority": 0.1855369412513049, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e44321c6f848a751af6db3e3069317f0ea93953371d008bf5520fc85f9ca2e0c", "run_seconds": 0.41199421882629395, "seed": 2024571707, "started_at": "2026-11-22T04:25:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.113496, 0.210368], "effect": 0.161932, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-45bbafd1302e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.965909091, 0.988636364, 0.988636364, 0.988636364], "frontier": [0.909090909, 1.0, 0.920454545, 0.977272727], "goal": [0.704545455, 0.795454545, 0.875, 0.806818182], "hypdisc": [0.761363636, 0.806818182, 0.818181818, 0.761363636], "infogain": [0.909090909, 1.0, 0.920454545, 0.977272727], "prederr": [0.909090909, 1.0, 0.920454545, 0.977272727], "random": [0.795454545, 0.795454545, 0.795454545, 0.772727273], "risk": [0.909090909, 1.0, 0.920454545, 0.977272727]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "a6c89a5085d4bbe03140f406863ceca69551d0113f9781b7e6a0bdd1d736ef9c", "primary_control": "random", "result_hash": "e44321c6f848a751af6db3e3069317f0ea93953371d008bf5520fc85f9ca2e0c", "signature": "63614efe0c67e10baa348e5cb4b666715c53d8ee2dd73b07d61af5a4673c637da9d022afa8a5b55f8cc893114456ecf2dd185605999a2dcd0ec479d182e7d808", "state": "VERIFIED", "summary": {"count": 0.982955, "frontier": 0.951705, "goal": 0.795455, "hypdisc": 0.786932, "infogain": 0.951705, "prederr": 0.951705, "random": 0.789773, "risk": 0.951705}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:47Z", "cpu_seconds": 0.6342097919999787, "created_at": "2026-11-22T04:25:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-04594a657763", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "919fc7bd98ebb315f5f26ccdf0852daaf9f8af2a30bbd222dd007123898c614e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-04594a657763", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:42Z", "seed": 1758194029, "treatment": "full"}, "priority": 0.19372441083448197, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1c06e18a74c11ee2f658f1186bb01f5c1442f4fc1b87a77a3c991ab73fc7852b", "run_seconds": 1.115046501159668, "seed": 1758194029, "started_at": "2026-11-22T04:25:45Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:36:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-04594a657763", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"e141_best": [1.0, 0.8125], "fixed_policy": [1.0, 0.125], "full": [1.0, 0.8125], "no_world_model": [1.0, 0.8125], "random": [1.0, 0.8125]}, "prereg_hash": "919fc7bd98ebb315f5f26ccdf0852daaf9f8af2a30bbd222dd007123898c614e", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "1c06e18a74c11ee2f658f1186bb01f5c1442f4fc1b87a77a3c991ab73fc7852b", "signature": "73fa08648fdb87c74a9f13158b8af73b51961937b2451ac4db243df7389b7d19c3f9cc2ec3000a701c3c5171155d2a6bfe38dfd0c86bf427a6eeae5079597700", "state": "REJECTED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.125, "full": 0.8125, "no_world_model": 0.8125, "random": 0.8125}}, "verified_at": "2026-11-22T04:36:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:51Z", "cpu_seconds": 2.6503559509999945, "created_at": "2026-11-22T04:25:42Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b7f1006ece9f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "08d97797ca8ac0bc52b4c7ea23099c43367bdf33c3217187a0cb96ab6e697d7d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b7f1006ece9f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:42Z", "seed": 900086499, "treatment": "full"}, "priority": 0.1881899402680707, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "67217d831ee6b196ac7b5251c704afc7d750f8c01f7ef48827c512ed1d8e2488", "run_seconds": 3.060624837875366, "seed": 900086499, "started_at": "2026-11-22T04:25:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.32793, 0.79707], "effect": 0.5625, "evaluated_at": "2026-11-22T04:36:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b7f1006ece9f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.5, 0.5, 0.0], "fixed_policy": [0.5, 0.75, 1.0, 0.25], "full": [0.75, 0.75, 1.0, 0.75], "no_world_model": [0.25, 0.25, 1.0, 1.0], "random": [1.0, 0.75, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "08d97797ca8ac0bc52b4c7ea23099c43367bdf33c3217187a0cb96ab6e697d7d", "primary_control": "e141_best", "result_hash": "67217d831ee6b196ac7b5251c704afc7d750f8c01f7ef48827c512ed1d8e2488", "signature": "952c2af8a8f8e28b17447011738c08b90ba1dee7581411713f7539d95efd7c43b9471bf9f1bec36842157f6ae37b656db5315841313bc17e6f9374518c26c40a", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.625, "full": 0.8125, "no_world_model": 0.625, "random": 0.9375}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:05Z", "cpu_seconds": 0.5831718039999885, "created_at": "2026-11-22T04:25:42Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0a62c0e17295", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "9a1d1ea63b0a6d4856313431d140a76603581bd151ca156a8d7f6274f124d19a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0a62c0e17295", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:42Z", "seed": 1441494294, "treatment": "full"}, "priority": 0.1848214862234132, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "72e5ad47ceb3807611f481a21fb01d4155756d96f95884f4472911ad13627333", "run_seconds": 0.585064172744751, "seed": 1441494294, "started_at": "2026-11-22T04:26:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.030758, 0.107673], "effect": 0.069216, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0a62c0e17295", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.531914894, 0.558333333, 0.714859438, 0.64893617], "frozen": [0.095744681, 0.1125, 0.108433735, 0.09929078], "full": [0.602836879, 0.6, 0.75502008, 0.773049645], "retrieval": [0.790780142, 0.804166667, 0.875502008, 0.865248227], "rules_nodetect": [0.602836879, 0.6, 0.75502008, 0.773049645]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "9a1d1ea63b0a6d4856313431d140a76603581bd151ca156a8d7f6274f124d19a", "primary_control": "facts", "result_hash": "72e5ad47ceb3807611f481a21fb01d4155756d96f95884f4472911ad13627333", "signature": "fb3ae76c054841fe6a2ae1bab9abcfa74c6d9b5961e4d723fef9f71d595dea451e136d64879df131665048b20f2a61a380e020cbe2e6a92cad2f3665ee45060b", "state": "VERIFIED", "summary": {"facts": 0.613511, "frozen": 0.103992, "full": 0.682727, "retrieval": 0.833924, "rules_nodetect": 0.682727}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:53Z", "cpu_seconds": 0.6164255759999833, "created_at": "2026-11-22T04:25:42Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c71e44bd197e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "3f2e7ca3cb8bf1e72d9332ba414f0d1ca181bfa31156fae456887b531262f48e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c71e44bd197e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:42Z", "seed": 1775476054, "treatment": "full"}, "priority": 0.18581186309523234, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "945ab714b73714f28d2140113e6985ab2c82986b458c1f21d8a695cfde34b3ff", "run_seconds": 0.7269644737243652, "seed": 1775476054, "started_at": "2026-11-22T04:25:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.034057, 0.057446], "effect": 0.045752, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-c71e44bd197e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.509803922, 0.816993464, 0.699346405, 0.705882353], "frozen": [0.08496732, 0.091503268, 0.091503268, 0.091503268], "full": [0.562091503, 0.849673203, 0.738562092, 0.764705882], "retrieval": [0.784313725, 0.836601307, 0.823529412, 0.830065359], "rules_nodetect": [0.562091503, 0.849673203, 0.738562092, 0.764705882]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "3f2e7ca3cb8bf1e72d9332ba414f0d1ca181bfa31156fae456887b531262f48e", "primary_control": "facts", "result_hash": "945ab714b73714f28d2140113e6985ab2c82986b458c1f21d8a695cfde34b3ff", "signature": "be07842d232fa94a6ee36256c095ba52d7c88b4591c266342fa5ddab17a19c7ca5c35f30e8cc17b5dfc77fa8e6903a476c5956d1043a7f376913f6f8274e510f", "state": "VERIFIED", "summary": {"facts": 0.683007, "frozen": 0.089869, "full": 0.728758, "retrieval": 0.818627, "rules_nodetect": 0.728758}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:48Z", "cpu_seconds": 0.2097367840000004, "created_at": "2026-11-22T04:25:42Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d499ba98acde", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "6f8009a9b45a5f06cfa324583cc09cb26877675bd35abccdf0fcdac62d277f99", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d499ba98acde", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:42Z", "seed": 1436877067, "treatment": "full"}, "priority": 0.1932983606916199, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4f38dae55502ffc8f103786cec4961a5c7694a2a87d2c64f680620efbcc4492f", "run_seconds": 0.35262250900268555, "seed": 1436877067, "started_at": "2026-11-22T04:25:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003636, 0.011212], "effect": 0.003788, "evaluated_at": "2026-11-22T04:36:24Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-d499ba98acde", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.96969697, 1.0, 0.984848485, 0.984848485], "frozen": [0.015151515, 0.03030303, 0.03030303, 0.03030303], "full": [0.96969697, 1.0, 0.984848485, 1.0], "retrieval": [0.96969697, 1.0, 0.984848485, 0.984848485], "rules_nodetect": [0.96969697, 1.0, 0.984848485, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "6f8009a9b45a5f06cfa324583cc09cb26877675bd35abccdf0fcdac62d277f99", "primary_control": "facts", "result_hash": "4f38dae55502ffc8f103786cec4961a5c7694a2a87d2c64f680620efbcc4492f", "signature": "fceff6006d2dd225e8d083067b4d4dccf6d3f34dca0d55b7ff8c49450401e46e1f81dcbb86eb6511f08a9842ba77320a3e33fb3dac6cb17c729cfe9b38fbe00d", "state": "VERIFIED", "summary": {"facts": 0.984848, "frozen": 0.026515, "full": 0.988636, "retrieval": 0.984848, "rules_nodetect": 0.988636}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:36:24Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:58Z", "cpu_seconds": 0.19110086300000262, "created_at": "2026-11-22T04:25:42Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3fe640d7a948", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "prereg_hash": "41e89629cfaec8a3bf63bd6b1a37702805b0096e00a3cd501cb891091e45536e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3fe640d7a948", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|math_oracle|math_oracle|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:25:42Z", "seed": 2029355882, "treatment": "gated"}, "priority": 0.18543919397966982, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "83fe8d935a5a3a0218d7e97e534fee2c875bade3306ee531aa81b052a0c3a19d", "run_seconds": 0.20743584632873535, "seed": 2029355882, "started_at": "2026-11-22T04:25:57Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-3fe640d7a948", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.363636364, 0.333333333, 0.106060606, 0.348484848], "gated": [0.363636364, 0.333333333, 0.106060606, 0.348484848], "naive": [0.106060606, 0.333333333, 0.106060606, 0.363636364], "scratch": [0.363636364, 0.333333333, 0.106060606, 0.348484848], "ungated": [0.106060606, 0.333333333, 0.106060606, 0.363636364]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "41e89629cfaec8a3bf63bd6b1a37702805b0096e00a3cd501cb891091e45536e", "primary_control": "scratch", "result_hash": "83fe8d935a5a3a0218d7e97e534fee2c875bade3306ee531aa81b052a0c3a19d", "signature": "f11939fd4082b6fe5da0d6370eef662cf64bc5fb2a4683ce72a873b65e592980bfa7545e813e6af215089fe8a47a28dbf53f1649a7cf49fdf691a0ab99853c05", "state": "PARTIAL", "summary": {"compressed": 0.287879, "gated": 0.287879, "naive": 0.227273, "scratch": 0.287879, "ungated": 0.227273}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:07Z", "cpu_seconds": 0.5281065540000043, "created_at": "2026-11-22T04:25:44Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-53c48ff55368", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "441426f7e4ecdacdb21ec4632b7a0c710fdc93ba19047e5ecc4e8a81ec1ea368", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-53c48ff55368", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:44Z", "seed": 1818166042, "treatment": "full"}, "priority": 0.18375528813360673, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "fcf8967b425a33f665eda91576d446277e5b77527b522557a45a1760c43da961", "run_seconds": 0.5784485340118408, "seed": 1818166042, "started_at": "2026-11-22T04:26:07Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.185, 0.06], "effect": -0.0625, "evaluated_at": "2026-11-22T04:36:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-53c48ff55368", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 0.75], "fixed_policy": [0.25, 0.25, 0.25, 0.25], "full": [1.0, 1.0, 0.75, 0.5], "no_world_model": [1.0, 1.0, 0.0, 0.0], "random": [1.0, 1.0, 0.75, 0.25]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "441426f7e4ecdacdb21ec4632b7a0c710fdc93ba19047e5ecc4e8a81ec1ea368", "primary_control": "e141_best", "result_hash": "fcf8967b425a33f665eda91576d446277e5b77527b522557a45a1760c43da961", "signature": "c4f3805d81252453f3a64e0be6484649c4c8cf5cd63d071d8d0ca3857451b62955001d5718148c9e99815962466945dae42fd51b280f76a6c9fb0cad6102b30f", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.25, "full": 0.8125, "no_world_model": 0.5, "random": 0.75}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:03Z", "cpu_seconds": 0.4554757850000044, "created_at": "2026-11-22T04:25:58Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f3f56d70157b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "298ecbee90164dc61b8b8824f6919e2c5c9e53cf332fd33eeea6de16872263a0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f3f56d70157b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:25:58Z", "seed": 7254511, "treatment": "frontier"}, "priority": 0.18521578054259702, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "3ec59e5f5d150d37d994ea2efb30ae96db956cb7faba828a473ac8e56dd3f655", "run_seconds": 0.45660948753356934, "seed": 7254511, "started_at": "2026-11-22T04:26:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.099651, 0.1655], "effect": 0.132576, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f3f56d70157b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 0.954545455, 0.984848485, 1.0], "frontier": [0.984848485, 0.96969697, 1.0, 1.0], "goal": [0.878787879, 0.803030303, 0.909090909, 0.863636364], "hypdisc": [0.848484848, 0.787878788, 0.909090909, 0.833333333], "infogain": [0.984848485, 0.96969697, 1.0, 1.0], "prederr": [0.984848485, 0.96969697, 1.0, 1.0], "random": [0.878787879, 0.848484848, 0.878787879, 0.818181818], "risk": [0.984848485, 0.96969697, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "298ecbee90164dc61b8b8824f6919e2c5c9e53cf332fd33eeea6de16872263a0", "primary_control": "random", "result_hash": "3ec59e5f5d150d37d994ea2efb30ae96db956cb7faba828a473ac8e56dd3f655", "signature": "92da9ca7db32b1e2d8fbbce1dd633b958e29c615ef75a31b0379baa4c68db55192eb28d4509f1631ef035f85a6463b81878ae3f75b73b3af981e84e0c3cd630a", "state": "VERIFIED", "summary": {"count": 0.984848, "frontier": 0.988636, "goal": 0.863636, "hypdisc": 0.844697, "infogain": 0.988636, "prederr": 0.988636, "random": 0.856061, "risk": 0.988636}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:10Z", "cpu_seconds": 0.17026210599999558, "created_at": "2026-11-22T04:25:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2df5b94df670", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "4d4cf278f5d1912f8a6e18f2867745d40fb29942350d71b616b2335cf011ae3b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2df5b94df670", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:58Z", "seed": 558666447, "treatment": "full"}, "priority": 0.18154522006202634, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "40e1ead16e904bfe952640b135d2b030be6766603528f249eeb993eabad4e81d", "run_seconds": 0.17039036750793457, "seed": 558666447, "started_at": "2026-11-22T04:26:10Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T04:36:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2df5b94df670", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.25, 1.0, 1.0], "fixed_policy": [0.0, 0.75, 0.0, 0.25], "full": [0.75, 0.25, 1.0, 0.25], "no_world_model": [0.75, 0.25, 1.0, 0.75], "random": [0.75, 0.25, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "4d4cf278f5d1912f8a6e18f2867745d40fb29942350d71b616b2335cf011ae3b", "primary_control": "e141_best", "result_hash": "40e1ead16e904bfe952640b135d2b030be6766603528f249eeb993eabad4e81d", "signature": "3b08fea2cc258f9998cbff91f5f26f4d9c4ecf6a06e79e06f98e259c8e2c4d7fe709f9902fc6e059accb11a60a37801259d0abe5e618fe0580a644ce537e5502", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.25, "full": 0.5625, "no_world_model": 0.6875, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:13Z", "cpu_seconds": 0.18855446500000994, "created_at": "2026-11-22T04:25:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-80c06b808c4a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "f1dee7ece1f680eba8286df3b2870194ff09a4b50e93715ce479beaeddf5949e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-80c06b808c4a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:58Z", "seed": 652979764, "treatment": "full"}, "priority": 0.18151223690730536, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b44f752c50916817d122936a17d264588f953f5d66dd6ab589ff8870012c16a0", "run_seconds": 0.20755338668823242, "seed": 652979764, "started_at": "2026-11-22T04:26:12Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-80c06b808c4a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.25, 0.75], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f1dee7ece1f680eba8286df3b2870194ff09a4b50e93715ce479beaeddf5949e", "primary_control": "e141_best", "result_hash": "b44f752c50916817d122936a17d264588f953f5d66dd6ab589ff8870012c16a0", "signature": "1c117638c5bfb77fe338377e3028143a1c21a376db0d33c9c350588b45c23e5790f43abfacb1000d3ad41a42259a8e65ff26b1760410364414127cfb91816b01", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.4375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:25:59Z", "cpu_seconds": 1.0542299340000056, "created_at": "2026-11-22T04:25:58Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b3996360df00", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "fd001bc4cf5bed1a83344a7a2210e6b6e20f665ce83fa1fc3cf67faed4998d9f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b3996360df00", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:25:58Z", "seed": 405339028, "treatment": "full"}, "priority": 0.20544653919506875, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-932c0b25a613", "result_hash": "207f38455467dcc62ec13b7f2e56eb6bc3cb76a170caf561f82cabe7b96f1f9c", "run_seconds": 1.1054818630218506, "seed": 405339028, "started_at": "2026-11-22T04:25:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.05742, 1.06758], "effect": 0.5625, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b3996360df00", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.75, 1.0], "fixed_policy": [1.0, 1.0, 0.75, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "fd001bc4cf5bed1a83344a7a2210e6b6e20f665ce83fa1fc3cf67faed4998d9f", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-932c0b25a613", "result_hash": "207f38455467dcc62ec13b7f2e56eb6bc3cb76a170caf561f82cabe7b96f1f9c", "signature": "4d67a0445033263f2fd4dd932ef6d26e3de923dba48e07424eeac8e3d97fc194343ad8d1fe7ddad87b3681f18ad260c923cc0a32832aaa41262c1b4c3d9eb005", "state": "VERIFIED", "summary": {"e141_best": 0.4375, "fixed_policy": 0.9375, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:02Z", "cpu_seconds": 0.32405488200001287, "created_at": "2026-11-22T04:25:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-813802c8a9cf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "b523b0fe9de78611f9008a0c92cfec12c862d03ffaa69f2bffca79b07187c814", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-813802c8a9cf", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:58Z", "seed": 107369092, "treatment": "full"}, "priority": 0.1862388349438339, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9e5cc57e0d541084945494c81ecf47e746c93e3142a6ee7e78d3d093560f8650", "run_seconds": 0.468799352645874, "seed": 107369092, "started_at": "2026-11-22T04:26:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.026029, 0.163555], "effect": 0.094792, "evaluated_at": "2026-11-22T04:36:26Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-813802c8a9cf", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.958333333, 0.791666667, 0.916666667, 0.85], "frozen": [0.25, 0.0, 0.25, 0.0], "full": [0.958333333, 0.9375, 1.0, 1.0], "retrieval": [0.958333333, 0.958333333, 1.0, 0.9], "rules_nodetect": [0.958333333, 0.9375, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b523b0fe9de78611f9008a0c92cfec12c862d03ffaa69f2bffca79b07187c814", "primary_control": "facts", "result_hash": "9e5cc57e0d541084945494c81ecf47e746c93e3142a6ee7e78d3d093560f8650", "signature": "81b54463f512ee4ff3fd586228ccd1aae67601984d3e9fc05ade893523c56e56c1d2adf9cf59199fb4a8ae1215d3b10ea0757f85debb9f04f5e20de0ec5a660e", "state": "VERIFIED", "summary": {"facts": 0.879167, "frozen": 0.125, "full": 0.973958, "retrieval": 0.954167, "rules_nodetect": 0.973958}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:36:26Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:14Z", "cpu_seconds": 0.3544424709999987, "created_at": "2026-11-22T04:25:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-93a1f2f68767", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "c66c5d2f82319e7d9a97b14badc946342bb64bed762c893367c3962da46eb8a4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-93a1f2f68767", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:58Z", "seed": 1058894400, "treatment": "full"}, "priority": 0.18125773652598104, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3d3ab6de594873f3b63cd86a916e356becfabe79764a9e2dc33b9fca07b76495", "run_seconds": 0.3589632511138916, "seed": 1058894400, "started_at": "2026-11-22T04:26:14Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-93a1f2f68767", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.897435897, 0.755555556, 0.927536232, 0.916666667], "frozen": [0.487179487, 0.7, 0.507246377, 0.5], "full": [0.897435897, 0.755555556, 0.927536232, 0.916666667], "retrieval": [0.897435897, 0.755555556, 0.927536232, 0.916666667], "rules_nodetect": [0.897435897, 0.833333333, 0.898550725, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "c66c5d2f82319e7d9a97b14badc946342bb64bed762c893367c3962da46eb8a4", "primary_control": "facts", "result_hash": "3d3ab6de594873f3b63cd86a916e356becfabe79764a9e2dc33b9fca07b76495", "signature": "03ab775faa8007e58304d4a74a5fad19d47255cfc25102e290fa49c818d08f6fc285aa6a427a4803e2c5b6cde303cb399cdf1f7b50bdebd4be7b3e1b99dcc602", "state": "PARTIAL", "summary": {"facts": 0.874299, "frozen": 0.548606, "full": 0.874299, "retrieval": 0.874299, "rules_nodetect": 0.886497}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:09Z", "cpu_seconds": 0.736864201000003, "created_at": "2026-11-22T04:25:58Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-dd0aebccbf47", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "189a1ea84fd8c4895f03f3c7b2d018c67609b94f18c6e51e252a8d5510c6fdd9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-dd0aebccbf47", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:25:58Z", "seed": 1494820148, "treatment": "full"}, "priority": 0.18182655969516104, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c19334f6c22186d7361b18a48ef2c29387da67a8a3329f1324370ec2a69d9200", "run_seconds": 0.7472460269927979, "seed": 1494820148, "started_at": "2026-11-22T04:26:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022222, 0.068519], "effect": 0.023148, "evaluated_at": "2026-11-22T04:36:27Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-dd0aebccbf47", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.796296296, 0.694444444, 0.846153846, 0.759259259], "frozen": [0.305555556, 0.300925926, 0.358974359, 0.283950617], "full": [0.796296296, 0.787037037, 0.846153846, 0.759259259], "retrieval": [0.796296296, 0.703703704, 0.846153846, 0.759259259], "rules_nodetect": [0.796296296, 0.773148148, 0.846153846, 0.759259259]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "189a1ea84fd8c4895f03f3c7b2d018c67609b94f18c6e51e252a8d5510c6fdd9", "primary_control": "facts", "result_hash": "c19334f6c22186d7361b18a48ef2c29387da67a8a3329f1324370ec2a69d9200", "signature": "4f2ebc3566d092454ee151592c976cffc204e403877cc4df976e10c8f204a3d1187eb0fd25704ccccef0c4e4dc6a2cd787486fddcb4485b223e7b047a728cb0b", "state": "VERIFIED", "summary": {"facts": 0.774038, "frozen": 0.312352, "full": 0.797187, "retrieval": 0.776353, "rules_nodetect": 0.793714}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:36:27Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:39Z", "cpu_seconds": 7.2188431339999966, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6eb0d987297d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 7, "n": 4}, "prereg_hash": "8e094c2ba6dcc5fe59955623d50a7a25ded6c5f555d5162fa0849139a5c0a7b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6eb0d987297d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:26:15Z", "seed": 78278441, "treatment": "frontier"}, "priority": 0.1817430556888729, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b946a375d9bc73d2e3377ae07557da07646f4d56c02d2f624903b1892786f8ff", "run_seconds": 7.491772413253784, "seed": 78278441, "started_at": "2026-11-22T04:26:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.100598, -0.035558], "effect": -0.068078, "evaluated_at": "2026-11-22T04:36:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6eb0d987297d", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.40625, 0.870833333, 0.480208333, 0.316666667], "frontier": [0.278571429, 0.691666667, 0.429166667, 0.270967742], "goal": [0.290178571, 0.566666667, 0.425, 0.269354839], "hypdisc": [0.372321429, 0.570833333, 0.440104167, 0.292473118], "infogain": [0.278571429, 0.691666667, 0.429166667, 0.270967742], "prederr": [0.278571429, 0.691666667, 0.429166667, 0.270967742], "random": [0.378571429, 0.783333333, 0.479166667, 0.301612903], "risk": [0.278571429, 0.691666667, 0.429166667, 0.270967742]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "8e094c2ba6dcc5fe59955623d50a7a25ded6c5f555d5162fa0849139a5c0a7b6", "primary_control": "random", "result_hash": "b946a375d9bc73d2e3377ae07557da07646f4d56c02d2f624903b1892786f8ff", "signature": "6fd6fa969c3b6779770d4216c86d4e5f6e9226e786502854f1095d5d0f7e58d727c67c5215870f6196d6bf222e1a7a1b236d77526e89d8e0c020b765501ed505", "state": "VERIFIED", "summary": {"count": 0.51849, "frontier": 0.417593, "goal": 0.3878, "hypdisc": 0.418933, "infogain": 0.417593, "prederr": 0.417593, "random": 0.485671, "risk": 0.417593}, "ties": 0, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:36:32Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:18Z", "cpu_seconds": 0.31150058500000455, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-07525dabca33", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "6e88ae88324eeb9c81eedbb64980856c8e5cdb4095231fb871e7ef45e4bc7d1e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-07525dabca33", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:15Z", "seed": 596699352, "treatment": "full"}, "priority": 0.19137940986936966, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "133a78d37a20ca7a3bb78d53d6880e554f59b974366931c833c5ac2785d4dba5", "run_seconds": 0.36788272857666016, "seed": 596699352, "started_at": "2026-11-22T04:26:18Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-07525dabca33", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.75, 0.5, 0.75, 0.5], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "6e88ae88324eeb9c81eedbb64980856c8e5cdb4095231fb871e7ef45e4bc7d1e", "primary_control": "e141_best", "result_hash": "133a78d37a20ca7a3bb78d53d6880e554f59b974366931c833c5ac2785d4dba5", "signature": "34af971103969f15711924563dd505aa91f334efbd6d4e15023c166f6194ad6e0dc154bdbc31735d07bd826e145dfa146fb5b539ec7dcbdd99fcc59d52492601", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.625, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:20Z", "cpu_seconds": 0.9926122509999971, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-309927e9c37f", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "0108024998fce436abb67053e28ac2e40f6583a6ab323c504468a7b8d560b2ee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-309927e9c37f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:15Z", "seed": 1410623049, "treatment": "full"}, "priority": 0.18880241473108333, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-6c02324cee30", "result_hash": "7ba04e014dcafa176158687424625dbf4c5135a854b8839d3c53dfef31e1a26c", "run_seconds": 1.0365707874298096, "seed": 1410623049, "started_at": "2026-11-22T04:26:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.24, 0.74], "effect": 0.25, "evaluated_at": "2026-11-22T04:36:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-309927e9c37f", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [0.5, 0.0, 1.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "0108024998fce436abb67053e28ac2e40f6583a6ab323c504468a7b8d560b2ee", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-6c02324cee30", "result_hash": "7ba04e014dcafa176158687424625dbf4c5135a854b8839d3c53dfef31e1a26c", "signature": "81ce6099d42d0555ec0087d99b63cea077d033f8e6df326ac883541c60f50405fd86c0e77376102015f6b22067b2520b091807fdbdda0f3eec357f72d5666b0b", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 1.0, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:36:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:24Z", "cpu_seconds": 0.5111268960000075, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-704e390f2ac6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "f38bed136106587d487bd394ed883d5aff2a56902f8c100c25af487d66d54638", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-704e390f2ac6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:15Z", "seed": 1192872439, "treatment": "full"}, "priority": 0.1881252765526595, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "10d0da96c4823fc80caeac18852ebdf7b324a32700693fba381c2248b0f711a4", "run_seconds": 0.5267822742462158, "seed": 1192872439, "started_at": "2026-11-22T04:26:23Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-704e390f2ac6", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.25, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 0.25, 1.0, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f38bed136106587d487bd394ed883d5aff2a56902f8c100c25af487d66d54638", "primary_control": "e141_best", "result_hash": "10d0da96c4823fc80caeac18852ebdf7b324a32700693fba381c2248b0f711a4", "signature": "153c51d537fa8c4e4aa1f090e1c64611e8f35e0e18cfe557983a35e527fdbebf0ea5c6a3e6441c79951524dc271b3397d581911a9124b71b6c13c68e474cd40e", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:28Z", "cpu_seconds": 1.3198349729999848, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a3f9e8548e3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "b600371138f7effd4f68205f91a646973c216c6f4b70a4ed2a267fd03f1c0429", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a3f9e8548e3b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:15Z", "seed": 307353521, "treatment": "full"}, "priority": 0.18379620779202452, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e73e5cd01ad09d46ef6781b4b9d7ac6deef927c16cc2482df118cc07a4f1b151", "run_seconds": 1.3265235424041748, "seed": 307353521, "started_at": "2026-11-22T04:26:26Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.200042, 0.200042], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-a3f9e8548e3b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.5, 1.0, 1.0], "fixed_policy": [0.75, 0.0, 1.0, 1.0], "full": [0.75, 0.75, 1.0, 0.75], "no_world_model": [0.25, 0.5, 1.0, 1.0], "random": [0.5, 0.5, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "b600371138f7effd4f68205f91a646973c216c6f4b70a4ed2a267fd03f1c0429", "primary_control": "e141_best", "result_hash": "e73e5cd01ad09d46ef6781b4b9d7ac6deef927c16cc2482df118cc07a4f1b151", "signature": "c018208c97f8974f7e6fa75aa9dca1806b18c9817d981fd5474f01eca9e0da57a4420a48ff5ed306776eb73fa81c07abdae3ac4d1b5124670915522802e9010f", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.6875, "full": 0.8125, "no_world_model": 0.6875, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:36:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:30Z", "cpu_seconds": 0.8975293430000022, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e8d164fd79ab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "0542f5278f24fd39cfabd3feb305b67588f778e4a65e2f6ee990fe4d8c15732b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e8d164fd79ab", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:15Z", "seed": 60077819, "treatment": "full"}, "priority": 0.18250749596549226, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e63a0e705fe1e53351c82ec116c5edf78ee16d1646dc9ad7c33614ed7dfe32d5", "run_seconds": 0.9448268413543701, "seed": 60077819, "started_at": "2026-11-22T04:26:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.233549, 0.516451], "effect": 0.375, "evaluated_at": "2026-11-22T04:36:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e8d164fd79ab", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.25, 0.75, 0.25], "fixed_policy": [1.0, 0.25, 1.0, 1.0], "full": [1.0, 0.75, 1.0, 0.75], "no_world_model": [0.75, 0.75, 0.75, 0.25], "random": [1.0, 0.75, 1.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "0542f5278f24fd39cfabd3feb305b67588f778e4a65e2f6ee990fe4d8c15732b", "primary_control": "e141_best", "result_hash": "e63a0e705fe1e53351c82ec116c5edf78ee16d1646dc9ad7c33614ed7dfe32d5", "signature": "d7d306b38a99aba959fe1faae2c90658c45397d58013f38190a9d25b17ef519bc4cc11a5cc11348d1a4bb86e7c4bc2a5716faf3e29411f30f29ae84bd7774f09", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.8125, "full": 0.875, "no_world_model": 0.625, "random": 0.8125}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:36:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:16Z", "cpu_seconds": 0.5878555980000044, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-dc6a43c3568e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "6d252f1a76ba804fab80fc0ad0a556615d4240cc8ed886ac8b772672a2514411", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-dc6a43c3568e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:26:15Z", "seed": 691779486, "treatment": "full"}, "priority": 0.1921638835512705, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c17e07d45c2ca547c856a1818863acbc12560bad2ed9e3b130f6dc5e474b87cd", "run_seconds": 0.613806962966919, "seed": 691779486, "started_at": "2026-11-22T04:26:15Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.005821, 0.152356], "effect": 0.073267, "evaluated_at": "2026-11-22T04:36:28Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-dc6a43c3568e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.455882353, 0.538461538, 0.570921986, 0.108333333], "frozen": [0.073529412, 0.102564103, 0.092198582, 0.080555556], "full": [0.529411765, 0.685897436, 0.531914894, 0.219444444], "retrieval": [0.637254902, 0.826923077, 0.641843972, 0.383333333], "rules_nodetect": [0.235294118, 0.384615385, 0.535460993, 0.236111111]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "6d252f1a76ba804fab80fc0ad0a556615d4240cc8ed886ac8b772672a2514411", "primary_control": "facts", "result_hash": "c17e07d45c2ca547c856a1818863acbc12560bad2ed9e3b130f6dc5e474b87cd", "signature": "a4c666079e9a2707502f57b025e71f885481dc52ea663e5c17bd49e3fd59cbfc2881990d47479157e42b8c2140256fd08e3716df8c8914b01492eba7d217a708", "state": "VERIFIED", "summary": {"facts": 0.4184, "frozen": 0.087212, "full": 0.491667, "retrieval": 0.622339, "rules_nodetect": 0.34787}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:36:28Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:26Z", "cpu_seconds": 0.674044533, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-9d9ac619111c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "8a78db80f9a6b32969867f947f82d1b8ed5b34b09d26386811f7d19bf36c6704", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-9d9ac619111c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:26:15Z", "seed": 118360888, "treatment": "gated"}, "priority": 0.18396501503585036, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "450043b9c8e07de7485f61e1c4eb71869ff39c0dff7f7f6122110546dd3e0aeb", "run_seconds": 0.7257237434387207, "seed": 118360888, "started_at": "2026-11-22T04:26:25Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-9d9ac619111c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.700980392, 0.724637681, 0.708333333, 0.793103448], "gated": [0.700980392, 0.724637681, 0.708333333, 0.793103448], "naive": [0.666666667, 0.702898551, 0.708333333, 0.75862069], "scratch": [0.700980392, 0.724637681, 0.708333333, 0.793103448], "ungated": [0.700980392, 0.702898551, 0.708333333, 0.701149425]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8a78db80f9a6b32969867f947f82d1b8ed5b34b09d26386811f7d19bf36c6704", "primary_control": "scratch", "result_hash": "450043b9c8e07de7485f61e1c4eb71869ff39c0dff7f7f6122110546dd3e0aeb", "signature": "3c9ce28ee4585317b1e82be7e4d2ef84c907532b75ecbcd81d060f17fa2ab1cfe9e0d4c510e95d06caaff23d6fea44437c6126115007432a8d53e211aadf6d01", "state": "PARTIAL", "summary": {"compressed": 0.731764, "gated": 0.731764, "naive": 0.70913, "scratch": 0.731764, "ungated": 0.70334}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:21Z", "cpu_seconds": 0.429461922999991, "created_at": "2026-11-22T04:26:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-f81a4587ee2b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "prereg_hash": "79ce570dcdf81572f55b14578424ed7f22624fc2b3073e33e58396c481bb47e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-f81a4587ee2b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:26:15Z", "seed": 601140896, "treatment": "gated"}, "priority": 0.18820660797548777, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "a38c5672f9332201d68c8176a9d6e46e0c2a7042dd81151ccb5cd1a58c65f07c", "run_seconds": 0.44652390480041504, "seed": 601140896, "started_at": "2026-11-22T04:26:21Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:29Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-f81a4587ee2b", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 0.833333333, 1.0], "gated": [1.0, 1.0, 0.833333333, 1.0], "naive": [0.588888889, 0.623655914, 0.301075269, 0.744444444], "scratch": [1.0, 1.0, 0.833333333, 1.0], "ungated": [0.655555556, 0.698924731, 0.370967742, 0.733333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "79ce570dcdf81572f55b14578424ed7f22624fc2b3073e33e58396c481bb47e2", "primary_control": "scratch", "result_hash": "a38c5672f9332201d68c8176a9d6e46e0c2a7042dd81151ccb5cd1a58c65f07c", "signature": "5960f71db534149f7fe9600015e75ecff9dae111cfb2e39ee82e4d95425be9f178e839568f88097171d6b756cb0faf3347ef1e491d345a79f7129b81771aad02", "state": "PARTIAL", "summary": {"compressed": 0.958333, "gated": 0.958333, "naive": 0.564516, "scratch": 0.958333, "ungated": 0.614695}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:36:29Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:50Z", "cpu_seconds": 0.6337732890000041, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-102527135ce0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "f297dcd2b3d76aa1322517dd04f26e9f6d7240cc6c1f599a1a69fc7af9f8cec9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-102527135ce0", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:32Z", "seed": 1175396272, "treatment": "full"}, "priority": 0.18270829978170217, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "fb8a6cc43e834c83377838a3c210e1cad71a550b11ad522c3f88319369f1a102", "run_seconds": 0.6624777317047119, "seed": 1175396272, "started_at": "2026-11-22T04:27:50Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-102527135ce0", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.0, 1.0], "fixed_policy": [0.0, 0.25, 0.0, 0.25], "full": [1.0, 1.0, 0.0, 1.0], "no_world_model": [1.0, 1.0, 0.0, 1.0], "random": [1.0, 1.0, 0.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "f297dcd2b3d76aa1322517dd04f26e9f6d7240cc6c1f599a1a69fc7af9f8cec9", "primary_control": "e141_best", "result_hash": "fb8a6cc43e834c83377838a3c210e1cad71a550b11ad522c3f88319369f1a102", "signature": "e60d6f615c68bc6fa56057fe1b31844e6106eb915ffc7f5636839b393e226242193f0c0b4c48332b34e87558d333928414dfe3e9f678eb6b092968fa6ae03601", "state": "PARTIAL", "summary": {"e141_best": 0.75, "fixed_policy": 0.125, "full": 0.75, "no_world_model": 0.75, "random": 0.5}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:52Z", "cpu_seconds": 0.3526521520000001, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-2276754df45b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "566cfb15a7b013bf13991c93f4a9f4b04b762881d0f59a50a3cb1fdda46db758", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-2276754df45b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:32Z", "seed": 895359228, "treatment": "full"}, "priority": 0.18255290577166347, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2a548284ac1d0874431e262eaedf4c6429ca7c6dd92248b02647ee35d3f055af", "run_seconds": 0.48793506622314453, "seed": 895359228, "started_at": "2026-11-22T04:27:52Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-2276754df45b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.75, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "566cfb15a7b013bf13991c93f4a9f4b04b762881d0f59a50a3cb1fdda46db758", "primary_control": "e141_best", "result_hash": "2a548284ac1d0874431e262eaedf4c6429ca7c6dd92248b02647ee35d3f055af", "signature": "8b8be68891453b9f17cc4c9f2f3116d8b20327acc9bd121098a301fd7b573a1c1b0879cafceeb9f3506d9ba117f7b9cd452cfa19479dbcd16766c853b8688300", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.3125, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:48Z", "cpu_seconds": 0.3751157590000105, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4a1f8f60c73d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "adedfc19e55aed930a582933fa8fa1d55912f3f9115c85faf6c5d7bf5d1bddf6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4a1f8f60c73d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:32Z", "seed": 54887632, "treatment": "full"}, "priority": 0.18536052294731406, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2f9590136cc793fbea0a97802dd1d82fe005a54a2bf2b3b6ea4ec5d4dd5c4ddb", "run_seconds": 0.4854457378387451, "seed": 54887632, "started_at": "2026-11-22T04:26:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:36:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-4a1f8f60c73d", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 0.0, 0.0, 0.25], "fixed_policy": [0.0, 0.0, 0.25, 0.0], "full": [0.5, 0.0, 0.0, 0.5], "no_world_model": [0.5, 0.0, 0.0, 0.25], "random": [0.0, 0.0, 0.0, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "adedfc19e55aed930a582933fa8fa1d55912f3f9115c85faf6c5d7bf5d1bddf6", "primary_control": "e141_best", "result_hash": "2f9590136cc793fbea0a97802dd1d82fe005a54a2bf2b3b6ea4ec5d4dd5c4ddb", "signature": "aa37241a10d1ce3e41fecad5c403396f2d65932642e21513907717af4b9d2d508fc71abe5970911c4a80b4eb425e8a34de729bcc3fec91a926e001c1d7317c0e", "state": "VERIFIED", "summary": {"e141_best": 0.1875, "fixed_policy": 0.0625, "full": 0.25, "no_world_model": 0.1875, "random": 0.125}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:36:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:44Z", "cpu_seconds": 0.5464980259999948, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-62cb71da0110", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "9804f27e685c6954d3c68b04880700df44ba2f8948f576c7f3ed524e60218a47", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-62cb71da0110", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:32Z", "seed": 1301797593, "treatment": "full"}, "priority": 0.18936978046371414, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "13793c795b494605e7793dc2dae2daf1a8bb6e1c6e8d4775947ea9af67453434", "run_seconds": 0.5693166255950928, "seed": 1301797593, "started_at": "2026-11-22T04:26:44Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:36:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-62cb71da0110", "family": "ow_goal", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"fixed_policy": [1.0, 0.25], "full": [1.0, 0.6875], "no_world_model": [1.0, 0.9375], "random": [1.0, 0.75]}, "prereg_hash": "9804f27e685c6954d3c68b04880700df44ba2f8948f576c7f3ed524e60218a47", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "13793c795b494605e7793dc2dae2daf1a8bb6e1c6e8d4775947ea9af67453434", "signature": "2c918324b74f79446df36a31c7617df11344a8bd8f7f7fe30225a6102ef46fc9f09f148b5d71157b95153a98e63eeb5ab559bc188dad37d9021476499eb9080f", "state": "REJECTED", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 0.6875, "no_world_model": 0.9375, "random": 0.75}}, "verified_at": "2026-11-22T04:36:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:52Z", "cpu_seconds": 0.5268841449999999, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-fcb932bed00c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "4e51d14408fc5f7be4af72bdaeab893e54faa4b43c21391c68b1a9a3ae9de1ba", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-fcb932bed00c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:32Z", "seed": 27617072, "treatment": "full"}, "priority": 0.18025424277423213, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6c0a8a7c4dbc13362ab101f51440c6092e7b007f908e8656e8fdd18589f77005", "run_seconds": 0.6357123851776123, "seed": 27617072, "started_at": "2026-11-22T04:27:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.555, 0.18], "effect": -0.1875, "evaluated_at": "2026-11-22T04:37:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-fcb932bed00c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.75, 0.75, 0.5], "fixed_policy": [0.5, 0.0, 0.25, 0.0], "full": [0.5, 0.25, 1.0, 0.5], "no_world_model": [0.5, 0.5, 0.75, 0.5], "random": [1.0, 0.75, 0.25, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "4e51d14408fc5f7be4af72bdaeab893e54faa4b43c21391c68b1a9a3ae9de1ba", "primary_control": "e141_best", "result_hash": "6c0a8a7c4dbc13362ab101f51440c6092e7b007f908e8656e8fdd18589f77005", "signature": "a7791399b6ad24025a83950dc5016eaab4efa264325b643eb8c2f7ecedc6334647206d86e8519ca17ad4b16e586464db6ad0a6c05be96c6176c46a4a6a1f9001", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.1875, "full": 0.5625, "no_world_model": 0.5625, "random": 0.625}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:42Z", "cpu_seconds": 1.535291744999995, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-84855be0074d", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "995d0d80c1d5d634f76ad1376ae931c5be4619cddc883478990bcfc6f7dfa401", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-84855be0074d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:26:32Z", "seed": 1274995413, "treatment": "full"}, "priority": 0.19981348000256385, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "05f954f8a665f143f5a6ed6195fca6b1355824b41036cf1b7a44ea106f4850db", "run_seconds": 1.8777210712432861, "seed": 1274995413, "started_at": "2026-11-22T04:26:41Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:36:32Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-84855be0074d", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.81875], "frozen": [1.0, 0.3625], "full": [1.0, 0.927083], "retrieval": [1.0, 0.939583], "rules_nodetect": [1.0, 0.927083]}, "prereg_hash": "995d0d80c1d5d634f76ad1376ae931c5be4619cddc883478990bcfc6f7dfa401", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "05f954f8a665f143f5a6ed6195fca6b1355824b41036cf1b7a44ea106f4850db", "signature": "e40e6ed9e3d2751387b13bddad9956c90f28bf8ef691432fd5d6c45d5bce8288b97109e1b839bcc177324ad8951d94032be43c5167f6eaad270723e6a5c67c0c", "state": "REJECTED", "summary": {"facts": 0.81875, "frozen": 0.3625, "full": 0.927083, "retrieval": 0.939583, "rules_nodetect": 0.927083}}, "verified_at": "2026-11-22T04:36:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:03Z", "cpu_seconds": 0.4158594919999814, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-eb3207863309", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "484150e4fa26156daf3b8aed83dd7dea3d96e79e9a2aba39fca2a98cab1572b7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-eb3207863309", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:26:32Z", "seed": 252192841, "treatment": "full"}, "priority": 0.18410642165283103, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "34db898925e46cc52eeca8090f24b022142ea9f44dc9df9b763f4e89aa61ef06", "run_seconds": 0.5354602336883545, "seed": 252192841, "started_at": "2026-11-22T04:27:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013173, 0.073173], "effect": 0.03, "evaluated_at": "2026-11-22T04:36:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-eb3207863309", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.88, 1.0, 1.0, 0.973333333], "frozen": [0.0, 0.0, 0.2, 0.0], "full": [0.973333333, 1.0, 1.0, 1.0], "retrieval": [0.973333333, 1.0, 1.0, 1.0], "rules_nodetect": [0.973333333, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "484150e4fa26156daf3b8aed83dd7dea3d96e79e9a2aba39fca2a98cab1572b7", "primary_control": "facts", "result_hash": "34db898925e46cc52eeca8090f24b022142ea9f44dc9df9b763f4e89aa61ef06", "signature": "739eedde36085d12b78eab73feb4400d483f0ab2a032ab74e8392f1a972672ffb4151697624838e5661b4ae5e7e6b064c494d2207f0824f8ac0f08c12ba26d0b", "state": "VERIFIED", "summary": {"facts": 0.963333, "frozen": 0.05, "full": 0.993333, "retrieval": 0.993333, "rules_nodetect": 0.993333}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:36:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:25Z", "cpu_seconds": 43.331611388, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-14e0d936a50a", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "6ecc4159eb5786d78302af83c9c157a42b1a2b9209d604449b0eb3bd790cac9a", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-14e0d936a50a", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:26:32Z", "seed": 910872348, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "2fc1207ff5eb8320986def82ede48c4777a9ed3be17665aac94dec90309f6ad0", "run_seconds": 52.40093779563904, "seed": 910872348, "started_at": "2026-11-22T04:26:33Z", "state": "PARTIAL", "verdict": {"accepted_generations": {"compressed": 0, "inherit_mem": 0, "inherit_nomem": 1, "random_search": 1, "reset_mem": 0, "retrieval": 0, "rootcause": 1, "xfer_mem": 1}, "better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:36:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-14e0d936a50a", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "random_search": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 10, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 20, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.233333333, 0.170138889, 0.958333333, 0.446296296, 0.262352941, 0.875, 0.871212121, 0.615277778, 0.921568627, 0.73989899], "compressed": [0.233333333, 0.170138889, 0.958333333, 0.446296296, 0.262352941, 0.875, 0.871212121, 0.615277778, 0.921568627, 0.73989899], "inherit_mem": [0.233333333, 0.170138889, 0.958333333, 0.446296296, 0.262352941, 0.875, 0.871212121, 0.615277778, 0.921568627, 0.73989899], "inherit_nomem": [0.233333333, 0.996527778, 0.958333333, 0.446296296, 0.262352941, 0.875, 0.871212121, 0.615277778, 0.74754902, 0.73989899], "random_search": [0.233333333, 0.996527778, 0.958333333, 0.424074074, 0.262352941, 0.857954545, 0.871212121, 0.6125, 0.921568627, 0.73989899], "reset_mem": [0.233333333, 0.170138889, 0.958333333, 0.446296296, 0.262352941, 0.875, 0.871212121, 0.615277778, 0.921568627, 0.73989899], "retrieval": [0.233333333, 0.170138889, 0.958333333, 0.446296296, 0.262352941, 0.875, 0.871212121, 0.615277778, 0.921568627, 0.73989899], "rootcause": [0.920833333, 0.3125, 1.0, 0.733333333, 0.276470588, 0.875, 0.992424242, 0.651388889, 0.921568627, 0.787878788], "xfer_mem": [0.233333333, 0.996527778, 0.958333333, 0.446296296, 0.262352941, 0.875, 0.871212121, 0.615277778, 0.74754902, 0.73989899]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "per_generation_sealed": [0.609341, 0.609341, 0.609341, 0.609341], "prereg_hash": "6ecc4159eb5786d78302af83c9c157a42b1a2b9209d604449b0eb3bd790cac9a", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "compressed", "xfer_mem"], "result_hash": "2fc1207ff5eb8320986def82ede48c4777a9ed3be17665aac94dec90309f6ad0", "signature": "202340ab552ea6859727ccf24d3fd67a3169cad556ac7d86a44c882afdb2fdbe36cab75fb5e4d0a201754824887c569d2f1679e7f96bb907ab7b5b00db1ed80a", "state": "PARTIAL", "summary": {"G0": 0.609341, "compressed": 0.609341, "inherit_mem": 0.609341, "inherit_nomem": 0.674578, "random_search": 0.687776, "reset_mem": 0.609341, "retrieval": 0.609341, "rootcause": 0.74714, "xfer_mem": 0.674578}, "ties": 10, "treatment": "inherit_mem", "wins": 0}, "verified_at": "2026-11-22T04:37:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:36Z", "cpu_seconds": 0.9195113840000033, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-02aa993c11ad", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "4012e922109a50a59e2ce6c30433d5eeea0f7dc8f5e0b22ebcf1c760d3d59feb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-02aa993c11ad", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:26:32Z", "seed": 968459113, "treatment": "gated"}, "priority": 0.31778533143312215, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "5e8bd9d96ba8bc7a2f8a942246baf6b097d0d536d856cbaa88cb03a2b932f9c2", "run_seconds": 0.9498500823974609, "seed": 968459113, "started_at": "2026-11-22T04:26:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007814, 0.12599], "effect": 0.059088, "evaluated_at": "2026-11-22T04:36:30Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-02aa993c11ad", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.339285714, 0.509615385, 0.45, 0.540322581], "gated": [0.339285714, 0.625, 0.45, 0.540322581], "naive": [0.6875, 0.711538462, 0.62, 0.661290323], "scratch": [0.339285714, 0.509615385, 0.45, 0.419354839], "ungated": [0.5, 0.711538462, 0.53, 0.661290323]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "4012e922109a50a59e2ce6c30433d5eeea0f7dc8f5e0b22ebcf1c760d3d59feb", "primary_control": "scratch", "result_hash": "5e8bd9d96ba8bc7a2f8a942246baf6b097d0d536d856cbaa88cb03a2b932f9c2", "signature": "c7d61dd9e54849e31863b07888414a006b83b51073f48db009237439962d7a381eaeee17950b414f5016bb925767eb64764978824f1fbc04378424d5b66d770e", "state": "VERIFIED", "summary": {"compressed": 0.459806, "gated": 0.488652, "naive": 0.670082, "scratch": 0.429564, "ungated": 0.600707}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:36:30Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:47Z", "cpu_seconds": 0.39338614800000027, "created_at": "2026-11-22T04:26:32Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-0e00d8075f7c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "prereg_hash": "74d2b7ca3f0615b290eab74dd34da412b0365caaaaf6ba082c04db7cd2802caa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-0e00d8075f7c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|causal_switch|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:26:32Z", "seed": 1118500802, "treatment": "gated"}, "priority": 0.18321978360611352, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "209a00749ff536f62dc456a20dc82309a76ea85c0a21ff1063e4b9999b5104f3", "run_seconds": 0.40946340560913086, "seed": 1118500802, "started_at": "2026-11-22T04:27:46Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-0e00d8075f7c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [1.0, 1.0, 1.0, 1.0], "gated": [1.0, 1.0, 1.0, 1.0], "naive": [1.0, 1.0, 0.709677419, 0.877777778], "scratch": [1.0, 1.0, 1.0, 1.0], "ungated": [1.0, 0.930107527, 0.784946237, 0.877777778]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "74d2b7ca3f0615b290eab74dd34da412b0365caaaaf6ba082c04db7cd2802caa", "primary_control": "scratch", "result_hash": "209a00749ff536f62dc456a20dc82309a76ea85c0a21ff1063e4b9999b5104f3", "signature": "020f105ed72bb4ea7fd8bb33e51d92dec84c9f551c77f731a2abd7f6e1e35552ab9cd11b1b2ddcc2a3564e06386063be7aeadbc2999082bf40e2345d79b3e502", "state": "PARTIAL", "summary": {"compressed": 1.0, "gated": 1.0, "naive": 0.896864, "scratch": 1.0, "ungated": 0.898208}, "ties": 4, "treatment": "gated", "wins": 0}, "verified_at": "2026-11-22T04:37:45Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:11Z", "cpu_seconds": 8.938764355000018, "created_at": "2026-11-22T04:26:48Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7e0ca38264e1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "5ce37d492dd59408a8f548894d12ba0087123a906ff4c4dcd154543266ab8330", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7e0ca38264e1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:26:48Z", "seed": 1680308205, "treatment": "frontier"}, "priority": 0.1848388939344253, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "567c2b6afda4711adf9e87bf4ae9430348b7e7a278237e026551004abc21eb60", "run_seconds": 10.505763053894043, "seed": 1680308205, "started_at": "2026-11-22T04:27:00Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.051979, 0.177022], "effect": 0.062522, "evaluated_at": "2026-11-22T04:36:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-7e0ca38264e1", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.538297872, 0.491489362, 0.485365854, 0.58021978], "frontier": [0.465957447, 0.382978723, 0.42195122, 0.628571429], "goal": [0.427659574, 0.342553191, 0.446341463, 0.415384615], "hypdisc": [0.153191489, 0.395744681, 0.509756098, 0.41978022], "infogain": [0.465957447, 0.382978723, 0.414634146, 0.650549451], "prederr": [0.465957447, 0.382978723, 0.414634146, 0.556043956], "random": [0.45106383, 0.438297872, 0.351219512, 0.408791209], "risk": [0.465957447, 0.382978723, 0.42195122, 0.628571429]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "5ce37d492dd59408a8f548894d12ba0087123a906ff4c4dcd154543266ab8330", "primary_control": "random", "result_hash": "567c2b6afda4711adf9e87bf4ae9430348b7e7a278237e026551004abc21eb60", "signature": "e023cd678efe6e08e599fd3d8e42f8d0aefd304940b956ff557d342a920c335b7a8d35aef9ac7b2fca872e6d096beb0bca2bddd31d4a13e075b9918c833f8306", "state": "VERIFIED", "summary": {"count": 0.523843, "frontier": 0.474865, "goal": 0.407985, "hypdisc": 0.369618, "infogain": 0.47853, "prederr": 0.454904, "random": 0.412343, "risk": 0.474865}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:36:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:55Z", "cpu_seconds": 0.5498877149999828, "created_at": "2026-11-22T04:26:48Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bb6aefac6421", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "ee8580ffb9627b4ba0da600f102361dae8adc98f00d6769930561231fecdf5f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bb6aefac6421", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:26:48Z", "seed": 513773156, "treatment": "full"}, "priority": 0.1862568242950274, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "94e50e863bfb313c707cca4ecaa6157a7d43f278541c2b945d499b0fe73c6136", "run_seconds": 0.7791967391967773, "seed": 513773156, "started_at": "2026-11-22T04:26:55Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-bb6aefac6421", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 0.5], "fixed_policy": [0.0, 0.0, 0.5, 0.0], "full": [1.0, 1.0, 1.0, 0.5], "no_world_model": [1.0, 1.0, 0.0, 0.0], "random": [0.0, 1.0, 0.75, 0.5]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "ee8580ffb9627b4ba0da600f102361dae8adc98f00d6769930561231fecdf5f4", "primary_control": "e141_best", "result_hash": "94e50e863bfb313c707cca4ecaa6157a7d43f278541c2b945d499b0fe73c6136", "signature": "1b4cb36e866c8d3fa30e958ec40bad30b963cff28cc223fd1811c4d1ebba0055b1e258c47ad4586fe9e0eab5e17a476ba193357c1241edf869d5fbb2251cba04", "state": "PARTIAL", "summary": {"e141_best": 0.875, "fixed_policy": 0.125, "full": 0.875, "no_world_model": 0.5, "random": 0.5625}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:50Z", "cpu_seconds": 0.12891332899999952, "created_at": "2026-11-22T04:26:48Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-51d397bfd22f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "1ea3ec5a671c46792e993023adb00fac28a4f20f38ffdb6687d21d73daf6b2c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-51d397bfd22f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:26:48Z", "seed": 547261133, "treatment": "full"}, "priority": 0.19212345620997034, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2a15fb3cb444a7d450d8accfcd7ba60a58d0eefc63e0b23a3f0595f1bd5cf45a", "run_seconds": 0.28075718879699707, "seed": 547261133, "started_at": "2026-11-22T04:26:49Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:36:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-51d397bfd22f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 1.0, 1.0, 1.0], "frozen": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 1.0, 1.0, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "1ea3ec5a671c46792e993023adb00fac28a4f20f38ffdb6687d21d73daf6b2c6", "primary_control": "facts", "result_hash": "2a15fb3cb444a7d450d8accfcd7ba60a58d0eefc63e0b23a3f0595f1bd5cf45a", "signature": "07043f5aca8d4c4b9a6a63c6c437eece7c988b708c920b5591f3bfce8beed75e09b556606aef18ad238a3eb9d874fc356da4436cf337c4bf0263d581774b9e02", "state": "PARTIAL", "summary": {"facts": 1.0, "frozen": 1.0, "full": 1.0, "retrieval": 1.0, "rules_nodetect": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:36:33Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:26:58Z", "cpu_seconds": 3.889860600999981, "created_at": "2026-11-22T04:26:48Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-2fdc5822995c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "166360d97d599640d27a230705841f67dbda35ff5b879ec32827b156d50ff089", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-2fdc5822995c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:26:48Z", "seed": 976884932, "treatment": "gated"}, "priority": 0.18828021368540826, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-a6385ba3503e", "result_hash": "127b904bd0993e30d41484e2b3152af676470b649a89ba283bdf5b80f09110c3", "run_seconds": 4.8375160694122314, "seed": 976884932, "started_at": "2026-11-22T04:26:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.013043, 0.040217], "effect": 0.013587, "evaluated_at": "2026-11-22T04:36:33Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-2fdc5822995c", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.65, 0.510869565, 0.625, 0.673913043], "gated": [0.65, 0.510869565, 0.625, 0.673913043], "naive": [0.65, 0.619565217, 0.701923077, 0.739130435], "scratch": [0.65, 0.510869565, 0.625, 0.619565217], "ungated": [0.566666667, 0.576086957, 0.634615385, 0.739130435]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "166360d97d599640d27a230705841f67dbda35ff5b879ec32827b156d50ff089", "primary_control": "scratch", "replicated": false, "replication_of": "E141-ow_transfer-a6385ba3503e", "result_hash": "127b904bd0993e30d41484e2b3152af676470b649a89ba283bdf5b80f09110c3", "signature": "68d6a75d2575612fb08f024a9a19554f1baeb97d90b280ab56269e84704d507d224612fc21eadb86739f50ee5bc94a5f2077ac29db3ac0c043d64b0135a1dd0a", "state": "VERIFIED", "summary": {"compressed": 0.614946, "gated": 0.614946, "naive": 0.677655, "scratch": 0.601359, "ungated": 0.629125}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:36:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:17Z", "cpu_seconds": 0.5623834359999762, "created_at": "2026-11-22T04:27:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3a33b7eafcba", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "45704ca52cbfa3dea394e161fae20cd1ab210cb7e7c703b57b37c3e534f6df45", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3a33b7eafcba", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:27:04Z", "seed": 2058614677, "treatment": "frontier"}, "priority": 0.1874114227886184, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a58130cf2bed8f3605869bd10a2a35e34780ac5577d50fa558e30413ad77b71e", "run_seconds": 0.6804239749908447, "seed": 2058614677, "started_at": "2026-11-22T04:27:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.05427, 0.06677], "effect": 0.00625, "evaluated_at": "2026-11-22T04:36:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-3a33b7eafcba", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.916666667, 0.775, 0.875, 0.833333333], "frontier": [0.9375, 0.775, 0.875, 0.791666667], "goal": [0.854166667, 0.75, 0.875, 0.770833333], "hypdisc": [0.9375, 0.75, 0.9, 0.833333333], "infogain": [0.958333333, 0.85, 0.975, 0.833333333], "prederr": [0.854166667, 0.85, 0.8, 0.770833333], "random": [0.895833333, 0.7, 0.925, 0.833333333], "risk": [0.9375, 0.775, 0.875, 0.791666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "45704ca52cbfa3dea394e161fae20cd1ab210cb7e7c703b57b37c3e534f6df45", "primary_control": "random", "result_hash": "a58130cf2bed8f3605869bd10a2a35e34780ac5577d50fa558e30413ad77b71e", "signature": "a0679d91a935fd366e22cff020a8ca1c85b95661194913ba8662d8072d89f4382956a7e0c82c85375ecf3bd9c65254ffa1c9ef75ecbd5cdaae7f8010b126580a", "state": "VERIFIED", "summary": {"count": 0.85, "frontier": 0.844792, "goal": 0.8125, "hypdisc": 0.855208, "infogain": 0.904167, "prederr": 0.81875, "random": 0.838542, "risk": 0.844792}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:36:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:33Z", "cpu_seconds": 11.550680678999981, "created_at": "2026-11-22T04:27:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-afba9c89f203", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 2, "n": 4}, "prereg_hash": "8f6db7a60431a4428cd0059c1f445dc596bf6257e47818c8b4ab21e8283c4b88", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-afba9c89f203", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:27:04Z", "seed": 1653037271, "treatment": "frontier"}, "priority": 0.18760236559270504, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-9cc4b1c2604c", "result_hash": "fa3fac4261bfe5e4c7c9c11552754dd2ec90eb784b76743a65cebe186167dbee", "run_seconds": 15.96030330657959, "seed": 1653037271, "started_at": "2026-11-22T04:27:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.00708, 0.117399], "effect": 0.06224, "evaluated_at": "2026-11-22T04:37:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-afba9c89f203", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.516634051, 0.505283757, 0.77007874, 1.0], "frontier": [0.516634051, 0.505283757, 0.743307087, 1.0], "goal": [0.507240705, 0.505283757, 0.628346457, 0.908571429], "hypdisc": [0.487279843, 0.490802348, 0.696062992, 1.0], "infogain": [0.516634051, 0.505283757, 0.743307087, 1.0], "prederr": [0.516634051, 0.505283757, 0.743307087, 1.0], "random": [0.489236791, 0.502935421, 0.644094488, 0.88], "risk": [0.516634051, 0.505283757, 0.743307087, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "8f6db7a60431a4428cd0059c1f445dc596bf6257e47818c8b4ab21e8283c4b88", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-9cc4b1c2604c", "result_hash": "fa3fac4261bfe5e4c7c9c11552754dd2ec90eb784b76743a65cebe186167dbee", "signature": "e70fb147b75ae81ce742ffdc7a5887e8cc13a001d50ac81ccd0c1f1835d389af1a56f44917f6d3d92f0f0e6b4323500ec4adbcbc17321a693e3056f53bc32e00", "state": "VERIFIED", "summary": {"count": 0.697999, "frontier": 0.691306, "goal": 0.637361, "hypdisc": 0.668536, "infogain": 0.691306, "prederr": 0.691306, "random": 0.629067, "risk": 0.691306}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:37:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:11Z", "cpu_seconds": 1.4463965390000055, "created_at": "2026-11-22T04:27:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c43c766bdb38", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "10569561070570a15979509659513a90b50b533d4ba185b7f34acc2e314f0a85", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c43c766bdb38", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:27:04Z", "seed": 348254486, "treatment": "frontier"}, "priority": 0.1877583865297454, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fefe027369da2b95eaaff485c7c5f44b7d45982f27c1ef25691a1428cb2ba3cb", "run_seconds": 2.451267719268799, "seed": 348254486, "started_at": "2026-11-22T04:27:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.03334, 0.03334], "effect": -0.0, "evaluated_at": "2026-11-22T04:36:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c43c766bdb38", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.966666667, 0.916666667, 0.866666667, 0.854166667], "frontier": [0.966666667, 0.916666667, 0.866666667, 0.729166667], "goal": [0.966666667, 0.916666667, 0.866666667, 0.729166667], "hypdisc": [0.966666667, 0.916666667, 0.866666667, 0.770833333], "infogain": [0.966666667, 0.916666667, 0.866666667, 0.729166667], "prederr": [0.966666667, 0.916666667, 0.866666667, 0.729166667], "random": [0.966666667, 0.875, 0.866666667, 0.770833333], "risk": [0.966666667, 0.916666667, 0.866666667, 0.729166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "10569561070570a15979509659513a90b50b533d4ba185b7f34acc2e314f0a85", "primary_control": "random", "result_hash": "fefe027369da2b95eaaff485c7c5f44b7d45982f27c1ef25691a1428cb2ba3cb", "signature": "f49094a3b29fa6d95a2dfca1088170d7a423ab7a04ccc031e723af527608d0c240a9b8c451baaf6412260e1a3247244c0b2c9acfd959226b66eb421624a3c004", "state": "VERIFIED", "summary": {"count": 0.901042, "frontier": 0.869792, "goal": 0.869792, "hypdisc": 0.880208, "infogain": 0.869792, "prederr": 0.869792, "random": 0.869792, "risk": 0.869792}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:36:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:45Z", "cpu_seconds": 0.2958888070000114, "created_at": "2026-11-22T04:27:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b49fb0e940a4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "785bedb1303d4283e4dedf2c1c4eaaa45d03ec29da2d60a4829a858d4e221d4f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b49fb0e940a4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:27:04Z", "seed": 514663747, "treatment": "full"}, "priority": 0.183518559008452, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "914fae8b06cd0f8f2fc64334aad36ad737fefca25667e949fefc576144aa4c32", "run_seconds": 0.4532475471496582, "seed": 514663747, "started_at": "2026-11-22T04:27:44Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:44Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b49fb0e940a4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "785bedb1303d4283e4dedf2c1c4eaaa45d03ec29da2d60a4829a858d4e221d4f", "primary_control": "e141_best", "result_hash": "914fae8b06cd0f8f2fc64334aad36ad737fefca25667e949fefc576144aa4c32", "signature": "6f45d04ee5435ee04f92e548789f81d81e273df2f5589446f3e7883677618a6b7c7e596e158609b8778439edf01c48dfda75e0762d9bb0601252c90f69834603", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:47Z", "cpu_seconds": 2.388660715, "created_at": "2026-11-22T04:27:04Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-879265461c2e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 1, "n": 4}, "prereg_hash": "d85b0012d672b5a86f80f1836a5fcfe5e3e169b464f0a8cb801d039097671ffe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-879265461c2e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:27:04Z", "seed": 1790203535, "treatment": "full"}, "priority": 0.18714362721992275, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d315640e138e03d2fdb5d0e571f7b6bcf0c58ea8f4add23d18e737efb8497951", "run_seconds": 2.574615478515625, "seed": 1790203535, "started_at": "2026-11-22T04:27:44Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.058768, 0.189384], "effect": 0.065308, "evaluated_at": "2026-11-22T04:37:45Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-879265461c2e", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.672727273, 0.806060606, 0.298850575, 0.8], "frozen": [0.0, 0.0, 0.103448276, 0.0], "full": [0.678787879, 0.806060606, 0.554022989, 0.8], "retrieval": [0.903030303, 0.937373737, 0.597701149, 0.949494949], "rules_nodetect": [0.678787879, 0.806060606, 0.554022989, 0.8]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "d85b0012d672b5a86f80f1836a5fcfe5e3e169b464f0a8cb801d039097671ffe", "primary_control": "facts", "result_hash": "d315640e138e03d2fdb5d0e571f7b6bcf0c58ea8f4add23d18e737efb8497951", "signature": "0b4d6ce4c000cf4b795665d09ec439a174841b247bc53af7344eace9dc5e9a914771c9f41014e6433564fb8b741b22fd20324d16884e191b3b875c7194dacc04", "state": "VERIFIED", "summary": {"facts": 0.64441, "frozen": 0.025862, "full": 0.709718, "retrieval": 0.8469, "rules_nodetect": 0.709718}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:37:46Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:22Z", "cpu_seconds": 0.22880578200002333, "created_at": "2026-11-22T04:27:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-fb3e68594b97", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "b1aaac0401baaa63338e685c5b22811dbb61899e2ee815de54ace9b379046cae", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-fb3e68594b97", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:27:19Z", "seed": 2027900110, "treatment": "full"}, "priority": 0.1888070723544839, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "041130fd12361563830e08012752bf24597dfa29f1be14e0892a3460ce2da930", "run_seconds": 0.3052353858947754, "seed": 2027900110, "started_at": "2026-11-22T04:27:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.094139, 0.844139], "effect": 0.375, "evaluated_at": "2026-11-22T04:36:34Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-fb3e68594b97", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 1.0, 0.5, 0.0], "fixed_policy": [0.0, 0.5, 0.75, 0.75], "full": [0.0, 1.0, 1.0, 1.0], "no_world_model": [0.0, 0.75, 1.0, 0.25], "random": [0.0, 0.75, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "b1aaac0401baaa63338e685c5b22811dbb61899e2ee815de54ace9b379046cae", "primary_control": "e141_best", "result_hash": "041130fd12361563830e08012752bf24597dfa29f1be14e0892a3460ce2da930", "signature": "d552f3d0d000f3f8e77ea544dae5f2f6f45e6c98e617f6ec41c198e1cc5ce89436e94aae4f5b873aaacf09ad913b0a244b4d56ddc59b0bf25a9eb3455fc3fa0d", "state": "VERIFIED", "summary": {"e141_best": 0.375, "fixed_policy": 0.5, "full": 0.75, "no_world_model": 0.5, "random": 0.6875}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:36:34Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:31Z", "cpu_seconds": 0.7002618439999999, "created_at": "2026-11-22T04:27:19Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-851abcfaee76", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "69b5dad3c34722659b4f485c964ca19a9302c10bea144ec69d1d60bf882903e6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-851abcfaee76", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:27:19Z", "seed": 1099474162, "treatment": "full"}, "priority": 0.18716179919920178, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ea32b58014450a08634f7a9b8d39b2636e01fcb1517bf8f10233ea06e53b73de", "run_seconds": 0.9274201393127441, "seed": 1099474162, "started_at": "2026-11-22T04:27:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.025, 0.077083], "effect": 0.026042, "evaluated_at": "2026-11-22T04:37:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-851abcfaee76", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.875, 1.0, 1.0], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 0.979166667, 1.0, 1.0], "retrieval": [1.0, 0.916666667, 1.0, 1.0], "rules_nodetect": [1.0, 0.979166667, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "69b5dad3c34722659b4f485c964ca19a9302c10bea144ec69d1d60bf882903e6", "primary_control": "facts", "result_hash": "ea32b58014450a08634f7a9b8d39b2636e01fcb1517bf8f10233ea06e53b73de", "signature": "1f92673e9cc152c165d5cfb68c44037ab4cf1db2ef374513704e3a31209491e6930c0ae93aa48d11432fe1c956e96d6330ab38e421ef61de8977ca4f5ca3e20b", "state": "VERIFIED", "summary": {"facts": 0.96875, "frozen": 0.0, "full": 0.994792, "retrieval": 0.979167, "rules_nodetect": 0.994792}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:31Z", "cpu_seconds": 0.36028602900000806, "created_at": "2026-11-22T04:27:19Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ab8a13f06fb9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "prereg_hash": "57f22c76817e6001dafb422014feb76d190fe001a1166c0ef264c45fa186273f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ab8a13f06fb9", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|causal_switch|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:27:19Z", "seed": 1871923382, "treatment": "gated"}, "priority": 0.1879721643053226, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "1b7d7679123ec8306bdbe45e65412b15ba10e85182d19dac7e00e59334e5148c", "run_seconds": 0.6220674514770508, "seed": 1871923382, "started_at": "2026-11-22T04:27:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016466, 0.091466], "effect": 0.0375, "evaluated_at": "2026-11-22T04:37:41Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-ab8a13f06fb9", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.766666667, 0.533333333, 0.5, 0.516666667], "gated": [0.766666667, 0.533333333, 0.533333333, 0.633333333], "naive": [0.316666667, 0.133333333, 0.516666667, 0.516666667], "scratch": [0.766666667, 0.533333333, 0.5, 0.516666667], "ungated": [0.316666667, 0.116666667, 0.516666667, 0.516666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "57f22c76817e6001dafb422014feb76d190fe001a1166c0ef264c45fa186273f", "primary_control": "scratch", "result_hash": "1b7d7679123ec8306bdbe45e65412b15ba10e85182d19dac7e00e59334e5148c", "signature": "afdbdfa90f728a373e925b12eac7b559677602ec5e938f19e799ef5564d5659c82b82c71f0f6c09dafe16d7d4dd83e025046f294d42e66a56dc942f4464e2401", "state": "VERIFIED", "summary": {"compressed": 0.579167, "gated": 0.616667, "naive": 0.370833, "scratch": 0.579167, "ungated": 0.366667}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:37:41Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:39Z", "cpu_seconds": 0.31246469100000007, "created_at": "2026-11-22T04:27:36Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-26317947cde5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "c4e3ed10956f20ae8578cd72afd7932484fe28d223a3129cfb64f0ddbbfb71af", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-26317947cde5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:27:36Z", "seed": 241826498, "treatment": "full"}, "priority": 0.188622728832753, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "0885636d4bf599575e06a4fbddb7845bf3779b88e09aaa69936267fc2e2e0a96", "run_seconds": 0.32129693031311035, "seed": 241826498, "started_at": "2026-11-22T04:27:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.245785, 0.370785], "effect": 0.0625, "evaluated_at": "2026-11-22T04:37:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-26317947cde5", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.5, 1.0, 1.0], "fixed_policy": [0.5, 0.5, 1.0, 1.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 0.75, 0.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "c4e3ed10956f20ae8578cd72afd7932484fe28d223a3129cfb64f0ddbbfb71af", "primary_control": "e141_best", "result_hash": "0885636d4bf599575e06a4fbddb7845bf3779b88e09aaa69936267fc2e2e0a96", "signature": "eedc272cb16dcaf25d65866d224a9396cad2d7c288e83fa298ec01c6c134656b62a13c3921137915cbb83dfe04770b4ee136fda815c6ec6de498d6bd9e9cdb05", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.75, "full": 0.9375, "no_world_model": 0.6875, "random": 1.0}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:39Z", "cpu_seconds": 1.1270880790000035, "created_at": "2026-11-22T04:27:36Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-16b42c8b733f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "8e82e67050a6ce9c1f6877759cf2346f6e635562ae0dd031ecc6e8952f99769f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-16b42c8b733f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:27:36Z", "seed": 2004042689, "treatment": "full"}, "priority": 0.18923341786043318, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7341f62669eec0b0521c1289dd9e384f4454e82f0d14fe0fa79fdd11ddbaa2a0", "run_seconds": 1.526249647140503, "seed": 2004042689, "started_at": "2026-11-22T04:27:38Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.041348, -0.019324], "effect": -0.030336, "evaluated_at": "2026-11-22T04:37:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-16b42c8b733f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.493527508, 0.457928803, 0.436011905, 0.398058252], "frozen": [0.181229773, 0.181229773, 0.166666667, 0.181229773], "full": [0.469255663, 0.415857605, 0.398809524, 0.3802589], "retrieval": [0.512944984, 0.482200647, 0.418154762, 0.420711974], "rules_nodetect": [0.747572816, 0.733009709, 0.398809524, 0.72815534]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "8e82e67050a6ce9c1f6877759cf2346f6e635562ae0dd031ecc6e8952f99769f", "primary_control": "facts", "result_hash": "7341f62669eec0b0521c1289dd9e384f4454e82f0d14fe0fa79fdd11ddbaa2a0", "signature": "0eda77683d307ea532c71e44ec7adecc053fa69b5d507be409d9a038b2fc21300990a1d1e2341e9ebf8afcd8e9761ef63ca0798fc2c7342d627c29d29c33e601", "state": "VERIFIED", "summary": {"facts": 0.446382, "frozen": 0.177589, "full": 0.416045, "retrieval": 0.458503, "rules_nodetect": 0.651887}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:43Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:41Z", "cpu_seconds": 2.451402669, "created_at": "2026-11-22T04:27:36Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-fb8ab3c2cd22", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "2b7b933188b1c9f883880c7974ea0763a08419f01bee53e57abb2eace93c88bc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-fb8ab3c2cd22", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:27:36Z", "seed": 1980899594, "treatment": "full"}, "priority": 0.1885974319038909, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-ec3046e0f5fa", "result_hash": "77b492305c36b5961a73fb39cf5e3940cd1feec823f7e4d291e78692d88a6dd5", "run_seconds": 2.541213035583496, "seed": 1980899594, "started_at": "2026-11-22T04:27:39Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.011644, 0.1728], "effect": 0.092222, "evaluated_at": "2026-11-22T04:37:43Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-fb8ab3c2cd22", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.125, 0.720833333, 0.309259259, 1.0], "frozen": [0.0, 0.3, 0.153703704, 0.36], "full": [0.205, 0.920833333, 0.398148148, 1.0], "retrieval": [0.718333333, 0.933333333, 0.522222222, 1.0], "rules_nodetect": [0.441666667, 0.433333333, 0.398148148, 0.792]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "2b7b933188b1c9f883880c7974ea0763a08419f01bee53e57abb2eace93c88bc", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-ec3046e0f5fa", "result_hash": "77b492305c36b5961a73fb39cf5e3940cd1feec823f7e4d291e78692d88a6dd5", "signature": "da565bc904fa99bf43c8555588ff021d5e30575c6bc1ad7d885e11105d953a58fcb3378f6781948b529902cbb9bfdb48d5bc6a6d5eeffadab3a76ca483d3f801", "state": "VERIFIED", "summary": {"facts": 0.538773, "frozen": 0.203426, "full": 0.630995, "retrieval": 0.793472, "rules_nodetect": 0.516287}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:37:44Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:01Z", "cpu_seconds": 3.432539356, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d14f3729ab30", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "fffbd1e0c2cbff4c93ff864ad2003669912615873f09691d254cee194a256146", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d14f3729ab30", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:27:53Z", "seed": 239934659, "treatment": "frontier"}, "priority": 0.18435937228131272, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fe28caef69fe4104c5c29e1923125085afb45b1d772b56279b2c2a1f4c42a6a6", "run_seconds": 3.771071434020996, "seed": 239934659, "started_at": "2026-11-22T04:27:57Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d14f3729ab30", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "fffbd1e0c2cbff4c93ff864ad2003669912615873f09691d254cee194a256146", "primary_control": "random", "result_hash": "fe28caef69fe4104c5c29e1923125085afb45b1d772b56279b2c2a1f4c42a6a6", "signature": "b842a44a9306995b956a2789d2c3c58e86f7790402312c385983d56492e389674fc7899e2238b47a051de4cc60735e458ef21c6a767f3faaf25d92f9724e5e02", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:37:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:52Z", "cpu_seconds": 0.48725476599999995, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e25e022f627a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "967217fb6f86b7fbb85436345b519503a7d58366a81a75ac045e5256b4655ae5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e25e022f627a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:27:53Z", "seed": 861828642, "treatment": "frontier"}, "priority": 0.17290757081618438, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "091979eda779f58942ec28de424c6e5eae22b35db7470285fb5299387681fb7f", "run_seconds": 0.7664220333099365, "seed": 861828642, "started_at": "2026-11-22T04:28:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.129749, 0.256615], "effect": 0.193182, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e25e022f627a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.795454545, 0.734848485, 0.878787879, 0.787878788], "frontier": [0.878787879, 0.553030303, 0.886363636, 0.818181818], "goal": [0.71969697, 0.492424242, 0.757575758, 0.643939394], "hypdisc": [0.583333333, 0.515151515, 0.689393939, 0.628787879], "infogain": [0.878787879, 0.553030303, 0.886363636, 0.818181818], "prederr": [0.878787879, 0.553030303, 0.886363636, 0.818181818], "random": [0.606060606, 0.431818182, 0.674242424, 0.651515152], "risk": [0.878787879, 0.553030303, 0.886363636, 0.818181818]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "967217fb6f86b7fbb85436345b519503a7d58366a81a75ac045e5256b4655ae5", "primary_control": "random", "result_hash": "091979eda779f58942ec28de424c6e5eae22b35db7470285fb5299387681fb7f", "signature": "90348aa74d37eccf68ce2ea9264a7ddca4b1cf259db67581a6b664dc7281b3c2f11226a7efc54caa8496d685602557f81bd230881bed0afe0a878c3248ec660a", "state": "VERIFIED", "summary": {"count": 0.799242, "frontier": 0.784091, "goal": 0.653409, "hypdisc": 0.604167, "infogain": 0.784091, "prederr": 0.784091, "random": 0.590909, "risk": 0.784091}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:02Z", "cpu_seconds": 0.5193807560000039, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0f7c63343392", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "8a19d0ac1756f030545c851c8324d88820e55f0cef7cc5f7e4dc6e485339ff15", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0f7c63343392", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:27:53Z", "seed": 1906154889, "treatment": "full"}, "priority": 0.18433766716901706, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b6f98800c772a48349d0c5e5324e72c120cab023221180afb3778e8809d8bacd", "run_seconds": 0.5247349739074707, "seed": 1906154889, "started_at": "2026-11-22T04:28:02Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0f7c63343392", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.0, 0.5, 0.25], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "8a19d0ac1756f030545c851c8324d88820e55f0cef7cc5f7e4dc6e485339ff15", "primary_control": "e141_best", "result_hash": "b6f98800c772a48349d0c5e5324e72c120cab023221180afb3778e8809d8bacd", "signature": "f7fbb66cd0f0dfc0d77bf9a2f5f62de3c6180a9453fcd8a855c9cb63d139f0b6778d36798a630830dabd9e94384d2674c1ecf55880895ece247ec5d9624c5f09", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:57Z", "cpu_seconds": 1.4166105090000087, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-41bfed7556ad", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "bf62000a0b8fd32fa2f81efadda07002881395844c5d6487373ea34dff6b9626", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-41bfed7556ad", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:27:53Z", "seed": 1476852828, "treatment": "full"}, "priority": 0.188110053285014, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b7fab5cc6319a3ed76125ddbdd15704d269cdbf831f7ecf129f86b84b2b00874", "run_seconds": 1.522770643234253, "seed": 1476852828, "started_at": "2026-11-22T04:27:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.697307, 0.197307], "effect": -0.25, "evaluated_at": "2026-11-22T04:37:46Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-41bfed7556ad", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.75, 0.75, 0.5], "fixed_policy": [1.0, 1.0, 0.25, 0.0], "full": [0.0, 0.75, 1.0, 0.0], "no_world_model": [0.0, 0.75, 0.75, 0.0], "random": [0.75, 0.75, 0.75, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.058, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.875", "prereg_hash": "bf62000a0b8fd32fa2f81efadda07002881395844c5d6487373ea34dff6b9626", "primary_control": "e141_best", "result_hash": "b7fab5cc6319a3ed76125ddbdd15704d269cdbf831f7ecf129f86b84b2b00874", "signature": "44a156ab92fbcf827351fc88805f3aaf4326f30aaee72017a147b18975e342b95c4ad4fc4c3ad4ae960350e7b9a19fda7e68a5a8977b4b1e5bca5c59635e240b", "state": "VERIFIED", "summary": {"e141_best": 0.6875, "fixed_policy": 0.5625, "full": 0.4375, "no_world_model": 0.375, "random": 0.75}, "ties": 1, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:49Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:27:58Z", "cpu_seconds": 0.6019892069999999, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-cb11fcf3f871", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "ed61b9e69628d57415d4126c9f20568f4fd169541ec2575d59d57dd693d56829", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-cb11fcf3f871", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:27:53Z", "seed": 1540566614, "treatment": "full"}, "priority": 0.18574726741178404, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d43ea775ca8fbc49251cf0bbc39287690e325f66f03644c520833384682b6adf", "run_seconds": 0.7070786952972412, "seed": 1540566614, "started_at": "2026-11-22T04:27:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.37, 0.12], "effect": -0.125, "evaluated_at": "2026-11-22T04:37:49Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-cb11fcf3f871", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 0.0, 0.0, 0.0], "full": [0.5, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [1.0, 0.5, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 1, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "ed61b9e69628d57415d4126c9f20568f4fd169541ec2575d59d57dd693d56829", "primary_control": "e141_best", "result_hash": "d43ea775ca8fbc49251cf0bbc39287690e325f66f03644c520833384682b6adf", "signature": "2e5a9755501f294abb2bd8d131a2fadbd662caab54b76e75209a3c21bf39cbd2c9f01ecd8ad219a47051a14a02703959991d779af2ebf3acd48d0074600a130d", "state": "VERIFIED", "summary": {"e141_best": 1.0, "fixed_policy": 0.25, "full": 0.875, "no_world_model": 0.9375, "random": 0.625}, "ties": 3, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:04Z", "cpu_seconds": 0.5630510210000006, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e447b0246f21", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "89b1bc56887008ffc621bc9c1f5340de49c3fc289abdc4335051edd33cecb8d3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e447b0246f21", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:27:53Z", "seed": 138225031, "treatment": "full"}, "priority": 0.18065962370224667, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "cc2d9ed7cbbce61360e8180d28203f27511a11f59a27e066ea4c2caadbed5582", "run_seconds": 0.652228593826294, "seed": 138225031, "started_at": "2026-11-22T04:28:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.195, 0.93], "effect": 0.5625, "evaluated_at": "2026-11-22T04:37:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e447b0246f21", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.0, 0.0], "fixed_policy": [0.0, 0.0, 0.0, 0.5], "full": [0.25, 0.75, 0.25, 1.0], "no_world_model": [0.5, 1.0, 0.75, 0.0], "random": [0.75, 0.75, 0.75, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "89b1bc56887008ffc621bc9c1f5340de49c3fc289abdc4335051edd33cecb8d3", "primary_control": "e141_best", "result_hash": "cc2d9ed7cbbce61360e8180d28203f27511a11f59a27e066ea4c2caadbed5582", "signature": "5c2fa21a137529998ebbd7b813dee696c077919ba307671407e0f558aa9f7f12c53e964bf0a74a53f023e29515ac878bedb530809a5831d04f809f8c4ea9960f", "state": "VERIFIED", "summary": {"e141_best": 0.0, "fixed_policy": 0.125, "full": 0.5625, "no_world_model": 0.5625, "random": 0.5625}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:37:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:38Z", "cpu_seconds": 0.8025940809999996, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-34d91276cc16", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "c2969353d3a6fb3c712b69a8cf9fc082e67a934480b894bee3a062c1dccbe48b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-34d91276cc16", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:27:53Z", "seed": 1924790920, "treatment": "full"}, "priority": 0.17652196297113362, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "6b7a45d43903c7e5dce3c730c5e15e6df5dc4659be5d0c0c5de45f20454c6691", "run_seconds": 0.81900954246521, "seed": 1924790920, "started_at": "2026-11-22T04:28:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.025973, 0.087045], "effect": 0.030536, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-34d91276cc16", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.916666667, 0.725, 0.598214286, 0.94], "frozen": [0.375, 0.325, 0.25, 0.22], "full": [0.916666667, 0.7, 0.705357143, 0.98], "retrieval": [0.916666667, 0.675, 0.714285714, 0.97], "rules_nodetect": [0.333333333, 0.625, 0.25, 0.28]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "c2969353d3a6fb3c712b69a8cf9fc082e67a934480b894bee3a062c1dccbe48b", "primary_control": "facts", "result_hash": "6b7a45d43903c7e5dce3c730c5e15e6df5dc4659be5d0c0c5de45f20454c6691", "signature": "1448eb791deb86aae0f7159095be94ddf3e8c098a1152d1222a7ae7c583d06ccfce3aacd7965106eb13558177e2b7a24460452eddca829310c8047ef606f6d06", "state": "VERIFIED", "summary": {"facts": 0.79497, "frozen": 0.2925, "full": 0.825506, "retrieval": 0.818988, "rules_nodetect": 0.372083}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:21Z", "cpu_seconds": 0.3851310269999999, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-94bd7064d56b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "prereg_hash": "6a22c7721aef0f95dd1f1cb0c12628a61ae69f8cbd0770e9fe6898fd2d9ee3fe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-94bd7064d56b", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|causal_switch|causal_switch|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:27:53Z", "seed": 1544777529, "treatment": "gated"}, "priority": 0.17971803944716142, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d8205d66791b0dcd1b0cdb469bd6e7cb8d57d46ae1a6a92bd20e6f1ab23985a9", "run_seconds": 0.7351479530334473, "seed": 1544777529, "started_at": "2026-11-22T04:28:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.020025, 0.220025], "effect": 0.1, "evaluated_at": "2026-11-22T04:37:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-94bd7064d56b", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.75, 0.65, 0.625, 0.6], "gated": [0.75, 0.8, 0.625, 0.85], "naive": [0.325, 0.125, 0.125, 0.125], "scratch": [0.75, 0.65, 0.625, 0.6], "ungated": [0.425, 0.45, 0.333333333, 0.425]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "6a22c7721aef0f95dd1f1cb0c12628a61ae69f8cbd0770e9fe6898fd2d9ee3fe", "primary_control": "scratch", "result_hash": "d8205d66791b0dcd1b0cdb469bd6e7cb8d57d46ae1a6a92bd20e6f1ab23985a9", "signature": "d765904fcfa0de4d8b7d8d52cc87c584d7aa16c81d8b947dd1628558fb366b06f3ad399e46750f3be9ccc82b904c7bd4b4f576c226ae63216c467b4699dafa00", "state": "VERIFIED", "summary": {"compressed": 0.65625, "gated": 0.75625, "naive": 0.175, "scratch": 0.65625, "ungated": 0.408333}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:37:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:26Z", "cpu_seconds": 4.270810097, "created_at": "2026-11-22T04:27:53Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ed0856fbde4f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "67bf522fc2112b7f62012a700da31d26fd92f28bcfc2dc963bd05c621c623b1d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ed0856fbde4f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:27:53Z", "seed": 2094664529, "treatment": "gated"}, "priority": 0.1786346939876309, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "d5963c79e588f29db52b16ae7cd2f7dd0278ca0c8a2143f9b51287b9f348cced", "run_seconds": 5.987194776535034, "seed": 2094664529, "started_at": "2026-11-22T04:28:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.009231, 0.028462], "effect": 0.009615, "evaluated_at": "2026-11-22T04:37:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-ed0856fbde4f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.556818182, 0.47826087, 0.451923077, 0.384615385], "gated": [0.556818182, 0.47826087, 0.451923077, 0.384615385], "naive": [0.681818182, 0.597826087, 0.721153846, 0.576923077], "scratch": [0.556818182, 0.47826087, 0.451923077, 0.346153846], "ungated": [0.613636364, 0.597826087, 0.644230769, 0.519230769]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "67bf522fc2112b7f62012a700da31d26fd92f28bcfc2dc963bd05c621c623b1d", "primary_control": "scratch", "result_hash": "d5963c79e588f29db52b16ae7cd2f7dd0278ca0c8a2143f9b51287b9f348cced", "signature": "8fb891e2a42e43c5663b483b060365db7d007afecce5662278eee2967090a7c9e05b5e6baca3cd29be0ce91bbb7d549f0078700066d5b6a2950cdab2c76a9d05", "state": "VERIFIED", "summary": {"compressed": 0.467904, "gated": 0.467904, "naive": 0.64443, "scratch": 0.458289, "ungated": 0.593731}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:37:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:15Z", "cpu_seconds": 0.7592498300000001, "created_at": "2026-11-22T04:28:08Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-78a7307a5907", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "4ee6490d4dce06cf12f293c6386f99de15f8306973e16f41baffdca30bbbfbcf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-78a7307a5907", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:28:08Z", "seed": 2043919936, "treatment": "full"}, "priority": 0.18460574131909707, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a93575a9e21485972abb724987514c430f19e50b2b04e427470f0f9f337b7ae3", "run_seconds": 1.0100927352905273, "seed": 2043919936, "started_at": "2026-11-22T04:28:14Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-78a7307a5907", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 1.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4ee6490d4dce06cf12f293c6386f99de15f8306973e16f41baffdca30bbbfbcf", "primary_control": "e141_best", "result_hash": "a93575a9e21485972abb724987514c430f19e50b2b04e427470f0f9f337b7ae3", "signature": "bc33c86179a82d4f1315f67ac9a85f0c6a5afd61fe36b7aed7af7aab6670fd16442f24c374fa11ffb8a90d1d02b8d67cd9f0131c26b162738e9626efacbbc008", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 1.0, "full": 1.0, "no_world_model": 1.0, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:15Z", "cpu_seconds": 0.8180797600000007, "created_at": "2026-11-22T04:28:08Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9f255d993deb", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "f8e6444ac3d521508e1711c65274859af757cc2550824ce505bf56f3aba9697d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9f255d993deb", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:28:08Z", "seed": 1780502726, "treatment": "full"}, "priority": 0.18694591488457374, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-95dbb50e0e41", "result_hash": "d085f914e426fb057cb0aff9a419f46fbbde0b0037bb5c49b0c638c170e32c7f", "run_seconds": 1.2430212497711182, "seed": 1780502726, "started_at": "2026-11-22T04:28:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016451, 0.266451], "effect": 0.125, "evaluated_at": "2026-11-22T04:37:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9f255d993deb", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 0.25, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 0.5, 1.0, 1.0], "no_world_model": [0.75, 0.25, 1.0, 1.0], "random": [1.0, 0.5, 1.0, 0.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "f8e6444ac3d521508e1711c65274859af757cc2550824ce505bf56f3aba9697d", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-95dbb50e0e41", "result_hash": "d085f914e426fb057cb0aff9a419f46fbbde0b0037bb5c49b0c638c170e32c7f", "signature": "8500d2b6434a0cfb97c8a345ba065cc8824405b785a795fba324346493aca7a4e9e1b3421dbb063b3ae047320e3b0643c0fd9d191d79a1f594fa9ca0c877f308", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.0, "full": 0.875, "no_world_model": 0.75, "random": 0.625}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:37:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:15Z", "cpu_seconds": 0.7009622879999999, "created_at": "2026-11-22T04:28:08Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-df2a0e330afa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "e373306943f1efa9d443e5d9efd36e5dec20d03672d60cb44e01bcd06aa44f80", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-df2a0e330afa", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:28:08Z", "seed": 906902985, "treatment": "full"}, "priority": 0.18424783569920306, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ba6a64d4798104c07419b61dec9a75de501953d9faec765daf28900a849db224", "run_seconds": 0.9631023406982422, "seed": 906902985, "started_at": "2026-11-22T04:28:14Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.62, -0.13], "effect": -0.375, "evaluated_at": "2026-11-22T04:37:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-df2a0e330afa", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 0.5, 1.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [0.0, 0.5, 0.0, 1.0], "no_world_model": [0.25, 1.0, 0.0, 0.0], "random": [0.5, 0.5, 0.5, 0.75]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e373306943f1efa9d443e5d9efd36e5dec20d03672d60cb44e01bcd06aa44f80", "primary_control": "e141_best", "result_hash": "ba6a64d4798104c07419b61dec9a75de501953d9faec765daf28900a849db224", "signature": "713b53198731e764ba904d379a1919a586c3fc4c42734eb40908ed5e681731983da0ec59277b19eb9b84b2b2b76dba443f54848179894f768ff1ffeb4faa660e", "state": "VERIFIED", "summary": {"e141_best": 0.75, "fixed_policy": 0.0625, "full": 0.375, "no_world_model": 0.3125, "random": 0.5625}, "ties": 1, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:21Z", "cpu_seconds": 0.7056406900000001, "created_at": "2026-11-22T04:28:08Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1cb80b0bbb7a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "2e61dddef39485d38e116b48c10215588e754cd3b370635750268e6d905395a2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1cb80b0bbb7a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:08Z", "seed": 235771718, "treatment": "full"}, "priority": 0.1828116942273409, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5585d5857a61f71c58eed15b017737764ab0de8e82510f1c66480d95a6c36cbd", "run_seconds": 1.1310875415802002, "seed": 235771718, "started_at": "2026-11-22T04:28:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.008774, 0.032205], "effect": 0.02049, "evaluated_at": "2026-11-22T04:37:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-1cb80b0bbb7a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.856060606, 0.727272727, 0.810457516, 0.519379845], "frozen": [0.098484848, 0.098484848, 0.08496732, 0.100775194], "full": [0.863636364, 0.757575758, 0.823529412, 0.550387597], "retrieval": [0.909090909, 0.78030303, 0.85620915, 0.720930233], "rules_nodetect": [0.863636364, 0.757575758, 0.823529412, 0.550387597]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2e61dddef39485d38e116b48c10215588e754cd3b370635750268e6d905395a2", "primary_control": "facts", "result_hash": "5585d5857a61f71c58eed15b017737764ab0de8e82510f1c66480d95a6c36cbd", "signature": "b74a25660c1f92374a04b37b046e4cb1b7b5891b9d151262156e1c08749507f5ca8898ebd1fdcbb5bf82e63732522440a5795721ab93621db8fdbda7ae09850a", "state": "VERIFIED", "summary": {"facts": 0.728293, "frozen": 0.095678, "full": 0.748782, "retrieval": 0.816633, "rules_nodetect": 0.748782}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:37:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:09Z", "cpu_seconds": 0.1597260579999995, "created_at": "2026-11-22T04:28:08Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-891f0204987b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 3, "n": 4}, "prereg_hash": "040ae05e985941c3eca66b27df34085cc09f7e94307927fd847f30c4fd09dc8e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-891f0204987b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:08Z", "seed": 2141948892, "treatment": "full"}, "priority": 0.1882789913900165, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "21884b0cb08bbe0569092c6111535dadf2e07cb2594c40b784c14f7d81618378", "run_seconds": 0.17133474349975586, "seed": 2141948892, "started_at": "2026-11-22T04:28:09Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.008182, 0.025227], "effect": 0.008523, "evaluated_at": "2026-11-22T04:37:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-891f0204987b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 1.0, 0.988636364, 0.795454545], "frozen": [0.022727273, 0.022727273, 0.034090909, 0.034090909], "full": [1.0, 1.0, 0.988636364, 0.829545455], "retrieval": [1.0, 1.0, 0.988636364, 0.829545455], "rules_nodetect": [1.0, 1.0, 0.988636364, 0.829545455]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "040ae05e985941c3eca66b27df34085cc09f7e94307927fd847f30c4fd09dc8e", "primary_control": "facts", "result_hash": "21884b0cb08bbe0569092c6111535dadf2e07cb2594c40b784c14f7d81618378", "signature": "9f4345ed05c78de4bd8be70b46b0bcc42a2c2e185396fb8f7f81f8e2fcbe02545cbfdcf7528652533ffdc0c5f1308828db44ce74f2ebf026181df7f9dcca7a0a", "state": "VERIFIED", "summary": {"facts": 0.946023, "frozen": 0.028409, "full": 0.954545, "retrieval": 0.954545, "rules_nodetect": 0.954545}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:53Z", "cpu_seconds": 3.786174868, "created_at": "2026-11-22T04:28:25Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d1651fb9cea2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 10, "n": 4}, "prereg_hash": "669e59279bf7916a64fe8b8f330b72f1e6b71429218d7623e92cc26f9168839d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d1651fb9cea2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:25Z", "seed": 2008637316, "treatment": "frontier"}, "priority": 0.17314145399003392, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6b5f6e6551d43c75b4538c107801b06792fe5439e728745c1668c3f6e0e8e2e9", "run_seconds": 4.342057943344116, "seed": 2008637316, "started_at": "2026-11-22T04:28:48Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035061, 0.03407], "effect": -0.000496, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d1651fb9cea2", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.232421875, 0.617391304, 0.312065972, 0.265151515], "frontier": [0.233072917, 0.52173913, 0.227864583, 0.261363636], "goal": [0.233072917, 0.530434783, 0.206163194, 0.263636364], "hypdisc": [0.20703125, 0.530434783, 0.198784722, 0.250757576], "infogain": [0.233072917, 0.417391304, 0.233072917, 0.265151515], "prederr": [0.233072917, 0.226086957, 0.26171875, 0.251515152], "random": [0.226236979, 0.573913043, 0.202690972, 0.243181818], "risk": [0.233072917, 0.52173913, 0.227864583, 0.261363636]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "669e59279bf7916a64fe8b8f330b72f1e6b71429218d7623e92cc26f9168839d", "primary_control": "random", "result_hash": "6b5f6e6551d43c75b4538c107801b06792fe5439e728745c1668c3f6e0e8e2e9", "signature": "678852c64c4914a378dba4b7c251b68036d0ff1d68c4dfcd2025896e7c9a104910d88b6d9d07567313d5416b13be6cca58c192c8faedbf5bf6ed910852067f07", "state": "VERIFIED", "summary": {"count": 0.356758, "frontier": 0.31101, "goal": 0.308327, "hypdisc": 0.296752, "infogain": 0.287172, "prederr": 0.243098, "random": 0.311506, "risk": 0.31101}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:37:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:35Z", "cpu_seconds": 0.6593981999999999, "created_at": "2026-11-22T04:28:25Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f9514398d2d5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "7bd5ae384cd529140891933608cf8c84428778b082a515a9dcb18b760926c28d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f9514398d2d5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:25Z", "seed": 70932691, "treatment": "frontier"}, "priority": 0.17970684109281432, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ca93be43c97fab956abbd781311c7d83186d9a2fe0f1ef8f48ba2d435623dbd9", "run_seconds": 0.666785478591919, "seed": 70932691, "started_at": "2026-11-22T04:28:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.030899, 0.020482], "effect": -0.005208, "evaluated_at": "2026-11-22T04:37:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-f9514398d2d5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 0.958333333], "frontier": [0.958333333, 0.979166667, 1.0, 0.916666667], "goal": [0.958333333, 0.958333333, 1.0, 0.854166667], "hypdisc": [1.0, 0.9375, 1.0, 0.895833333], "infogain": [0.958333333, 0.979166667, 1.0, 0.916666667], "prederr": [0.958333333, 0.979166667, 1.0, 0.916666667], "random": [1.0, 0.979166667, 1.0, 0.895833333], "risk": [0.958333333, 0.979166667, 1.0, 0.916666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "7bd5ae384cd529140891933608cf8c84428778b082a515a9dcb18b760926c28d", "primary_control": "random", "result_hash": "ca93be43c97fab956abbd781311c7d83186d9a2fe0f1ef8f48ba2d435623dbd9", "signature": "82b83092050ae27b3cf8717a377c2dec1b06a162fe69346e21f3987e03380714c23eebea9c84334f3bfefacf391d52081954dafb7e777bc485dfb9df0f747f01", "state": "VERIFIED", "summary": {"count": 0.989583, "frontier": 0.963542, "goal": 0.942708, "hypdisc": 0.958333, "infogain": 0.963542, "prederr": 0.963542, "random": 0.96875, "risk": 0.963542}, "ties": 2, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:37:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:53Z", "cpu_seconds": 0.38064125599999965, "created_at": "2026-11-22T04:28:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2042ee751c5b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "1f2a86ed5c60f5b29db9a9aa74a59b0d7e2e9a4f3fbfd86cccf4cd7ca639fea4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2042ee751c5b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:25Z", "seed": 1762958515, "treatment": "full"}, "priority": 0.17278509420946558, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "21cf16f17bbc37d0f6818eecf1d62b3cf72b6197e5842b51414fad2a376e7345", "run_seconds": 0.46196770668029785, "seed": 1762958515, "started_at": "2026-11-22T04:28:53Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.006675, 0.073342], "effect": 0.033333, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2042ee751c5b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.916666667, 0.783333333, 1.0], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 0.966666667, 0.866666667, 1.0], "retrieval": [1.0, 0.966666667, 0.916666667, 1.0], "rules_nodetect": [1.0, 0.966666667, 0.866666667, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "1f2a86ed5c60f5b29db9a9aa74a59b0d7e2e9a4f3fbfd86cccf4cd7ca639fea4", "primary_control": "facts", "result_hash": "21cf16f17bbc37d0f6818eecf1d62b3cf72b6197e5842b51414fad2a376e7345", "signature": "926a300a5985f56e7a80533036d2e4d57ec71bbba5fe9c28356ba1207d095f731e5bd202b589cc97a976eae6baddca5d54039d569ec264026370a7c051c3590f", "state": "VERIFIED", "summary": {"facts": 0.925, "frozen": 0.0, "full": 0.958333, "retrieval": 0.970833, "rules_nodetect": 0.958333}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:32Z", "cpu_seconds": 0.5088258460000006, "created_at": "2026-11-22T04:28:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-21820a2c5de2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "9d850f5ff246430d406f0454739aa864b4f36c72d98945e79cb4f32bf3d28b76", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-21820a2c5de2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:25Z", "seed": 1489063480, "treatment": "full"}, "priority": 0.18065885441804908, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ebc44347dced45eb48e70408a78a16ed230b15a656a3514f0e88f2cad0834f95", "run_seconds": 0.5198192596435547, "seed": 1489063480, "started_at": "2026-11-22T04:28:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.02214, 0.117582], "effect": 0.047721, "evaluated_at": "2026-11-22T04:37:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-21820a2c5de2", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.796296296, 0.731481481, 0.743589744, 0.615384615], "frozen": [0.37037037, 0.361111111, 0.358974359, 0.314102564], "full": [0.814814815, 0.75, 0.743589744, 0.769230769], "retrieval": [0.796296296, 0.722222222, 0.769230769, 0.717948718], "rules_nodetect": [0.814814815, 0.796296296, 0.743589744, 0.769230769]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "9d850f5ff246430d406f0454739aa864b4f36c72d98945e79cb4f32bf3d28b76", "primary_control": "facts", "result_hash": "ebc44347dced45eb48e70408a78a16ed230b15a656a3514f0e88f2cad0834f95", "signature": "fef53110cd5afaf3e5fc223a639bbf8230f7fbf9e0b58be81118141958007735cdacdcded712a3be74a79f0f7431efb6d92ff0f28e895dce04b7579a4be28b09", "state": "VERIFIED", "summary": {"facts": 0.721688, "frozen": 0.35114, "full": 0.769409, "retrieval": 0.751425, "rules_nodetect": 0.780983}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:37:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:37Z", "cpu_seconds": 0.9575997479999998, "created_at": "2026-11-22T04:28:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-236e944e5461", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 10, "n": 4}, "prereg_hash": "3771ca6b3d36f880c5f29c4d0ad08d7168c692b227bddd0020a2edfbc170adf8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-236e944e5461", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:25Z", "seed": 1833638129, "treatment": "full"}, "priority": 0.1770202977575248, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a1c7e2b7869a91473186f0147f80a5e2e04da1a4fd6cd3fbb97f7a61d544c6cf", "run_seconds": 0.9625296592712402, "seed": 1833638129, "started_at": "2026-11-22T04:28:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00346, 0.100399], "effect": 0.04847, "evaluated_at": "2026-11-22T04:37:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-236e944e5461", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.844444444, 0.787326389, 0.427083333, 0.674479167], "frozen": [0.674074074, 0.341145833, 0.337862319, 0.427083333], "full": [0.844444444, 0.796875, 0.538496377, 0.747395833], "retrieval": [0.918518519, 0.809027778, 0.787137681, 0.825520833], "rules_nodetect": [0.655555556, 0.796875, 0.538496377, 0.5546875]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "3771ca6b3d36f880c5f29c4d0ad08d7168c692b227bddd0020a2edfbc170adf8", "primary_control": "facts", "result_hash": "a1c7e2b7869a91473186f0147f80a5e2e04da1a4fd6cd3fbb97f7a61d544c6cf", "signature": "648c0cf483a6ca53286275f8ad16b32f6fa3fe40523bf7dfadfc3afcca0549ab3591c477c0c7adf95eb48cf7cfc5a9559b657270991cb052c7922b0267b50005", "state": "VERIFIED", "summary": {"facts": 0.683333, "frozen": 0.445041, "full": 0.731803, "retrieval": 0.835051, "rules_nodetect": 0.636404}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:31Z", "cpu_seconds": 1.240226229, "created_at": "2026-11-22T04:28:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-451a4aa449d0", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "000dfed597acb7b8baa87e5c5e790f1994a7be79631567e80a203d644fbad443", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-451a4aa449d0", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:25Z", "seed": 1001700161, "treatment": "full"}, "priority": 0.18343654911784116, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-df732dd9e95a", "result_hash": "3f6bfa3edf18ad6c22b6a415ae4eca0dfed997e8978114f0c580656c6d0d14b9", "run_seconds": 1.2652525901794434, "seed": 1001700161, "started_at": "2026-11-22T04:28:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.034905, 0.021344], "effect": -0.00678, "evaluated_at": "2026-11-22T04:37:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-451a4aa449d0", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.570921986, 0.423076923, 0.455673759, 0.673758865], "frozen": [0.177304965, 0.10989011, 0.108156028, 0.131205674], "full": [0.604609929, 0.406593407, 0.421985816, 0.663120567], "retrieval": [0.59751773, 0.426739927, 0.407801418, 0.679078014], "rules_nodetect": [0.604609929, 0.606227106, 0.421985816, 0.663120567]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "000dfed597acb7b8baa87e5c5e790f1994a7be79631567e80a203d644fbad443", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-df732dd9e95a", "result_hash": "3f6bfa3edf18ad6c22b6a415ae4eca0dfed997e8978114f0c580656c6d0d14b9", "signature": "23a61d962c98a5dba552d4200bd9b06029a15c56d6f667d3154fd22fae41d50dad23807b8d6d753144d550a5674b3b617d96b3683de9a5d2eaf7947c5fb1cb09", "state": "VERIFIED", "summary": {"facts": 0.530858, "frozen": 0.131639, "full": 0.524077, "retrieval": 0.527784, "rules_nodetect": 0.573986}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:30Z", "cpu_seconds": 0.20993487099999997, "created_at": "2026-11-22T04:28:25Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-651bca6dd59f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "89ed41dccf8535d687c5849f50934099fa7a2f12a4b400038148541c88785bff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-651bca6dd59f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:25Z", "seed": 1247036836, "treatment": "full"}, "priority": 0.18367867953469463, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ca735e2d16c40a446cecb54ecaf08cafc076afd2012e9eaaaa6b25beb7d3bbc2", "run_seconds": 0.221207857131958, "seed": 1247036836, "started_at": "2026-11-22T04:28:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.021818, 0.067273], "effect": 0.022727, "evaluated_at": "2026-11-22T04:37:53Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-651bca6dd59f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.666666667, 1.0, 0.984848485], "frozen": [0.03030303, 0.03030303, 0.03030303, 0.03030303], "full": [1.0, 0.757575758, 1.0, 0.984848485], "retrieval": [1.0, 0.742424242, 1.0, 0.984848485], "rules_nodetect": [1.0, 0.757575758, 1.0, 0.984848485]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "89ed41dccf8535d687c5849f50934099fa7a2f12a4b400038148541c88785bff", "primary_control": "facts", "result_hash": "ca735e2d16c40a446cecb54ecaf08cafc076afd2012e9eaaaa6b25beb7d3bbc2", "signature": "a1675e8a52fed3ffc1472a4d89f2875b933f4ac60a9f583c9440c2dea94736f86f8515667050f85e54d9fc849d801f7e1428ebe8c9e1c70edc69b46a957b3a0a", "state": "VERIFIED", "summary": {"facts": 0.912879, "frozen": 0.030303, "full": 0.935606, "retrieval": 0.931818, "rules_nodetect": 0.935606}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:53Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:36Z", "cpu_seconds": 1.174573088999999, "created_at": "2026-11-22T04:28:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5d54da897029", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "d0edc24f7ed9d572bf7746688970c2886d4e83cc10e169a08f9aed501f39876c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5d54da897029", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:40Z", "seed": 1666896626, "treatment": "frontier"}, "priority": 0.17211050834253752, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e01784fc695f95db40f4a73676d3a829b1f3806a71d003e2cb4163d6befc85e8", "run_seconds": 1.4001984596252441, "seed": 1666896626, "started_at": "2026-11-22T04:31:35Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.036345, 0.06811], "effect": 0.015883, "evaluated_at": "2026-11-22T04:38:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-5d54da897029", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.807017544, 0.703703704, 0.777777778, 0.882352941], "frontier": [0.824561404, 0.675925926, 0.797979798, 0.882352941], "goal": [0.824561404, 0.731481481, 0.797979798, 0.882352941], "hypdisc": [0.824561404, 0.675925926, 0.797979798, 0.882352941], "infogain": [0.824561404, 0.722222222, 0.797979798, 0.882352941], "prederr": [0.824561404, 0.722222222, 0.797979798, 0.882352941], "random": [0.877192982, 0.62037037, 0.737373737, 0.882352941], "risk": [0.824561404, 0.675925926, 0.797979798, 0.882352941]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d0edc24f7ed9d572bf7746688970c2886d4e83cc10e169a08f9aed501f39876c", "primary_control": "random", "result_hash": "e01784fc695f95db40f4a73676d3a829b1f3806a71d003e2cb4163d6befc85e8", "signature": "ae438174a331e9d4e4fbb66ef4fe2f9b301ae279da770f6143473a8fc99a23d300ae7817a9f0d3adf219d3bfcf23b2d474dbb8908f1601418d97f01082558a07", "state": "VERIFIED", "summary": {"count": 0.792713, "frontier": 0.795205, "goal": 0.809094, "hypdisc": 0.795205, "infogain": 0.806779, "prederr": 0.806779, "random": 0.779323, "risk": 0.795205}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:48Z", "cpu_seconds": 0.29050681899999997, "created_at": "2026-11-22T04:28:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6ab6070af76b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "362d6d8023fcf576bf8396ad12d2c66b4e7cb702ff60af6e5996ac0a7895f7a3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6ab6070af76b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:40Z", "seed": 1977896538, "treatment": "frontier"}, "priority": 0.17679903156148785, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "2781b2d5c65ca2e7bcda3edd5a53f82bc10f417a77c7ae029f6b83e813e3d49a", "run_seconds": 0.29117393493652344, "seed": 1977896538, "started_at": "2026-11-22T04:28:47Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00938, 0.115441], "effect": 0.05303, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-6ab6070af76b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.984848485, 1.0, 0.984848485, 1.0], "frontier": [0.984848485, 0.96969697, 0.984848485, 0.893939394], "goal": [0.772727273, 0.787878788, 0.954545455, 0.818181818], "hypdisc": [0.909090909, 0.909090909, 0.954545455, 0.893939394], "infogain": [0.984848485, 0.96969697, 0.984848485, 0.893939394], "prederr": [0.984848485, 0.96969697, 0.984848485, 0.893939394], "random": [0.893939394, 0.954545455, 0.863636364, 0.909090909], "risk": [0.984848485, 0.96969697, 0.984848485, 0.893939394]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "362d6d8023fcf576bf8396ad12d2c66b4e7cb702ff60af6e5996ac0a7895f7a3", "primary_control": "random", "result_hash": "2781b2d5c65ca2e7bcda3edd5a53f82bc10f417a77c7ae029f6b83e813e3d49a", "signature": "d1a2a9aa78bc2c4e984c8de5a7abcd0a0ddac1f7f4f06a9b4bda146f336d303ab1e5237b308343f888e36da8d8141bed4aa6ec7b6f865d17fc64b8e5611e550b", "state": "VERIFIED", "summary": {"count": 0.992424, "frontier": 0.958333, "goal": 0.833333, "hypdisc": 0.916667, "infogain": 0.958333, "prederr": 0.958333, "random": 0.905303, "risk": 0.958333}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:43Z", "cpu_seconds": 0.42433965199999957, "created_at": "2026-11-22T04:28:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a5eef821509b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "6a226165a9f0c64f1995c89e92c15fbc793b17337cfa05e96fe7ea38aa4f7c24", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a5eef821509b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:40Z", "seed": 1322340734, "treatment": "frontier"}, "priority": 0.17805630131158778, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c4a4c73bf24baacf1c219a0f09371b6e71f8cf8a3ebd7b51592dac20e79c4c14", "run_seconds": 0.4778149127960205, "seed": 1322340734, "started_at": "2026-11-22T04:28:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.097774, 0.190105], "effect": 0.143939, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a5eef821509b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 0.96969697], "goal": [0.803030303, 0.757575758, 0.833333333, 0.787878788], "hypdisc": [0.909090909, 0.818181818, 0.909090909, 0.909090909], "infogain": [1.0, 1.0, 1.0, 0.96969697], "prederr": [1.0, 1.0, 1.0, 0.96969697], "random": [0.878787879, 0.803030303, 0.833333333, 0.878787879], "risk": [1.0, 1.0, 1.0, 0.96969697]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "6a226165a9f0c64f1995c89e92c15fbc793b17337cfa05e96fe7ea38aa4f7c24", "primary_control": "random", "result_hash": "c4a4c73bf24baacf1c219a0f09371b6e71f8cf8a3ebd7b51592dac20e79c4c14", "signature": "ea7835d8feaa205ac8d144ba305284b633ccd47ac22c5cad6434e50b99c3c41179c59eb252f0c57655600018a936a526316f3839096e9b7b584ca13e921b6505", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.992424, "goal": 0.795455, "hypdisc": 0.886364, "infogain": 0.992424, "prederr": 0.992424, "random": 0.848485, "risk": 0.992424}, "ties": 0, "treatment": "frontier", "wins": 4}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:42Z", "cpu_seconds": 0.31633739999999966, "created_at": "2026-11-22T04:28:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0a1c53194903", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "79e90dd987f4f5687b9473491548cf6e96b1d8278dda805c39b946de62991030", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0a1c53194903", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:40Z", "seed": 251622007, "treatment": "full"}, "priority": 0.18567447912060558, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cc44997e961567b9647a8d583db3746fcb6d315eb299387be43a979feacf2257", "run_seconds": 0.330474853515625, "seed": 251622007, "started_at": "2026-11-22T04:28:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.00119, 0.138393], "effect": 0.069792, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-0a1c53194903", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.833333333, 0.979166667, 0.916666667, 0.925], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 0.979166667, 0.979166667, 0.975], "retrieval": [0.979166667, 1.0, 0.979166667, 0.975], "rules_nodetect": [1.0, 0.979166667, 0.979166667, 0.975]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "79e90dd987f4f5687b9473491548cf6e96b1d8278dda805c39b946de62991030", "primary_control": "facts", "result_hash": "cc44997e961567b9647a8d583db3746fcb6d315eb299387be43a979feacf2257", "signature": "6604bf2c0ee11aeda109d5985c73e9017dc56f19b5ea5f3eb3a518748b70cecf3cc98f6c609e34fcc27cb4278f9caf7cf3cb047db61e278d0df4dcd6c1f8b205", "state": "VERIFIED", "summary": {"facts": 0.913542, "frozen": 0.0, "full": 0.983333, "retrieval": 0.983333, "rules_nodetect": 0.983333}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:41Z", "cpu_seconds": 0.4215371240000001, "created_at": "2026-11-22T04:28:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-65431feb37c6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "07ad883d86571af706be186f3a9f88b316db0c275ce16555e62a8c87b3cf1070", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-65431feb37c6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:40Z", "seed": 1508056324, "treatment": "full"}, "priority": 0.18662860572760614, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cc4693f83ef7676017ffa869fea10891f7b76236376f95446aaba9c1d571f5d2", "run_seconds": 0.43682241439819336, "seed": 1508056324, "started_at": "2026-11-22T04:28:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058791, 0.230792], "effect": 0.144792, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-65431feb37c6", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.71875, 0.828125, 0.766666667, 0.508333333], "frozen": [0.0, 0.125, 0.2, 0.0], "full": [0.96875, 0.890625, 0.95, 0.591666667], "retrieval": [0.921875, 0.890625, 0.95, 0.6], "rules_nodetect": [0.96875, 0.34375, 0.65, 0.591666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "07ad883d86571af706be186f3a9f88b316db0c275ce16555e62a8c87b3cf1070", "primary_control": "facts", "result_hash": "cc4693f83ef7676017ffa869fea10891f7b76236376f95446aaba9c1d571f5d2", "signature": "bbe7401e663d74a9958fcb099c6441c3710e690f5c85fefc9ecf5530bd713d4e71181babb3af4ef4f3342e59c4c0a9b37dcffaa2204d3acc50ce29d3e2fd7b07", "state": "VERIFIED", "summary": {"facts": 0.705469, "frozen": 0.08125, "full": 0.85026, "retrieval": 0.840625, "rules_nodetect": 0.638542}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:47Z", "cpu_seconds": 0.7132253319999997, "created_at": "2026-11-22T04:28:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ef27844bc354", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "7380ef1cfc2e01b54236e93ab6b7b3b7eb874c163e6d454ddabdbffcaaf61dac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ef27844bc354", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:40Z", "seed": 1885877152, "treatment": "full"}, "priority": 0.17804388918632894, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "18682bdc261338b60308f7ca6ce39b06a7b08ba58217f5bff33289f2498c5fda", "run_seconds": 0.7397787570953369, "seed": 1885877152, "started_at": "2026-11-22T04:28:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.035712, -0.008274], "effect": -0.021993, "evaluated_at": "2026-11-22T04:37:55Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-ef27844bc354", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.371428571, 0.435616438, 0.819354839, 0.247619048], "frozen": [0.123809524, 0.131506849, 0.135483871, 0.123809524], "full": [0.352380952, 0.397260274, 0.793548387, 0.242857143], "retrieval": [0.395238095, 0.449315068, 0.793548387, 0.223809524], "rules_nodetect": [0.404761905, 0.17260274, 0.374193548, 0.095238095]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "7380ef1cfc2e01b54236e93ab6b7b3b7eb874c163e6d454ddabdbffcaaf61dac", "primary_control": "facts", "result_hash": "18682bdc261338b60308f7ca6ce39b06a7b08ba58217f5bff33289f2498c5fda", "signature": "50b7c20da2e814afe652ac3af5b1a2d8ee49048009701c91a23e41d3025593ee44a694f1cc5e9c8e5b515f8bcea5511c517d8b854a19be5459073dee5fe3db0d", "state": "VERIFIED", "summary": {"facts": 0.468505, "frozen": 0.128652, "full": 0.446512, "retrieval": 0.465478, "rules_nodetect": 0.261699}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:37:55Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:32Z", "cpu_seconds": 0.85099408, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-725afcfa5111", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "b256379e48c73dd309c7a2d2adf12a4e5b939979a81342e0ebee02dc4301d94f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-725afcfa5111", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:56Z", "seed": 1560889226, "treatment": "frontier"}, "priority": 0.1776621122767684, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "995e0de864d9aefa7f5618bddf3a7dad1ef599b253286f19e907b9b407251b5f", "run_seconds": 1.533139944076538, "seed": 1560889226, "started_at": "2026-11-22T04:29:31Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.001027, 0.058973], "effect": 0.03, "evaluated_at": "2026-11-22T04:38:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-725afcfa5111", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 0.973333333, 1.0, 1.0], "goal": [1.0, 0.92, 0.973333333, 0.986666667], "hypdisc": [1.0, 1.0, 0.986666667, 0.973333333], "infogain": [1.0, 0.973333333, 1.0, 1.0], "prederr": [1.0, 0.973333333, 1.0, 1.0], "random": [1.0, 0.96, 0.96, 0.933333333], "risk": [1.0, 0.973333333, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "b256379e48c73dd309c7a2d2adf12a4e5b939979a81342e0ebee02dc4301d94f", "primary_control": "random", "result_hash": "995e0de864d9aefa7f5618bddf3a7dad1ef599b253286f19e907b9b407251b5f", "signature": "a86d1284e44daf5927558ef6ca9c19bd15932634c663ac63af38a4bfa87fd491757aead50d338a762fb4235eb4aea8709a14d97131659c0ac84424e2544c3400", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.993333, "goal": 0.97, "hypdisc": 0.99, "infogain": 0.993333, "prederr": 0.993333, "random": 0.963333, "risk": 0.993333}, "ties": 1, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:38:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:04Z", "cpu_seconds": 4.616727976, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-9ecbe13f366b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 2, "n": 4}, "prereg_hash": "37bf04125388efb51b3df36c90e6443ed36e33ae330c30969f52f9e485e7e394", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-9ecbe13f366b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:56Z", "seed": 1658236304, "treatment": "frontier"}, "priority": 0.17932478095965845, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "685e61418f02373e230a5f8f20cdc081d1358ca7aa4703c4f78aa5289249469b", "run_seconds": 5.884188652038574, "seed": 1658236304, "started_at": "2026-11-22T04:28:58Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:37:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-9ecbe13f366b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [1.0, 1.0, 1.0, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "37bf04125388efb51b3df36c90e6443ed36e33ae330c30969f52f9e485e7e394", "primary_control": "random", "result_hash": "685e61418f02373e230a5f8f20cdc081d1358ca7aa4703c4f78aa5289249469b", "signature": "55b1e316df0f675fd72e959fa2dc8d1778f4877c8cec8b2480d8d5ca5b9344c1ae35044dcda4e045ab7e175ee921865eee2779c48a57140f0324b9a09a1dd509", "state": "PARTIAL", "summary": {"count": 1.0, "frontier": 1.0, "goal": 1.0, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 1.0, "risk": 1.0}, "ties": 4, "treatment": "frontier", "wins": 0}, "verified_at": "2026-11-22T04:37:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:41Z", "cpu_seconds": 1.2378986870000013, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d031bc6fb639", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "b2f34a65947a8c4fe027c795b88664fd39ce62e89bc24c530a62021c151860c2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d031bc6fb639", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:28:56Z", "seed": 930230418, "treatment": "frontier"}, "priority": 0.17638559439297516, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b8a223c542f06192d1e990e89293718fab73b4a80fe8c7df928c10ed023de935", "run_seconds": 1.2461543083190918, "seed": 930230418, "started_at": "2026-11-22T04:29:40Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.061615, 0.329689], "effect": 0.195652, "evaluated_at": "2026-11-22T04:38:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d031bc6fb639", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 0.972826087, 0.983695652], "frontier": [1.0, 0.994565217, 1.0, 0.793478261], "goal": [0.782608696, 0.711956522, 0.695652174, 0.597826087], "hypdisc": [0.684782609, 0.684782609, 0.804347826, 0.744565217], "infogain": [1.0, 0.994565217, 1.0, 0.793478261], "prederr": [1.0, 0.994565217, 1.0, 0.793478261], "random": [0.701086957, 0.739130435, 0.766304348, 0.798913043], "risk": [1.0, 0.994565217, 1.0, 0.793478261]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "b2f34a65947a8c4fe027c795b88664fd39ce62e89bc24c530a62021c151860c2", "primary_control": "random", "result_hash": "b8a223c542f06192d1e990e89293718fab73b4a80fe8c7df928c10ed023de935", "signature": "2a7e0a08783ac4cc3039ecc52c22cb3d7482eda3af12319d7085b5b7da28db84eb00c42c292b8feec7ffef2033ed6b97f8aeb2e03f212c88bf07e992ce52b70b", "state": "VERIFIED", "summary": {"count": 0.98913, "frontier": 0.947011, "goal": 0.697011, "hypdisc": 0.72962, "infogain": 0.947011, "prederr": 0.947011, "random": 0.751359, "risk": 0.947011}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:38:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:11Z", "cpu_seconds": 11.312418059, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0ba4fe5cca5c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "225420670a303998362dcaadfaf71c469e2fd6ddf72e4425b40742c8fcb28dda", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0ba4fe5cca5c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:28:56Z", "seed": 506571459, "treatment": "full"}, "priority": 0.17859118641309032, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "93c2633217e8600a0cd9b48d3be387f46fe79b7d42fcc3d5dcaae8b01853f63b", "run_seconds": 12.720495462417603, "seed": 506571459, "started_at": "2026-11-22T04:28:58Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.058706, 0.691294], "effect": 0.375, "evaluated_at": "2026-11-22T04:37:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-0ba4fe5cca5c", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.25, 0.5, 0.0, 0.25], "fixed_policy": [0.0, 1.0, 0.0, 0.0], "full": [0.25, 1.0, 0.25, 1.0], "no_world_model": [0.25, 0.25, 0.25, 1.0], "random": [0.25, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "225420670a303998362dcaadfaf71c469e2fd6ddf72e4425b40742c8fcb28dda", "primary_control": "e141_best", "result_hash": "93c2633217e8600a0cd9b48d3be387f46fe79b7d42fcc3d5dcaae8b01853f63b", "signature": "4c1a6e0dfb32fdd0747a55d30a423c09f1961bfde9064197cd1d4e995501cfe827c6328d3c6b2dc1599ef2cca155219c7ed8e761d5f1ab26c8d87b6ffe60ed06", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.25, "full": 0.625, "no_world_model": 0.4375, "random": 0.5625}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:37:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:47Z", "cpu_seconds": 1.2975390329999996, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ea33752c3c73", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "8c97227f8a3caea1a9a8db8530d757e6facfd47356600630e0aeb51c499611f3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ea33752c3c73", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:28:56Z", "seed": 1763478263, "treatment": "full"}, "priority": 0.1761154344998014, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d00884c7be048ebeb983b695bcf4c76230a71111a9e903c436176c2164cd26c4", "run_seconds": 1.308222770690918, "seed": 1763478263, "started_at": "2026-11-22T04:29:46Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.019083, 0.855917], "effect": 0.4375, "evaluated_at": "2026-11-22T04:38:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-ea33752c3c73", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.25, 0.75], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 1.0, 0.25, 0.75], "random": [0.5, 0.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "8c97227f8a3caea1a9a8db8530d757e6facfd47356600630e0aeb51c499611f3", "primary_control": "e141_best", "result_hash": "d00884c7be048ebeb983b695bcf4c76230a71111a9e903c436176c2164cd26c4", "signature": "cda546f0a81b76df543ecd90804654c18ab7147734189f77d62d22c9e45a615632fcec8965df2307833cd4b3e2d064cc043c4dcab9b6eb8aa5ad1fe4e8295c01", "state": "VERIFIED", "summary": {"e141_best": 0.5, "fixed_policy": 0.0625, "full": 0.9375, "no_world_model": 0.75, "random": 0.375}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:38:06Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:28:59Z", "cpu_seconds": 0.790388272, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-368d38a2c355", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "8fa1a3fb2454e87bceb09790ee3104ba0e474061a5acd7e665b3e227bf44c4f9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-368d38a2c355", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:56Z", "seed": 1723535690, "treatment": "full"}, "priority": 0.19018143236852855, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "016992c8c64cf1e3511e9e99fe6f127cff3a3df210146beca0b64e155d4ee90c", "run_seconds": 1.3870375156402588, "seed": 1723535690, "started_at": "2026-11-22T04:28:57Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.002344, 0.015432], "effect": 0.006544, "evaluated_at": "2026-11-22T04:37:58Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-368d38a2c355", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.923076923, 0.604166667, 0.817307692, 0.8125], "frozen": [0.435897436, 0.291666667, 0.269230769, 0.535714286], "full": [0.923076923, 0.611111111, 0.836538462, 0.8125], "retrieval": [0.923076923, 0.618055556, 0.875, 0.8125], "rules_nodetect": [0.871794872, 0.722222222, 0.836538462, 0.839285714]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "8fa1a3fb2454e87bceb09790ee3104ba0e474061a5acd7e665b3e227bf44c4f9", "primary_control": "facts", "result_hash": "016992c8c64cf1e3511e9e99fe6f127cff3a3df210146beca0b64e155d4ee90c", "signature": "bb114aa8170dca32b875dd84684d810a9f117c571a3a119f83f2fa5e83143e26f684a6f052f516f7468b879f6044467642658186fe3067ef2f3922d0b269c301", "state": "VERIFIED", "summary": {"facts": 0.789263, "frozen": 0.383127, "full": 0.795807, "retrieval": 0.807158, "rules_nodetect": 0.81746}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:37:58Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:59Z", "cpu_seconds": 0.8203299929999979, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7262ff0f9658", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "ff7213bb048a3d219c3b3b4c4845e7b4acfd4bbedb9d917030566826397a46b8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7262ff0f9658", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:28:56Z", "seed": 1196309236, "treatment": "full"}, "priority": 0.1737116465669414, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "20cd273e0e0c56ccef5622b71dff6837600bc8490ce5e7dd6ddeda1cc67c14c3", "run_seconds": 0.915515661239624, "seed": 1196309236, "started_at": "2026-11-22T04:29:58Z", "state": "REJECTED", "verdict": {"evaluated_at": "2026-11-22T04:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7262ff0f9658", "family": "ow_model", "kind": "adversarial", "lab": "AGI_LAB", "mismatch": {"facts": [1.0, 0.895565], "frozen": [1.0, 0.395774], "full": [1.0, 0.915457], "retrieval": [1.0, 0.917742], "rules_nodetect": [1.0, 0.915457]}, "prereg_hash": "ff7213bb048a3d219c3b3b4c4845e7b4acfd4bbedb9d917030566826397a46b8", "reason": "claimed summary does not match recomputation from raw rows", "result_hash": "20cd273e0e0c56ccef5622b71dff6837600bc8490ce5e7dd6ddeda1cc67c14c3", "signature": "61cf57102c1d6a5c508be2469694c5e8034bfcb3bd55cb7e80333524df454242e62d3598c73802336ccfa8c89dd2fefc3f21918bb7aaf4a62282e29a1aeda508", "state": "REJECTED", "summary": {"facts": 0.895565, "frozen": 0.395774, "full": 0.915457, "retrieval": 0.917742, "rules_nodetect": 0.915457}}, "verified_at": "2026-11-22T04:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:53Z", "cpu_seconds": 0.9272195629999977, "created_at": "2026-11-22T04:28:56Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-009b26d37c2f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "prereg_hash": "d0b6fcd0830c62531f74f4fbaf9ae43799711984a1515f31cd3e5aa8239f98a8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-009b26d37c2f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:28:56Z", "seed": 1667348674, "treatment": "gated"}, "priority": 0.17471458892615493, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "1f22a546a89ff0653353bb9c90bd348ef6da1c05e2a757dbae8455fa3d565e26", "run_seconds": 1.0009286403656006, "seed": 1667348674, "started_at": "2026-11-22T04:29:52Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.025629, 0.06108], "effect": 0.017726, "evaluated_at": "2026-11-22T04:38:09Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-009b26d37c2f", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.382352941, 0.714285714, 0.464285714, 0.529411765], "gated": [0.382352941, 0.714285714, 0.464285714, 0.529411765], "naive": [0.367647059, 0.607142857, 0.330357143, 0.558823529], "scratch": [0.382352941, 0.75, 0.401785714, 0.485294118], "ungated": [0.529411765, 0.785714286, 0.589285714, 0.705882353]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "d0b6fcd0830c62531f74f4fbaf9ae43799711984a1515f31cd3e5aa8239f98a8", "primary_control": "scratch", "result_hash": "1f22a546a89ff0653353bb9c90bd348ef6da1c05e2a757dbae8455fa3d565e26", "signature": "97d19ce4e8f0b4c28120316e11060e9a1eac2b35806cdc7e843b72a5921687a4299641670b8681e6247df3a8ce706f77f951e96d031b817989d5d87c9f659b0a", "state": "VERIFIED", "summary": {"compressed": 0.522584, "gated": 0.522584, "naive": 0.465993, "scratch": 0.504858, "ungated": 0.652574}, "ties": 1, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:28Z", "cpu_seconds": 0.8179985859999981, "created_at": "2026-11-22T04:29:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-24f75d14ce81", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "891532498174511542b6bbbbaa3885394cbdbb4416040916367f635fd7e5e775", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-24f75d14ce81", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:29:11Z", "seed": 1977872082, "treatment": "full"}, "priority": 0.17878353503375538, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a42c8b5907aceca8c585e41d53f2be59fa5446c48d9a59db592f98420ca867e1", "run_seconds": 1.201880931854248, "seed": 1977872082, "started_at": "2026-11-22T04:29:27Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.42207, 0.04707], "effect": -0.1875, "evaluated_at": "2026-11-22T04:37:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-24f75d14ce81", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 1.0], "fixed_policy": [0.25, 0.5, 0.0, 0.0], "full": [0.5, 0.5, 0.75, 1.0], "no_world_model": [0.5, 1.0, 0.5, 1.0], "random": [0.0, 1.0, 0.75, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 2, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "891532498174511542b6bbbbaa3885394cbdbb4416040916367f635fd7e5e775", "primary_control": "e141_best", "result_hash": "a42c8b5907aceca8c585e41d53f2be59fa5446c48d9a59db592f98420ca867e1", "signature": "c7af4f275071ab9d76d7090bbad2b6331db449b884303f9dcf311d2ff1dbd2cd81b42fa965da1a47522f1743a1b43cb85da54ebb4248716b9fe90e368c21a300", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.1875, "full": 0.6875, "no_world_model": 0.75, "random": 0.6875}, "ties": 2, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:38:00Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:22Z", "cpu_seconds": 0.6225243319999993, "created_at": "2026-11-22T04:29:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-68503b55133c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "8697fb27e91ed70226908da04a1b05506395164be92cb00ce13a4f3e14e137e0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-68503b55133c", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:29:11Z", "seed": 1593752979, "treatment": "full"}, "priority": 0.18000430298747516, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b627c9d4aed53d3635d9c05c926d3624604d1c32610704f711611e6670277f7f", "run_seconds": 0.6643269062042236, "seed": 1593752979, "started_at": "2026-11-22T04:29:21Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.016714, 0.021079], "effect": 0.002183, "evaluated_at": "2026-11-22T04:37:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-68503b55133c", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.388888889, 0.761904762, 0.846153846, 0.857142857], "frozen": [0.333333333, 0.361904762, 0.358974359, 0.428571429], "full": [0.416666667, 0.742857143, 0.846153846, 0.857142857], "retrieval": [0.416666667, 0.742857143, 0.846153846, 0.857142857], "rules_nodetect": [0.444444444, 0.4, 0.487179487, 0.666666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "8697fb27e91ed70226908da04a1b05506395164be92cb00ce13a4f3e14e137e0", "primary_control": "facts", "result_hash": "b627c9d4aed53d3635d9c05c926d3624604d1c32610704f711611e6670277f7f", "signature": "a028cbb56fae2c661da9d5d56e58d2f08f7929853d633dac0fd6704042ecfcdb086f37b2d0453a351a0d0fc8127e2dd540d3e10bb1ff0dbdf2a4dc6440b1b706", "state": "VERIFIED", "summary": {"facts": 0.713523, "frozen": 0.370696, "full": 0.715705, "retrieval": 0.715705, "rules_nodetect": 0.499573}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:16Z", "cpu_seconds": 0.5983073240000003, "created_at": "2026-11-22T04:29:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b096b53d4b92", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "719f49fcc827279899c3c60612c313c081e1b899eb2cfacc5454e2f811b580c1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b096b53d4b92", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:29:11Z", "seed": 1953706162, "treatment": "full"}, "priority": 0.18702614498753442, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b8470584c5a6f0ca7e6cd3b173b08cea62ad93db5e97d6470190ca45fd08fa09", "run_seconds": 0.6017773151397705, "seed": 1953706162, "started_at": "2026-11-22T04:29:16Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.201592, 0.101661], "effect": -0.049966, "evaluated_at": "2026-11-22T04:37:59Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b096b53d4b92", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.4, 0.445652174, 0.443548387, 0.477272727], "frozen": [0.313333333, 0.358695652, 0.379032258, 0.272727273], "full": [0.23, 0.293478261, 0.399193548, 0.643939394], "retrieval": [0.316666667, 0.315217391, 0.375, 0.53030303], "rules_nodetect": [0.746666667, 0.836956522, 0.798387097, 0.696969697]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "719f49fcc827279899c3c60612c313c081e1b899eb2cfacc5454e2f811b580c1", "primary_control": "facts", "result_hash": "b8470584c5a6f0ca7e6cd3b173b08cea62ad93db5e97d6470190ca45fd08fa09", "signature": "112bfaf5580f63674dd185ee2b7ae7662dbe24c5727d41b370aa52a1f3b893ba67edd3b70db9a55d099d603864ad143d4ff0112e9f87933dca601873530d5a07", "state": "VERIFIED", "summary": {"facts": 0.441618, "frozen": 0.330947, "full": 0.391653, "retrieval": 0.384297, "rules_nodetect": 0.769745}, "ties": 0, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:37:59Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:08Z", "cpu_seconds": 4.120184608, "created_at": "2026-11-22T04:29:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1df97423366b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "e5fac1257985e82102d1a16c6d6b0554c779bfc6bf3b89bbf9cdd5fc43ce7ff1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1df97423366b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:29:28Z", "seed": 409290259, "treatment": "frontier"}, "priority": 0.17292225941162892, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1f35d727e94293fb2129554ec396e75db2eb5df75aea148c3375f5be2b3381ef", "run_seconds": 5.048072814941406, "seed": 409290259, "started_at": "2026-11-22T04:31:03Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.08488, 0.092213], "effect": 0.003666, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1df97423366b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 0.772727273], "frontier": [1.0, 0.865384615, 1.0, 0.806818182], "goal": [0.928571429, 0.846153846, 1.0, 0.818181818], "hypdisc": [1.0, 0.903846154, 1.0, 0.852272727], "infogain": [1.0, 0.865384615, 1.0, 0.806818182], "prederr": [1.0, 0.865384615, 1.0, 0.806818182], "random": [1.0, 0.980769231, 0.972222222, 0.704545455], "risk": [1.0, 0.865384615, 1.0, 0.806818182]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "e5fac1257985e82102d1a16c6d6b0554c779bfc6bf3b89bbf9cdd5fc43ce7ff1", "primary_control": "random", "result_hash": "1f35d727e94293fb2129554ec396e75db2eb5df75aea148c3375f5be2b3381ef", "signature": "c9952ff349094b592602f5f7a7a8e093091c4659c7ddb5cbf4df78e9663f0e58d0763f7384ea463b2c74e50fa14c745e8e516ecea093b67a2cd00e0b9dbe900d", "state": "VERIFIED", "summary": {"count": 0.943182, "frontier": 0.918051, "goal": 0.898227, "hypdisc": 0.93903, "infogain": 0.918051, "prederr": 0.918051, "random": 0.914384, "risk": 0.918051}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:38Z", "cpu_seconds": 0.8886862989999997, "created_at": "2026-11-22T04:29:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-06ec7c52a39f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 9, "n": 4}, "prereg_hash": "23eef6c7828109638284e21cab7b84d4aa41779e7f54e574535f517a8c252d33", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-06ec7c52a39f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:29:28Z", "seed": 252635364, "treatment": "full"}, "priority": 0.17645233830082396, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "19835b9ed9c6ccfb70b81ee91ea2ef6ab312c2463155a5f0e553533e508e632c", "run_seconds": 1.015852689743042, "seed": 252635364, "started_at": "2026-11-22T04:29:37Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.00723, 0.129739], "effect": 0.061254, "evaluated_at": "2026-11-22T04:38:01Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-06ec7c52a39f", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.529661017, 0.915625, 0.515833333, 0.545614035], "frozen": [0.421186441, 0.5125, 0.370833333, 0.435964912], "full": [0.569491525, 0.93125, 0.680833333, 0.570175439], "retrieval": [0.811864407, 0.95625, 0.8075, 0.812280702], "rules_nodetect": [0.569491525, 0.93125, 0.680833333, 0.570175439]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "23eef6c7828109638284e21cab7b84d4aa41779e7f54e574535f517a8c252d33", "primary_control": "facts", "result_hash": "19835b9ed9c6ccfb70b81ee91ea2ef6ab312c2463155a5f0e553533e508e632c", "signature": "e269ad3c7d379c0ca50a37500889c2a05b32e8e7d732f6bfaab256e4333f1b97ee3f995871f7d4e72750ed854990a483b2afa4848a831d8e7d225f18894d8c07", "state": "VERIFIED", "summary": {"facts": 0.626683, "frozen": 0.435121, "full": 0.687938, "retrieval": 0.846974, "rules_nodetect": 0.687938}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:38:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:35Z", "cpu_seconds": 0.6973008510000014, "created_at": "2026-11-22T04:29:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5eace14cf18a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 4, "n": 4}, "prereg_hash": "2af4748d8da1bf98874587a37dc0dd4997a5eaf8727ad6f8322da401fa0a2062", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5eace14cf18a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:29:28Z", "seed": 1649587177, "treatment": "full"}, "priority": 0.17744315895198318, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5e0c8ebb2bd9fb917838664f8e98d61afa83f8b4d7124c7958eb9f2e56cb2bd0", "run_seconds": 1.2616093158721924, "seed": 1649587177, "started_at": "2026-11-22T04:29:33Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.045245, 0.186005], "effect": 0.115625, "evaluated_at": "2026-11-22T04:38:00Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-5eace14cf18a", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.543333333, 0.65625, 0.505833333, 0.544166667], "frozen": [0.384166667, 0.42, 0.433333333, 0.411666667], "full": [0.5725, 0.74375, 0.696666667, 0.699166667], "retrieval": [0.830833333, 0.88625, 0.813333333, 0.848333333], "rules_nodetect": [0.5725, 0.74375, 0.696666667, 0.699166667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "2af4748d8da1bf98874587a37dc0dd4997a5eaf8727ad6f8322da401fa0a2062", "primary_control": "facts", "result_hash": "5e0c8ebb2bd9fb917838664f8e98d61afa83f8b4d7124c7958eb9f2e56cb2bd0", "signature": "28770c33e6e5d3a8704ee9bc68b9bbf830e913a319ee259f6ee37b90a38726bcdc1a6cd6b5e11d017b8d45663ee49aa3dce132bf05d27a52960fb3ff24efdd08", "state": "VERIFIED", "summary": {"facts": 0.562396, "frozen": 0.412292, "full": 0.678021, "retrieval": 0.844688, "rules_nodetect": 0.678021}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:38:01Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:40Z", "cpu_seconds": 112.83557683999999, "created_at": "2026-11-22T04:29:28Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-970a0ebe698c", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "89dafc2c72b41c1968bcab0c9a66c1905914da3544c6f4646a2c2cadce8aad01", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-970a0ebe698c", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:29:28Z", "seed": 1182083023, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "ae00ac665dc7c00821a27b3c98e6fe75d574ba9315e26acbc8466c38c075e1eb", "run_seconds": 129.09354400634766, "seed": 1182083023, "started_at": "2026-11-22T04:29:31Z", "state": "VERIFIED", "verdict": {"accepted_generations": {"compressed": 2, "inherit_mem": 2, "inherit_nomem": 2, "random_search": 1, "reset_mem": 2, "retrieval": 0, "rootcause": 1, "xfer_mem": 2}, "better": "higher", "ci95": [-0.040999, 0.229698], "effect": 0.09435, "evals": {"compressed": 13, "inherit_mem": 13, "inherit_nomem": 13, "random_search": 13, "reset_mem": 13, "retrieval": 13, "rootcause": 13, "xfer_mem": 13}, "evaluated_at": "2026-11-22T04:38:22Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_rsi-970a0ebe698c", "family": "ow_rsi", "final_configs": {"compressed": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "inherit_nomem": {"abstain_threshold": 0.35, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.75}, "random_search": {"abstain_threshold": 0.368, "change_detect": true, "change_window": 8, "goal_bias": 0.522, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "reset_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "retrieval": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "rootcause": {"abstain_threshold": 0.5, "change_detect": false, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"frontier": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}, "xfer_mem": {"abstain_threshold": 0.5, "change_detect": true, "change_window": 8, "goal_bias": 0.3, "history2": false, "mix": {"prederr": 0.6, "random": 1.0}, "planner_depth": 12, "reset_patience": 0, "rule_threshold": 0.9}}, "hypothesis_supported": false, "instance_scores": {"G0": [0.694444444, 1.0, 0.9625, 0.765625, 0.240849673, 0.370792564, 0.704545455, 0.451636905, 0.363333333, 0.250316456], "compressed": [0.416666667, 1.0, 0.9875, 0.932291667, 0.629248366, 0.245596869, 0.897727273, 0.553571429, 0.39, 0.694936709], "inherit_mem": [0.416666667, 1.0, 0.9875, 0.932291667, 0.629248366, 0.245596869, 0.897727273, 0.553571429, 0.39, 0.694936709], "inherit_nomem": [0.666666667, 1.0, 0.9375, 0.765625, 0.20751634, 0.370792564, 0.704545455, 0.451636905, 0.316666667, 0.720253165], "random_search": [0.666666667, 1.0, 0.9625, 0.765625, 0.240849673, 0.370792564, 0.704545455, 0.451636905, 0.333333333, 0.675949367], "reset_mem": [0.416666667, 1.0, 0.9875, 0.932291667, 0.629248366, 0.245596869, 0.897727273, 0.553571429, 0.39, 0.694936709], "retrieval": [0.694444444, 1.0, 0.9625, 0.765625, 0.240849673, 0.370792564, 0.704545455, 0.451636905, 0.363333333, 0.250316456], "rootcause": [0.694444444, 1.0, 0.9875, 0.932291667, 0.629248366, 0.245596869, 0.897727273, 0.618303571, 0.383333333, 0.365189873], "xfer_mem": [0.416666667, 1.0, 0.9875, 0.932291667, 0.629248366, 0.245596869, 0.897727273, 0.614583333, 0.491666667, 0.697468354]}, "kind": "primary", "lab": "RSI_LAB", "log10_p_sign": -1.0465, "losses": 2, "n": 10, "n_sealed": 10, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0898", "per_generation_sealed": [0.580404, 0.580404, 0.675364, 0.674754], "prereg_hash": "89dafc2c72b41c1968bcab0c9a66c1905914da3544c6f4646a2c2cadce8aad01", "primary_control": "G0", "recomputed_arms": ["inherit_mem", "xfer_mem", "compressed"], "result_hash": "ae00ac665dc7c00821a27b3c98e6fe75d574ba9315e26acbc8466c38c075e1eb", "signature": "1ce48cfe6d41b9be2736aa7645be087595294456192ba91955aa6c9181b198442c89f5ecfd1948adaa8c23158086f8f681a09d89b1087f6acb9a9c5d0b29c507", "state": "VERIFIED", "summary": {"G0": 0.580404, "compressed": 0.674754, "inherit_mem": 0.674754, "inherit_nomem": 0.61412, "random_search": 0.61719, "reset_mem": 0.674754, "retrieval": 0.580404, "rootcause": 0.675364, "xfer_mem": 0.691275}, "ties": 1, "treatment": "inherit_mem", "wins": 7}, "verified_at": "2026-11-22T04:40:51Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:44Z", "cpu_seconds": 0.6052968300000003, "created_at": "2026-11-22T04:29:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-44f07b0ced3b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "1056bf97b629473f8aadd9424a1c0cd8dd14ae9a0a38f0a3c5bef4dce28bf33e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-44f07b0ced3b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:29:43Z", "seed": 490639348, "treatment": "frontier"}, "priority": 0.18574936590767022, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6bbc88ccd2215c90c7dbf0f9b7785b1a3bf638931b64d2a87db933f5ad669dd8", "run_seconds": 0.6876885890960693, "seed": 490639348, "started_at": "2026-11-22T04:29:43Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.131884, 0.131572], "effect": -0.000156, "evaluated_at": "2026-11-22T04:38:02Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-44f07b0ced3b", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.81, 0.9375, 0.84375, 0.783333333], "frontier": [0.85, 0.916666667, 0.90625, 0.633333333], "goal": [0.86, 0.770833333, 0.75, 0.75], "hypdisc": [0.76, 0.895833333, 0.78125, 0.85], "infogain": [0.85, 0.916666667, 0.875, 0.683333333], "prederr": [0.88, 0.854166667, 0.59375, 0.666666667], "random": [0.76, 0.854166667, 0.859375, 0.833333333], "risk": [0.85, 0.916666667, 0.90625, 0.633333333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "1056bf97b629473f8aadd9424a1c0cd8dd14ae9a0a38f0a3c5bef4dce28bf33e", "primary_control": "random", "result_hash": "6bbc88ccd2215c90c7dbf0f9b7785b1a3bf638931b64d2a87db933f5ad669dd8", "signature": "b1e318b1544e1388671baf07da2430c8e28f2a16dadca50077374d72f0b9cf619d1a9baebda0543eac53d1b8aa57dad8ed99ddee50cbcc85437a8f6fb28e1304", "state": "VERIFIED", "summary": {"count": 0.843646, "frontier": 0.826562, "goal": 0.782708, "hypdisc": 0.821771, "infogain": 0.83125, "prederr": 0.748646, "random": 0.826719, "risk": 0.826562}, "ties": 0, "treatment": "frontier", "wins": 3}, "verified_at": "2026-11-22T04:38:02Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:50Z", "cpu_seconds": 1.2028934260000002, "created_at": "2026-11-22T04:29:43Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5f129e4c3606", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "ae90da8fdf954da41deab29e72a9d1a1cb5190f92cb32217417461866a8f2f49", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5f129e4c3606", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:29:43Z", "seed": 1635313353, "treatment": "full"}, "priority": 0.1751868798441499, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b46f4b0ab0e15863a587bb4f9ccc54194a04699bc1ab7fe8cbce7e597c8634af", "run_seconds": 1.285881757736206, "seed": 1635313353, "started_at": "2026-11-22T04:29:49Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:38:06Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5f129e4c3606", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.75, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 0.75, 0.75], "random": [0.75, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "ae90da8fdf954da41deab29e72a9d1a1cb5190f92cb32217417461866a8f2f49", "primary_control": "e141_best", "result_hash": "b46f4b0ab0e15863a587bb4f9ccc54194a04699bc1ab7fe8cbce7e597c8634af", "signature": "f1e547f8699f2e2187e6b1c3327a4b3574329dadc3fe6172a00b510c683cde5f04d82b6da6d2580de923849b6f7858563ee9a25662ec5137a5046d5f986c2509", "state": "VERIFIED", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 0.8125, "random": 0.6875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:38:09Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:01Z", "cpu_seconds": 0.889789597, "created_at": "2026-11-22T04:29:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7aa10a25dd97", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "02b270b14f478ee06fcf0e0721592b6fcbe0e2959d3dee40ab2c721aa4dbaf1c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7aa10a25dd97", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:29:43Z", "seed": 1657959879, "treatment": "full"}, "priority": 0.17295444897967704, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "80081054cfc9bcd4a74d54bc6252fe915fb2ee070194fa7249306617fb0f0143", "run_seconds": 1.540113925933838, "seed": 1657959879, "started_at": "2026-11-22T04:30:59Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003535, 0.02541], "effect": 0.010938, "evaluated_at": "2026-11-22T04:38:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-7aa10a25dd97", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.825, 1.0, 0.5625, 0.966666667], "frozen": [0.325, 0.283333333, 0.40625, 0.35], "full": [0.8375, 1.0, 0.59375, 0.966666667], "retrieval": [0.8625, 1.0, 0.75, 0.966666667], "rules_nodetect": [0.8375, 1.0, 0.59375, 0.966666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "02b270b14f478ee06fcf0e0721592b6fcbe0e2959d3dee40ab2c721aa4dbaf1c", "primary_control": "facts", "result_hash": "80081054cfc9bcd4a74d54bc6252fe915fb2ee070194fa7249306617fb0f0143", "signature": "3bc012f6efb471552c78747cca22fff227b21523df47a9383e24dbd295d5169b49d1fe112e70c13bad69e256b092eeb53fb4dd8bb4e53792d173c29d88047e04", "state": "VERIFIED", "summary": {"facts": 0.838542, "frozen": 0.341146, "full": 0.849479, "retrieval": 0.894792, "rules_nodetect": 0.849479}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:38:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:29:56Z", "cpu_seconds": 0.7624278319999993, "created_at": "2026-11-22T04:29:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b3b079e254ba", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 6, "n": 4}, "prereg_hash": "7ddda339d29c940419be4429da6d6a429a5022cf6881beff488ca3bddfa98161", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b3b079e254ba", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:29:43Z", "seed": 1354344458, "treatment": "full"}, "priority": 0.17382069684153237, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7695f176d3335db347b907eb4aed39a05917d87a57d8baac54fd5ae02760f267", "run_seconds": 1.0141730308532715, "seed": 1354344458, "started_at": "2026-11-22T04:29:55Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.013434, 0.133833], "effect": 0.073633, "evaluated_at": "2026-11-22T04:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-b3b079e254ba", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.498305085, 0.6875, 0.9, 0.49], "frozen": [0.439830508, 0.55, 0.55, 0.418333333], "full": [0.570338983, 0.8, 0.8875, 0.6125], "retrieval": [0.756779661, 0.75, 0.975, 0.795833333], "rules_nodetect": [0.453389831, 0.5, 0.625, 0.366666667]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "7ddda339d29c940419be4429da6d6a429a5022cf6881beff488ca3bddfa98161", "primary_control": "facts", "result_hash": "7695f176d3335db347b907eb4aed39a05917d87a57d8baac54fd5ae02760f267", "signature": "f33b87902426a8bceec827ed93e2a0c0ffd9a055fb17053b4044b8cd366de6b40fdc0bf990ac9027a6b4bc961367bf3c324b45eec68f7956345620624e95a30c", "state": "VERIFIED", "summary": {"facts": 0.643951, "frozen": 0.489541, "full": 0.717585, "retrieval": 0.819403, "rules_nodetect": 0.486264}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:38:10Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:15Z", "cpu_seconds": 0.88408262, "created_at": "2026-11-22T04:30:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ae5f4a05483a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "bbbf4fe339a97a11d9b0885fef8f04b87202dd83aca65f7739e3ca08ebe748ce", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ae5f4a05483a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:30:00Z", "seed": 1792083540, "treatment": "frontier"}, "priority": 0.17342331924270937, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "2b424f251e25b00789b7e45f065902a288e8e86cea3ed2d50fef672d27018b3d", "run_seconds": 1.7603580951690674, "seed": 1792083540, "started_at": "2026-11-22T04:30:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.027762, 0.127762], "effect": 0.05, "evaluated_at": "2026-11-22T04:38:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-ae5f4a05483a", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.95, 1.0, 0.983333333, 1.0], "hypdisc": [1.0, 1.0, 1.0, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.833333333, 1.0, 0.966666667, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "bbbf4fe339a97a11d9b0885fef8f04b87202dd83aca65f7739e3ca08ebe748ce", "primary_control": "random", "result_hash": "2b424f251e25b00789b7e45f065902a288e8e86cea3ed2d50fef672d27018b3d", "signature": "8ce998855583f2a974cc81824e64ccd177500046d8e02b4256ebcf802861bbe79fac85c85d58e9e6ec990b3c679526560977617615774cab3530c17cdd08180b", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 1.0, "goal": 0.983333, "hypdisc": 1.0, "infogain": 1.0, "prederr": 1.0, "random": 0.95, "risk": 1.0}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:08Z", "cpu_seconds": 0.729961522, "created_at": "2026-11-22T04:30:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-d522a085f8e5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "171a25102b7a91815d2d35e37ffce6851ba41e9969ebfab5a705e31e5fb605ac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-d522a085f8e5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:30:00Z", "seed": 302016984, "treatment": "frontier"}, "priority": 0.17680136853258582, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1af23635d7266ff0e4c1ec8f062d9da36b1fa41b7bb55caacdc787472f0060c8", "run_seconds": 0.876392126083374, "seed": 302016984, "started_at": "2026-11-22T04:30:08Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003337, 0.036671], "effect": 0.016667, "evaluated_at": "2026-11-22T04:38:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-d522a085f8e5", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [1.0, 0.979166667, 1.0, 0.975], "goal": [1.0, 0.895833333, 1.0, 0.925], "hypdisc": [1.0, 0.958333333, 1.0, 0.95], "infogain": [1.0, 0.979166667, 1.0, 0.975], "prederr": [1.0, 0.979166667, 1.0, 0.975], "random": [1.0, 0.9375, 1.0, 0.95], "risk": [1.0, 0.979166667, 1.0, 0.975]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "171a25102b7a91815d2d35e37ffce6851ba41e9969ebfab5a705e31e5fb605ac", "primary_control": "random", "result_hash": "1af23635d7266ff0e4c1ec8f062d9da36b1fa41b7bb55caacdc787472f0060c8", "signature": "51c65a67323445edbca3b06fc6681f4a3dc811e9f31c66e9e43a3007fb92cebf03f6b6299a5a6719f199fb8a953c8e18939ffe7687f1d4e5cd4e7f7a62ae160a", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.988542, "goal": 0.955208, "hypdisc": 0.977083, "infogain": 0.988542, "prederr": 0.988542, "random": 0.971875, "risk": 0.988542}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:32:09Z", "cpu_seconds": 5.065611634000007, "created_at": "2026-11-22T04:30:00Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f5994d68b87e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "bcc37fea6aafb74cb223fefbaa29ec2945642679e9a8078d776b2262a7a6ce6c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f5994d68b87e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:30:00Z", "seed": 651408016, "treatment": "frontier"}, "priority": 0.17191555111410978, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "31d4c91d9ab0e726fa1fa9f62d751c1d28593d2508b50e607df20874eb641589", "run_seconds": 5.573391914367676, "seed": 651408016, "started_at": "2026-11-22T04:32:03Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:30:12Z", "cpu_seconds": 0.7427889520000015, "created_at": "2026-11-22T04:30:00Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b8b63518bff4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "4060bb703bf40bc8f4aafa68f9bcfa510e4cf18559d9ef55e4463bc0969f7ed0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b8b63518bff4", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:30:00Z", "seed": 1421236814, "treatment": "full"}, "priority": 0.17460585101002846, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "665bc12ff70e9ee35fdabf59f3d9492ec25b60d63e289680a7e4369e50753400", "run_seconds": 0.7711596488952637, "seed": 1421236814, "started_at": "2026-11-22T04:30:11Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:38:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-b8b63518bff4", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 1.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 0.75, 1.0], "no_world_model": [1.0, 1.0, 0.75, 1.0], "random": [1.0, 1.0, 0.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "4060bb703bf40bc8f4aafa68f9bcfa510e4cf18559d9ef55e4463bc0969f7ed0", "primary_control": "e141_best", "result_hash": "665bc12ff70e9ee35fdabf59f3d9492ec25b60d63e289680a7e4369e50753400", "signature": "679b28e7f8f6f282e32a1e712088fe803979a22171fcf959070d870a450d0deab280a393bfe37e274c8913fe39f56bc87a65d648984038a6a688ac6ab3200308", "state": "PARTIAL", "summary": {"e141_best": 0.9375, "fixed_policy": 0.0625, "full": 0.9375, "no_world_model": 0.9375, "random": 0.75}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:38:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:06Z", "cpu_seconds": 1.3201669799999998, "created_at": "2026-11-22T04:30:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-09f173d229a1", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "e5a7868e541ff5ad22faf292dc8f277292bb89717a929de469cfd9c5f2cbd6fa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-09f173d229a1", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:30:00Z", "seed": 382213179, "treatment": "full"}, "priority": 0.17723084944824272, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-df732dd9e95a", "result_hash": "9a1af1eda9d5167f7cad39ff3e399a0c46d11334ef3bfd56bb64da5e72b2db7c", "run_seconds": 1.8067889213562012, "seed": 382213179, "started_at": "2026-11-22T04:30:04Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.02103, -0.00952], "effect": -0.015275, "evaluated_at": "2026-11-22T04:38:12Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-09f173d229a1", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.70952381, 0.711340206, 0.50177305, 0.726190476], "frozen": [0.178571429, 0.113402062, 0.106382979, 0.180952381], "full": [0.688095238, 0.692439863, 0.492907801, 0.714285714], "retrieval": [0.695238095, 0.697594502, 0.484042553, 0.723809524], "rules_nodetect": [0.688095238, 0.692439863, 0.734042553, 0.714285714]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.0, "losses": 4, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "1", "prereg_hash": "e5a7868e541ff5ad22faf292dc8f277292bb89717a929de469cfd9c5f2cbd6fa", "primary_control": "facts", "replicated": true, "replication_of": "E141-ow_model-df732dd9e95a", "result_hash": "9a1af1eda9d5167f7cad39ff3e399a0c46d11334ef3bfd56bb64da5e72b2db7c", "signature": "41276dde563919773395c63e2d20401e4b247e579db136d4d2cf93fe9d6ef84bb7aa6341445e1185eac8b4067711733d4d931aae873f121bb90e0eece4b0a60f", "state": "VERIFIED", "summary": {"facts": 0.662207, "frozen": 0.144827, "full": 0.646932, "retrieval": 0.650171, "rules_nodetect": 0.707216}, "ties": 0, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:38:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:03Z", "cpu_seconds": 0.7936699890000005, "created_at": "2026-11-22T04:30:00Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-dd1e53c2e1a5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 3, "n": 4}, "prereg_hash": "1cf7ca10e1fcf715bec958659545a2a93ad88d974c792bcc6d26f45c15bfdaf2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-dd1e53c2e1a5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:30:00Z", "seed": 677017688, "treatment": "full"}, "priority": 0.17807732163299786, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c1695248cb3e9435e29fad61a7253ab8e7250f08c38c5a003e5bd5270a26a700", "run_seconds": 0.9702551364898682, "seed": 677017688, "started_at": "2026-11-22T04:30:01Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.027807, 0.130036], "effect": 0.078922, "evaluated_at": "2026-11-22T04:38:10Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-dd1e53c2e1a5", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.459477124, 0.521568627, 0.453594771, 0.435294118], "frozen": [0.313071895, 0.416993464, 0.312418301, 0.335947712], "full": [0.555555556, 0.621568627, 0.57124183, 0.437254902], "retrieval": [0.766013072, 0.78627451, 0.798039216, 0.730065359], "rules_nodetect": [0.555555556, 0.621568627, 0.57124183, 0.437254902]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "1cf7ca10e1fcf715bec958659545a2a93ad88d974c792bcc6d26f45c15bfdaf2", "primary_control": "facts", "result_hash": "c1695248cb3e9435e29fad61a7253ab8e7250f08c38c5a003e5bd5270a26a700", "signature": "726a350be11167b04918478a970202ebdb898f754d1b7bb9f44d612e55463d9640654b7ee8a383befbebf95a5ee774fbe8a5019407a8109e1e19dd6b024a4b05", "state": "VERIFIED", "summary": {"facts": 0.467484, "frozen": 0.344608, "full": 0.546405, "retrieval": 0.770098, "rules_nodetect": 0.546405}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:38:12Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:25Z", "cpu_seconds": 4.079179728999999, "created_at": "2026-11-22T04:30:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e6b683047741", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "1152d2f40ed99cfd8e7a303a3bba823d58e6158139870e40f6256c8d15ceeb83", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e6b683047741", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:30:15Z", "seed": 1561693805, "treatment": "frontier"}, "priority": 0.17971432321800418, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "de7a89159a62b804815f4e61046fe38a1d7c01c23b3c9e0716dab8484fc3535c", "run_seconds": 5.235518455505371, "seed": 1561693805, "started_at": "2026-11-22T04:30:20Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.188266, 0.087245], "effect": -0.050511, "evaluated_at": "2026-11-22T04:38:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-e6b683047741", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.235714286, 0.400826446, 0.56043956, 0.268398268], "frontier": [0.317857143, 0.163911846, 0.538461538, 0.245670996], "goal": [0.317857143, 0.163911846, 0.472527473, 0.19047619], "hypdisc": [0.257142857, 0.314049587, 0.604395604, 0.228354978], "infogain": [0.328571429, 0.239669421, 0.538461538, 0.261904762], "prederr": [0.192857143, 0.283746556, 0.406593407, 0.251082251], "random": [0.342857143, 0.41046832, 0.450549451, 0.264069264], "risk": [0.317857143, 0.163911846, 0.538461538, 0.245670996]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.028, "losses": 3, "n": 4, "outcome": "CONTRADICTED", "p_sign_display": "0.937", "prereg_hash": "1152d2f40ed99cfd8e7a303a3bba823d58e6158139870e40f6256c8d15ceeb83", "primary_control": "random", "result_hash": "de7a89159a62b804815f4e61046fe38a1d7c01c23b3c9e0716dab8484fc3535c", "signature": "44e56d0a7aea1316fdf4eea9a0f090c7e3f16b1c66cb51dcea34b86b55a63445bf700bbb2301c9a308236c3c58c9718062cd3b9677d99356c1129737a6435a04", "state": "VERIFIED", "summary": {"count": 0.366345, "frontier": 0.316475, "goal": 0.286193, "hypdisc": 0.350986, "infogain": 0.342152, "prederr": 0.28357, "random": 0.366986, "risk": 0.316475}, "ties": 0, "treatment": "frontier", "wins": 1}, "verified_at": "2026-11-22T04:38:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:29Z", "cpu_seconds": 4.983157005999999, "created_at": "2026-11-22T04:30:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1a0ab4bc7fb3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "63436129fb6cd79e6858e6de223f2c34eb10c608c06c4dddf365e13053b60dd0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1a0ab4bc7fb3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:30:15Z", "seed": 1624339279, "treatment": "full"}, "priority": 0.1730087588210867, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d509e8397475cac3cdc422d250f33c75fd9c9974d63f0c0c7c43f19301dd32ad", "run_seconds": 6.943320035934448, "seed": 1624339279, "started_at": "2026-11-22T04:30:22Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.12, 0.37], "effect": 0.125, "evaluated_at": "2026-11-22T04:38:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-1a0ab4bc7fb3", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.0, 0.0, 0.75, 0.0], "fixed_policy": [0.0, 0.0, 0.25, 0.25], "full": [0.5, 0.0, 0.75, 0.0], "no_world_model": [0.0, 0.0, 1.0, 0.5], "random": [0.5, 1.0, 0.75, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "63436129fb6cd79e6858e6de223f2c34eb10c608c06c4dddf365e13053b60dd0", "primary_control": "e141_best", "result_hash": "d509e8397475cac3cdc422d250f33c75fd9c9974d63f0c0c7c43f19301dd32ad", "signature": "104bc3b8835f673bdfe63fd6ce389a8d4a8f66866d26280eaeeb241b4ff42b959ef284911481c7ecb21f659dde34f4e8dda51de724fb7664e349d75fd5d31803", "state": "VERIFIED", "summary": {"e141_best": 0.1875, "fixed_policy": 0.125, "full": 0.3125, "no_world_model": 0.375, "random": 0.5625}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:38:15Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:32:07Z", "cpu_seconds": 0.7915249039999992, "created_at": "2026-11-22T04:30:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-375f992888fc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "0ece15b6179cd1568c88b37e0b7a43e136dec9f24dd0f57769ec4811129d416a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-375f992888fc", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:30:15Z", "seed": 581226667, "treatment": "full"}, "priority": 0.17156736057707306, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "6912e51747075667cbc78b7498ed808dbd512316927835da726369ed9c0facd0", "run_seconds": 0.8381516933441162, "seed": 581226667, "started_at": "2026-11-22T04:32:05Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:30:17Z", "cpu_seconds": 0.3787593180000002, "created_at": "2026-11-22T04:30:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-425f89992744", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 1, "n": 4}, "prereg_hash": "79e548be95ea5d10bd04bf53dcf46c9f7ec320e8d6fee79f43622c5d82176911", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-425f89992744", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:30:15Z", "seed": 1229524385, "treatment": "full"}, "priority": 0.18363029479027002, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0dec840d3e714d868e63f5452c737f75d6266c636e2cb7d4c7789f6bf4ba133d", "run_seconds": 0.41570067405700684, "seed": 1229524385, "started_at": "2026-11-22T04:30:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.000281, 0.041386], "effect": 0.020833, "evaluated_at": "2026-11-22T04:38:13Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-425f89992744", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.983333333, 0.95, 0.983333333, 1.0], "frozen": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "retrieval": [1.0, 0.983333333, 1.0, 1.0], "rules_nodetect": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "79e548be95ea5d10bd04bf53dcf46c9f7ec320e8d6fee79f43622c5d82176911", "primary_control": "facts", "result_hash": "0dec840d3e714d868e63f5452c737f75d6266c636e2cb7d4c7789f6bf4ba133d", "signature": "92e5e53a9e8eadee77ba8343d03e30dd3073e0b7f6b4a138e43c5bd2ac073accfc29f414e6c531ae9f6e57eda204f8406aa9bf70f80b9ffa74c8dd60a7e69b02", "state": "VERIFIED", "summary": {"facts": 0.979167, "frozen": 0.0, "full": 1.0, "retrieval": 0.995833, "rules_nodetect": 1.0}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:38:13Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:32:12Z", "cpu_seconds": 4.507153998, "created_at": "2026-11-22T04:30:15Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1ed1843b78aa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "f004f7473d06f2579f0f07d8701db7ec612834836d31ab604e276f2f61ce1372", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1ed1843b78aa", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:30:15Z", "seed": 1618549660, "treatment": "gated"}, "priority": 0.17123990272080736, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "85f2874c21fe62a89734f812640a63b2fadd55972d89894b8761e4f7fd59c4c2", "run_seconds": 4.992833614349365, "seed": 1618549660, "started_at": "2026-11-22T04:32:07Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:30:46Z", "cpu_seconds": 0.940935192, "created_at": "2026-11-22T04:30:31Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a5f227c91059", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "1fa2d5de090c3215a5d4a7ff9ecb8cb99e3326deed3a1af1d3343e7ed5ea6a8b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a5f227c91059", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:30:31Z", "seed": 1873196578, "treatment": "frontier"}, "priority": 0.1808837159764482, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "125577968e629186f38ac7c53f081dc600aedb967bc505603fa8bfea4ed8c32e", "run_seconds": 1.095444679260254, "seed": 1873196578, "started_at": "2026-11-22T04:30:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.24402, 0.17866], "effect": -0.03268, "evaluated_at": "2026-11-22T04:38:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-a5f227c91059", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.294117647, 0.908496732, 0.431372549, 0.908496732], "frontier": [0.300653595, 0.718954248, 0.431372549, 0.470588235], "goal": [0.300653595, 0.516339869, 0.431372549, 0.784313725], "hypdisc": [0.307189542, 0.718954248, 0.424836601, 0.843137255], "infogain": [0.333333333, 0.718954248, 0.405228758, 0.725490196], "prederr": [0.307189542, 0.300653595, 0.424836601, 0.830065359], "random": [0.320261438, 0.529411765, 0.405228758, 0.797385621], "risk": [0.287581699, 0.692810458, 0.437908497, 0.470588235]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "1fa2d5de090c3215a5d4a7ff9ecb8cb99e3326deed3a1af1d3343e7ed5ea6a8b", "primary_control": "random", "result_hash": "125577968e629186f38ac7c53f081dc600aedb967bc505603fa8bfea4ed8c32e", "signature": "9634dfbcf467a6d4e5058bc664a289ee0002de86e102a5c98bded019c00ffb36f56b6ba525fcda54e84668aec5daf73514f8e0641fcdf1dde276066571d0490c", "state": "VERIFIED", "summary": {"count": 0.635621, "frontier": 0.480392, "goal": 0.50817, "hypdisc": 0.573529, "infogain": 0.545752, "prederr": 0.465686, "random": 0.513072, "risk": 0.472222}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:58Z", "cpu_seconds": 0.9955133959999998, "created_at": "2026-11-22T04:30:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-267b05d62f6e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "9af6f2b72d2124b05cc3ad061d557909aa41ebb995750f7ab92b74e84accf499", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-267b05d62f6e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:30:31Z", "seed": 2003665593, "treatment": "full"}, "priority": 0.17424989819069517, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "62eb7580c501ef796c4471a873cbd2c39d5553c8dc176994c541cbdf725f87cf", "run_seconds": 1.2747836112976074, "seed": 2003665593, "started_at": "2026-11-22T04:30:56Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.06, 0.185], "effect": 0.0625, "evaluated_at": "2026-11-22T04:38:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-267b05d62f6e", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 1.0], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [0.75, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [0.5, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "9af6f2b72d2124b05cc3ad061d557909aa41ebb995750f7ab92b74e84accf499", "primary_control": "e141_best", "result_hash": "62eb7580c501ef796c4471a873cbd2c39d5553c8dc176994c541cbdf725f87cf", "signature": "e1d1a03fd51df22a792896ed9c5e958d6bfae7a2e5b8c4c5ec0eb953e18ed8b3f34aa35e2bfb80e53516154f8d8811d482073fe97ef28e0dfeaba5c605daba06", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.0, "full": 0.9375, "no_world_model": 0.9375, "random": 0.875}, "ties": 3, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:38:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:46Z", "cpu_seconds": 0.6535901000000001, "created_at": "2026-11-22T04:30:31Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-98787a356309", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "30dc4368e2bd0b879df75ec15923840a6165f8ee036ab77ea00f7eafaf9d943b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-98787a356309", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:30:31Z", "seed": 1538713855, "treatment": "full"}, "priority": 0.18048953166579515, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "90781f891069266da4934ef22a9531a948697dcba87019f86f97e1d87e40541f", "run_seconds": 0.8456542491912842, "seed": 1538713855, "started_at": "2026-11-22T04:30:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.03705, 0.163734], "effect": 0.100392, "evaluated_at": "2026-11-22T04:38:15Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-98787a356309", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.541176471, 0.563137255, 0.543529412, 0.576470588], "frozen": [0.35372549, 0.436862745, 0.381176471, 0.391372549], "full": [0.676078431, 0.567843137, 0.661960784, 0.72], "retrieval": [0.823529412, 0.861176471, 0.838431373, 0.849411765], "rules_nodetect": [0.676078431, 0.567843137, 0.661960784, 0.72]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "30dc4368e2bd0b879df75ec15923840a6165f8ee036ab77ea00f7eafaf9d943b", "primary_control": "facts", "result_hash": "90781f891069266da4934ef22a9531a948697dcba87019f86f97e1d87e40541f", "signature": "056f4a378d28449f41d601e971a2027dd5a74b1772268cc39056aa88e6f58a1f5bb044b9603a4e808ba4bf02bfba171d6ad0e1a5086f405cb577d9ae6a443500", "state": "VERIFIED", "summary": {"facts": 0.556078, "frozen": 0.390784, "full": 0.656471, "retrieval": 0.843137, "rules_nodetect": 0.656471}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:38:17Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:54Z", "cpu_seconds": 3.4657688290000004, "created_at": "2026-11-22T04:30:47Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1124c10d7553", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "26e3315ad90c2886c7aa3d66b1267bd6dc7272cca045c4acf7f2015dbaf274db", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1124c10d7553", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:30:47Z", "seed": 1676234439, "treatment": "frontier"}, "priority": 0.33678607638726377, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a0b16ea9352cf3f63225cd518e176617f49a532662738a9a4e7eaaaca96fd535", "run_seconds": 3.7879724502563477, "seed": 1676234439, "started_at": "2026-11-22T04:30:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.201631, 0.080576], "effect": -0.060528, "evaluated_at": "2026-11-22T04:38:18Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-1124c10d7553", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.75, 0.311497326, 1.0, 0.303289474], "frontier": [0.359375, 0.308823529, 0.982142857, 0.298026316], "goal": [0.604166667, 0.284759358, 0.928571429, 0.297368421], "hypdisc": [0.609375, 0.302139037, 1.0, 0.303289474], "infogain": [0.395833333, 0.309491979, 1.0, 0.307236842], "prederr": [0.520833333, 0.292112299, 0.714285714, 0.286842105], "random": [0.635416667, 0.290775401, 0.964285714, 0.3], "risk": [0.359375, 0.308823529, 0.982142857, 0.298026316]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "26e3315ad90c2886c7aa3d66b1267bd6dc7272cca045c4acf7f2015dbaf274db", "primary_control": "random", "result_hash": "a0b16ea9352cf3f63225cd518e176617f49a532662738a9a4e7eaaaca96fd535", "signature": "97aeb770672c3aa8720075c985bc0f02a2adb974cb65e9d5a197a6c53ee7ec464a15e9b76856041a1dcd0b11f031a3fa336968f27eaae0aabfea6d3b72accd08", "state": "VERIFIED", "summary": {"count": 0.591197, "frontier": 0.487092, "goal": 0.528716, "hypdisc": 0.553701, "infogain": 0.503141, "prederr": 0.453518, "random": 0.547619, "risk": 0.487092}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:19Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:30:51Z", "cpu_seconds": 0.7313102160000002, "created_at": "2026-11-22T04:30:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e1b6fdbff68b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "fba298e098eb03161b8d44fbd8d7fc1cee693a6cd79928d8a9348d396672cc90", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e1b6fdbff68b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:30:47Z", "seed": 1845624894, "treatment": "full"}, "priority": 0.1907716922015615, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "cf741d5440f1073477e4d000306c46efa95bace66fb678a41fdb172ea36199a4", "run_seconds": 0.7492437362670898, "seed": 1845624894, "started_at": "2026-11-22T04:30:51Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.04707, 0.42207], "effect": 0.1875, "evaluated_at": "2026-11-22T04:38:17Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-e1b6fdbff68b", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [0.5, 1.0, 1.0, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 1.0, 0.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "fba298e098eb03161b8d44fbd8d7fc1cee693a6cd79928d8a9348d396672cc90", "primary_control": "e141_best", "result_hash": "cf741d5440f1073477e4d000306c46efa95bace66fb678a41fdb172ea36199a4", "signature": "3c34ffad0f909317fe01055000283225f38216741999ecad48ae8c656db6879f2b4a00bf7b82a4e90d3c7f34e5cf7b274d014e9c90acf9b6da55653456bea309", "state": "VERIFIED", "summary": {"e141_best": 0.8125, "fixed_policy": 0.0, "full": 1.0, "no_world_model": 1.0, "random": 0.75}, "ties": 2, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:38:18Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:14Z", "cpu_seconds": 0.9852434109999999, "created_at": "2026-11-22T04:31:03Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c0d3e090b627", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "596134335aa5a307aa9769e6731c8577c4c7090efe6da168e8d4fe3d81c9cf1c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c0d3e090b627", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:31:03Z", "seed": 1719978244, "treatment": "frontier"}, "priority": 0.18037720622728287, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d92d2790ec944244d69cf5bf1e29d8da3077a46bfa1c49b3ee6773d3f351c50f", "run_seconds": 0.9983270168304443, "seed": 1719978244, "started_at": "2026-11-22T04:31:13Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.130626, 0.054869], "effect": -0.037879, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-c0d3e090b627", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.717948718, 0.882352941, 0.696969697, 0.743589744], "frontier": [0.384615385, 0.882352941, 0.696969697, 0.653846154], "goal": [0.615384615, 0.882352941, 0.636363636, 0.448717949], "hypdisc": [0.615384615, 0.882352941, 0.606060606, 0.423076923], "infogain": [0.717948718, 0.843137255, 0.651515152, 0.551282051], "prederr": [0.743589744, 0.843137255, 0.287878788, 0.705128205], "random": [0.564102564, 0.882352941, 0.681818182, 0.641025641], "risk": [0.384615385, 0.882352941, 0.696969697, 0.653846154]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "596134335aa5a307aa9769e6731c8577c4c7090efe6da168e8d4fe3d81c9cf1c", "primary_control": "random", "result_hash": "d92d2790ec944244d69cf5bf1e29d8da3077a46bfa1c49b3ee6773d3f351c50f", "signature": "8329810f3b49728b19d288ae4e0cfc62da555df290bd5fbad016828b1498583755fe47ba673935a0eb8e2a092ad360b43c35184280d15d1b1f78d2f3694f1704", "state": "VERIFIED", "summary": {"count": 0.760215, "frontier": 0.654446, "goal": 0.645705, "hypdisc": 0.631719, "infogain": 0.690971, "prederr": 0.644933, "random": 0.692325, "risk": 0.654446}, "ties": 1, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:12Z", "cpu_seconds": 0.6169542159999999, "created_at": "2026-11-22T04:31:03Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-5ba229f3ee18", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "c5d25648cc539ff2d25282f5b27f425a79842803c35bf60447697595f5ce8dee", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5ba229f3ee18", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:31:03Z", "seed": 821207636, "treatment": "full"}, "priority": 0.1847981854403386, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7a700f231941b2d89dbb59486aa84e0b76a527988ef4271e3721229edd5d3311", "run_seconds": 0.6173465251922607, "seed": 821207636, "started_at": "2026-11-22T04:31:12Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.370785, 0.245785], "effect": -0.0625, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-5ba229f3ee18", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 0.75, 0.75], "fixed_policy": [0.0, 0.0, 0.0, 0.75], "full": [1.0, 1.0, 0.25, 1.0], "no_world_model": [1.0, 1.0, 0.25, 0.75], "random": [1.0, 0.75, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "c5d25648cc539ff2d25282f5b27f425a79842803c35bf60447697595f5ce8dee", "primary_control": "e141_best", "result_hash": "7a700f231941b2d89dbb59486aa84e0b76a527988ef4271e3721229edd5d3311", "signature": "193953c6d5fbec83947f414b76940945a739be54594242305551d76a9e1b3f56552c2e1da5bf9fff42bad1983c65fcd39535fa15bd306dfa88f649c9844d000b", "state": "VERIFIED", "summary": {"e141_best": 0.875, "fixed_policy": 0.1875, "full": 0.8125, "no_world_model": 0.75, "random": 0.9375}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:07Z", "cpu_seconds": 0.48840327100000014, "created_at": "2026-11-22T04:31:03Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-393129c1c4cc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "1c3af5f0c1de03d8aee78794369ecc35a6178b1733ac776cd8d1084e9bbf45ac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-393129c1c4cc", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:03Z", "seed": 230522204, "treatment": "full"}, "priority": 0.19103518840890027, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "8415ba1c3e91dc29d086caefdafaaf013f9bc47cc7f5bbf62d0dacdd0bacf9b1", "run_seconds": 0.6033689975738525, "seed": 230522204, "started_at": "2026-11-22T04:31:06Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.073944, 0.030312], "effect": -0.021816, "evaluated_at": "2026-11-22T04:38:19Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-393129c1c4cc", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.711340206, 0.777777778, 0.464646465, 0.843137255], "frozen": [0.281786942, 0.277777778, 0.343434343, 0.352941176], "full": [0.725085911, 0.777777778, 0.363636364, 0.843137255], "retrieval": [0.738831615, 0.777777778, 0.484848485, 0.862745098], "rules_nodetect": [0.725085911, 0.777777778, 0.757575758, 0.843137255]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "1c3af5f0c1de03d8aee78794369ecc35a6178b1733ac776cd8d1084e9bbf45ac", "primary_control": "facts", "result_hash": "8415ba1c3e91dc29d086caefdafaaf013f9bc47cc7f5bbf62d0dacdd0bacf9b1", "signature": "cf0b0fbdf97ad64ce436ca07bcb0e5fb15a8aa8696d8eeedf7d5c261917d19357a420db462bd6a8e4bc502a5eb02c44c73f66bedee5b6bd7b2f6b5d999b1da09", "state": "VERIFIED", "summary": {"facts": 0.699225, "frozen": 0.313985, "full": 0.677409, "retrieval": 0.716051, "rules_nodetect": 0.775894}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:18Z", "cpu_seconds": 0.6608408599999995, "created_at": "2026-11-22T04:31:03Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-cc341d0c9437", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "cbe03ac94e076d0a2351b4fc117607359c9bf424db3b024bebd4e6c7ae51dd85", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-cc341d0c9437", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:03Z", "seed": 1698041960, "treatment": "full"}, "priority": 0.17651145025265885, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f4a18e19d6605af1c981759ff7b62692a69273b26d7b0c38767f4be1560629b7", "run_seconds": 0.711249828338623, "seed": 1698041960, "started_at": "2026-11-22T04:31:17Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.028712, 0.050934], "effect": 0.011111, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-cc341d0c9437", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.486111111, 1.0, 0.764705882, 0.65], "frozen": [0.333333333, 0.357142857, 0.632352941, 0.35], "full": [0.555555556, 1.0, 0.764705882, 0.625], "retrieval": [0.513888889, 1.0, 0.764705882, 0.75], "rules_nodetect": [0.555555556, 1.0, 0.764705882, 0.625]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.1249, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.75", "prereg_hash": "cbe03ac94e076d0a2351b4fc117607359c9bf424db3b024bebd4e6c7ae51dd85", "primary_control": "facts", "result_hash": "f4a18e19d6605af1c981759ff7b62692a69273b26d7b0c38767f4be1560629b7", "signature": "320f273cc34c50ea441c24ee346ed5de5bb7b3acdc9186bb05db50a536a8e88c07e62afa6e0ff20ae3b0e754f62174a6c4742eff70a070370a3690370676620b", "state": "VERIFIED", "summary": {"facts": 0.725204, "frozen": 0.418207, "full": 0.736315, "retrieval": 0.757149, "rules_nodetect": 0.736315}, "ties": 2, "treatment": "full", "wins": 1}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:31Z", "cpu_seconds": 0.7786802510000008, "created_at": "2026-11-22T04:31:03Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-fab5874a8feb", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "23ce970df363ced2689d9f01a88e84a2fb3564f91b832de18ee6efb81366ae90", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-fab5874a8feb", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:31:03Z", "seed": 1673430767, "treatment": "gated"}, "priority": 0.17404161402723828, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "73b0bff40bb5fc867e8bd6bd17f23098a485f6a17667ff268b82f2a965682590", "run_seconds": 0.9620277881622314, "seed": 1673430767, "started_at": "2026-11-22T04:31:30Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.021, 0.06475], "effect": 0.021875, "evaluated_at": "2026-11-22T04:38:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-fab5874a8feb", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.75625, 0.469148936, 0.695833333, 0.658333333], "gated": [0.84375, 0.469148936, 0.695833333, 0.658333333], "naive": [0.75625, 0.385106383, 0.503125, 0.658333333], "scratch": [0.75625, 0.469148936, 0.695833333, 0.658333333], "ungated": [0.8125, 0.530851064, 0.6, 0.564583333]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "23ce970df363ced2689d9f01a88e84a2fb3564f91b832de18ee6efb81366ae90", "primary_control": "scratch", "result_hash": "73b0bff40bb5fc867e8bd6bd17f23098a485f6a17667ff268b82f2a965682590", "signature": "eb7c6191f518629e84c617d61b1646b8aaf003e0c1212d3a144da35017847b8de26f7c951336d41281d8bbd0fd0a35b905d93775cc89c4ec2a06a746e2061a00", "state": "VERIFIED", "summary": {"compressed": 0.644891, "gated": 0.666766, "naive": 0.575704, "scratch": 0.644891, "ungated": 0.626984}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:38:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:24Z", "cpu_seconds": 0.7445469249999999, "created_at": "2026-11-22T04:31:18Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-38ffe1d2755e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "9a5bb7e0deffed2eda5efb9ad56362d8655e94aee49ca789459def1fc0382451", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-38ffe1d2755e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:31:18Z", "seed": 385012772, "treatment": "frontier"}, "priority": 0.1878123906269446, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fcbdee11dc343d104eb603c6ea08696014453ef6ca38ece9f81264c58f2de10b", "run_seconds": 0.8459315299987793, "seed": 385012772, "started_at": "2026-11-22T04:31:23Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.003667, 0.045333], "effect": 0.020833, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-38ffe1d2755e", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [0.983333333, 1.0, 1.0, 1.0], "frontier": [1.0, 1.0, 1.0, 1.0], "goal": [0.966666667, 0.85, 0.95, 1.0], "hypdisc": [1.0, 1.0, 0.933333333, 1.0], "infogain": [1.0, 1.0, 1.0, 1.0], "prederr": [1.0, 1.0, 1.0, 1.0], "random": [0.966666667, 1.0, 0.95, 1.0], "risk": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "9a5bb7e0deffed2eda5efb9ad56362d8655e94aee49ca789459def1fc0382451", "primary_control": "random", "result_hash": "fcbdee11dc343d104eb603c6ea08696014453ef6ca38ece9f81264c58f2de10b", "signature": "266bd52136e2c4d368a9aafd43499e15a9c15f3282ee46c10c773fcf1d7b6d9d81d08258aa388cc26d8eccf4767feb2d2c0982a486c6b21042579c23e052c70e", "state": "VERIFIED", "summary": {"count": 0.995833, "frontier": 1.0, "goal": 0.941667, "hypdisc": 0.983333, "infogain": 1.0, "prederr": 1.0, "random": 0.979167, "risk": 1.0}, "ties": 2, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:20Z", "cpu_seconds": 0.6534582060000016, "created_at": "2026-11-22T04:31:18Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-426d459e56af", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "f789d452fcc965325204e85b3de5eb3680071ffd6273394a80342f98e85b9397", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-426d459e56af", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:31:18Z", "seed": 536014729, "treatment": "frontier"}, "priority": 0.1906386212522758, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-8a80b6767ccc", "result_hash": "1dd09904de88aa15287e5e50cf0adc158f1c6dff50782f74325e357db0da18de", "run_seconds": 0.713719367980957, "seed": 536014729, "started_at": "2026-11-22T04:31:19Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.036882, 0.047298], "effect": 0.005208, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_explore-426d459e56af", "family": "ow_explore", "hypothesis_supported": false, "instance_scores": {"count": [1.0, 1.0, 1.0, 1.0], "frontier": [0.9375, 0.979166667, 0.958333333, 0.958333333], "goal": [0.958333333, 0.875, 0.958333333, 0.770833333], "hypdisc": [0.979166667, 0.958333333, 0.958333333, 0.958333333], "infogain": [0.9375, 0.979166667, 0.958333333, 0.958333333], "prederr": [0.9375, 0.979166667, 0.958333333, 0.958333333], "random": [0.958333333, 0.9375, 1.0, 0.916666667], "risk": [0.9375, 0.979166667, 0.958333333, 0.958333333]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.1627, "losses": 2, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.687", "prereg_hash": "f789d452fcc965325204e85b3de5eb3680071ffd6273394a80342f98e85b9397", "primary_control": "random", "replicated": true, "replication_of": "E141-ow_explore-8a80b6767ccc", "result_hash": "1dd09904de88aa15287e5e50cf0adc158f1c6dff50782f74325e357db0da18de", "signature": "e685a1bc1f1342972db8a5fe8394881e10a955e15cccbb8b449f0c8cb248ee2b36ba13877d0cbb3b9081de72a09734a62c26fd9e7946fd1ae6f902998949a508", "state": "VERIFIED", "summary": {"count": 1.0, "frontier": 0.958333, "goal": 0.890625, "hypdisc": 0.963542, "infogain": 0.958333, "prederr": 0.958333, "random": 0.953125, "risk": 0.958333}, "ties": 0, "treatment": "frontier", "wins": 2}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:25Z", "cpu_seconds": 0.5929551960000001, "created_at": "2026-11-22T04:31:18Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2dc9e5795892", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "21010954b595cefbc376440e893c49a8994ddcb2e090f53f75c049b94dd68f69", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2dc9e5795892", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:18Z", "seed": 1780377756, "treatment": "full"}, "priority": 0.18576160854835339, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ba1f0e8f27b809c3e28c392d46fd924ce1991c9f9bbd9eebafa2bc175594ed64", "run_seconds": 0.6402449607849121, "seed": 1780377756, "started_at": "2026-11-22T04:31:25Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.007836, 0.030056], "effect": 0.01111, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-2dc9e5795892", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.815602837, 0.901960784, 0.928104575, 0.615384615], "frozen": [0.092198582, 0.091503268, 0.091503268, 0.083333333], "full": [0.808510638, 0.908496732, 0.934640523, 0.653846154], "retrieval": [0.836879433, 0.928104575, 0.928104575, 0.769230769], "rules_nodetect": [0.808510638, 0.908496732, 0.934640523, 0.653846154]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.5051, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.312", "prereg_hash": "21010954b595cefbc376440e893c49a8994ddcb2e090f53f75c049b94dd68f69", "primary_control": "facts", "result_hash": "ba1f0e8f27b809c3e28c392d46fd924ce1991c9f9bbd9eebafa2bc175594ed64", "signature": "528a260879e2b1f3e3daeceb57e20767716095db1c11da9a21b5abb3c59fe29f5e47d65bc78ac487b5d51a06f32b41f942934c142fed77f088c9fa59144cce02", "state": "VERIFIED", "summary": {"facts": 0.815263, "frozen": 0.089635, "full": 0.826374, "retrieval": 0.86558, "rules_nodetect": 0.826374}, "ties": 0, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:38:20Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:30Z", "cpu_seconds": 0.6916933519999997, "created_at": "2026-11-22T04:31:18Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-6d43f8b5f82a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "482ee1e3a4b28828cd3d63c98f4ccd97b28d074a64d4d5a00608183009176c70", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-6d43f8b5f82a", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:31:18Z", "seed": 1785171328, "treatment": "gated"}, "priority": 0.18107086562750213, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "476e0bbe7e7fc3c8724b220e5099bb8fcf40a0540c377afe40769722a2757430", "run_seconds": 0.773303747177124, "seed": 1785171328, "started_at": "2026-11-22T04:31:29Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.028235, 0.087059], "effect": 0.029412, "evaluated_at": "2026-11-22T04:38:20Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-6d43f8b5f82a", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.39869281, 0.241830065, 0.411764706, 0.39869281], "gated": [0.39869281, 0.241830065, 0.411764706, 0.39869281], "naive": [0.39869281, 0.222222222, 0.261437908, 0.241830065], "scratch": [0.39869281, 0.241830065, 0.411764706, 0.281045752], "ungated": [0.222222222, 0.196078431, 0.418300654, 0.346405229]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "482ee1e3a4b28828cd3d63c98f4ccd97b28d074a64d4d5a00608183009176c70", "primary_control": "scratch", "result_hash": "476e0bbe7e7fc3c8724b220e5099bb8fcf40a0540c377afe40769722a2757430", "signature": "6ce90112a8c21e674ff4f4362cc3ba82423e35f16ff5044538b247c8d82e4eeec43e20be4afffe5c1c62855fef6aac5dfe21c7c3a746db41c9dfff0aa5746903", "state": "VERIFIED", "summary": {"compressed": 0.362745, "gated": 0.362745, "naive": 0.281046, "scratch": 0.333333, "ungated": 0.295752}, "ties": 3, "treatment": "gated", "wins": 1}, "verified_at": "2026-11-22T04:38:21Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:54Z", "cpu_seconds": 6.467370029, "created_at": "2026-11-22T04:31:35Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-012290944169", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 9, "n": 4}, "prereg_hash": "e7f376b9eda9a052efeacfff1e5e400720516e4cfc3650642a26dba502648a66", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-012290944169", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:31:35Z", "seed": 107556166, "treatment": "frontier"}, "priority": 0.17824270226719602, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "92342f724e76e1b7cdfdd4dde46b10b2479edc00dcab4159f9d5541591bf1e85", "run_seconds": 7.1268532276153564, "seed": 107556166, "started_at": "2026-11-22T04:31:47Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:31:50Z", "cpu_seconds": 1.8770411320000004, "created_at": "2026-11-22T04:31:35Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8f970be9e461", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "5d6aa5820cbfdadbf41d35e647ad521451c3e1017b0e0a7454d3e5617aa10c17", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8f970be9e461", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:31:35Z", "seed": 1682997071, "treatment": "full"}, "priority": 0.17683369174424485, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c54266179113c3c1ecda196788b5fb40021ead01bec5f0cb3a99e9ec44a0cb2a", "run_seconds": 2.1578447818756104, "seed": 1682997071, "started_at": "2026-11-22T04:31:47Z", "state": "PARTIAL", "verdict": {"better": "higher", "ci95": [0.0, 0.0], "effect": 0.0, "evaluated_at": "2026-11-22T04:40:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-8f970be9e461", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 1.0, 1.0, 1.0], "fixed_policy": [1.0, 1.0, 0.0, 1.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [1.0, 1.0, 1.0, 0.75], "random": [1.0, 1.0, 1.0, 1.0]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": 0.0, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "1", "prereg_hash": "5d6aa5820cbfdadbf41d35e647ad521451c3e1017b0e0a7454d3e5617aa10c17", "primary_control": "e141_best", "result_hash": "c54266179113c3c1ecda196788b5fb40021ead01bec5f0cb3a99e9ec44a0cb2a", "signature": "f0344871049fb0f5c5b78f0a97bb5290f1e635653f5a6726b8f4cbbe50e9c60adf6fbba48c5125d44c6967dd9726d4ca7ff043d5c424723597ca8242d95a0601", "state": "PARTIAL", "summary": {"e141_best": 1.0, "fixed_policy": 0.75, "full": 1.0, "no_world_model": 0.9375, "random": 1.0}, "ties": 4, "treatment": "full", "wins": 0}, "verified_at": "2026-11-22T04:40:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:37Z", "cpu_seconds": 0.5769420529999998, "created_at": "2026-11-22T04:31:35Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-9b86bd04187a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "3c51c166b3b9936e32b1ee783dfbab7697e6453db627d22bd782dc0cd55294f0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-9b86bd04187a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:31:35Z", "seed": 154564534, "treatment": "full"}, "priority": 0.3249592262438934, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-f57036e0d736", "result_hash": "f3967c3e9f8a3d256b8b46597dab0d0f2dc0a95931151dbfbe512f1c749b8e25", "run_seconds": 0.7343664169311523, "seed": 154564534, "started_at": "2026-11-22T04:31:36Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.26, 1.24], "effect": 0.75, "evaluated_at": "2026-11-22T04:38:21Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_goal-9b86bd04187a", "family": "ow_goal", "hypothesis_supported": false, "instance_scores": {"e141_best": [1.0, 0.0, 0.0, 0.0], "fixed_policy": [0.25, 0.0, 0.0, 0.0], "full": [1.0, 1.0, 1.0, 1.0], "no_world_model": [0.75, 1.0, 1.0, 1.0], "random": [1.0, 1.0, 0.0, 1.0]}, "kind": "replication", "lab": "AGI_LAB", "log10_p_sign": -0.9031, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.125", "prereg_hash": "3c51c166b3b9936e32b1ee783dfbab7697e6453db627d22bd782dc0cd55294f0", "primary_control": "e141_best", "replicated": true, "replication_of": "E141-ow_goal-f57036e0d736", "result_hash": "f3967c3e9f8a3d256b8b46597dab0d0f2dc0a95931151dbfbe512f1c749b8e25", "signature": "cb540be2a59c52bbf65954ced85dfee221bc3aafb9b3f9f1030a8cd84d9ba7d2b4d35a4ebfa1adc3ab61f55469f2d07df840ad768b1dfa32440819ef7644f803", "state": "VERIFIED", "summary": {"e141_best": 0.25, "fixed_policy": 0.0625, "full": 1.0, "no_world_model": 0.9375, "random": 0.75}, "ties": 1, "treatment": "full", "wins": 3}, "verified_at": "2026-11-22T04:38:22Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:42Z", "cpu_seconds": 0.4657859240000004, "created_at": "2026-11-22T04:31:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-414e8d46b831", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "3f00e01e2aa828340eb9176f87b4d6092a6475753ca5cd546ea3d1530e8d134e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-414e8d46b831", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:35Z", "seed": 656867232, "treatment": "full"}, "priority": 0.1800131847017844, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "957d9d3fba4c480a8325b5100da83a9f1724220115791c67fb0a995e768ecb49", "run_seconds": 0.5173625946044922, "seed": 656867232, "started_at": "2026-11-22T04:31:42Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [0.078588, 0.128739], "effect": 0.103663, "evaluated_at": "2026-11-22T04:40:51Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-414e8d46b831", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [0.178294574, 0.60130719, 0.410852713, 0.620155039], "frozen": [0.108527132, 0.091503268, 0.100775194, 0.108527132], "full": [0.317829457, 0.705882353, 0.496124031, 0.705426357], "retrieval": [0.356589147, 0.758169935, 0.674418605, 0.837209302], "rules_nodetect": [0.209302326, 0.346405229, 0.372093023, 0.387596899]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -1.2041, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.0625", "prereg_hash": "3f00e01e2aa828340eb9176f87b4d6092a6475753ca5cd546ea3d1530e8d134e", "primary_control": "facts", "result_hash": "957d9d3fba4c480a8325b5100da83a9f1724220115791c67fb0a995e768ecb49", "signature": "00bcfde13e5eddd914752bd77fb4905d9353abf6063e2e127292b11372abd9bad76f7879d396c0dc832b13ffd9dc7bfb0fc32a0eb9c0819edf83271d79c33c0d", "state": "VERIFIED", "summary": {"facts": 0.452652, "frozen": 0.102333, "full": 0.556316, "retrieval": 0.656597, "rules_nodetect": 0.328849}, "ties": 0, "treatment": "full", "wins": 4}, "verified_at": "2026-11-22T04:40:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:42Z", "cpu_seconds": 1.036450833, "created_at": "2026-11-22T04:31:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-763130c81c7b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "cd591d355f9d3da9521408c36c0b6114479dd2d4229497963be9cf05a5177079", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-763130c81c7b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:35Z", "seed": 1626042164, "treatment": "full"}, "priority": 0.18097445321378236, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "6903a6dffa126a041d258b280b74ea724dae2de4de6f1dbc45ab7c57ecc28939", "run_seconds": 1.0567691326141357, "seed": 1626042164, "started_at": "2026-11-22T04:31:41Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.022554, 0.077717], "effect": 0.027582, "evaluated_at": "2026-11-22T04:40:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_model-763130c81c7b", "family": "ow_model", "hypothesis_supported": false, "instance_scores": {"facts": [1.0, 0.89375, 0.7875, 0.961956522], "frozen": [0.270833333, 0.31875, 0.35625, 0.260869565], "full": [1.0, 0.94375, 0.875, 0.934782609], "retrieval": [1.0, 0.96875, 0.86875, 0.945652174], "rules_nodetect": [1.0, 0.94375, 0.875, 0.934782609]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.301, "losses": 1, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.5", "prereg_hash": "cd591d355f9d3da9521408c36c0b6114479dd2d4229497963be9cf05a5177079", "primary_control": "facts", "result_hash": "6903a6dffa126a041d258b280b74ea724dae2de4de6f1dbc45ab7c57ecc28939", "signature": "41a1d01b4dea2973037a75b55d3fdff51f64c8871c14ac29f93db8181b5de54f429b230a0c19af0d27f42561d9a1196c5f033e28889998611f0de7997cfaaa0f", "state": "VERIFIED", "summary": {"facts": 0.910802, "frozen": 0.301676, "full": 0.938383, "retrieval": 0.945788, "rules_nodetect": 0.938383}, "ties": 1, "treatment": "full", "wins": 2}, "verified_at": "2026-11-22T04:40:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:45Z", "cpu_seconds": 0.5776799309999916, "created_at": "2026-11-22T04:31:35Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-c90a81474a4e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "de67dfbca5d43e9b225bbb37f93ef05230383149e7c67351b3302f5d722eb015", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-c90a81474a4e", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:31:35Z", "seed": 606303958, "treatment": "gated"}, "priority": 0.17921916177636604, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "6cbb9fb6e899330e72c4d8b200393be03697f201337a6b72fb246503cb7a53df", "run_seconds": 0.5802350044250488, "seed": 606303958, "started_at": "2026-11-22T04:31:45Z", "state": "VERIFIED", "verdict": {"better": "higher", "ci95": [-0.024094, 0.145009], "effect": 0.060458, "evaluated_at": "2026-11-22T04:40:52Z", "evaluator_pubkey": "0b00fe53501740b32c0dfd3e3e87bf81aa976e1155fb4cbb4f744062d1b9a739", "experiment_id": "E141-ow_transfer-c90a81474a4e", "family": "ow_transfer", "hypothesis_supported": false, "instance_scores": {"compressed": [0.215686275, 0.405228758, 0.267973856, 0.39869281], "gated": [0.215686275, 0.405228758, 0.267973856, 0.39869281], "naive": [0.065359477, 0.202614379, 0.202614379, 0.189542484], "scratch": [0.215686275, 0.405228758, 0.209150327, 0.215686275], "ungated": [0.183006536, 0.405228758, 0.202614379, 0.248366013]}, "kind": "primary", "lab": "AGI_LAB", "log10_p_sign": -0.6021, "losses": 0, "n": 4, "outcome": "INCONCLUSIVE", "p_sign_display": "0.25", "prereg_hash": "de67dfbca5d43e9b225bbb37f93ef05230383149e7c67351b3302f5d722eb015", "primary_control": "scratch", "result_hash": "6cbb9fb6e899330e72c4d8b200393be03697f201337a6b72fb246503cb7a53df", "signature": "559653a6f165d68beff2f9a233cec09f0c26c560b7fe38d2189a210c634b32de104c7d9325c7aef2c68a4fb5b94b5e8fabb7c9e0288d1b592a4e4b1a72d7dc0c", "state": "VERIFIED", "summary": {"compressed": 0.321895, "gated": 0.321895, "naive": 0.165033, "scratch": 0.261438, "ungated": 0.259804}, "ties": 2, "treatment": "gated", "wins": 2}, "verified_at": "2026-11-22T04:40:52Z"}
{"attempts": 1, "completed_at": "2026-11-22T04:31:52Z", "cpu_seconds": 1.1293810509999958, "created_at": "2026-11-22T04:31:50Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e78a758815f7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 3, "n": 4}, "prereg_hash": "d49f103a8e2fee7a05e382ed2b680a74c55865fc40da79e392b69dea108f54ae", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e78a758815f7", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:31:50Z", "seed": 1910138053, "treatment": "full"}, "priority": 0.3382915834049599, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "275465ce41320319d5213c9aa71de04c49054bcedc126f24fd8f15490cc38c04", "run_seconds": 1.4310741424560547, "seed": 1910138053, "started_at": "2026-11-22T04:31:50Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:31:58Z", "cpu_seconds": 1.0723040320000052, "created_at": "2026-11-22T04:31:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4a0accc62f21", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "3ac84ff05b672e0522edbc41a6e6e71c7ef8c86c4ef1c3b8a3704786754457f5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4a0accc62f21", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:50Z", "seed": 319753141, "treatment": "full"}, "priority": 0.181120212117045, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5bfa65974de449c6d30aa3d1df8d5638fb4cd58d9cda7bb371fd3885d019b946", "run_seconds": 1.2370498180389404, "seed": 319753141, "started_at": "2026-11-22T04:31:57Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:31:56Z", "cpu_seconds": 0.9041385900000005, "created_at": "2026-11-22T04:31:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-af412773ad0b", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "afa7a8874f7f9a20d0b972f96359f7d59b123f6572347f29fd20e01a09d39c14", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-af412773ad0b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:50Z", "seed": 1544103920, "treatment": "full"}, "priority": 0.18888938577650666, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-fe66afdbaa3b", "result_hash": "a5616c9e575c64aae6fe2095dee3f91d432d4f0fdfb673942bf27921f2a332d6", "run_seconds": 1.0225803852081299, "seed": 1544103920, "started_at": "2026-11-22T04:31:55Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:02Z", "cpu_seconds": 0.7663263800000006, "created_at": "2026-11-22T04:31:50Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d214b87f6373", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 2, "n": 4}, "prereg_hash": "9537df42adb9431103cd03342074ee993bc912dc5e4a41f3ee4146a2f4278172", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d214b87f6373", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:31:50Z", "seed": 1946838880, "treatment": "full"}, "priority": 0.1734570885955003, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "235d697dad09654e1056d08caab1098e226f15ee1f1519a65ad491e907c3b456", "run_seconds": 0.8432486057281494, "seed": 1946838880, "started_at": "2026-11-22T04:32:01Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:18Z", "cpu_seconds": 0.6432092000000011, "created_at": "2026-11-22T04:31:50Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-401d203f8b76", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "4590ba4b4a1af85e77c22a27437aa22b58882c79d1e0578e641270031943e79e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-401d203f8b76", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:31:50Z", "seed": 413158881, "treatment": "gated"}, "priority": 0.17058461244788414, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f7307b0993e1fa9a106baf228ce1d8469b7d8666e85dc33ec3f223f012349a48", "run_seconds": 0.7565109729766846, "seed": 413158881, "started_at": "2026-11-22T04:32:17Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:00Z", "cpu_seconds": 0.7292784279999971, "created_at": "2026-11-22T04:31:50Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-bd87838972f3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "2fd8e7e1f8dba28fbec947066deb2d450d51c5b5192bde8a2ba57ed5c9a308fe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-bd87838972f3", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:31:50Z", "seed": 257132915, "treatment": "gated"}, "priority": 0.17774995797252433, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "7b9c9a431af2e4b801933d75866dc143b2a1e7b5ed74d58c2df7c5727480b4be", "run_seconds": 0.8116538524627686, "seed": 257132915, "started_at": "2026-11-22T04:31:59Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:37Z", "cpu_seconds": 1.0642491430000014, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-020233773dd6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "f55cf9aad6421e10cffb2892399dccd92d8f2aa322ce8274355a326aac466655", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-020233773dd6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:07Z", "seed": 577630210, "treatment": "frontier"}, "priority": 0.16504294640087502, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "7a19251f260f4a59b77837ae6b00bdd76fd1a4d52b84d2c1065cae7608397a8a", "run_seconds": 1.1701819896697998, "seed": 577630210, "started_at": "2026-11-22T04:32:36Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:24Z", "cpu_seconds": 1.4183439500000006, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-87db7cee53dc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "bc91c4d1106ab923b9c3a3f96756b659726746193699c22dfdb4b1382c47c68e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-87db7cee53dc", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:07Z", "seed": 1990310616, "treatment": "frontier"}, "priority": 0.16847403283224754, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "fbe85a4e9eab132133bed4975d7ba5e4cbb427a8504064553eb7079d16dd9f38", "run_seconds": 1.435429573059082, "seed": 1990310616, "started_at": "2026-11-22T04:32:22Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:12Z", "cpu_seconds": 0.5400442890000008, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0aa1f75cf226", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "1528c82a7b16ed7de2c63fad48d8decbb8b04a8be230422360a5c8bf98a20e42", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0aa1f75cf226", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:07Z", "seed": 1052695678, "treatment": "full"}, "priority": 0.1770793360053573, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ea44fb73389778d220a0c9b3de0e7ad7ae3a55e289814d7a6db81967caf56ad8", "run_seconds": 0.5479137897491455, "seed": 1052695678, "started_at": "2026-11-22T04:32:12Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:15Z", "cpu_seconds": 0.46740103499999464, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-8ec72d445402", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "d06d5801ba21c34ea708ff07fbc71f9ba945eaa619654772413ea1b7d1c8f1a7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-8ec72d445402", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:07Z", "seed": 1346113019, "treatment": "full"}, "priority": 0.1762757413548508, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c323bbbcebd2bd3ad1208b664a8ccf1b3437835fbfd7c7c1c38af828a93fd0af", "run_seconds": 0.4712073802947998, "seed": 1346113019, "started_at": "2026-11-22T04:32:14Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:37Z", "cpu_seconds": 2.1644941339999946, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-2b55da67618e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 2, "n": 4}, "prereg_hash": "9e6b6b382a0d6de01546dc82d75aabdb733cc367289096485e57698cd2bbb2c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2b55da67618e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:32:07Z", "seed": 895893111, "treatment": "full"}, "priority": 0.16832325661442946, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9334a19e56651ec2ad676743f550deeec2a0d866dd5f18c307b273bc557b2c2a", "run_seconds": 2.251105546951294, "seed": 895893111, "started_at": "2026-11-22T04:32:35Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:17Z", "cpu_seconds": 0.48389038199999845, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7a054a64a17e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "c49716cd86a636f5ebff84a41e3f6c262929ed0305dafaf891c231ce6a45cfb2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7a054a64a17e", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:32:07Z", "seed": 1028067953, "treatment": "full"}, "priority": 0.1761188855390687, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d82c0b8daa64ae3ffd64af023ccbdf5f905f534e71df387cbefb5de0e874f6f5", "run_seconds": 0.5238523483276367, "seed": 1028067953, "started_at": "2026-11-22T04:32:17Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:55Z", "cpu_seconds": 0.9298635430000104, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-59e4c27761aa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "prereg_hash": "a888c67027cef13a30bb8c2df1b8609e2efa387ba4985b7f7610e44735e725c0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-59e4c27761aa", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|grid_nav|grid_nav|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:07Z", "seed": 1168277685, "treatment": "gated"}, "priority": 0.1650379767625239, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "052ded550a9d8c0cdc36b85bd4cf62b7acc7751e7959c55bffc80027dd2cfb30", "run_seconds": 0.96038818359375, "seed": 1168277685, "started_at": "2026-11-22T04:32:54Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:21Z", "cpu_seconds": 1.009539195000002, "created_at": "2026-11-22T04:32:07Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-aa01e998d3d3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|crafting|crafting|1|1"}, "prereg_hash": "e2cfd2bdbecce9fb68053a7c724de03f3069a839398178e0bb971c7248f9dd9f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-aa01e998d3d3", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|crafting|crafting|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:07Z", "seed": 2146422801, "treatment": "gated"}, "priority": 0.1702148014879551, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "bdb32b0ce0d17c28b82a2daf53b9c8d2f09ad4ac93f4a8939777cfeeda271ca3", "run_seconds": 1.0491712093353271, "seed": 2146422801, "started_at": "2026-11-22T04:32:20Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:31Z", "cpu_seconds": 1.1445263709999978, "created_at": "2026-11-22T04:32:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b22aca75a9ab", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "11b9d67ba566c6a7d60bff72176f40c2b7a0ddee9baa3a7c31cfdeaa10123db3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b22aca75a9ab", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:22Z", "seed": 560123820, "treatment": "full"}, "priority": 0.17031172519045684, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2508e3024734217bc9eab49e671eb6c8e7fa869734079a3e06508a0fa0d1b4e2", "run_seconds": 1.1648945808410645, "seed": 560123820, "started_at": "2026-11-22T04:32:30Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:29Z", "cpu_seconds": 0.6152058040000021, "created_at": "2026-11-22T04:32:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bec0ac8d6b5b", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "3580ce7f72263b1f4768a28883021d3c83d38754f5d5ccf8afe74640c1880716", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bec0ac8d6b5b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:22Z", "seed": 1863918655, "treatment": "full"}, "priority": 0.17036939491017264, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-2c276c20e04d", "result_hash": "0b99eb9c755361086c901981945bfefe7b332df5f122ecb5004a1f3f627ce905", "run_seconds": 0.6277492046356201, "seed": 1863918655, "started_at": "2026-11-22T04:32:29Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:25Z", "cpu_seconds": 1.9237892789999975, "created_at": "2026-11-22T04:32:22Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-dff09045ab71", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "f84ef7e14a26ee856129f57a69540ca6c430c6bcfeb7a4cfc8a28a636658f832", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-dff09045ab71", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:22Z", "seed": 933662785, "treatment": "full"}, "priority": 0.17946478516271872, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "566866af8e6c2ab2455a884b05555930814666a72563040cbeced9d0d0e6c9b8", "run_seconds": 2.0749645233154297, "seed": 933662785, "started_at": "2026-11-22T04:32:23Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:30Z", "cpu_seconds": 4.0034885660000015, "created_at": "2026-11-22T04:32:22Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8458accc23ed", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "dd6887c788a5748be2d28c5f99f5cdb50a69c3768b42660f2ea5ed2e86397d36", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8458accc23ed", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:22Z", "seed": 297686450, "treatment": "gated"}, "priority": 0.17204145176418698, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-58969546a769", "result_hash": "c0bf56bf1adfc2231803dcb4e73cdf12ef37011059152fe9cd5efb9201185cd2", "run_seconds": 4.1330578327178955, "seed": 297686450, "started_at": "2026-11-22T04:32:26Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:35Z", "cpu_seconds": 0.5461500010000009, "created_at": "2026-11-22T04:32:22Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-933a1326dd55", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "9c368baa8650505f23d610b178a8d918b437f2e23a8245019f9e871e1eed3ab5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-933a1326dd55", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:22Z", "seed": 1287216883, "treatment": "gated"}, "priority": 0.1694382562987251, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "cfb1b193b45c8745405c88bcb0edc4cefa8be786c660f707f9c091799a6be15c", "run_seconds": 0.5486257076263428, "seed": 1287216883, "started_at": "2026-11-22T04:32:34Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:41Z", "cpu_seconds": 0.802953475999999, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a724e6be4cc8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "acbe6de64665d2348765d8b63dd806f0f64f9425828226f0e1e6ea79835af12f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a724e6be4cc8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:39Z", "seed": 915682340, "treatment": "frontier"}, "priority": 0.18727866957491815, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9c03ab392d2816b77ffd97142bd3bc2cd361a937e148226effd2a88c3e59835f", "run_seconds": 0.8336856365203857, "seed": 915682340, "started_at": "2026-11-22T04:32:40Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:59Z", "cpu_seconds": 4.488543291000003, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e9d974554801", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 4, "n": 4}, "prereg_hash": "a3c7c01da7e8715b51c1e3ff01a9b5464650ab384244019c69b716f96b8bbd1b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e9d974554801", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:39Z", "seed": 701393444, "treatment": "frontier"}, "priority": 0.16575952294260524, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "46d69f645bf2bbea96cb60ed00153c136accd41de4d3784217d366674fec7572", "run_seconds": 4.9797704219818115, "seed": 701393444, "started_at": "2026-11-22T04:32:54Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:43Z", "cpu_seconds": 0.6165564649999986, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f71e9a952a38", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "6aad068b3c652eff4802f584bae76154dc18f13058db56c0643a7d36bb064ff0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f71e9a952a38", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:39Z", "seed": 2091506448, "treatment": "frontier"}, "priority": 0.17438542294314546, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "43fa3a0a5c54fb6c1a92b896738b394953296c69e3559a95b8b577444306af37", "run_seconds": 0.6216964721679688, "seed": 2091506448, "started_at": "2026-11-22T04:32:42Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:49Z", "cpu_seconds": 1.0530086489999917, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a973332ecfc9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 4, "n": 4}, "prereg_hash": "72a94220a4d745bfb3bccdaaa25dc727f33da3c85a0bd04f5d103c508f624e2a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a973332ecfc9", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:39Z", "seed": 1296260478, "treatment": "full"}, "priority": 0.16848697840963633, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "731a5db9425aa8cef1c3dedd94ced581bd070a591a1043f0ec53e68f54d5bb67", "run_seconds": 1.160412311553955, "seed": 1296260478, "started_at": "2026-11-22T04:32:48Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:43Z", "cpu_seconds": 0.7409896549999928, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c2a4d2bdd865", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "871554215ae2056b2a7fbf3da1fdcb3557ef628f1e875b3e767af132640a3cf8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c2a4d2bdd865", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:32:39Z", "seed": 186485626, "treatment": "full"}, "priority": 0.17358582741813885, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "76502c1d46e78301deb6b36f4004562a664362b17422783e909217ec560a9ef8", "run_seconds": 0.7541217803955078, "seed": 186485626, "started_at": "2026-11-22T04:32:42Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:49Z", "cpu_seconds": 0.8122135830000019, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-dde1ce66aa1f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "40aae9d4d59fe0c696c8250c9384088386c2dd5a03a14903330f52e33d0a124b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-dde1ce66aa1f", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:32:39Z", "seed": 2130845662, "treatment": "full"}, "priority": 0.17273987899535237, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "cd30c7d0306770860e571c38ec4b493dd0215887f9888908886b4afef6419187", "run_seconds": 0.8932499885559082, "seed": 2130845662, "started_at": "2026-11-22T04:32:48Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:53Z", "cpu_seconds": 0.9745752059999973, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ecab82aea5da", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "7c383316184b9cef5ef9fcca67012a4b2712060117fcd23711c04dd2c33806a8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ecab82aea5da", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:32:39Z", "seed": 699942349, "treatment": "full"}, "priority": 0.16654641575863274, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b041a61e02f65edf01beb0a6ab6fdbb5d5c6385f793b3819f4c8afa3c5120398", "run_seconds": 1.0090737342834473, "seed": 699942349, "started_at": "2026-11-22T04:32:52Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:47Z", "cpu_seconds": 1.0865282620000016, "created_at": "2026-11-22T04:32:39Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-6f183c4900cc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "c1439731369cf9099c4e20142fb8c79ffb97a75608881efe14e956c4c3d1b71a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-6f183c4900cc", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:39Z", "seed": 730074633, "treatment": "gated"}, "priority": 0.1732403324434712, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "447f5e6b75fc15779b0f5c83f4ef6e28eba3480a8764fccf48c6b931d8e70aa6", "run_seconds": 1.1069798469543457, "seed": 730074633, "started_at": "2026-11-22T04:32:46Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:30Z", "cpu_seconds": 1.3463624959999976, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0f8bfb96b3e6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "fcc2a003d5cd67cc86928c210bffa375ce22f0e266c805d5ba3a099146c61d07", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0f8bfb96b3e6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:56Z", "seed": 1774228878, "treatment": "frontier"}, "priority": 0.16691187263907267, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "96669227794c2a487e7cfbfa7a72015fdefb5b12db566943d31c641ee5ac7222", "run_seconds": 1.4201607704162598, "seed": 1774228878, "started_at": "2026-11-22T04:33:28Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:11Z", "cpu_seconds": 3.8637916389999987, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-301ce8647fb5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "ecb1ad646bf75fd4b1bf06148eea295ea80d5db4001a09de2399660aae0449f8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-301ce8647fb5", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:56Z", "seed": 992325237, "treatment": "frontier"}, "priority": 0.16776606494958576, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1f52df48c04c1a12f8f821ff2b3bbc700abfa0f13d530ca14d41de225b360005", "run_seconds": 4.319502592086792, "seed": 992325237, "started_at": "2026-11-22T04:33:06Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:07Z", "cpu_seconds": 2.3445599109999975, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e4721f05b96b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 8, "n": 4}, "prereg_hash": "6e8dfebc2a1b23756671bd253b0fcadb681f74670d94c03c9a6c882984e51328", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e4721f05b96b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:32:56Z", "seed": 833444259, "treatment": "frontier"}, "priority": 0.17140918530184976, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "716fd9e0604549fc53f9d0016fc4a6bf8ec460919553140bf56867c31c35a785", "run_seconds": 3.1004297733306885, "seed": 833444259, "started_at": "2026-11-22T04:33:04Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:29Z", "cpu_seconds": 0.688090735000003, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-15ad8a36470b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "6381385b713cf7d4031aea53fcd15493aad6f874306bd9272e80e51b7204fbd4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-15ad8a36470b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:56Z", "seed": 1199017335, "treatment": "full"}, "priority": 0.16669832977933313, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "a0c504b2a29f9a5f7d9c04ebdda8cca29a222566d1ed449490ffecb117dd4a51", "run_seconds": 0.7109766006469727, "seed": 1199017335, "started_at": "2026-11-22T04:33:28Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:32:59Z", "cpu_seconds": 1.1089415440000003, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-814964ce1887", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 1, "n": 4}, "prereg_hash": "8d46cad9ec5f2f6e361a417ed26f1bc14363f1295984e611281d2864a52f2b09", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-814964ce1887", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:32:56Z", "seed": 167060415, "treatment": "full"}, "priority": 0.18353021120175825, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "62738e79accbf8f986a2980f129ecef82b6ea93eaa0f72d24e929c1d284a2057", "run_seconds": 1.294346570968628, "seed": 167060415, "started_at": "2026-11-22T04:32:58Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:14Z", "cpu_seconds": 1.0361819029999992, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-564a94cc54f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 2, "n": 4}, "prereg_hash": "135d4e0253cd16df3b80e2fc5500ec08ba1b61181cfa60f5a99f7351d0a2e9ca", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-564a94cc54f0", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:32:56Z", "seed": 652619075, "treatment": "full"}, "priority": 0.16761670916122345, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d4f6c287403d4f67a5417fafbfa63271197541fde674fda45cc5b5970479c8ba", "run_seconds": 1.0405795574188232, "seed": 652619075, "started_at": "2026-11-22T04:33:12Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:01Z", "cpu_seconds": 0.846551183999992, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5589835e43a6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "prereg_hash": "36f1dfd5456cfae3b5ce09cc4c5cafc1a378efc0addbfa1b31e37a44d0770ba5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5589835e43a6", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|stack_machine|stack_machine|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:56Z", "seed": 1273432076, "treatment": "gated"}, "priority": 0.17462469295631103, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9f7406c1dc5e11c924bcc8f78b5ffdc6c74bc5ac2b21cee1e9bd7b2e4da7f643", "run_seconds": 0.8701889514923096, "seed": 1273432076, "started_at": "2026-11-22T04:33:00Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:10Z", "cpu_seconds": 4.648000223, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-58d342050900", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "prereg_hash": "a1fe698e96a39e03effddbe14ae6f64d216a7d50be3881350143ffc428a849c6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-58d342050900", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|coop_agents|coop_agents|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:56Z", "seed": 1075904762, "treatment": "gated"}, "priority": 0.16817597430258804, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "e6224894aae4e3c49c3880e5bdd5037fcca1acc8d9b1fad778f9696b5c652546", "run_seconds": 5.8177642822265625, "seed": 1075904762, "started_at": "2026-11-22T04:33:04Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:42Z", "cpu_seconds": 1.1021027350000026, "created_at": "2026-11-22T04:32:56Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-da1a875e6ce3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "prereg_hash": "c5deeb1e281c733d837b6afab04e9c54fae097601309fae6edb73888c0461650", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-da1a875e6ce3", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:32:56Z", "seed": 573714182, "treatment": "gated"}, "priority": 0.16498019478387146, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "24335cc37c552344aeabbdc2996b1e497a8ab820e311ac42ce35f69ccb32371b", "run_seconds": 1.266333818435669, "seed": 573714182, "started_at": "2026-11-22T04:33:40Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:23Z", "cpu_seconds": 4.0173048129999955, "created_at": "2026-11-22T04:33:13Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7f6ed26ff2cf", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "4b255f24f17171f3b1ae5270dffdda4d7b71c5a766627092631cff81286bb829", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7f6ed26ff2cf", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:33:13Z", "seed": 480408931, "treatment": "frontier"}, "priority": 0.17293786608708528, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "9e9b254caaa5131c8d708569f573e7414aadc0801f7dc9d6228298a44025894f", "run_seconds": 4.383673429489136, "seed": 480408931, "started_at": "2026-11-22T04:33:19Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:23Z", "cpu_seconds": 1.5352777030000198, "created_at": "2026-11-22T04:33:13Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-a1ce85a9b0f1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "e795567902c5f4965f170aea9ae4d4c85979eb63a41da2bfe0394091632aacfa", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-a1ce85a9b0f1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:33:13Z", "seed": 969496567, "treatment": "frontier"}, "priority": 0.1686130605060121, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ad722249a64000c27e8a56920bc241d4ddb8a8b4521d04112e46f7fe0afe05ff", "run_seconds": 1.7127211093902588, "seed": 969496567, "started_at": "2026-11-22T04:33:22Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:22Z", "cpu_seconds": 0.7154256649999979, "created_at": "2026-11-22T04:33:13Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0ce9182b9e47", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "de680484eb6e2e5ea405b47f398d444d5f50f173ceb12eea2981b364267ac2ce", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0ce9182b9e47", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:13Z", "seed": 487505183, "treatment": "full"}, "priority": 0.17233861899063285, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "63db7c2e74cfd10da48c01f5a56580395deec208655209d32559650a610b0799", "run_seconds": 0.9863176345825195, "seed": 487505183, "started_at": "2026-11-22T04:33:21Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:17Z", "cpu_seconds": 0.8575457239999764, "created_at": "2026-11-22T04:33:13Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0d44b6f16ea8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "c35d891d4aabd99ad7f1a10e94e8e00812cfcaaaf791bc14be9ace3588120bbf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0d44b6f16ea8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:13Z", "seed": 54886992, "treatment": "full"}, "priority": 0.17350220396096067, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0b5b9bcbb12b7646317c9be18e7bbb854919b4ce1c2eaf1a877aaa5b511b151d", "run_seconds": 0.8593432903289795, "seed": 54886992, "started_at": "2026-11-22T04:33:16Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:29Z", "cpu_seconds": 1.285544508000001, "created_at": "2026-11-22T04:33:13Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8f789a683261", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 4, "n": 4}, "prereg_hash": "ecf57266aa19f9ce6690de2725031b33c980f110fff9d732d8c5522502339165", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8f789a683261", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:13Z", "seed": 1163869111, "treatment": "full"}, "priority": 0.1679458696221751, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ff595e44045d299bbc84a48bd0774d3f15f4d91c9dfce8421901ea61c57466e0", "run_seconds": 1.333127737045288, "seed": 1163869111, "started_at": "2026-11-22T04:33:28Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:16Z", "cpu_seconds": 0.8415298340000028, "created_at": "2026-11-22T04:33:13Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-087b32790f4d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|crafting|crafting|1|1"}, "prereg_hash": "0f700a358acbc9163b117eff741256c128ff6f94d420390eaa131d5339aef56a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-087b32790f4d", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|crafting|crafting|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:33:13Z", "seed": 1508322880, "treatment": "gated"}, "priority": 0.1758255642358813, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "66f868f955d0a57de0827f92a3fcf271f8dbad47e651b71347d71c42b10e6ca2", "run_seconds": 0.8434410095214844, "seed": 1508322880, "started_at": "2026-11-22T04:33:15Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:36Z", "cpu_seconds": 1.0777758580000025, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4c2034df9813", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "cf8b918bf345ac0875ab981f397611adf1f49c30cea643b6bd4a6a9511d8a117", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4c2034df9813", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:33:30Z", "seed": 1078751075, "treatment": "frontier"}, "priority": 0.1661955909935468, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "93e1572dd2111db6dfb4f45794493fcbf4adb419608a78e2ffbdf8352512b8df", "run_seconds": 1.448697805404663, "seed": 1078751075, "started_at": "2026-11-22T04:33:35Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:42Z", "cpu_seconds": 1.6091865669999947, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5f7dfd59c83a", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "b2c4db1cec1372d062f3991ea301555b4532db73c2ac412002decdcb264a1e92", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5f7dfd59c83a", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:33:30Z", "seed": 219786679, "treatment": "frontier"}, "priority": 0.16507960800615565, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-c846f04ef3ee", "result_hash": "bcdeac32d6113cf284293cfa75d24db10f0f65a56c9c5828263b2f6fbe73b808", "run_seconds": 1.752786636352539, "seed": 219786679, "started_at": "2026-11-22T04:33:40Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:35Z", "cpu_seconds": 0.9924991780000028, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c7f3cdc5fefc", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "ef4ff2ea9653bc1ec68c8d7de00637db92d5a43ca559596cceca14e065746593", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c7f3cdc5fefc", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:33:30Z", "seed": 1254554118, "treatment": "frontier"}, "priority": 0.17135839826095292, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-5b23263d7a11", "result_hash": "d5bed12e3fa00f5e33e3cf5c8179b7a96dc0961dd5642c6de32cecaab1ec564f", "run_seconds": 1.1847341060638428, "seed": 1254554118, "started_at": "2026-11-22T04:33:34Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:22Z", "cpu_seconds": 0.6373352770000054, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e281ffc9ecca", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "b14e77445e0fa6e3fb28d2397984158a40bcded9d63e541e4432f5ec04b0c5fe", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e281ffc9ecca", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:33:30Z", "seed": 410522893, "treatment": "frontier"}, "priority": 0.15383145581923338, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-4c30b0094249", "result_hash": "5394f4ea7f81a1f7a42f8c62db538dd3025ec9707b05594b06aba7e0f0f32c6a", "run_seconds": 0.7105269432067871, "seed": 410522893, "started_at": "2026-11-22T04:35:21Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:50Z", "cpu_seconds": 2.6142015850000178, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0fa422db79ca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "55c4e4848c1694c3591b7774cde6f002facc6e1c5e52a28576f12b245d774b62", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0fa422db79ca", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:33:30Z", "seed": 1370508202, "treatment": "full"}, "priority": 0.15414446848271407, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1fab14dca033afe6a97a565a99c6f63a7b6cb9ce9403d0b2fd30093b540dec44", "run_seconds": 2.6818244457244873, "seed": 1370508202, "started_at": "2026-11-22T04:33:47Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:42Z", "cpu_seconds": 1.209877198000001, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a1b61d73e8ed", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "a5c565beabcb7b3de63cd791621adb734f3cb0dcaf1aebf2d43edd63256fb2a9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a1b61d73e8ed", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:33:30Z", "seed": 2030890436, "treatment": "full"}, "priority": 0.1641226842334609, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "70d8c845b7e0da6a4a1dc6e4f5de4db4d365a83e0beb6ee95f1551cc0fd3d1ca", "run_seconds": 1.2923119068145752, "seed": 2030890436, "started_at": "2026-11-22T04:33:41Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:48Z", "cpu_seconds": 1.3454296139999968, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c8170b601485", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "47792cfac45c4020e400937bbbc3359313a5033a90476d8e231a16711ae8a89a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c8170b601485", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:30Z", "seed": 1288078974, "treatment": "full"}, "priority": 0.15822633450118853, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "58812a73034dedf58c24331860aa2973b48566d0cb36e55126916a764b096d10", "run_seconds": 1.3550052642822266, "seed": 1288078974, "started_at": "2026-11-22T04:33:47Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:35Z", "cpu_seconds": 1.0197694439999978, "created_at": "2026-11-22T04:33:30Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-e8e0db33e029", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "9adae81033328b005261ea30818b994ba4e3f2e7ca57b0703bfae2c8548b2baf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-e8e0db33e029", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:33:30Z", "seed": 1731522711, "treatment": "gated"}, "priority": 0.16847456506060132, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "25e787d1502741d189235bcdbadfe6e0928553f7ee4a6e30aa868ccf8081eb57", "run_seconds": 1.1564602851867676, "seed": 1731522711, "started_at": "2026-11-22T04:33:34Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:03Z", "cpu_seconds": 1.0751435159999971, "created_at": "2026-11-22T04:33:47Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-2fb2c79b87cd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "5981705db7864f2f02cceac0c5120e1d2f33021a2c02d0081d13f50aac138c8a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2fb2c79b87cd", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:33:47Z", "seed": 1658835171, "treatment": "frontier"}, "priority": 0.15775570713753553, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "2fddbaf96e2628a171b52be3b9831854b9a9345abe00dfac171e79cd51ec1aed", "run_seconds": 1.119863510131836, "seed": 1658835171, "started_at": "2026-11-22T04:34:02Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:01Z", "cpu_seconds": 1.3356218109999958, "created_at": "2026-11-22T04:33:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c9d9ba65d868", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "2ef10a044a430bd15819d382363368c055e5daa52bc86712f930fd3b65763837", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c9d9ba65d868", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:33:47Z", "seed": 2110884248, "treatment": "full"}, "priority": 0.16129973684791166, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "976a1cdbaa1fd17c8afc747d845db663cb9e7799a4adaf95b05d58e1c4e10b64", "run_seconds": 1.5924808979034424, "seed": 2110884248, "started_at": "2026-11-22T04:33:59Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:48Z", "cpu_seconds": 1.0695054279999994, "created_at": "2026-11-22T04:33:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-693e98b85005", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "58043b656921daa0f0e3eb3a02768808c4b1550bc1aac2d21a3a08cf52b47edd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-693e98b85005", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:47Z", "seed": 1905701022, "treatment": "full"}, "priority": 0.173616686690662, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "b2626f519431ef90910326746cd14f65e8cb59127447b3fbc3485da969eae34b", "run_seconds": 1.1157634258270264, "seed": 1905701022, "started_at": "2026-11-22T04:33:47Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:55Z", "cpu_seconds": 1.1689638950000045, "created_at": "2026-11-22T04:33:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-77987745ccf2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "b20fbd156bc1bee3dbfd32e92cb8d9833f965ae776e1c1bd37c6725a4ce62092", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-77987745ccf2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:47Z", "seed": 170366974, "treatment": "full"}, "priority": 0.1667514872401567, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "59ddf2d496f6bf27880ed287da87db4d1b26571ae97e6c1de24b5bd9d11c68ab", "run_seconds": 1.2345526218414307, "seed": 170366974, "started_at": "2026-11-22T04:33:53Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:01Z", "cpu_seconds": 1.1956621230000053, "created_at": "2026-11-22T04:33:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-8a9db10496b4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "d472dc9a8a05f17baafba552bef724527481bb949705c0b425763d87a3dba0d4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-8a9db10496b4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:47Z", "seed": 1675882644, "treatment": "full"}, "priority": 0.15782863260009247, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "9f624f98858a4f9a0ff347f379e50c6e445b2bfae8a4f802215cf9a973bfa750", "run_seconds": 1.2883591651916504, "seed": 1675882644, "started_at": "2026-11-22T04:34:00Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:57Z", "cpu_seconds": 1.81314755599999, "created_at": "2026-11-22T04:33:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a0fee1889d56", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "64f183ad7cd7918a677fe05625d4a76d95a96e5c63b9d4c38b16b691e8711ccf", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a0fee1889d56", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:33:47Z", "seed": 837153006, "treatment": "full"}, "priority": 0.1639540915588564, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "acd8bfd67ed2c455b830fe5e5e7b3bad65994c76edb66bbf2d06d579290673a6", "run_seconds": 1.8559186458587646, "seed": 837153006, "started_at": "2026-11-22T04:33:55Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:33:54Z", "cpu_seconds": 0.8385478200000023, "created_at": "2026-11-22T04:33:47Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-aac55aad00da", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "f9aca9ec59430393a2af65b38d724c3f3bdafb0e3def81722346b7c3bfdc477d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-aac55aad00da", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:33:47Z", "seed": 1862025762, "treatment": "gated"}, "priority": 0.17108660876800683, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "83219110b39f3f754444876dc90c5f071f3060853d928b42925893913cdce0d8", "run_seconds": 0.8722019195556641, "seed": 1862025762, "started_at": "2026-11-22T04:33:53Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:07Z", "cpu_seconds": 0.9631375009999985, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-116c7278de7d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 6, "n": 4}, "prereg_hash": "024a4fa39a423ef60e6d7b48d17865509574c7149bd313a54413d2cfbf03c73a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-116c7278de7d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:04Z", "seed": 1642313909, "treatment": "frontier"}, "priority": 0.16991841631908217, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "6970706f1964f2a1873a6b4f0bf872e87a06dd257c49c85ce9f482d66ae201b6", "run_seconds": 1.0736873149871826, "seed": 1642313909, "started_at": "2026-11-22T04:34:06Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:34Z", "cpu_seconds": 1.841619245000004, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-5c421cb4cb4d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 6, "n": 4}, "prereg_hash": "a18e26237a0c02c4526295b6271c71df8125f58743254a6db165059c35408701", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-5c421cb4cb4d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:04Z", "seed": 1495782183, "treatment": "frontier"}, "priority": 0.157327181457157, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "05bc080d63de68658ea9ed07d1c61f44911ae62f07c36e70cb4787b589fcf868", "run_seconds": 2.2654664516448975, "seed": 1495782183, "started_at": "2026-11-22T04:34:32Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:22Z", "cpu_seconds": 8.241079733999982, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-633947e346ec", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "8f44c90932d926c1f0dc6501045029acc57cb2b70cc68880422bf84a004a27c1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-633947e346ec", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:04Z", "seed": 196625058, "treatment": "frontier"}, "priority": 0.15928178000394974, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "03e56f810f3ce191628772f2357574b74fc453ae78ecfce05c23ff9768abeff0", "run_seconds": 8.654233694076538, "seed": 196625058, "started_at": "2026-11-22T04:34:13Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:07Z", "cpu_seconds": 0.9352260149999978, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4a43c1b9ab1d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 5, "n": 4}, "prereg_hash": "b231fd7938b559697a457a2d959fb48ec61cb1316ab9740a51a93d0f4e2a6a7e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4a43c1b9ab1d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:34:04Z", "seed": 783912369, "treatment": "full"}, "priority": 0.16497468935708665, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "abd3149d82e70693434e11303f9325de9d3a38bb163cd07862c20c032e92e5b5", "run_seconds": 1.4033019542694092, "seed": 783912369, "started_at": "2026-11-22T04:34:06Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:19Z", "cpu_seconds": 1.2420768229999979, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bb540b8ea52e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "53e06d40b0a34dfee0572339554456d86024fb9f15469289d50b8b3af27c99b3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bb540b8ea52e", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:34:04Z", "seed": 1804827104, "treatment": "full"}, "priority": 0.15917210381900676, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "5f0a1082e525acd5337c490309d144e7a1596597a9999dbdf62d212e4e3cb8bd", "run_seconds": 1.428802490234375, "seed": 1804827104, "started_at": "2026-11-22T04:34:18Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:08Z", "cpu_seconds": 0.4253164320000167, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bd0263e99caa", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "c4d29e0d653c0f17f8ecff89a5e5f4fd43d101fd71df213481cbc094430ebe2f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bd0263e99caa", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:34:04Z", "seed": 1803549817, "treatment": "full"}, "priority": 0.16459130237155362, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-c5336f3c6b1d", "result_hash": "cef67dce7c81880457b3711ca29b85c873ad0d442eb96a42fa24e59941dc4cb1", "run_seconds": 0.47870588302612305, "seed": 1803549817, "started_at": "2026-11-22T04:34:08Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:14Z", "cpu_seconds": 1.1161360169999952, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-09bd9152f61d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "50ebd6766d27f08139efcff9e3ca733f5129392383d879e6d553bc0ec96d394e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-09bd9152f61d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:34:04Z", "seed": 764993081, "treatment": "full"}, "priority": 0.1596493582289488, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "726cd3c5bdf1cd4e1834044f5e87e50f7701de0a5280ef49139ae6909a11dc17", "run_seconds": 1.2468323707580566, "seed": 764993081, "started_at": "2026-11-22T04:34:12Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:13Z", "cpu_seconds": 1.188032579999998, "created_at": "2026-11-22T04:34:04Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-7b4e783a8ef7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "0be8afd6f6103191c8a78ff78ee4d2d573588aa5e55a1d71d9ff1884b7de0503", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7b4e783a8ef7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:34:04Z", "seed": 244504335, "treatment": "full"}, "priority": 0.1643699706219288, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "d93b2aa286fa834572eda2a231eaffb96399d5bb41671f1b4ed2d858bd9d1705", "run_seconds": 1.3357772827148438, "seed": 244504335, "started_at": "2026-11-22T04:34:12Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:09Z", "cpu_seconds": 2.2796942909999984, "created_at": "2026-11-22T04:34:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-363fd31bcff7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 8, "n": 4}, "prereg_hash": "f2a261cb7115d66b2958a958083f1e16c9f5ba75efb026943c909834d3f8b965", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-363fd31bcff7", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:19Z", "seed": 849609962, "treatment": "frontier"}, "priority": 0.15434350234465666, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "22171e9f8b00186a48aff417a7f859318e747b57a86d39bf4b73e36e1c286d8a", "run_seconds": 2.4656755924224854, "seed": 849609962, "started_at": "2026-11-22T04:35:07Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:22Z", "cpu_seconds": 2.1624539000000027, "created_at": "2026-11-22T04:34:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-74eb58da0ef0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "deefe50ec91a967b46553535034f8754281e98f3488cd227b1f87ffd48bd68d4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-74eb58da0ef0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:19Z", "seed": 60929072, "treatment": "frontier"}, "priority": 0.15405557811463327, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "ab92cd60e43d492b43966d6225d1abbd6297ef9b9d9a2f5f0ebb5314a6ab5bf5", "run_seconds": 2.5383522510528564, "seed": 60929072, "started_at": "2026-11-22T04:35:19Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:29Z", "cpu_seconds": 3.041669609000003, "created_at": "2026-11-22T04:34:19Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-f68a2b075e5d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "5b58862d67dca4c8c576d33e059a50484102a256e06239b117adcf7d8f037cc2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f68a2b075e5d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:19Z", "seed": 2016360744, "treatment": "frontier"}, "priority": 0.16078658441763993, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "34712fa130be0b5ae98fda19515bc6f9646cfc18e5fc40a7f0cf2c861d93f37c", "run_seconds": 3.859581470489502, "seed": 2016360744, "started_at": "2026-11-22T04:34:25Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:27Z", "cpu_seconds": 1.9822812219999975, "created_at": "2026-11-22T04:34:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1733ab444e56", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "7e0d64c7e2b3892536d09e3e7452e606f46ab96d9d36966675e59c5fa223dbe3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1733ab444e56", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:34:19Z", "seed": 2030148649, "treatment": "full"}, "priority": 0.1613106348796218, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "da9494042879f92ee84429d816d8e2f52497f1a49952d2dccd48c796c58c5a10", "run_seconds": 2.056847095489502, "seed": 2030148649, "started_at": "2026-11-22T04:34:24Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:37Z", "cpu_seconds": 2.832850481999998, "created_at": "2026-11-22T04:34:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-6284ca3e3e35", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 7, "n": 4}, "prereg_hash": "e585a17daf31df5a8d024551b6e992c108b3d4a0bd819d073f5aa4817ad734bd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-6284ca3e3e35", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:34:19Z", "seed": 1759625876, "treatment": "full"}, "priority": 0.15714366429724913, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c0beba85c8b527fd3429a5c23cf4049181aa706f3e7b7118ebcaa75263403956", "run_seconds": 3.1024866104125977, "seed": 1759625876, "started_at": "2026-11-22T04:34:34Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:43Z", "cpu_seconds": 12.622501416000006, "created_at": "2026-11-22T04:34:19Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-69665c2a07d5", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 7, "n": 4}, "prereg_hash": "d7d99af80eff1842a216bdb78e938e2439894714908a619cdf9a15767a87800d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-69665c2a07d5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:34:19Z", "seed": 1226428286, "treatment": "full"}, "priority": 0.1582795787662912, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-22fc75aa7ed3", "result_hash": "a661d98c43bbd6421c362be088fb95f6ba34506a0039b0d5f007d93f66f32346", "run_seconds": 15.778353929519653, "seed": 1226428286, "started_at": "2026-11-22T04:34:27Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:20Z", "cpu_seconds": 0.7119255580000043, "created_at": "2026-11-22T04:34:19Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-05c4a05ca073", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "60e53da0876aa619a91479859beae8b8dfdee174ca38183e7df6a50e4f6768c3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-05c4a05ca073", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:34:19Z", "seed": 464884760, "treatment": "gated"}, "priority": 0.16704198612167084, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "4553fa88f42654e039ff072360b2fe8ab515157f45edb26f0032c0f9dfec97d8", "run_seconds": 0.8073327541351318, "seed": 464884760, "started_at": "2026-11-22T04:34:19Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:43Z", "cpu_seconds": 3.023016108, "created_at": "2026-11-22T04:34:35Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-2fba82f03b0f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "7bf2e5dc7764a268fc2d7f265e99ee142d85fe9c6531e1a4eb82ac0411f4e6d2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2fba82f03b0f", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:35Z", "seed": 107816934, "treatment": "frontier"}, "priority": 0.31212041322818973, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "404ffd8b17ac008e96182d324eef031c6eaa70bd353844fdea1e9cf5b8a90bff", "run_seconds": 4.396040678024292, "seed": 107816934, "started_at": "2026-11-22T04:34:39Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:51Z", "cpu_seconds": 1.3896630839999986, "created_at": "2026-11-22T04:34:35Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-53d164a87e17", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "8437fc9d335ac7ef1e185721be5105bbe665f4a08001a316f69905cd3c83d6d8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-53d164a87e17", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:34:35Z", "seed": 763202355, "treatment": "full"}, "priority": 0.16132923183149178, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4cec9f5e158f897f28696c7d2af33dae67e7155bb001006420d30579f3f3d073", "run_seconds": 1.8798964023590088, "seed": 763202355, "started_at": "2026-11-22T04:34:49Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:58Z", "cpu_seconds": 1.3104880980000004, "created_at": "2026-11-22T04:34:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4175648c83f2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 10, "n": 4}, "prereg_hash": "997ab91a3b9ffeec788bd6d0aad20a13d49426803e4f382f5e681b2c76df27d7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4175648c83f2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:34:35Z", "seed": 1839715513, "treatment": "full"}, "priority": 0.1599013768287606, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fede0fa7f5f1df0746cf62459cec05648c54fc22a16bedf7f4a90fd9fcf39e5d", "run_seconds": 1.3917598724365234, "seed": 1839715513, "started_at": "2026-11-22T04:34:56Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:50Z", "cpu_seconds": 1.2720062409999855, "created_at": "2026-11-22T04:34:35Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-5683c281373b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "7dddb2be6ca47012e683b3efc1a9a387bcc2a173e6e6b855d21b469b1c97da42", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-5683c281373b", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:34:35Z", "seed": 233826846, "treatment": "full"}, "priority": 0.1626085810482708, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5eaf25afcfbafbb0c094b407fde91dad992cb9c610a7f7b76c7ab552063a70fb", "run_seconds": 1.7806084156036377, "seed": 233826846, "started_at": "2026-11-22T04:34:48Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:39Z", "cpu_seconds": 101.85760007799999, "created_at": "2026-11-22T04:34:35Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-f1e268e021b0", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "ec6d685aca669d710ff6a545eee7477fdb703c0354c6a177a160f3009c93e438", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-f1e268e021b0", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:34:35Z", "seed": 1407395734, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "9a20b122137c6da4676bb6db1956e63d48f08bad3f68f450c4104e48629337ac", "run_seconds": 121.36288452148438, "seed": 1407395734, "started_at": "2026-11-22T04:34:37Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:44Z", "cpu_seconds": 1.4068675450000043, "created_at": "2026-11-22T04:34:35Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3382a90efea5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "prereg_hash": "8a0d776c0537925126582be93afec199b164e8dc7cc665c5b609f5dc940cb932", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3382a90efea5", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:34:35Z", "seed": 1284074168, "treatment": "gated"}, "priority": 0.1705491591613193, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "cc628c324ba74b4ad6fdcce3f3f95da76e9d34e0becdf3a84172ca1c18f411e3", "run_seconds": 2.0530080795288086, "seed": 1284074168, "started_at": "2026-11-22T04:34:42Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:50Z", "cpu_seconds": 1.0792513729999982, "created_at": "2026-11-22T04:34:35Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-3f9f51349610", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "prereg_hash": "39ff2071ebe29e3a76d13ac84431e973601a7d3144985dbbaf9dd0e3d8616c48", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-3f9f51349610", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|grid_nav|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:34:35Z", "seed": 810659504, "treatment": "gated"}, "priority": 0.16164763531988496, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "f1714fe8233ef0865fe12b5a1fa293c074b871f4408d1a6af820113e77746fdf", "run_seconds": 1.6102986335754395, "seed": 810659504, "started_at": "2026-11-22T04:34:48Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:05Z", "cpu_seconds": 1.5253528129999978, "created_at": "2026-11-22T04:34:51Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-2ff5dbe55253", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "726803044851a6cfde56646baf86491a67ef195f33d27cf6b5d3421814b8a67e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2ff5dbe55253", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:34:51Z", "seed": 1277822185, "treatment": "frontier"}, "priority": 0.15666968816116073, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "7317d0e428ead75396a9e409c672553a6d66c8b15c535638e6b76eda255abbf4", "run_seconds": 2.2024457454681396, "seed": 1277822185, "started_at": "2026-11-22T04:35:03Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:55Z", "cpu_seconds": 0.48402977899999655, "created_at": "2026-11-22T04:34:51Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-81b877751f10", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "d59789dc7dd9e458a00711cf5ec0622510a4e99b783a3ef0187fd8806ae5aa9b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-81b877751f10", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:34:51Z", "seed": 253675764, "treatment": "full"}, "priority": 0.3826590388208314, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-a9bf348dbf33", "result_hash": "35f67fb688482d65dd0da36aa60d72d952b7932ca873e5222e7095ea66b864a5", "run_seconds": 0.5192410945892334, "seed": 253675764, "started_at": "2026-11-22T04:34:55Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:02Z", "cpu_seconds": 0.9067400010000028, "created_at": "2026-11-22T04:34:51Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b6d36a2b79f5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 8, "n": 4}, "prereg_hash": "f8eae9eb6afe46b106f05c7bf24639f9ae0b0b7aa7decfc6501b30045314ef0c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b6d36a2b79f5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:34:51Z", "seed": 1794826727, "treatment": "full"}, "priority": 0.15959648195319137, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "7b606f54c2e3edf972f24e012f2d3a6f1c291ee0f3077b0eefe63d531a35742a", "run_seconds": 1.1482291221618652, "seed": 1794826727, "started_at": "2026-11-22T04:35:00Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:05Z", "cpu_seconds": 2.202820708999994, "created_at": "2026-11-22T04:34:51Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1468a7414264", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "prereg_hash": "02b0ba742f5de1709ec9d8c2ceea994779651c55ac94be50dd96da0293813c78", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1468a7414264", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|arc_grid|arc_grid|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:34:51Z", "seed": 7373559, "treatment": "gated"}, "priority": 0.15943490938703742, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "158dbdb7477e498274cc33b9f815335e45038cdb9d1f94fc692eb120ce5abd0d", "run_seconds": 2.7930634021759033, "seed": 7373559, "started_at": "2026-11-22T04:35:03Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:34:58Z", "cpu_seconds": 2.8576121299999997, "created_at": "2026-11-22T04:34:51Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-9b9dfceb6e15", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "prereg_hash": "c0f490191a140b6e205ceed578daedea6b818d6473b76da7746367bd57869323", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-9b9dfceb6e15", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|coop_agents|coop_agents|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:34:51Z", "seed": 82896333, "treatment": "gated"}, "priority": 0.16415294956490328, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "2b73a12a194614cb72aafbefafecbe0bc3481fb877c96a4faa026846ce2260bc", "run_seconds": 3.081040859222412, "seed": 82896333, "started_at": "2026-11-22T04:34:55Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:20Z", "cpu_seconds": 2.958361388, "created_at": "2026-11-22T04:35:08Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-033af90173f2", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 5, "n": 4}, "prereg_hash": "199cc7aa9d291f505a03d9907a81bacf2c4b89a452b81765b31d52f8fff33852", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-033af90173f2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:08Z", "seed": 1750770057, "treatment": "frontier"}, "priority": 0.15526563001130778, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "0c92add2ccd4b329c1549d0b09ac99b7e804f8512a53f5e2fc6e3e2b1c943c1d", "run_seconds": 3.279210090637207, "seed": 1750770057, "started_at": "2026-11-22T04:35:17Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:16Z", "cpu_seconds": 1.798440768000006, "created_at": "2026-11-22T04:35:08Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c681fa8183f2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 3, "n": 4}, "prereg_hash": "893bb6a4009bafdcd0ce97a42536c46763056883ab8295b37f7836c77778387a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c681fa8183f2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:08Z", "seed": 835318036, "treatment": "frontier"}, "priority": 0.15810708203872975, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "60f55d9ee3388b0d5705dae1a914eeb257f5695fe7f5c36ec9ab6ca921a9940c", "run_seconds": 2.0984060764312744, "seed": 835318036, "started_at": "2026-11-22T04:35:14Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:33Z", "cpu_seconds": 3.539623915, "created_at": "2026-11-22T04:35:08Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0e839e4c1cd0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 7, "n": 4}, "prereg_hash": "a31c38a30890e7827becd4724ffdff30e74bc3cbc13aa5d74d099c1c489e6019", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0e839e4c1cd0", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:08Z", "seed": 1694785094, "treatment": "full"}, "priority": 0.15268197533110256, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "6727a8517736e097de92ce22164015307b0472fd1a2546c964a93c4c6586b8b5", "run_seconds": 4.610551357269287, "seed": 1694785094, "started_at": "2026-11-22T04:35:28Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:35Z", "cpu_seconds": 1.896976715000008, "created_at": "2026-11-22T04:35:08Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-620a61ab5465", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 9, "n": 4}, "prereg_hash": "da0b361d760bc2b12936dbbf4a247cdf0345b9d766544af103eddc7283bbb921", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-620a61ab5465", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:08Z", "seed": 411126836, "treatment": "full"}, "priority": 0.15239897508190497, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2bc5465a5f34538ff781d123131f8c429519601cdf26c85f2c5a1cba604bc634", "run_seconds": 2.826671600341797, "seed": 411126836, "started_at": "2026-11-22T04:35:32Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:14Z", "cpu_seconds": 2.5447153610000015, "created_at": "2026-11-22T04:35:08Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-91dc207a31fc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "d0a4cad1f3488f19a48f0576319697320f1380d04f3440abb97a3f9d6e56622c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-91dc207a31fc", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:08Z", "seed": 896211388, "treatment": "full"}, "priority": 0.3693358217287218, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "169f3456d3ccc64988abd1ff901297ea0b83b9f83769abed58facef4ced4bfa2", "run_seconds": 3.516541004180908, "seed": 896211388, "started_at": "2026-11-22T04:35:10Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:39Z", "cpu_seconds": 1.4161282820000025, "created_at": "2026-11-22T04:35:08Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-5c20491e1891", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "prereg_hash": "ff52b6fdbf097933748ff1745712cb6a64b9a843ec9a4da77508d13757d99289", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-5c20491e1891", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|arc_grid|arc_grid|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:35:08Z", "seed": 1111722901, "treatment": "gated"}, "priority": 0.15189791219654072, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "de33e8c4085c33a71285c17519aef47a4ec21a4bc8945300c8830c878c9c5d1f", "run_seconds": 1.4618525505065918, "seed": 1111722901, "started_at": "2026-11-22T04:35:38Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:12Z", "cpu_seconds": 1.0056719250000015, "created_at": "2026-11-22T04:35:08Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-8abb2c6e0e5c", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "a1e9b9e2992197e40dae6b7afc96a60266457349285e3ad68db2520ba56fc8f9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-8abb2c6e0e5c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:35:08Z", "seed": 195722754, "treatment": "gated"}, "priority": 0.17594974853466933, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-9a78c70b9f09", "result_hash": "d221eedf945cdef1f5c02de99e325f717d346ec3c89eeaf57a2fc9558fa83dce", "run_seconds": 1.5288076400756836, "seed": 195722754, "started_at": "2026-11-22T04:35:10Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:27Z", "cpu_seconds": 1.4952945219999947, "created_at": "2026-11-22T04:35:23Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-134616681506", "kind": "adversarial", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "b9aef7b56577823abaac5ed14cae32f8338ddac90f8b093428cb87facab0cc97", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-134616681506", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "adversarial", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:23Z", "seed": 1292418858, "treatment": "frontier"}, "priority": 0.15506522407822043, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d9db1a1f85ab682cbfad80f14d1d3fe78e29983f1691b3ad27fe935c30737245", "run_seconds": 1.7570013999938965, "seed": 1292418858, "started_at": "2026-11-22T04:35:25Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:35:23Z", "error": null, "family": "ow_goal", "fence": 0, "id": "E141-ow_goal-5461879f7bb9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "05bcbea5f10962a23784b75fa78214af69d99f878e20e10389be7bf79e6719af", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-5461879f7bb9", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:35:23Z", "seed": 437088834, "treatment": "full"}, "priority": 0.1429921232764136, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 437088834, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:35:23Z", "error": null, "family": "ow_model", "fence": 0, "id": "E141-ow_model-2351c0f4dbea", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 8, "n": 4}, "prereg_hash": "737d0df29dde2268b6a1262278ce90dd4b6fc70a6d395f58ae6cf66464f93af3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-2351c0f4dbea", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:23Z", "seed": 1285560853, "treatment": "full"}, "priority": 0.14470900858728308, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 1285560853, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:35:23Z", "error": null, "family": "ow_model", "fence": 0, "id": "E141-ow_model-a65cde207841", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 1, "n": 4}, "prereg_hash": "49db8313e8ef7aea074a7014749dd2aa32974d56f9a968875cb5f6fe9f79e137", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a65cde207841", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:23Z", "seed": 124904717, "treatment": "full"}, "priority": 0.14357533646290216, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-f8854cb83ae8", "result_hash": null, "run_seconds": null, "seed": 124904717, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:45Z", "cpu_seconds": 1.9499347279999881, "created_at": "2026-11-22T04:35:23Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d96ec0d8ec88", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "9005b123d9ba2f13a18f85a2e2fcf3dca2b32e1a47aabc4047f96bf0ec1b4dda", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d96ec0d8ec88", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:23Z", "seed": 5910704, "treatment": "full"}, "priority": 0.14699860732761627, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "2e0abfb07ef2246177ace463f15bac8774c1e9bcb86e1317522ba06fe9411641", "run_seconds": 2.1124460697174072, "seed": 5910704, "started_at": "2026-11-22T04:36:43Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:35:23Z", "error": null, "family": "ow_model", "fence": 0, "id": "E141-ow_model-e03ef4b13bc5", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 7, "n": 4}, "prereg_hash": "c2e3fa2e5381d8aae7f6682ce9f47d72760b415b4aa83ed186b4b6a30285902f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-e03ef4b13bc5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:23Z", "seed": 923383402, "treatment": "full"}, "priority": 0.1458236609170364, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-018159a912fc", "result_hash": null, "run_seconds": null, "seed": 923383402, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:05Z", "cpu_seconds": 1.8302979689999859, "created_at": "2026-11-22T04:35:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-00b1d522f245", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "37add223bb8f751c94c4105b99099350f5fe9018aa65b7d1019e5dbdfdb0f1e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-00b1d522f245", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:40Z", "seed": 751904298, "treatment": "frontier"}, "priority": 0.14889249566684315, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c5ec9ea1c54cdd7f3db97f0b851ae0171f7bafee53230da283ba9a15e37f397d", "run_seconds": 2.7759690284729004, "seed": 751904298, "started_at": "2026-11-22T04:36:02Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:45Z", "cpu_seconds": 2.7772758620000104, "created_at": "2026-11-22T04:35:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0ff8f70241ca", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 6, "n": 4}, "prereg_hash": "3d8a04bdecf006c1114d63f7a75c011757ad44864c2e02ede60fb2d04e98d401", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0ff8f70241ca", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:40Z", "seed": 243634421, "treatment": "frontier"}, "priority": 0.15282276827113353, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5da3c449f15b9d0eb47504e91f6f7bce9c967a501be4ea181999e219814cbda0", "run_seconds": 3.9638755321502686, "seed": 243634421, "started_at": "2026-11-22T04:35:41Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:43Z", "cpu_seconds": 1.7570623969999986, "created_at": "2026-11-22T04:35:40Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-82986daf2500", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "9ac2ed27fe2f1f00f5faae920bc2680222d65f466d94a466585c719e37b8dd45", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-82986daf2500", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:40Z", "seed": 2066572347, "treatment": "frontier"}, "priority": 0.1538817464533587, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "d9d057dbe57d852a1a16c6a1051a37b662db849ba5ed1aaea1e2559be153d5eb", "run_seconds": 2.5199172496795654, "seed": 2066572347, "started_at": "2026-11-22T04:35:40Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:18Z", "cpu_seconds": 0.5098885470000027, "created_at": "2026-11-22T04:35:40Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1db4e57e003d", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 4, "n": 4}, "prereg_hash": "08ac12e11f9aabfac714f1380e0f7b0ad48d8095bd2441d95ad45cb48d00d0ff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1db4e57e003d", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:35:40Z", "seed": 823966473, "treatment": "full"}, "priority": 0.1476935568490012, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-bd2a647d87f6", "result_hash": "e61c0da524ecddb33e86436b7d4a767cf83a8bce2dcdef997f64923e9e6a1722", "run_seconds": 0.5427484512329102, "seed": 823966473, "started_at": "2026-11-22T04:36:17Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:45Z", "cpu_seconds": 0.13046718500000054, "created_at": "2026-11-22T04:35:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-1e7f342794a2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "2122386aaf595ba6e8d4aa5d69976a3d3f692817c4949e8f2842b91553b6c6b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-1e7f342794a2", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:40Z", "seed": 1496729322, "treatment": "full"}, "priority": 0.1519965331496644, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "4d9051e379643a67e64cd63aeede6b138693ff28e0f543f0b95d880df43967d1", "run_seconds": 0.45790576934814453, "seed": 1496729322, "started_at": "2026-11-22T04:35:44Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:51Z", "cpu_seconds": 1.6792225960000025, "created_at": "2026-11-22T04:35:40Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a57d49353274", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 4, "n": 4}, "prereg_hash": "c3141b7d2a99bbdfd6dfb4464ac8763c3afa63c86f21895644aeaf2733d60dfc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a57d49353274", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:40Z", "seed": 1484782928, "treatment": "full"}, "priority": 0.15006986995077412, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "589e0f5fca1d46df4f4e6a6148f86160265fd3b307f9e2b636fdcb2eb46958b2", "run_seconds": 1.7605400085449219, "seed": 1484782928, "started_at": "2026-11-22T04:35:50Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:57Z", "cpu_seconds": 0.6602286359999994, "created_at": "2026-11-22T04:35:55Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0c117c1cf5f1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 8, "n": 4}, "prereg_hash": "3e3f69d3eca38104f8bd56428c2cbf7fb7ac8035d1a16a8b3c6c532b24f6df01", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0c117c1cf5f1", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:55Z", "seed": 1499599827, "treatment": "frontier"}, "priority": 0.16238958624492633, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "64fb6253cad0ccbb1849d806384b2e9aae9bc7cf8e7738b305184bb53f6d4bb2", "run_seconds": 0.8037569522857666, "seed": 1499599827, "started_at": "2026-11-22T04:35:56Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:07Z", "cpu_seconds": 5.266348472999994, "created_at": "2026-11-22T04:35:55Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-40efd36cb887", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 4, "n": 4}, "prereg_hash": "2e798265ec79319a74c98c85abb060944de8265dbecb981172edb19059f5974c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-40efd36cb887", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:55Z", "seed": 2112717236, "treatment": "frontier"}, "priority": 0.1497642859211653, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "1a4407f4f55367fc00007eb3fd227284a340c57d1042e66cf5a21141836c8058", "run_seconds": 7.166576385498047, "seed": 2112717236, "started_at": "2026-11-22T04:35:59Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:35:57Z", "cpu_seconds": 0.24671831399999178, "created_at": "2026-11-22T04:35:55Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-c092ba683b95", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 5, "n": 4}, "prereg_hash": "28758671df6b7e2a56030898688436fd4073e4f29062155de550ed45ef35e105", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-c092ba683b95", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:35:55Z", "seed": 465020608, "treatment": "frontier"}, "priority": 0.15068607758971808, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "2977f2faf40208324c6c99f3ca7a6025828f038027dd1d25f3fb011d7160061b", "run_seconds": 0.2679755687713623, "seed": 465020608, "started_at": "2026-11-22T04:35:57Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:23Z", "cpu_seconds": 1.844813895999991, "created_at": "2026-11-22T04:35:55Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1c31f0b5f100", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "3290d6c75db10ec8ddd8c2eeae4759a54761e95aa77709d76c00daf045f85b37", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1c31f0b5f100", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:35:55Z", "seed": 841620231, "treatment": "full"}, "priority": 0.14765737539041143, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "28cb3eda490be5c2d1d4813773c89f012374692796843193900dd4adf14800a3", "run_seconds": 1.870943546295166, "seed": 841620231, "started_at": "2026-11-22T04:36:21Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:24Z", "cpu_seconds": 1.0927614920000082, "created_at": "2026-11-22T04:35:55Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a87073c5eb2f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "94938b095849ff435b3abb15a7acef022d2de51e66f9e23dd2282f9f8ac046f4", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a87073c5eb2f", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:35:55Z", "seed": 767442501, "treatment": "full"}, "priority": 0.14762875003935927, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f38cb8f82ab2b2d5c2d08d201d6317a7f8fff13399c18c3769ad93a4624e9565", "run_seconds": 1.1385245323181152, "seed": 767442501, "started_at": "2026-11-22T04:36:23Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:11Z", "cpu_seconds": 0.1650952910000001, "created_at": "2026-11-22T04:35:55Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e3fcc3ea4895", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 9, "n": 4}, "prereg_hash": "d062837faa980f74fcf47e4d8d8542a23494608c2060162b8405997ce034d1e0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e3fcc3ea4895", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:35:55Z", "seed": 1445581077, "treatment": "full"}, "priority": 0.14803664046589748, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2c2bfba9ef40afaba86da4c760b25b3a375c56200df2823cadc70e2c6aae5e10", "run_seconds": 0.17037296295166016, "seed": 1445581077, "started_at": "2026-11-22T04:36:10Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:10Z", "cpu_seconds": 0.4385580440000183, "created_at": "2026-11-22T04:35:55Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-03b1ee51540d", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 5, "n": 4}, "prereg_hash": "794dbf0a48b522fce43adc70e0932c347e62ce161f7d352b0e1a343e50991225", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-03b1ee51540d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:35:55Z", "seed": 1782109130, "treatment": "full"}, "priority": 0.14833608328760922, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-273989b63419", "result_hash": "097ec14b8df64d858b2d8c41a2275c3808678c02cab467b3eccc4ff6550c734a", "run_seconds": 0.46724510192871094, "seed": 1782109130, "started_at": "2026-11-22T04:36:10Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_explore", "fence": 0, "id": "E141-ow_explore-2f301f75c374", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 3, "n": 4}, "prereg_hash": "9e56bf9e7347a3d8f0ea082be94890fb2140a5b97000693ef57ae490f7a506dd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-2f301f75c374", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:36:12Z", "seed": 1806958698, "treatment": "frontier"}, "priority": 0.1417608792150996, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 1806958698, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_explore", "fence": 0, "id": "E141-ow_explore-60e0c03bb5d2", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 4, "n": 4}, "prereg_hash": "3aea16239f27949fb4cf31aef56dd1dea9b4ded7612f410dbf0cd03c566f2e9c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-60e0c03bb5d2", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:36:12Z", "seed": 551089808, "treatment": "frontier"}, "priority": 0.14570121682666776, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-13d559243a9b", "result_hash": null, "run_seconds": null, "seed": 551089808, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_goal", "fence": 0, "id": "E141-ow_goal-0d6a659ba7ce", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 8, "n": 4}, "prereg_hash": "bcde8e49b39b3787a0fac91ffe6f62df1dde704ee2a8e42ed7793d6ac9e3346f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0d6a659ba7ce", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:12Z", "seed": 196062844, "treatment": "full"}, "priority": 0.14296235035871432, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 196062844, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:16Z", "cpu_seconds": 0.15430761899999368, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6b0e03fd6294", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "7999e91b148949790e9d906b85c91e6742d262dbc6b9cc7d80c9766eae1e6138", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6b0e03fd6294", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:12Z", "seed": 2131611419, "treatment": "full"}, "priority": 0.1496432784127916, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "c2556027321b297cb35ae31f0ca8ba14b62756c2f202bc81d5496030204dcc2e", "run_seconds": 0.28948235511779785, "seed": 2131611419, "started_at": "2026-11-22T04:36:16Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_model", "fence": 0, "id": "E141-ow_model-7b4986c103b5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 10, "n": 4}, "prereg_hash": "fce40442d76fa10eed7dac5823b36559e307a05e796b8dffbe227d46c231e096", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-7b4986c103b5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:12Z", "seed": 941786062, "treatment": "full"}, "priority": 0.1401203047990613, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 941786062, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:18Z", "cpu_seconds": 2.7412792869999976, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9ae850cac855", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 5, "n": 4}, "prereg_hash": "815ed66327434a38fa793a749456d1ff4876e4c59c76fa001e5040a5572b3579", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9ae850cac855", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:12Z", "seed": 1032743187, "treatment": "full"}, "priority": 0.15455487177931893, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3588750c80f1cdbddf9c29c0f44b56a0a1be8194eda661845ddadbdb1e729318", "run_seconds": 3.2449698448181152, "seed": 1032743187, "started_at": "2026-11-22T04:36:15Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_model", "fence": 0, "id": "E141-ow_model-bcb53c3e2ee1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "4f467e0f18f503b3c7a89d3f6e2bb46b326af2c1bcb6afe9fc57b2e5678568d8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-bcb53c3e2ee1", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:12Z", "seed": 2123414173, "treatment": "full"}, "priority": 0.14598263108639442, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 2123414173, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:36:12Z", "error": null, "family": "ow_transfer", "fence": 0, "id": "E141-ow_transfer-4d23f732d8be", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "prereg_hash": "21a7295d8150fd65dbdc72aed1b5454f69cf473b225cf59cee660771406bb4e9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-4d23f732d8be", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|physics_1d|physics_1d|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:36:12Z", "seed": 1583515393, "treatment": "gated"}, "priority": 0.14619173596856214, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-b3c98494b060", "result_hash": null, "run_seconds": null, "seed": 1583515393, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:32Z", "cpu_seconds": 1.7784326219999969, "created_at": "2026-11-22T04:36:27Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-47ee97856978", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 7, "n": 4}, "prereg_hash": "a2a770efc332227978be821812dcb52869c499b5c7913cea328dc17af6feec4d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-47ee97856978", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:36:27Z", "seed": 1545390834, "treatment": "frontier"}, "priority": 0.2822168193899564, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "eab7e74b78800617f5d92a0eea06fefc519e7f34af0a5d06ea37e778e50c586a", "run_seconds": 2.0382580757141113, "seed": 1545390834, "started_at": "2026-11-22T04:36:30Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:41Z", "cpu_seconds": 2.182428686999998, "created_at": "2026-11-22T04:36:27Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-febe70d8bf25", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 6, "n": 4}, "prereg_hash": "6c79330a7bef6edd9af682a81e754d7031246abaff99c11d6526eab44eb2df62", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-febe70d8bf25", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:36:27Z", "seed": 1109536599, "treatment": "frontier"}, "priority": 0.2556059012558969, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "042622c783e0d17b785f65633e3c7e5285156b463f8a6637ba3c1d202a56ea24", "run_seconds": 3.4991238117218018, "seed": 1109536599, "started_at": "2026-11-22T04:36:37Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:32Z", "cpu_seconds": 2.4460418019999963, "created_at": "2026-11-22T04:36:27Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-1466762cfbfe", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 1, "n": 4}, "prereg_hash": "99ee7f053d31afd17b32c4fe476599a1264750c5b49ea0e98bad45db96f1072a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-1466762cfbfe", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:27Z", "seed": 511949351, "treatment": "full"}, "priority": 0.28450249439187636, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "1691a982cb8c8667ea1dd72c73ae7253006356ae8b714e9d33d6e330f5e3de63", "run_seconds": 3.1753573417663574, "seed": 511949351, "started_at": "2026-11-22T04:36:29Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:40Z", "cpu_seconds": 1.3993252200000086, "created_at": "2026-11-22T04:36:27Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-809427046b08", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "8c389f237d232e93ab4ac67b6bcd29071a039a0e974b555c89e3f2d8a87b4c43", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-809427046b08", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:27Z", "seed": 1425265591, "treatment": "full"}, "priority": 0.22400800205249477, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b43a561a3bf5683cf42b90d6cb6df3031acff4546d323f0fb34d0b16a0b3f55e", "run_seconds": 2.6091859340667725, "seed": 1425265591, "started_at": "2026-11-22T04:36:37Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:31Z", "cpu_seconds": 2.666761852999997, "created_at": "2026-11-22T04:36:27Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a249d9da6718", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 7, "n": 4}, "prereg_hash": "e94b415aec2acfecdfc2cd2c9790c1379088f20c784dfd00029c4f65732e9b41", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a249d9da6718", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:27Z", "seed": 966557093, "treatment": "full"}, "priority": 0.3667642392387206, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "edf6036f03385f08846d95e677d4583d96038c09f2948328512ace0d465e950d", "run_seconds": 3.262479066848755, "seed": 966557093, "started_at": "2026-11-22T04:36:28Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:38Z", "cpu_seconds": 1.2553382749999997, "created_at": "2026-11-22T04:36:27Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ee84f98c2f8f", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "d4bbf810a5d3c050d3f8c6354c0fa397caa7b54fea362be0d347a4d53e6fd5b3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ee84f98c2f8f", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:36:27Z", "seed": 1430426423, "treatment": "gated"}, "priority": 0.26446022683288384, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "5aa0e783f7236dc270d6ef6e7775d3cb16b479966316b0811bcfc23eb8e59188", "run_seconds": 1.5714607238769531, "seed": 1430426423, "started_at": "2026-11-22T04:36:36Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:51Z", "cpu_seconds": 0.3847954780000009, "created_at": "2026-11-22T04:36:44Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0d8b4d3db204", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 5, "n": 4}, "prereg_hash": "d22cb0a3f3cb01202d7c638bea088417895990541b118bc019c75dd232ddf067", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0d8b4d3db204", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:44Z", "seed": 2004081779, "treatment": "full"}, "priority": 0.17613491726523073, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7c88e6a76e4ef58b20aa7c7ce606e3236755d3d50d330f2cb9228d112d5ffba3", "run_seconds": 0.39449357986450195, "seed": 2004081779, "started_at": "2026-11-22T04:36:51Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:46Z", "cpu_seconds": 0.5222824479999986, "created_at": "2026-11-22T04:36:44Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-ccb5f1b696bc", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "b23efe01ba97433f63557d0afdbf1af5c6b2c8b77bab762f2568ae7809910a92", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-ccb5f1b696bc", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:44Z", "seed": 14148460, "treatment": "full"}, "priority": 0.3367497828628422, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b397441809ebeeafeb439bb02b8a2fe11c38094958757fa971007ffd93f7d040", "run_seconds": 0.599877119064331, "seed": 14148460, "started_at": "2026-11-22T04:36:45Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:46Z", "cpu_seconds": 0.22868152400000952, "created_at": "2026-11-22T04:36:44Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-d4416a0f549b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 10, "n": 4}, "prereg_hash": "077805f1e8e1963346e423a8253f024e3a31c2ef44ed34cf0280ca4cd578813c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-d4416a0f549b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:44Z", "seed": 2058923884, "treatment": "full"}, "priority": 0.33180524861128546, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "b452e167cb3207768732a74b2525d94f9b2bb7da5953c842c8c8732d73603c80", "run_seconds": 0.236283540725708, "seed": 2058923884, "started_at": "2026-11-22T04:36:46Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:57Z", "cpu_seconds": 0.6053898910000015, "created_at": "2026-11-22T04:36:44Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-63546535fa67", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 10, "n": 4}, "prereg_hash": "e25e4a1449557195a03e89656bb07019143f1a411d207767b6bf666e87fe9aa0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-63546535fa67", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:44Z", "seed": 1957430695, "treatment": "full"}, "priority": 0.16382219395218023, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e841a3f239e6bb17f0e43690664d062253c433aa1ec19812ab78eb446bf11c83", "run_seconds": 0.822873592376709, "seed": 1957430695, "started_at": "2026-11-22T04:36:56Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:52Z", "cpu_seconds": 1.1838132920000106, "created_at": "2026-11-22T04:36:44Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-6b83265e8e34", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 9, "n": 4}, "prereg_hash": "e110d360c752c3a855a3fe0f620449d77d8ef7341e50d5bf5a66897e74617891", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-6b83265e8e34", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:44Z", "seed": 748202568, "treatment": "full"}, "priority": 0.17228171676662735, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "5818eeda5a5e6ad3e798b88b0f295f4cb86d727c1b9bf59bff28cb56cc24b2f8", "run_seconds": 1.1871919631958008, "seed": 748202568, "started_at": "2026-11-22T04:36:51Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:57Z", "cpu_seconds": 0.4348158450000028, "created_at": "2026-11-22T04:36:44Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-b58c817977e2", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "prereg_hash": "429c110ad2fd2a18faade9037bed8358c081c5ac65f8f0f0cfa8a54b518f64ac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-b58c817977e2", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lever_chain|lever_chain|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:36:44Z", "seed": 1789465329, "treatment": "gated"}, "priority": 0.16841467486460043, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "813b74acdb98a2832263d32cd8d9c4df93bed164e93940fb7911f9d12fae0add", "run_seconds": 0.7111260890960693, "seed": 1789465329, "started_at": "2026-11-22T04:36:56Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:36:51Z", "cpu_seconds": 1.2772456209999916, "created_at": "2026-11-22T04:36:44Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ba26217b7211", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "prereg_hash": "d2ff3b4adea0d42dedb3e715809f496734fb5034dc3d1167a7d2db488a809787", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ba26217b7211", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|grid_nav|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:36:44Z", "seed": 863643373, "treatment": "gated"}, "priority": 0.24154145747291236, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "193abf28685b21ea6ebef265408323ae63a64206b1db6c7bd01ae03e6a6e9e68", "run_seconds": 1.2798855304718018, "seed": 863643373, "started_at": "2026-11-22T04:36:50Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:04Z", "cpu_seconds": 1.229209477000012, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-8fd4e5f358e8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "6caefcbb50166715e88897da96c8c48a8872e7186141201673b65e203be15a60", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8fd4e5f358e8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:36:59Z", "seed": 843944318, "treatment": "frontier"}, "priority": 0.1531868241249538, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "a1d6bb91c33ea21d66f14e0168b4c4794e2d013d6a70264f74c3dd056f0f7596", "run_seconds": 1.2550427913665771, "seed": 843944318, "started_at": "2026-11-22T04:37:03Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:15Z", "cpu_seconds": 1.65261012900001, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0e0abba0e47b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 5, "n": 4}, "prereg_hash": "c496757ad8df3b7f2bb72ddacf3a1db7405138712767206729775e2ebb7bec8a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0e0abba0e47b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:59Z", "seed": 786840474, "treatment": "full"}, "priority": 0.14955341578336173, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c651e330d1cd793a296329b1355ed762debeb006a5ec31d885cd31e8aba26a0c", "run_seconds": 2.3245699405670166, "seed": 786840474, "started_at": "2026-11-22T04:37:12Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:02Z", "cpu_seconds": 0.42953226199998085, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4c5d3803f7b5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 9, "n": 4}, "prereg_hash": "4ae7c78a17ba85040071b4342e91004029dd4eed4d3498700101021fb224171a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4c5d3803f7b5", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:59Z", "seed": 1252321338, "treatment": "full"}, "priority": 0.17658145218117666, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "8bf88d03a081afe0b6cf92d3e6f7c3079b8314a49b62a3464cb735596c8e2a7e", "run_seconds": 0.43291473388671875, "seed": 1252321338, "started_at": "2026-11-22T04:37:02Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:30Z", "cpu_seconds": 0.171349089, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-75b87953737a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 2, "n": 4}, "prereg_hash": "c1de514ff571135e38ea200b78dd013d0c0992373dbb6ccbd94048ccc1291dd8", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-75b87953737a", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:36:59Z", "seed": 520063017, "treatment": "full"}, "priority": 0.14881204154514127, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "6a334cc54e47686a595c60bfb0d8617b6901fd7703b28a89f10645f9762d32e3", "run_seconds": 0.17357230186462402, "seed": 520063017, "started_at": "2026-11-22T04:39:30Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:07Z", "cpu_seconds": 0.12431662700001311, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-4a363e52f1d4", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 8, "n": 4}, "prereg_hash": "979e44f5ed40f4014c631e6c6f67d44fae3e5507bbc5eb9843f8d2ec4b8ae52c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-4a363e52f1d4", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:59Z", "seed": 602179992, "treatment": "full"}, "priority": 0.15238266455675015, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0313e5726d7b19abc4054e11318385a333ef346f0f8aff3649ed5a7e3d21ce4c", "run_seconds": 0.13323330879211426, "seed": 602179992, "started_at": "2026-11-22T04:37:07Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:12Z", "cpu_seconds": 2.3752457619999916, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ca05a55bf191", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "8772d2eee51f64e01fa9109a7ad94497eb300684b5013f6fc560321b35ec3f22", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ca05a55bf191", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:36:59Z", "seed": 1838856305, "treatment": "full"}, "priority": 0.15044270591502562, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f5484ca1a4cf5d1b2e912fc89eaaa3749d2a78d89e29cd29bd171ec05e6c32ad", "run_seconds": 3.309525728225708, "seed": 1838856305, "started_at": "2026-11-22T04:37:09Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:11Z", "cpu_seconds": 2.6508802060000107, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-31dd96ac3f79", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lights|lights|1|1"}, "prereg_hash": "47f42d413fa54069123cfc5f364f1ae4e57fa7fb8748ced766a3e99a06f269b6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-31dd96ac3f79", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "A|lights|lights|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:36:59Z", "seed": 951879055, "treatment": "gated"}, "priority": 0.15092503342222524, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "9d643a3a69b541a65ea187a417bab55226853c79e732fbbbfe30846e7f4f62e9", "run_seconds": 3.2914834022521973, "seed": 951879055, "started_at": "2026-11-22T04:37:08Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:03Z", "cpu_seconds": 1.1155435800000078, "created_at": "2026-11-22T04:36:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-ced351172cf4", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "prereg_hash": "1c40023005856cf7fcdf2b5bb23eabe8a94f4f5b1b668fb942c7e25dcb01d82d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-ced351172cf4", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|stack_machine|stack_machine|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:36:59Z", "seed": 265665217, "treatment": "gated"}, "priority": 0.1721853166675201, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-1ce8f6045ef3", "result_hash": "b5634c0fcf9a360170ea185126ce0aee5d67db4cd7792fa0ba83695bb5a25cce", "run_seconds": 1.1268854141235352, "seed": 265665217, "started_at": "2026-11-22T04:37:02Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:19Z", "cpu_seconds": 1.2121881289999976, "created_at": "2026-11-22T04:37:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-4a2bfc10a1ca", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "33183052886e2595c9fb8e9797ac4be567dcb495d84fe8589e0bf32730cb533e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-4a2bfc10a1ca", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:37:15Z", "seed": 138456848, "treatment": "frontier"}, "priority": 0.16127079095275493, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": "E141-ow_explore-6e9af831ef69", "result_hash": "9c630ad45fbeec2f9faa177ab0336d9b61957d2e16192cf24c9c5f7f31507029", "run_seconds": 1.4503538608551025, "seed": 138456848, "started_at": "2026-11-22T04:37:17Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:26Z", "cpu_seconds": 3.207050073000005, "created_at": "2026-11-22T04:37:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-7f867e5caa7b", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 7, "n": 4}, "prereg_hash": "7d9f6b00449db43963cc9cec9f428615a1be8086d249e4d42a6b093d1f6fa59e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-7f867e5caa7b", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:37:15Z", "seed": 1426569235, "treatment": "frontier"}, "priority": 0.15516694832407868, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "cef6143e45d2d81f4e982c21ce064ece91933fa3a9509e6146eaabe490a33cf1", "run_seconds": 3.2465288639068604, "seed": 1426569235, "started_at": "2026-11-22T04:37:23Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:21Z", "cpu_seconds": 1.143258904999982, "created_at": "2026-11-22T04:37:15Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-e3ce2916aec9", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 10, "n": 4}, "prereg_hash": "134580c478abc388d98dff092864b67355bc010939fbd21a501a9853a220be76", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-e3ce2916aec9", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:37:15Z", "seed": 95152743, "treatment": "frontier"}, "priority": 0.15626535231859257, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5f3da03614d80f71b6dcd24cd3a5892fc2b60ad24e3abbb1c2d0eef17f02e7d2", "run_seconds": 1.1445136070251465, "seed": 95152743, "started_at": "2026-11-22T04:37:20Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:30Z", "cpu_seconds": 3.436408975000006, "created_at": "2026-11-22T04:37:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-182a013e7478", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 7, "n": 4}, "prereg_hash": "d6ea1e63007d4f00949fe43a8d4a7c00c858f12909bac591a59c7072421f94bc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-182a013e7478", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:37:15Z", "seed": 1961457091, "treatment": "full"}, "priority": 0.14961582625510392, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c3f1131b2b78b4f7b7b88c760e0fb5f1bea30a5261bb127949f1e9be1ecb7c58", "run_seconds": 3.7706398963928223, "seed": 1961457091, "started_at": "2026-11-22T04:37:26Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:45Z", "cpu_seconds": 0.176868939000002, "created_at": "2026-11-22T04:37:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-21ac61da7c72", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 1, "n": 4}, "prereg_hash": "e1414d30930fc4d121a2022d39804f6696d64947267ddd76c528c49945b4c492", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-21ac61da7c72", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:37:15Z", "seed": 1536358265, "treatment": "full"}, "priority": 0.14893836343395356, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f2a35119219c8693f39a1b233420e6abe0954aa078bc062fa9a32c0d208c6cc6", "run_seconds": 0.39327239990234375, "seed": 1536358265, "started_at": "2026-11-22T04:37:44Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:04Z", "cpu_seconds": 86.55916534, "created_at": "2026-11-22T04:37:15Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e270380cdca3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 1, "n": 4}, "prereg_hash": "dbe5a4c966b613bd3560c1a91029ba9162135195aa7b5263b36479393bd13ccd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e270380cdca3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:37:15Z", "seed": 1921157981, "treatment": "full"}, "priority": 0.1499233482330562, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "d768a813e14743ca55ab060df86febd09fb5e47c1f09f47e36ebf7afe9ed06da", "run_seconds": 100.37754011154175, "seed": 1921157981, "started_at": "2026-11-22T04:37:24Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:18Z", "cpu_seconds": 1.0444050080000125, "created_at": "2026-11-22T04:37:15Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-33a4d71208e8", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 3, "n": 4}, "prereg_hash": "617283390e41ab13a62733a36ca926bf2e81be6e3b0b3a3089099bff4aea16e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-33a4d71208e8", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:37:15Z", "seed": 1668832886, "treatment": "full"}, "priority": 0.16252673472849433, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-df732dd9e95a", "result_hash": "f6fb1cbea52fe36cf0da5f915dd6a45df718451b6bb5048a910f95aee05fc608", "run_seconds": 1.1886253356933594, "seed": 1668832886, "started_at": "2026-11-22T04:37:16Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:33Z", "cpu_seconds": 0.5954027309999996, "created_at": "2026-11-22T04:37:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-18e4adab79e1", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 4, "n": 4}, "prereg_hash": "d186eaee8d85429c42bdc260157bb98dfdd5aedb125d889aa52aab6448e63b10", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-18e4adab79e1", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:37:31Z", "seed": 2140962832, "treatment": "full"}, "priority": 0.14855314303780037, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "44e7da5ef4e17696ea916dfddf6a04b7ae4f7c88d58bcb1f3a45c5b573d214de", "run_seconds": 0.6057388782501221, "seed": 2140962832, "started_at": "2026-11-22T04:39:33Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:39Z", "cpu_seconds": 0.24588622199999577, "created_at": "2026-11-22T04:37:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-81319cf26148", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "80cdd9d8c451aa8e64a836fe73031ea2eb22c0702f96fde12ef5b049cad23445", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-81319cf26148", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:37:31Z", "seed": 1948742673, "treatment": "full"}, "priority": 0.1512434090076187, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "ec615b316fe9c3f83e7dc9225dc5a42b06b2a35eb719ee2067785a8dd2b7c922", "run_seconds": 0.43067216873168945, "seed": 1948742673, "started_at": "2026-11-22T04:37:39Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:49Z", "cpu_seconds": 4.345293545000004, "created_at": "2026-11-22T04:37:31Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c3daa94da58b", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 2, "n": 4}, "prereg_hash": "b0b583872ec0cb15aba39bd652a9cbf60893bfb532444dfef6e2b50e26cc7bac", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c3daa94da58b", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:37:31Z", "seed": 977128454, "treatment": "full"}, "priority": 0.15023712680034876, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-c2a940d67696", "result_hash": "53c9c44953a87f9ee59b8ae3b1ec2c427fef17ecef3aa0d95b9b682c11d369af", "run_seconds": 7.243741035461426, "seed": 977128454, "started_at": "2026-11-22T04:37:42Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:37Z", "cpu_seconds": 1.560855042, "created_at": "2026-11-22T04:37:31Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-aab6276fdf5a", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 9, "n": 4}, "prereg_hash": "58e351eef0c18a1f04010a7c9bb69c089ee24f35e2a81c502a10e79d67e4e65f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-aab6276fdf5a", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:37:31Z", "seed": 1469350001, "treatment": "full"}, "priority": 0.14752213347314505, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "f96da736f1282ce3c5bc556d7e6bbb664192728c4f598d0b424c704206b4ded7", "run_seconds": 1.6765999794006348, "seed": 1469350001, "started_at": "2026-11-22T04:39:35Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:34Z", "cpu_seconds": 2.5713774889999996, "created_at": "2026-11-22T04:37:31Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-d4307df64cd7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 7, "n": 4}, "prereg_hash": "56c08ba5523c8962d59401bece36c995f851141e86d917b1c0a45dbc2ebacb61", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-d4307df64cd7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:37:31Z", "seed": 1153343383, "treatment": "full"}, "priority": 0.1530224569125977, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "fdce7d2a1ee64e9caa99eecc10803a2378330f82ae6f7ada13d2ff5ab791e858", "run_seconds": 2.9412076473236084, "seed": 1153343383, "started_at": "2026-11-22T04:37:31Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:37:37Z", "cpu_seconds": 1.8776801310000053, "created_at": "2026-11-22T04:37:31Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-056438efa664", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "prereg_hash": "f0df229393681c6271d17ecf888f699ede13a5c98df15e8e20b7e186e923b121", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-056438efa664", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|lights|causal_switch|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:37:31Z", "seed": 690963936, "treatment": "gated"}, "priority": 0.1513765329840491, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "b05689c4039cd672469dbaf52b6371376af59d45871e44deb0f738a1f0dff58f", "run_seconds": 2.2203094959259033, "seed": 690963936, "started_at": "2026-11-22T04:37:35Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:39Z", "cpu_seconds": 1.4667224279999989, "created_at": "2026-11-22T04:37:47Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-0a1ff4a33b3c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "dd80095e90135d63422e4a3156e116ab05da5bb5d40b7d0d03b02b4f9d185dfd", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-0a1ff4a33b3c", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:37:47Z", "seed": 790671705, "treatment": "frontier"}, "priority": 0.1474713574522298, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c16921e13c8a78cfae3f65f1d225cc4e551077bdaa955bc213d7f1d6e4d62a20", "run_seconds": 1.5286343097686768, "seed": 790671705, "started_at": "2026-11-22T04:39:38Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:25Z", "cpu_seconds": 0.5651853870000001, "created_at": "2026-11-22T04:37:47Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-26788284af31", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 2, "n": 4}, "prereg_hash": "68174dd9bb0a11c5abaa7b8433e800f596edd7d5cfc56357366a69d3e2d121e2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-26788284af31", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:37:47Z", "seed": 1597233856, "treatment": "full"}, "priority": 0.15284576481368206, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2a2bcfaba0400bbd738dbb493672c6d1eae8f34b59dca1ebd189803e59a53d99", "run_seconds": 0.6906425952911377, "seed": 1597233856, "started_at": "2026-11-22T04:39:24Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:38:42Z", "cpu_seconds": 1.023190696, "created_at": "2026-11-22T04:37:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-9f8964bf3af7", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "5665e59cd5f5df8c3e1eaf7155dfb6702f2cfdd49f7809d7e5a0132b4f63af9e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-9f8964bf3af7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:37:47Z", "seed": 1963258872, "treatment": "full"}, "priority": 0.15508629139371966, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-576d77d1045b", "result_hash": "d8f25a85332dad0c5d622b1c4b1cf4a8e866267c2cc2c074e9a1300f7baec7ce", "run_seconds": 1.4917230606079102, "seed": 1963258872, "started_at": "2026-11-22T04:38:40Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:40Z", "cpu_seconds": 1.9986422939999997, "created_at": "2026-11-22T04:37:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-a702956839b6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 5, "n": 4}, "prereg_hash": "d2ad01344fef41d6c8073b7751860a64643073c81227c46dd42009d591f49ed3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-a702956839b6", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:37:47Z", "seed": 1540364346, "treatment": "full"}, "priority": 0.14724086058092573, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "a4f2f1eec7beddb0c64dc54d4f9f6927e7f4e481d9c2f59a10391e50260ad0bd", "run_seconds": 2.163388252258301, "seed": 1540364346, "started_at": "2026-11-22T04:39:38Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:38:43Z", "cpu_seconds": 1.640684817, "created_at": "2026-11-22T04:37:47Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-ed2c963df165", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "24ac95101d5fbbb3e70b02895ce09e8565640389bb9708b982e78aaccc274876", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-ed2c963df165", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:37:47Z", "seed": 628101566, "treatment": "full"}, "priority": 0.25697106628968297, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "6e43dd86c66de99bfac9aff0d159bd37de39c957f08aa1701daaf97c4b60cdd9", "run_seconds": 2.3733959197998047, "seed": 628101566, "started_at": "2026-11-22T04:38:40Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:09Z", "cpu_seconds": 76.048123089, "created_at": "2026-11-22T04:37:47Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-0258bbeb087f", "kind": "primary", "lab": "RSI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "b2418d5b78cce5435ec35c88b5457016055d21bc459ad55014644084c7750b12", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-0258bbeb087f", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:37:47Z", "seed": 511179648, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": "c81d85c2e2a7e7fbe21fe2601649d1f395012e03124e43603d404dfb9a5c364f", "run_seconds": 88.59214234352112, "seed": 511179648, "started_at": "2026-11-22T04:38:40Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:28Z", "cpu_seconds": 3.0206315869999996, "created_at": "2026-11-22T04:38:52Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-70e2ccf3cad6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 10, "n": 4}, "prereg_hash": "6ccc862b6c43fba28bdba8f78cb345dfa55c2cd2648da700356601d85cf7b1fc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-70e2ccf3cad6", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:38:52Z", "seed": 717691654, "treatment": "frontier"}, "priority": 0.3229517051988273, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "e84b619a74de9f0da10730606fce951188ea26410d079b45ba444a1102a11c85", "run_seconds": 3.3751721382141113, "seed": 717691654, "started_at": "2026-11-22T04:39:24Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:33Z", "cpu_seconds": 7.694023742, "created_at": "2026-11-22T04:38:52Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-aae88116fe72", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "crafting", "level": 5, "n": 4}, "prereg_hash": "ee540cc9e42924286fae32f15c106fa46f23c4582b61668f6d61c98226797e49", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-aae88116fe72", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "crafting", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:38:52Z", "seed": 762481560, "treatment": "full"}, "priority": 0.33354194132335535, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "405cfd29edc588209b4e81923e85530baa994ea6554f31fdda8edbbbbe1c90f9", "run_seconds": 8.38442611694336, "seed": 762481560, "started_at": "2026-11-22T04:39:24Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:45Z", "cpu_seconds": 0.7110645309999999, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4ec35eb688dd", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 9, "n": 4}, "prereg_hash": "248f5d14b913d3fccec0e9b9d1207ba1caabd9c3db91e845290eb4edff8316fb", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4ec35eb688dd", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:39:39Z", "seed": 2002783783, "treatment": "full"}, "priority": 0.3316829257707976, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4a4e61c1501c15af382e42ff5b5f4cd1832b8684948d6f479033baeef74aafdf", "run_seconds": 0.890460729598999, "seed": 2002783783, "started_at": "2026-11-22T04:39:44Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:47Z", "cpu_seconds": 0.7843165860000001, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bca9a9b55654", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 2, "n": 4}, "prereg_hash": "eb43894f89acb6e64b623ddac724316b408dfdae215bda967c52c7a4ff0f234b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bca9a9b55654", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:39:39Z", "seed": 1535700321, "treatment": "full"}, "priority": 0.3154905355223411, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "3c9ce7657e24e1e843ce280efaf8f64d46a07dda7bd3e19ffabd795775be39bf", "run_seconds": 1.5042426586151123, "seed": 1535700321, "started_at": "2026-11-22T04:39:45Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:58Z", "cpu_seconds": 0.21232212299999986, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-c390ba5f1ea3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 4, "n": 4}, "prereg_hash": "4d50a909a3064c158193a6263276e706e05f9e702278c1f095fd811ff41d27d0", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c390ba5f1ea3", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:39:39Z", "seed": 529360860, "treatment": "full"}, "priority": 0.1746070612237734, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4485d5cbef367b151167bb6c18ab9457a9cf8517cf11f8664c4af0d8ee0c453c", "run_seconds": 0.21583056449890137, "seed": 529360860, "started_at": "2026-11-22T04:39:58Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:52Z", "cpu_seconds": 0.7594023809999992, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e821cfeaf361", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 8, "n": 4}, "prereg_hash": "b4cf8429362b98f0c112fe9451867c4f4904a74ece293239f3407693228d9fc1", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e821cfeaf361", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:39:39Z", "seed": 477320187, "treatment": "full"}, "priority": 0.30238932254909245, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "e4d2d3996f8e4d21ab77c85d3e95fbbdfb3aaf18fa67088c568833cd13a3d257", "run_seconds": 1.5455212593078613, "seed": 477320187, "started_at": "2026-11-22T04:39:50Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:49Z", "cpu_seconds": 0.8106400059999999, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-01c3733a11d5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 3, "n": 4}, "prereg_hash": "4db6a5952acd6eee79ddc993d01256ed7d5ccbc30dd88c9d90fe17dd363ab995", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-01c3733a11d5", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:39:39Z", "seed": 858506428, "treatment": "full"}, "priority": 0.3091238417667963, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "e0b49df484a9dc873898602ca7e0088d2f406c7adc8fcb0307ab6e09d20ba0ab", "run_seconds": 1.2720537185668945, "seed": 858506428, "started_at": "2026-11-22T04:39:48Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:55Z", "cpu_seconds": 0.6677204019999996, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-34dea10b7da3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "physics_1d", "level": 2, "n": 4}, "prereg_hash": "46dd139abd38b92cd5cc792347a121ab0922f1835eceaa419afa58bb4d6312f2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-34dea10b7da3", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "physics_1d", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:39:39Z", "seed": 436965529, "treatment": "full"}, "priority": 0.1779085490035768, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "08a5ff15b4b1eb52608049b3211b5ab67508dd661c71570e2757a632bc4b6b0d", "run_seconds": 0.6710550785064697, "seed": 436965529, "started_at": "2026-11-22T04:39:54Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:43Z", "cpu_seconds": 0.9623573940000001, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-effb09488759", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "6b4d5d71262851453232501ecfbfd4c8d74b995cdddf4f5817c53a48db1119e5", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-effb09488759", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:39:39Z", "seed": 324575198, "treatment": "full"}, "priority": 0.33346869900527, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "65aced3ba436e1e0ac5e056b6feee7de36ed033c83202d040b0ac763801dbc2c", "run_seconds": 1.0624310970306396, "seed": 324575198, "started_at": "2026-11-22T04:39:42Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:39:53Z", "cpu_seconds": 0.32070262299999985, "created_at": "2026-11-22T04:39:39Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-f57a5e60c655", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 9, "n": 4}, "prereg_hash": "f753344c1aaf927ba17a85769f9085f140bc0f4e2722c98696fa842292ee05d6", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-f57a5e60c655", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:39:39Z", "seed": 1221263652, "treatment": "full"}, "priority": 0.286284427993555, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-0a1c53194903", "result_hash": "ba72c6c84aeee21a154fdfaf5b9dea3f2b31dd0c5221c1c821b86ed3a08c174a", "run_seconds": 0.6258265972137451, "seed": 1221263652, "started_at": "2026-11-22T04:39:52Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:08Z", "cpu_seconds": 1.1688460579999997, "created_at": "2026-11-22T04:39:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-575dafb13de3", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 1, "n": 4}, "prereg_hash": "c9641b2fab2697e1ab115735474bc4dbd67458c92adce078bb20c7832267894a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-575dafb13de3", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:39:56Z", "seed": 1055596840, "treatment": "frontier"}, "priority": 0.14970433449904014, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "c8bf4df351a8bbc00899f3051e6ad741a0c0d36465a795ab944a69dbc41da1f7", "run_seconds": 1.606233835220337, "seed": 1055596840, "started_at": "2026-11-22T04:40:06Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:00Z", "cpu_seconds": 3.1167742349999994, "created_at": "2026-11-22T04:39:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-6bbf2ff49f47", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 6, "n": 4}, "prereg_hash": "15fb5be6d52a63d0c9d50abf3cefdc553b791a5c7cd8dc508f6401a1ea36e6b9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-6bbf2ff49f47", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:39:56Z", "seed": 1362363025, "treatment": "frontier"}, "priority": 0.28114901112635704, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "2b796f65f649498999a78c9bdb6aad6690c2b51014fdad8014e543583c229036", "run_seconds": 3.1720848083496094, "seed": 1362363025, "started_at": "2026-11-22T04:39:57Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:15Z", "cpu_seconds": 5.354435798999999, "created_at": "2026-11-22T04:39:56Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-ba1e68d9f22d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 1, "n": 4}, "prereg_hash": "2bbaf2e8225ee94165dc82062cef4257ffeaca40ed2480d6265963f09b3bba68", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-ba1e68d9f22d", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:39:56Z", "seed": 1396458691, "treatment": "frontier"}, "priority": 0.1477735035394389, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "f6d752af4586352e27168f9e8810ff07353e981cf8eb5959faf1e8f4a34fa80d", "run_seconds": 5.570657014846802, "seed": 1396458691, "started_at": "2026-11-22T04:40:09Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:26Z", "cpu_seconds": 0.2624651809999996, "created_at": "2026-11-22T04:39:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-080ae9ef53af", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "1cf128c7f27cab268584ef70809fe781520e1237e94d5873290198b92878b5ff", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-080ae9ef53af", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:39:56Z", "seed": 741803479, "treatment": "full"}, "priority": 0.14770859420955446, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "4bf4940f3755fc0835e5ca14f7a2dba1c5304974857d5b9794d7d057a7b7486a", "run_seconds": 0.2643580436706543, "seed": 741803479, "started_at": "2026-11-22T04:40:26Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:01Z", "cpu_seconds": 1.1331871819999995, "created_at": "2026-11-22T04:39:56Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-48d6238783d6", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "d825154877fad9500fe776a7cb52dff32fdf0deb7fb706b207a9cabe27c33032", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-48d6238783d6", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:39:56Z", "seed": 972427749, "treatment": "full"}, "priority": 0.15411717727714472, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "adbe2aa270fc110f30e204870920e63468a8762fc2903b4d6eab052ff9b844a6", "run_seconds": 1.2582933902740479, "seed": 972427749, "started_at": "2026-11-22T04:40:00Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:04Z", "cpu_seconds": 1.0889001930000006, "created_at": "2026-11-22T04:39:56Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-b6076ab1a74d", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "794451818ca7eb3a46eb59e16f125ad2f81671179084de36a78ed1765f14788f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-b6076ab1a74d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:39:56Z", "seed": 719989437, "treatment": "full"}, "priority": 0.15411599459469885, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "ee2ff044a53cd53cb62371b8d642abb55e16248564659971175842b4d8e5c133", "run_seconds": 1.3939893245697021, "seed": 719989437, "started_at": "2026-11-22T04:40:03Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:09Z", "cpu_seconds": 3.2720964789999982, "created_at": "2026-11-22T04:39:56Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-54ec523024f0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "62a234a2db32e441896afff6f20a9cafbd56a6e2da820f6b37bdb740354a4586", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-54ec523024f0", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:39:56Z", "seed": 1035614777, "treatment": "gated"}, "priority": 0.1510661642671292, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "58669e719ffb42e0f322690c73d84123381444b22a02b218d1722944d3638a98", "run_seconds": 3.9875307083129883, "seed": 1035614777, "started_at": "2026-11-22T04:40:05Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:19Z", "cpu_seconds": 0.834583898, "created_at": "2026-11-22T04:40:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-17b846726f22", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 2, "n": 4}, "prereg_hash": "5f09d12f4f8d0ff4400bf563bac8d298adb3b26694d7152413df9e4ac5a4983e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-17b846726f22", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:11Z", "seed": 1058194777, "treatment": "frontier"}, "priority": 0.1646667443132727, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "41992d3f97ab80983eb632c3f9480f23023def953bd4121ec3d8fba9b26652a1", "run_seconds": 1.02962064743042, "seed": 1058194777, "started_at": "2026-11-22T04:40:18Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:15Z", "cpu_seconds": 0.7736027760000006, "created_at": "2026-11-22T04:40:11Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-1a674e73fef0", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 10, "n": 4}, "prereg_hash": "863a78beb9922504f0e4a361a3b4c4d9f5f99122f2adc2cc23063b70097b113f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-1a674e73fef0", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:11Z", "seed": 1845630123, "treatment": "frontier"}, "priority": 0.16804793004058136, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "084ad5e72dcccf9c465fb98381facf9bbd04ade7aedcd88712f89ab8b0f30745", "run_seconds": 0.8636314868927002, "seed": 1845630123, "started_at": "2026-11-22T04:40:14Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:26Z", "cpu_seconds": 0.9169968259999983, "created_at": "2026-11-22T04:40:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-a245198dfd09", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 4, "n": 4}, "prereg_hash": "af23f310e0989c2fc480610a2d819d394437fe1343764de7c959c123accb4004", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-a245198dfd09", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:11Z", "seed": 686701097, "treatment": "full"}, "priority": 0.15044985573819536, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "80e029892cca87993dbfb8633270c6cfffb7cacb9a39bf7e0541357416bedb6a", "run_seconds": 0.9395279884338379, "seed": 686701097, "started_at": "2026-11-22T04:40:26Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:13Z", "cpu_seconds": 0.6439279400000002, "created_at": "2026-11-22T04:40:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-b11a32bbea64", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 9, "n": 4}, "prereg_hash": "a942c40d432e3c319bc7b00166441fa074c6c9131ee687eca8a915c32293f685", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-b11a32bbea64", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:11Z", "seed": 652266188, "treatment": "full"}, "priority": 0.1712640535141358, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "2a79477400adb4badb8fb4b4c73a3eb3cda34c6b51c83bcceb307c89bd7948f3", "run_seconds": 0.6531338691711426, "seed": 652266188, "started_at": "2026-11-22T04:40:13Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:21Z", "cpu_seconds": 0.5788263350000022, "created_at": "2026-11-22T04:40:11Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-f861bbea4253", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "d6b868f2306a766858507e253d81fe3c97ec874f6c1b6ba81221492c60e48152", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-f861bbea4253", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:11Z", "seed": 1510345236, "treatment": "full"}, "priority": 0.16141200679930956, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "c0e39ff5b98e3a0c9b9aff06a9580d0cd78e581ab66ae6b276293f2373d26df1", "run_seconds": 0.7515044212341309, "seed": 1510345236, "started_at": "2026-11-22T04:40:20Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:25Z", "cpu_seconds": 0.9615313829999987, "created_at": "2026-11-22T04:40:11Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-77934b8a8c31", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lever_chain", "level": 10, "n": 4}, "prereg_hash": "c89cc01756ee2a40315aaf3d58c5253e68a96fa1af02bc633c821a959c929151", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-77934b8a8c31", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lever_chain", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:40:11Z", "seed": 56948122, "treatment": "full"}, "priority": 0.15085754140181903, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "862f780bd423faae63fad82380330c101516809430d464670a9993c69b54fb84", "run_seconds": 0.9929385185241699, "seed": 56948122, "started_at": "2026-11-22T04:40:24Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:21Z", "cpu_seconds": 1.0207962850000012, "created_at": "2026-11-22T04:40:11Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-1701e15649d5", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "prereg_hash": "d51750af2a888ffdb556c9bf8fa852e0eaea0f9fd39654e54c0f84dae5df310d", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-1701e15649d5", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "S|crafting|crafting|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:40:11Z", "seed": 2076342177, "treatment": "gated"}, "priority": 0.1560830796999894, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "3287ff552605b50aa634cbea1328f5bc02ec46338142d0558fa491fa5d2eb328", "run_seconds": 1.2092182636260986, "seed": 2076342177, "started_at": "2026-11-22T04:40:20Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:31Z", "cpu_seconds": 0.7497464300000001, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-3856b0165e08", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 5, "n": 4}, "prereg_hash": "a2ae8756fe63c9b9c388dcc93f14ee0cc9b76ee549e85f406baa12f6d4b7939b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-3856b0165e08", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:28Z", "seed": 2104372795, "treatment": "frontier"}, "priority": 0.16670652954502013, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "5a003ab16ecd7d2cf810909771829f56c15dbe501bced4d63b41733df72bdaf2", "run_seconds": 0.8065059185028076, "seed": 2104372795, "started_at": "2026-11-22T04:40:30Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:34Z", "cpu_seconds": 2.3905296189999987, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-753f0e306d94", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 2, "n": 4}, "prereg_hash": "0b9eec501df254b3f8c65ddb2f22441f4d7e08bd32db831c02805bbb171b2004", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-753f0e306d94", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 2, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:28Z", "seed": 2036883019, "treatment": "frontier"}, "priority": 0.16041571514563652, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "4ab61a6b216341c13996fd98f91fb7f8ef6450afc7875747f831aa1819cb97ea", "run_seconds": 2.4955883026123047, "seed": 2036883019, "started_at": "2026-11-22T04:40:31Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:44Z", "cpu_seconds": 2.4872366459999995, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-926565c212ec", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 5, "n": 4}, "prereg_hash": "440f6eec4e0df6355b671335cacc9b52b00509c80a929890db77ae9a4ffca31b", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-926565c212ec", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:28Z", "seed": 1583399187, "treatment": "frontier"}, "priority": 0.1483291918331457, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "b34b26a3dbe88c5f0bb74bc49ae111992a6de60fdcdcc79116558b82a52b5d46", "run_seconds": 2.5354115962982178, "seed": 1583399187, "started_at": "2026-11-22T04:40:42Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:33Z", "cpu_seconds": 1.4238830219999983, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-4c7b15023cc8", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "62eb67fe3aa3894376db3337b9c56b191fb853312d2c9c46c327373d7d3333d9", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-4c7b15023cc8", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:28Z", "seed": 1609221515, "treatment": "full"}, "priority": 0.15178509501384185, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": "E141-ow_goal-55e6be2ad704", "result_hash": "00c8363d133c19fcf496f6690cff93b15b5b328add5190e52690cb2da921f01e", "run_seconds": 1.4760537147521973, "seed": 1609221515, "started_at": "2026-11-22T04:40:31Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:56Z", "cpu_seconds": 2.3561319570000023, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-26772fc6d746", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 8, "n": 4}, "prereg_hash": "08d74bf034c1fca6a87933b9174266ee3d428b0daa87929c23242259086a3d4e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-26772fc6d746", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:40:28Z", "seed": 508978472, "treatment": "full"}, "priority": 0.14825664806805205, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "52c4c457150001a42adecdc9bbf1536f0487e448cf03ddd57d5786074f6b19b1", "run_seconds": 2.713686466217041, "seed": 508978472, "started_at": "2026-11-22T04:40:53Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:39Z", "cpu_seconds": 0.33789538199999924, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-36c4dce639f7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 4, "n": 4}, "prereg_hash": "61f713470d2fbeb6c842243fd5273b1cf132a71e59f9336c5402c01ed3998535", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-36c4dce639f7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 4, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:40:28Z", "seed": 1097165882, "treatment": "full"}, "priority": 0.149053957882582, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "0b43148a58e8aa522399e7f4522e0c372ceb000efe001dfbb8208752045e03ad", "run_seconds": 0.3383677005767822, "seed": 1097165882, "started_at": "2026-11-22T04:40:39Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:37Z", "cpu_seconds": 0.44803629400000133, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-c1c487d27353", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 5, "n": 4}, "prereg_hash": "c16fca7d6c54137fd5f400856ffeab500a1235f770699edf5c0ae476d336b08e", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-c1c487d27353", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 5, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:40:28Z", "seed": 1927771176, "treatment": "full"}, "priority": 0.1507712955783852, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-0cef5d6a8750", "result_hash": "4ab8a8ab3306ee03a89db9072b491a80e174f205285e1d639eb27054865a3b09", "run_seconds": 0.4537017345428467, "seed": 1927771176, "started_at": "2026-11-22T04:40:36Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:38Z", "cpu_seconds": 0.4453314529999979, "created_at": "2026-11-22T04:40:28Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-39461256d159", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "prereg_hash": "440b63c2c3d9d136cb0c50c2455a1915cc18065324ad9da90cc1313a700993f7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-39461256d159", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|lever_chain|lever_chain|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:40:28Z", "seed": 871387614, "treatment": "gated"}, "priority": 0.14995410484366428, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-77b3137bc021", "result_hash": "fc116a370355dc0741520275d9763c88f62b6b29018410e4ebf4d4b0ef95a919", "run_seconds": 0.4507429599761963, "seed": 871387614, "started_at": "2026-11-22T04:40:38Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:41:00Z", "cpu_seconds": 3.6723546450000004, "created_at": "2026-11-22T04:40:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-301b903c8142", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 8, "n": 4}, "prereg_hash": "cae94c1f50c879d19d2f2c1c1d96856168bf616c63bbe9b4045a33c32d61d56a", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-301b903c8142", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 8, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:43Z", "seed": 1040303366, "treatment": "frontier"}, "priority": 0.14757558468095988, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "664975c95e13cab76f576b68f970191b23ddffac7d98b1cc8dfd0d208a85bc44", "run_seconds": 3.6868345737457275, "seed": 1040303366, "started_at": "2026-11-22T04:40:56Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:48Z", "cpu_seconds": 3.404684336999999, "created_at": "2026-11-22T04:40:43Z", "error": null, "family": "ow_explore", "fence": 1, "id": "E141-ow_explore-906fee8df7f8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "eb8897e9d94fd93a01c23631613af2bbf37f4bbdfd342c4ca521c55186d45da2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-906fee8df7f8", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:43Z", "seed": 2111280407, "treatment": "frontier"}, "priority": 0.15272930428873244, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": "bf83ee33ead6ee672d8bf85c6fca5ee5e5a8fe38705b35e31fc696c7100ffd2b", "run_seconds": 3.64068341255188, "seed": 2111280407, "started_at": "2026-11-22T04:40:44Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:46Z", "cpu_seconds": 2.264971797000001, "created_at": "2026-11-22T04:40:43Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-bcc89982f564", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "lights", "level": 3, "n": 4}, "prereg_hash": "ddb4c7e55de1bfcce4f8b59124d80d13c4bcd9d59163012451e40451c94f7693", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-bcc89982f564", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "lights", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:43Z", "seed": 1032572675, "treatment": "full"}, "priority": 0.1532532704506427, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "977db656b2dc1c09a536e093e8130c3dcdcfb09bd0e383c97cb4a9a8ccc4939c", "run_seconds": 2.4338982105255127, "seed": 1032572675, "started_at": "2026-11-22T04:40:43Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:51Z", "cpu_seconds": 0.19304553200000285, "created_at": "2026-11-22T04:40:43Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-e1c3086444f8", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 6, "n": 4}, "prereg_hash": "b7c9ef6c5772fb613695966c5a51dcc39827ebe0ed37029625aca9fb2f08e6f2", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-e1c3086444f8", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:43Z", "seed": 1817188588, "treatment": "full"}, "priority": 0.14948767926897366, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "f4618bf0915d05f468e78a38efd19a6e16f1f6ab9b3b0375fc314325676f7c52", "run_seconds": 0.19346880912780762, "seed": 1817188588, "started_at": "2026-11-22T04:40:51Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:57Z", "cpu_seconds": 0.4514152629999977, "created_at": "2026-11-22T04:40:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-0e54396c56c7", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "math_oracle", "level": 7, "n": 4}, "prereg_hash": "e6f7232d5047aa0aebfc4da51b9c9370d5eae816e9d18ca0e6ea67a62ab4db85", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-0e54396c56c7", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "math_oracle", "level": 7, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:40:43Z", "seed": 2040332705, "treatment": "full"}, "priority": 0.14720623185543918, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": null, "result_hash": "3129288e34b3ec28b94ca4464a3089875e17c94b0517b2d6c96c669d935bd397", "run_seconds": 0.46540236473083496, "seed": 2040332705, "started_at": "2026-11-22T04:40:56Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:40:51Z", "cpu_seconds": 1.5028641020000002, "created_at": "2026-11-22T04:40:43Z", "error": null, "family": "ow_model", "fence": 1, "id": "E141-ow_model-aee8c1caa215", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "coop_agents", "level": 10, "n": 4}, "prereg_hash": "c2dcb6cb1ed246a371cabd90ab263808a58d70637c6dbbf82282b9766a7b7b9f", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-aee8c1caa215", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "coop_agents", "level": 10, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:40:43Z", "seed": 1210955885, "treatment": "full"}, "priority": 0.1516290948980829, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-d5fe5939322c", "result_hash": "708d7f382ebd2c88daa9b47b1f5f4d063bbb0049edbee8af3b24c12ca28a80ed", "run_seconds": 1.6016311645507812, "seed": 1210955885, "started_at": "2026-11-22T04:40:49Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:40:43Z", "error": null, "family": "ow_transfer", "fence": 0, "id": "E141-ow_transfer-d6bf5f937a98", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "0b071b20bce22738798ceee047ed766ba9cc0c4f795d2bbf5d056b8b8f4e8469", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-d6bf5f937a98", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:40:43Z", "seed": 918375779, "treatment": "gated"}, "priority": 0.1448968228087988, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-1fbe62cc0fef", "result_hash": null, "run_seconds": null, "seed": 918375779, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_explore", "fence": 0, "id": "E141-ow_explore-8e44b8468f6e", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "arc_grid", "level": 9, "n": 4}, "prereg_hash": "59d31f75f980824d86e2a37271b44244e25079bf3a933fc615cfa1730dc0601c", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-8e44b8468f6e", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "arc_grid", "level": 9, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:59Z", "seed": 305181807, "treatment": "frontier"}, "priority": 0.14292512763683465, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 305181807, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_explore", "fence": 0, "id": "E141-ow_explore-f93115999897", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "grid_nav", "level": 1, "n": 4}, "prereg_hash": "43bb2a59f857df240203fdb3bcdb9d124c62cd834fa7f959e69f516ff4de3472", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["random", "count", "prederr", "infogain", "hypdisc", "goal", "risk"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "open-world exploration (10 families x 10 levels)", "evolution": "E142", "experiment_id": "E141-ow_explore-f93115999897", "family": "ow_explore", "hypothesis": "OW1: next-observation accuracy (all reachable states x actions) of frontier > random, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "grid_nav", "level": 1, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "random", "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "registered_at": "2026-11-22T04:40:59Z", "seed": 1961180141, "treatment": "frontier"}, "priority": 0.14304052512674814, "question": "Does model-based frontier exploration learn a more accurate world model than random exploration at the same interaction budget?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 1961180141, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:41:05Z", "cpu_seconds": 0.28002430900000164, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_goal", "fence": 1, "id": "E141-ow_goal-0d0112fbeb5c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 6, "n": 4}, "prereg_hash": "3106e60f1be939d57dbdc014a4fa0db2879c9467e6a5ba19c63a1d23360ee3dc", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-0d0112fbeb5c", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:59Z", "seed": 1516337188, "treatment": "full"}, "priority": 0.15077934349573663, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": "7e9a6919a0361f54d71dc0ff81521a334b38d048c9d08312a18775362d3211e4", "run_seconds": 0.4351928234100342, "seed": 1516337188, "started_at": "2026-11-22T04:41:05Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_goal", "fence": 0, "id": "E141-ow_goal-c836004924aa", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "causal_switch", "level": 3, "n": 4}, "prereg_hash": "e79433988bcedce4f0ebb8f5de80e7808abe211e6a0fc128864a6588a7dcf508", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["e141_best", "random", "no_world_model", "fixed_policy"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "hidden-goal discovery, planning and abstention", "evolution": "E142", "experiment_id": "E141-ow_goal-c836004924aa", "family": "ow_goal", "hypothesis": "OW4: evaluator-executed goal completion (unsolvable: correct abstention) of full > e141_best, same budget.", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "causal_switch", "level": 3, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "e141_best", "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "registered_at": "2026-11-22T04:40:59Z", "seed": 1547921234, "treatment": "full"}, "priority": 0.14873155110105862, "question": "Does the E142 agent (frontier + goal-directed exploration, rule model, abstention) complete hidden goals more often than the E141-style count-curiosity tabular agent?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 1547921234, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_model", "fence": 0, "id": "E141-ow_model-416f01a72b6d", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"env": "stack_machine", "level": 6, "n": 4}, "prereg_hash": "72f3d142db1193ba821a72b37b987d04a4effcc9f502d2f4f008d1f62cc5c0b7", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["facts", "rules_nodetect", "frozen", "retrieval"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "persistent world model: induced rules + revision", "evolution": "E142", "experiment_id": "E141-ow_model-416f01a72b6d", "family": "ow_model", "hypothesis": "OW2: accuracy of full model (rules + change detection + history) > facts-only model, same trajectory budget.", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L1-L10", "metric_direction": "higher", "params": {"env": "stack_machine", "level": 6, "n": 4}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "facts", "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "registered_at": "2026-11-22T04:40:59Z", "seed": 705721761, "treatment": "full"}, "priority": 0.1451103215001392, "question": "Do induced rules plus tested revision predict unseen transitions better than a fact table from the same experience?", "replicates": "E141-ow_model-f2bf98c8ddb4", "result_hash": null, "run_seconds": null, "seed": 705721761, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_rsi", "fence": 1, "id": "E141-ow_rsi-06c5bfb6d6cd", "kind": "primary", "lab": "RSI_LAB", "lease_owner": "w-any-3f27a83e", "lease_until": 1795322524.865438, "max_attempts": 3, "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "prereg_hash": "ef528e62d43e8e6d5efac36b04b95198755cc59d811b3a10dfee47c3f81efd3c", "preregistration": {"analysis": "experiment-level sign tests (confirmatory, Holm across OW1-OW7)", "compute": "same seed; 3 generations x 4 children x 6 dev instances per lineage arm; random_search gets the same 12 candidate evaluations", "controls": ["G0", "random_search", "inherit_nomem", "reset_mem", "rootcause", "retrieval", "compressed", "xfer_mem"], "evolution": "E142", "experiment_id": "E141-ow_rsi-06c5bfb6d6cd", "family": "ow_rsi", "heldout": "10 sealed instances generated by the evaluator from its secret salt", "hypothesis_id": "OW5-OW7", "lab": "RSI_LAB", "params": {"generations": 3, "k": 4, "n_dev": 6, "n_sealed": 10}, "pathway": "2 (agent harness / search / memory parameters; no code generation, no weights)", "preregistration": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "G0", "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "registered_at": "2026-11-22T04:40:59Z", "seed": 122249637, "treatment": "inherit_mem"}, "priority": 4.5, "question": "Does a governed lineage that inherits its proposer and keeps research memory produce an open-world agent configuration that beats G0, compute-matched random search, the no-memory lineage and the reset-proposer lineage on sealed instances?", "replicates": null, "result_hash": null, "run_seconds": null, "seed": 122249637, "started_at": "2026-11-22T04:41:01Z", "state": "RUNNING", "verdict": null, "verified_at": null}
{"attempts": 1, "completed_at": "2026-11-22T04:41:02Z", "cpu_seconds": 0.6164809790000021, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_transfer", "fence": 1, "id": "E141-ow_transfer-c02e57855f7c", "kind": "primary", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "prereg_hash": "3c3378147e530f5376e8eaaf3c159c455b15164121958d49a352b4bff89d5f76", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-c02e57855f7c", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "primary", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.05, "n": 4, "pair": "S|physics_1d|physics_1d|1|9"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:40:59Z", "seed": 1468503982, "treatment": "gated"}, "priority": 0.16139431429115253, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": null, "result_hash": "32158a8232a3b1d92eb42914bc40620ad6c185aacfa7a69628cc0e893f2ad579", "run_seconds": 0.6282672882080078, "seed": 1468503982, "started_at": "2026-11-22T04:41:02Z", "state": "VERIFICATION_PENDING", "verdict": null, "verified_at": null}
{"attempts": 0, "completed_at": null, "cpu_seconds": null, "created_at": "2026-11-22T04:40:59Z", "error": null, "family": "ow_transfer", "fence": 0, "id": "E141-ow_transfer-eb96ddef48dd", "kind": "replication", "lab": "AGI_LAB", "lease_owner": null, "lease_until": null, "max_attempts": 3, "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "prereg_hash": "07bf04015a18523e8b9499dc88b8c01b6c318844e07a41b2d727572baafd2df3", "preregistration": {"analysis": "paired per-instance difference vs primary control; 95% normal interval; one-sided exact sign test (log space); SUPPORTED iff wins > losses and p < 0.05; Holm correction applied across experiments in the report; ties excluded from the sign test", "completion": "every arm produces a prediction for every instance; evaluator recomputes all scores", "compute": {"cpu_only": true, "model_calls": 0, "paid_api": 0}, "controls": ["scratch", "ungated", "naive", "compressed"], "dataset": "constructed environment regenerated from (params, seed); freshness: generated per experiment", "domain": "cross-configuration / interface / domain transfer", "evolution": "E142", "experiment_id": "E141-ow_transfer-eb96ddef48dd", "family": "ow_transfer", "hypothesis": "OW3: target accuracy gated > scratch (budget_mult 0.05); OW3b: gated > ungated (negative transfer rejected).", "kind": "replication", "lab": "AGI_LAB", "ladder_level": "L9", "metric_direction": "higher", "params": {"budget_mult": 0.1, "n": 4, "pair": "C|coop_agents|grid_nav|1|1"}, "preregistration_commit": "f77d2db2f", "preregistration_doc": "docs/evolution/E142_OPEN_WORLD_PREREGISTRATION.md", "primary_control": "scratch", "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "registered_at": "2026-11-22T04:40:59Z", "seed": 944761467, "treatment": "gated"}, "priority": 0.14468187859696569, "question": "Does an evidence-gated, interface-aligned rule prior from a source environment improve the target world model at a scarce budget?", "replicates": "E141-ow_transfer-709201c95c69", "result_hash": null, "run_seconds": null, "seed": 944761467, "started_at": null, "state": "PREREGISTERED", "verdict": null, "verified_at": null}
