{"edges": [{"from": "H:H1", "rel": "addresses", "to": "Q:Q1"}, {"from": "H:H2", "rel": "addresses", "to": "Q:Q2"}, {"from": "H:H3", "rel": "addresses", "to": "Q:Q3"}, {"from": "H:H3s", "rel": "addresses", "to": "Q:Q3"}, {"from": "H:H7", "rel": "addresses", "to": "Q:Q7"}, {"from": "H:H7b", "rel": "addresses", "to": "Q:Q7"}, {"from": "H:H5", "rel": "addresses", "to": "Q:Q5"}, {"from": "H:H8", "rel": "addresses", "to": "Q:Q8"}, {"from": "H:H9", "rel": "addresses", "to": "Q:Q9"}, {"from": "Q:Q1", "rel": "serves", "to": "O:O1"}, {"from": "Q:Q2", "rel": "serves", "to": "O:O1"}, {"from": "Q:Q2", "rel": "depends_on", "to": "Q:Q1"}, {"from": "Q:Q3", "rel": "serves", "to": "O:O1"}, {"from": "Q:Q3", "rel": "depends_on", "to": "Q:Q2"}, {"from": "Q:Q4", "rel": "serves", "to": "O:O1"}, {"from": "Q:Q5", "rel": "serves", "to": "O:O2"}, {"from": "Q:Q6", "rel": "serves", "to": "O:O3"}, {"from": "Q:Q7", "rel": "serves", "to": "O:O4"}, {"from": "Q:Q7", "rel": "depends_on", "to": "Q:Q4"}, {"from": "Q:Q8", "rel": "serves", "to": "O:O3"}, {"from": "Q:Q8", "rel": "depends_on", "to": "Q:Q6"}, {"from": "Q:Q9", "rel": "serves", "to": "O:O1"}, {"from": "Q:Q9", "rel": "depends_on", "to": "Q:Q4"}, {"from": "X:active.dev", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.sealed.random", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.sealed.sweep", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.sealed.eig", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.sealed.eig_memory", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.sealed.eig_poisoned_memory", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.transfer.eig", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.transfer.eig_memory", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.transfer.random", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.shift.eig", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.shift.eig_memory", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.adversarial.eig", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:active.adversarial.random", "rel": "executed_by", "to": "A:agent-investigator"}, {"from": "X:worldmodel.val", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:worldmodel.sealed", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:worldmodel.shift", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:worldmodel.adversarial", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:continual.dev", "rel": "executed_by", "to": "A:agent-learner"}, {"from": "X:continual.sealed", "rel": "executed_by", "to": "A:agent-learner"}, {"from": "X:skills.high_diversity", "rel": "executed_by", "to": "A:agent-skill-engineer"}, {"from": "X:skills.low_diversity", "rel": "executed_by", "to": "A:agent-skill-engineer"}, {"from": "X:selfimprove.C1_mixture_hybrid", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:selfimprove.C2_strict_switch_0.95", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:selfimprove.C3_lenient_evaluator", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:selfimprove.C4_read_sealed_in_dev", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:selfimprove.C5_self_signed_promotion", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:cycle2.B2.gated_skills", "rel": "executed_by", "to": "A:agent-skill-engineer"}, {"from": "X:cycle2.B1.grammar_growth", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:cycle2.B1.grammar_growth_breakdown", "rel": "executed_by", "to": "A:agent-modeler"}, {"from": "X:active.sealed.eig", "rel": "tests", "to": "H:H1"}, {"from": "X:active.sealed.eig", "rel": "tests", "to": "H:H2"}, {"from": "X:active.sealed.eig", "rel": "tests", "to": "H:H3"}, {"from": "C:C1_mixture_hybrid", "rel": "tests", "to": "H:H7"}, {"from": "C:C2_strict_switch_0.95", "rel": "tests", "to": "H:H7"}, {"from": "C:C3_lenient_evaluator", "rel": "tests", "to": "H:H7"}, {"from": "C:C4_read_sealed_in_dev", "rel": "tests", "to": "H:H7"}, {"from": "C:C5_self_signed_promotion", "rel": "tests", "to": "H:H7"}, {"from": "P:AMA-Bench", "rel": "informs (not reproduced)", "to": "O:O2"}, {"from": "P:LongMemEval", "rel": "informs (not reproduced)", "to": "O:O2"}, {"from": "P:LoCoMo", "rel": "informs (not reproduced)", "to": "O:O2"}, {"from": "P:Agent-World", "rel": "informs (not reproduced)", "to": "Q:Q6"}, {"from": "P:ACS", "rel": "design reference (no interop claimed)", "to": "O:O4"}, {"from": "O:O1", "rel": "builds_on", "to": "E:E112"}], "nodes": [{"id": "P:AMA-Bench", "label": "AMA-Bench: Evaluating Long-Horizon Memory for Agentic Applications (arXiv:2602.22769)", "status": "REVIEWED_NOT_REPRODUCED", "type": "paper"}, {"id": "P:Agent-World", "label": "Agent-World: Scaling Real-World Environment Synthesis (arXiv:2604.18292)", "status": "REVIEWED_NOT_REPRODUCED", "type": "paper"}, {"id": "P:LongMemEval", "label": "LongMemEval (arXiv:2410.10813)", "status": "REVIEWED_NOT_REPRODUCED", "type": "paper"}, {"id": "P:LoCoMo", "label": "LoCoMo (Maharana et al., NAACL 2024)", "status": "REVIEWED_NOT_REPRODUCED", "type": "paper"}, {"id": "P:ACS", "label": "Microsoft Agent Control Specification 0.3.1-beta (draft, design reference)", "status": "REVIEWED_NOT_REPRODUCED", "type": "paper"}, {"id": "H:H1", "label": "EIG needs fewer experiments than random at equal identification", "status": "SUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "H:H2", "label": "Validated family prior reduces experiments vs uniform-prior EIG", "status": "UNSUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "H:H3", "label": "Memory prior transfers to 6-variable worlds", "status": "SUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "H:H3s", "label": "Memory prior does not hurt on shifted worlds", "status": "UNSUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "H:H7", "label": "Mixture hybrid lowers sealed Brier vs switch baseline", "status": "UNSUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "H:H7b", "label": "Stricter switch threshold lowers sealed Brier", "status": "UNSUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "H:H5", "label": "Selective specialization (experts) retains best per unit compute", "status": "STALLED", "type": "hypothesis"}, {"id": "H:H8", "label": "Gated top-4 library loses fewer shift tasks than ungated top-4, keeping sealed gains", "status": "UNSUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "H:H9", "label": "Grammar growth lowers shift Brier vs the switch hybrid without in-class regression", "status": "UNSUPPORTED_HYPOTHESIS", "type": "hypothesis"}, {"id": "Q:Q1", "label": "Does EIG experiment selection beat random selection on sealed worlds?", "type": "question"}, {"id": "Q:Q2", "label": "Does a prior from validated memory of earlier worlds reduce experiments further?", "type": "question"}, {"id": "Q:Q3", "label": "Does that memory prior transfer to a different environment size, and does it hurt under shift?", "type": "question"}, {"id": "Q:Q4", "label": "Which world model predicts interventions best in- and out-of-distribution?", "type": "question"}, {"id": "Q:Q5", "label": "Which continual learner retains best per unit compute?", "type": "question"}, {"id": "Q:Q6", "label": "Do mined skills raise held-out solve rate; does environment diversity matter?", "type": "question"}, {"id": "Q:Q7", "label": "Can a candidate configuration beat the frozen world-model baseline on sealed data?", "type": "question"}, {"id": "Q:Q8", "label": "Does context-gating make a broader skill library safe under shift? (backlog B2)", "type": "question"}, {"id": "Q:Q9", "label": "Does growing the rule grammar after misspecification fix shift prediction? (backlog B1)", "type": "question"}, {"id": "O:O1", "label": "Learn causal structure of unfamiliar environments with fewer experiments", "type": "objective"}, {"id": "O:O2", "label": "Learn continually without silently destroying earlier capability", "type": "objective"}, {"id": "O:O3", "label": "Acquire reusable skills that help on held-out tasks and are rolled back when they hurt", "type": "objective"}, {"id": "O:O4", "label": "Improve the lab itself only through sealed, independently verified, authorized changes", "type": "objective"}, {"id": "X:active.dev", "label": "active.dev", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.sealed.random", "label": "active.sealed.random", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.sealed.sweep", "label": "active.sealed.sweep", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.sealed.eig", "label": "active.sealed.eig", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.sealed.eig_memory", "label": "active.sealed.eig_memory", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.sealed.eig_poisoned_memory", "label": "active.sealed.eig_poisoned_memory", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.transfer.eig", "label": "active.transfer.eig", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.transfer.eig_memory", "label": "active.transfer.eig_memory", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.transfer.random", "label": "active.transfer.random", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.shift.eig", "label": "active.shift.eig", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.shift.eig_memory", "label": "active.shift.eig_memory", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.adversarial.eig", "label": "active.adversarial.eig", "status": "ADMITTED", "type": "experiment"}, {"id": "X:active.adversarial.random", "label": "active.adversarial.random", "status": "ADMITTED", "type": "experiment"}, {"id": "X:worldmodel.val", "label": "worldmodel.val", "status": "ADMITTED", "type": "experiment"}, {"id": "X:worldmodel.sealed", "label": "worldmodel.sealed", "status": "ADMITTED", "type": "experiment"}, {"id": "X:worldmodel.shift", "label": "worldmodel.shift", "status": "ADMITTED", "type": "experiment"}, {"id": "X:worldmodel.adversarial", "label": "worldmodel.adversarial", "status": "ADMITTED", "type": "experiment"}, {"id": "X:continual.dev", "label": "continual.dev", "status": "ADMITTED", "type": "experiment"}, {"id": "X:continual.sealed", "label": "continual.sealed", "status": "ADMITTED", "type": "experiment"}, {"id": "X:skills.high_diversity", "label": "skills.high_diversity", "status": "ADMITTED", "type": "experiment"}, {"id": "X:skills.low_diversity", "label": "skills.low_diversity", "status": "ADMITTED", "type": "experiment"}, {"id": "X:selfimprove.C1_mixture_hybrid", "label": "selfimprove.C1_mixture_hybrid", "status": "ADMITTED", "type": "experiment"}, {"id": "X:selfimprove.C2_strict_switch_0.95", "label": "selfimprove.C2_strict_switch_0.95", "status": "ADMITTED", "type": "experiment"}, {"id": "X:selfimprove.C3_lenient_evaluator", "label": "selfimprove.C3_lenient_evaluator", "status": "ADMITTED", "type": "experiment"}, {"id": "X:selfimprove.C4_read_sealed_in_dev", "label": "selfimprove.C4_read_sealed_in_dev", "status": "ADMITTED", "type": "experiment"}, {"id": "X:selfimprove.C5_self_signed_promotion", "label": "selfimprove.C5_self_signed_promotion", "status": "ADMITTED", "type": "experiment"}, {"id": "X:cycle2.B2.gated_skills", "label": "cycle2.B2.gated_skills", "status": "ADMITTED", "type": "experiment"}, {"id": "X:cycle2.B1.grammar_growth", "label": "cycle2.B1.grammar_growth", "status": "ADMITTED", "type": "experiment"}, {"id": "X:cycle2.B1.grammar_growth_breakdown", "label": "cycle2.B1.grammar_growth_breakdown", "status": "ADMITTED", "type": "experiment"}, {"id": "A:agent-skill-engineer", "label": "agent-skill-engineer", "type": "agent"}, {"id": "A:agent-investigator", "label": "agent-investigator", "type": "agent"}, {"id": "A:agent-learner", "label": "agent-learner", "type": "agent"}, {"id": "A:agent-modeler", "label": "agent-modeler", "type": "agent"}, {"id": "M:6bf9ff280e87@1", "label": "world-family-prior", "status": "ACTIVE", "type": "memory:semantic"}, {"id": "M:86f5eed908d3@1", "label": "skill:high_diversity:skill0_sq_sort", "status": "RETIRED", "type": "memory:procedural"}, {"id": "M:c4284707233d@1", "label": "backlog:B1", "status": "ACTIVE", "type": "memory:strategic"}, {"id": "M:fb51b51f0b97@1", "label": "backlog:B2", "status": "ACTIVE", "type": "memory:strategic"}, {"id": "M:1a5664e020e2@1", "label": "backlog:B3", "status": "ACTIVE", "type": "memory:strategic"}, {"id": "M:8ddcc18f08d9@1", "label": "backlog:B4", "status": "ACTIVE", "type": "memory:strategic"}, {"id": "M:3c5c95f41add@1", "label": "backlog:B5", "status": "ACTIVE", "type": "memory:strategic"}, {"id": "M:d82b11953aa6@1", "label": "backlog:B6", "status": "ACTIVE", "type": "memory:strategic"}, {"id": "C:C1_mixture_hybrid", "label": "C1_mixture_hybrid", "status": "RETAIN_BASELINE", "type": "candidate"}, {"id": "C:C2_strict_switch_0.95", "label": "C2_strict_switch_0.95", "status": "RETAIN_BASELINE", "type": "candidate"}, {"id": "C:C3_lenient_evaluator", "label": "C3_lenient_evaluator", "status": "REJECTED", "type": "candidate"}, {"id": "C:C4_read_sealed_in_dev", "label": "C4_read_sealed_in_dev", "status": "REJECTED", "type": "candidate"}, {"id": "C:C5_self_signed_promotion", "label": "C5_self_signed_promotion", "status": "RETAIN_BASELINE", "type": "candidate"}, {"id": "E:E112", "label": "E112 Lab 2.0 (baseline)", "type": "evolution"}, {"id": "E:E113", "label": "E113 model-backed collective (peer; pending)", "type": "evolution"}]}