{
 "answers": [
  {
   "a": "E71 at 16d8d1673 plus a concurrent partial E72 commit set 1c928a4ca..99d5f5972; all inherited E68/E69/E70/E71 suites pass. The prior E72 docs/audit are preserved under docs/evolution-72/history and artifacts/e72/history.",
   "q": "What was inherited?"
  },
  {
   "a": "Adaptive scheduling, real task-completion model evaluation and routing, five real-source missions with independent proof, cross-host execution on two VMs, worker/DB/cold-restart recovery, a durable 1-hour long-horizon controller, held-out memory evaluation, two governed experiments, an external benchmark, measured economics, a local ClawX view, a 93-scenario security lab, a gate registry and a standalone offline verifier.",
   "q": "What was implemented?"
  },
  {
   "a": "56 real local transformer calls across SmolLM2-135M, SmolLM2-360M and Qwen2.5-0.5B; 0 remote calls. Remote providers are BLOCKED (no credential/endpoint).",
   "q": "Which models actually executed?"
  },
  {
   "a": "All five: repository engineering, defensive security, arXiv research, Iris data analysis, host reliability. 5/5 accepted from artifacts and tests.",
   "q": "Which genuine missions completed?"
  },
  {
   "a": "A verified patch + independent regression test, a security report with before/after exploit evidence, a cited research brief, a validated data analysis, a host-reliability report, five signed v5 evidence bundles, a 1-hour mission journal and a signed E72 evidence bundle.",
   "q": "What real deliverables were produced?"
  },
  {
   "a": "Scheduling tested 2..64 real processes (max busy 8 under the controller); economics ran 240 tasks at 4 and 8 workers; cross-host ran 4 workers on host A and 2 on host B; recovery killed one live worker. No million-agent claim.",
   "q": "How many workers actually executed?"
  },
  {
   "a": "1,000,000 from E71 are preserved as historical records, not processes. E72 did not re-run a logical-identity scale test.",
   "q": "How many logical identities were tested?"
  },
  {
   "a": "Every mission EXECUTE was recomputed by a different worker (cross-field where set); five mission bundles and two recovery bundles pass the E68 offline verifier; cross-host evidence verifies on both hosts; the E72 bundle verifier rechecks signatures, digests and key arithmetic.",
   "q": "What was independently verified?"
  },
  {
   "a": "Long-horizon: 0.0174 h verified productive vs 0.6078 h process-alive. Economics: 4 workers 1.24 h productive vs 1.53 h alive; 8 workers measured separately. E70's 2.03 h interval union is preserved in the reconciliation, replacing the misleading 163 h figure.",
   "q": "What were verified productive worker-hours?"
  },
  {
   "a": "Scheduling levels: 2w 0.8656, 4w 0.7438, 8w 0.376, 16w 0.236, 32w 0.0988, 64w 0.0605. Economics: 4w 0.8132 (peak PSS 25.6 MB), 8w 0.4614 (peak PSS 46.6 MB). Long-horizon is duty-cycled: 0.02863.",
   "q": "What was utilization under each workload?"
  },
  {
   "a": "Oversubscription/lease failure reproduced and fixed (naive 32 workers p95 3.478 s and 36 lease expiries at 1 s; controller p95 0.544 s and 0 lease expiries). Five real-source missions with stronger independent proof. Cross-host execution moved from BLOCKED to demonstrated. Recovery added DB-lock and cold-restart tests. A 1-hour durable mission replaced short demonstrations.",
   "q": "What improvements were demonstrated against E71?"
  },
  {
   "a": "Yes, twice. (1) SSH-tunnel/SQLite: two distinct VMs, 18 verified tasks on host A and 22 on host B, 0 duplicates. (2) PostgreSQL shared state: mission M1 completed with 12 tasks produced on one host and independently verified on the other; a 972-event chain verifies; 0 duplicates. Coordinator failover succeeded on a repeated run: host B became leader, fencing advanced 2->3, M2 completed with 0 duplicates. The first attempt failed (b_took_over=false) and is recorded. Network partition was not induced.",
   "q": "Was cross-host execution demonstrated?"
  },
  {
   "a": "Yes, on the development replica B: boot_id changed (32cacb7c... -> e245c1ff...), uptime reset to 120 s, mission M4 resumed and completed 100/100 verified tasks, and 5 tasks holding a 12 s expired grant were refused after recovery. This is one reboot, not a soak test, and it is not a production disaster-recovery proof.",
   "q": "Was actual host reboot recovery demonstrated?"
  },
  {
   "a": "One real 1-hour campaign, cycled every 20 s with a restart at the phase boundary: 172 cycles, 688 verified EXECUTE tasks, journal chain intact. 6/24/72/168-hour missions are BLOCKED.",
   "q": "Which long-running missions completed?"
  },
  {
   "a": "Worker SIGKILL: retried (max attempt 2), mission completed, 0 duplicates. SQLite exclusive lock: independent write failed closed; integrity ok. Coordinator cold restart: prior mission preserved; new mission completed. Host B reboot: boot_id changed, mission M4 resumed 100/100, expired grants refused. Coordinator failover attempt: B did NOT take over; not claimed. Provider outage: classified BLOCKED. Naive scheduling with a 1 s lease: 36 expiries, mission FAILED.",
   "q": "What happened during failure injection?"
  },
  {
   "a": "Deterministic retrieval 0/20 -> 20/20 (retrieval availability); real local model 0/6 -> 4/6 with context. Memory authority violations 0; injection quarantined; supersession correct.",
   "q": "Did memory improve held-out task performance?"
  },
  {
   "a": "Two measured gains: routing +0.25 and scheduling +0.667 on held-out criteria; history chain verifies; promotion is withheld pending owner approval. Guard tests refused a protected target, no-held-out, evaluator substitution and agent self-promotion.",
   "q": "Did governed self-improvement produce reproducible gains?"
  },
  {
   "a": "Model cost $0 (local, unbilled); remote cost BLOCKED. 4 workers: CPU 0.727 s/verified task, peak PSS 25.6 MB, 81.3% utilization. 8 workers: 0.715 s, 46.6 MB, 46.1%. Cross-host host B had 2 vCPU / ~1 GB available. All figures are measured from the process tree.",
   "q": "What were the actual costs and resource requirements?"
  },
  {
   "a": "The lab is same-project, not an audit. PostgreSQL is a single primary, not Byzantine fault tolerant, and network partition was not induced. No remote-provider path, hardware attestation, SOC 2, third-party audit or independent trust anchor. ClawX is local. The evidence key is self-generated. Heuristic detectors can miss novel defects.",
   "q": "What security weaknesses remain?"
  },
  {
   "a": "{\"BLOCKED\": 3, \"FAIL\": 0, \"PARTIAL\": 4, \"PASS\": 21}",
   "q": "Which gates are PASS, PARTIAL, BLOCKED, or FAIL?"
  },
  {
   "a": "Yes: commit 3fa6c6fb8 pushed to the established development replica, clean clone, 18 E72 tests and the standalone verifier passed; remote HEAD matched that commit. The final docs/evidence commit is one commit ahead and is pushed after this audit; see replication.json.",
   "q": "Was the remote replica independently verified?"
  }
 ],
 "branch": "evolution-72",
 "facts": {
  "benchmark": {
   "families": {
    "arithmetic": {
     "accuracy": 0.1,
     "correct": 1,
     "n": 10,
     "rows": [
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "14.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 14.694,
       "task_id": "gsm0",
       "text": "14",
       "tokens_out": 3
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "540.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 1.054,
       "task_id": "gsm1",
       "text": "1800",
       "tokens_out": 5
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "50.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 5.582,
       "task_id": "gsm2",
       "text": "To determine the percentage chance that a baby wearing a bow is wearing purple, we need to follow these steps:\n\n1.",
       "tokens_out": 24
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "220.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 1.369,
       "task_id": "gsm3",
       "text": "2240",
       "tokens_out": 5
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "64800.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 9.503,
       "task_id": "gsm4",
       "text": "To determine how much more money Fern will make by choosing the WidgetMaster 9000 over the Widgeteer",
       "tokens_out": 24
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "2.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 2.389,
       "task_id": "gsm5",
       "text": "16",
       "tokens_out": 3
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "410.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 2.464,
       "task_id": "gsm6",
       "text": "80 feet",
       "tokens_out": 4
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "5.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 7.395,
       "task_id": "gsm7",
       "text": "To determine how many candy bars George bought, we need to follow these steps:\n\n1. Calculate the total amount of money",
       "tokens_out": 24
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "77.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 2.792,
       "task_id": "gsm8",
       "text": "Elvira will have \u20ac865 left for her clothing.",
       "tokens_out": 15
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "112.0",
       "family": "arithmetic",
       "keyword_score": 0.0,
       "latency_s": 4.331,
       "task_id": "gsm9",
       "text": "To determine how much money Paul spent on driving his car over the two-week period, we need to calculate the total cost",
       "tokens_out": 24
      }
     ],
     "wilson95": [
      0.01787574951572113,
      0.4041563854975721
     ]
    },
    "citation_extract": {
     "accuracy": 0.8,
     "correct": 4,
     "n": 5,
     "rows": [
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "2611.15207",
       "family": "citation_extract",
       "keyword_score": 1.0,
       "latency_s": 5.098,
       "task_id": "cite0",
       "text": "http://arxiv.org/abs/2611.15207v1",
       "tokens_out": 21
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "2610.34634",
       "family": "citation_extract",
       "keyword_score": 1.0,
       "latency_s": 8.721,
       "task_id": "cite1",
       "text": "http://arxiv.org/abs/2610.34634v1",
       "tokens_out": 21
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "2610.33633",
       "family": "citation_extract",
       "keyword_score": 1.0,
       "latency_s": 6.583,
       "task_id": "cite2",
       "text": "http://arxiv.org/abs/2610.33633v1",
       "tokens_out": 21
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "2610.31130",
       "family": "citation_extract",
       "keyword_score": 0.5,
       "latency_s": 3.9,
       "task_id": "cite3",
       "text": "http://arxiv.org/abs/2610.29722v1",
       "tokens_out": 21
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "2610.29722",
       "family": "citation_extract",
       "keyword_score": 1.0,
       "latency_s": 3.295,
       "task_id": "cite4",
       "text": "http://arxiv.org/abs/2610.29722v1",
       "tokens_out": 21
      }
     ],
     "wilson95": [
      0.3755282641185388,
      0.9637768390302125
     ]
    },
    "data_analysis_deterministic": {
     "accuracy": 1.0,
     "correct": 5,
     "method": "exact recomputation over the public CSV",
     "n": 5,
     "wilson95": [
      0.5655085052479191,
      1.0
     ]
    },
    "data_analysis_model": {
     "accuracy": 0.2,
     "correct": 1,
     "n": 5,
     "rows": [
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "150.0",
       "family": "data_analysis_model",
       "keyword_score": 0.0,
       "latency_s": 2.018,
       "task_id": "data0",
       "text": "100000",
       "tokens_out": 7
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "3.0",
       "family": "data_analysis_model",
       "keyword_score": 0.0,
       "latency_s": 0.925,
       "task_id": "data1",
       "text": "10",
       "tokens_out": 3
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "5.84",
       "family": "data_analysis_model",
       "keyword_score": 0.0,
       "latency_s": 1.12,
       "task_id": "data2",
       "text": "5.84",
       "tokens_out": 5
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "4.4",
       "family": "data_analysis_model",
       "keyword_score": 0.0,
       "latency_s": 2.632,
       "task_id": "data3",
       "text": "5.1",
       "tokens_out": 4
      },
      {
       "correct": false,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "1.0",
       "family": "data_analysis_model",
       "keyword_score": 0.0,
       "latency_s": 0.848,
       "task_id": "data4",
       "text": "3",
       "tokens_out": 2
      }
     ],
     "wilson95": [
      0.03622316096978745,
      0.6244717358814613
     ]
    },
    "memory_retrieval": {
     "accuracy": 1.0,
     "correct": 20,
     "method": "deterministic retrieval over verified memory",
     "n": 20,
     "wilson95": [
      0.8389,
      1.0
     ]
    },
    "planning": {
     "accuracy": 1.0,
     "correct": 5,
     "n": 5,
     "rows": [
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "design",
       "family": "planning",
       "keyword_score": 1.0,
       "latency_s": 0.61,
       "task_id": "plan0",
       "text": "Design",
       "tokens_out": 2
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "test",
       "family": "planning",
       "keyword_score": 1.0,
       "latency_s": 0.553,
       "task_id": "plan1",
       "text": "Test",
       "tokens_out": 2
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "deploy",
       "family": "planning",
       "keyword_score": 1.0,
       "latency_s": 0.869,
       "task_id": "plan2",
       "text": "Deploy",
       "tokens_out": 2
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "test",
       "family": "planning",
       "keyword_score": 1.0,
       "latency_s": 0.918,
       "task_id": "plan3",
       "text": "Test",
       "tokens_out": 2
      },
      {
       "correct": true,
       "error": null,
       "execution_class": "REAL_LOCAL_MODEL",
       "expected": "design",
       "family": "planning",
       "keyword_score": 1.0,
       "latency_s": 1.589,
       "task_id": "plan4",
       "text": "Design",
       "tokens_out": 2
      }
     ],
     "wilson95": [
      0.5655085052479191,
      1.0
     ]
    },
    "repository_defect": {
     "accuracy": 1.0,
     "clean_false_positives": 0,
     "correct": 5,
     "method": "deterministic AST+mutation detector",
     "n": 5,
     "rows": [
      {
       "detected": true,
       "findings": [
        "TRUNCATED_DIGEST"
       ],
       "mutant": 0,
       "test_fails_before": true,
       "test_passes_after": true
      },
      {
       "detected": true,
       "findings": [
        "WEAK_HASH"
       ],
       "mutant": 1,
       "test_fails_before": true,
       "test_passes_after": true
      },
      {
       "detected": true,
       "findings": [
        "UNREACHABLE_RETURN"
       ],
       "mutant": 2,
       "test_fails_before": true,
       "test_passes_after": true
      },
      {
       "detected": true,
       "findings": [
        "TYPE_ERROR"
       ],
       "mutant": 3,
       "test_fails_before": true,
       "test_passes_after": true
      },
      {
       "detected": true,
       "findings": [
        "BYTES_NOT_HEX"
       ],
       "mutant": 4,
       "test_fails_before": true,
       "test_passes_after": true
      }
     ],
     "wilson95": [
      0.5655085052479191,
      1.0
     ]
    },
    "test_generation": {
     "accuracy": 1.0,
     "correct": 5,
     "method": "detector-authored regression test + before/after execution",
     "n": 5,
     "wilson95": [
      0.5655085052479191,
      1.0
     ]
    }
   },
   "model": {
    "calls": 25,
    "exact_accuracy": 0.44,
    "execution_class": "REAL_LOCAL_MODEL",
    "keyword_overstatement": -0.06,
    "latency_p50_s": 2.464,
    "mean_keyword_score": 0.38,
    "name": "qwen2.5-0.5b",
    "params_m": 494.0,
    "tokens_out": 267,
    "wilson95": [
     0.2666537494700256,
     0.6293298298043557
    ]
   }
  },
  "branch": "evolution-72",
  "clawx": {
   "deployed": false,
   "tenant_isolation": true
  },
  "cross_host": {
   "failover": "PASS (repeated run; first attempt failed)",
   "partition": "BLOCKED",
   "postgres_demo": {
    "boot_id_changed": true,
    "chain": {
     "bad_sequences": [],
     "events": 155,
     "head": "781a741e32f16184cc35e6d3bd43e360930ac6788ccfc584ae52b40b2ad17e95",
     "ok": true
    },
    "duplicate_results": 0,
    "failover_demonstrated": true,
    "m1_cross_host_verified": 8,
    "m1_state": "COMPLETED",
    "reboot_mission_tasks": 16,
    "reboot_mission_verified": 16,
    "uptime_reset": true
   },
   "ssh_demo": {
    "duplicate_results": 0,
    "evidence_verified": true,
    "mission_state": "COMPLETED",
    "verified_execute_by_host": {
     "host-a": 18,
     "host-b": 22
    }
   }
  },
  "e70_reconciliation": {
   "alternative_definitions": {
    "all_workers_created_to_stop": 163.894,
    "busy_union_per_worker_hours": 2.0293,
    "final_alive_at_measurement": 163.3935,
    "observed_process_lifetime": 163.6772,
    "verified_productive_hours": 2.0293,
    "wall_union_busy_hours": 0.026
   },
   "delta_matches_excluded_workers": true,
   "detail": {
    "busy_seconds_sum_by_task": 7305.45,
    "busy_seconds_union_per_worker": 7305.45,
    "busy_seconds_wall_union": 93.73,
    "definitions": {
     "busy_seconds_union_per_worker": "per-worker interval union of verified EXECUTE+VERIFY tasks",
     "verified_productive_seconds": "sum of EXECUTE+VERIFY intervals of VERIFIED tasks in COMPLETED missions",
     "worker_hours_all": "sum(stopped_at - created_at) for every worker row",
     "worker_hours_final_alive": "E70 definition: sum(stopped_at - created_at) for workers whose final state is not DEAD/RETIRED/REVOKED"
    },
    "elapsed_s": 602.0,
    "excluded_worker_hours": 0.5005,
    "execution_task_seconds": 3676.05,
    "idle_seconds": 580911.0,
    "idle_seconds_definition": "worker_hours_final_alive - busy_seconds_union_per_worker",
    "idle_share": 0.98758,
    "schema": "cain72.accounting.recompute.v1",
    "source_db": "/tmp/opencode/e70-endurance-1000-1794611918/coord.db",
    "started_at": "2026-11-13T23:18:38Z",
    "stopped_at": "2026-11-13T23:28:40Z",
    "task_p50_s": 3.542,
    "tasks_per_worker_max": 3,
    "tasks_per_worker_p50": 1.0,
    "verification_task_seconds": 3629.4,
    "verified_execute_tasks": 1000,
    "verified_productive_seconds": 7305.45,
    "verified_verify_tasks": 1000,
    "worker_hours_all": 163.894,
    "worker_hours_final_alive": 163.3935,
    "worker_hours_observed_process": 163.6772,
    "workers_excluded_final_state": [
     {
      "seconds": 600.64,
      "state": "DEAD",
      "worker_id": "w-4a8848cae6"
     },
     {
      "seconds": 600.64,
      "state": "DEAD",
      "worker_id": "w-198783d9a5"
     },
     {
      "seconds": 600.64,
      "state": "DEAD",
      "worker_id": "w-5004aea0bc"
     }
    ],
    "workers_final_alive": 997,
    "workers_total": 1000
   },
   "e71_reported_hours": 163.894,
   "excluded_workers": [
    {
     "seconds": 600.64,
     "state": "DEAD",
     "worker_id": "w-4a8848cae6"
    },
    {
     "seconds": 600.64,
     "state": "DEAD",
     "worker_id": "w-198783d9a5"
    },
    {
     "seconds": 600.64,
     "state": "DEAD",
     "worker_id": "w-5004aea0bc"
    }
   ],
   "explanation": "E70 summed created_at->stop for workers still alive when it measured (997 rows); the E71 audit summed every worker row (1,000), including the 3 workers killed and replaced mid-run. The delta is exactly those replacement rows. Counting each process only until its observed ended_at gives the third figure, which is the most defensible process-alive measure. All three are arithmetically correct under their stated definitions; none measures productive work.",
   "note": "No original value is overwritten. The reconciliation is a computation over raw rows.",
   "original_e70_hours": 163.4645,
   "raw_db": "/tmp/opencode/e70-endurance-1000-1794611918/coord.db",
   "recomputed_all_workers_hours": 163.894,
   "recomputed_delta_hours": 0.5005,
   "recomputed_final_alive_hours": 163.3935,
   "recomputed_observed_process_hours": 163.6772,
   "schema": "cain72.e70_reconciliation.v1",
   "status": "RECONCILED",
   "summary_stopped_at": "2026-11-13T23:28:40Z",
   "timestamp_precision_uncertainty_hours": 0.1
  },
  "economics": {
   "cpu_seconds_per_verified_task": {
    "4": 0.7273,
    "8": 0.7153
   },
   "interpretation": "More workers did not reduce CPU seconds per verified task; they raised peak memory and coordinator overhead. For CPU-bound verified work the smallest team that keeps latency inside the lease is the economical choice.",
   "peak_pss_mb": {
    "4": 25.6,
    "8": 46.6
   },
   "task_p95_s": {
    "4": 0.448,
    "8": 0.471
   },
   "throughput_tasks_per_min": {
    "4": 254.4,
    "8": 283.9
   }
  },
  "final_head": "3fa6c6fb826e6a51fa74cb7af51caaaaefb6c7d3",
  "gates": {
   "BLOCKED": 3,
   "FAIL": 0,
   "PARTIAL": 4,
   "PASS": 21
  },
  "inherited": {
   "concurrent_e72_commits": "1c928a4ca..99d5f5972 (prior partial E72, preserved)",
   "e68_worker_hours": 1018.29,
   "e70_worker_hours_original": 163.4645,
   "e71_verified_productive_hours": 5.3181,
   "e71_worker_hours": 6.65,
   "history_branch_start": "16d8d1673 (E71 replication)"
  },
  "long_horizon": {
   "actual_elapsed_seconds": 3618.66,
   "campaign_mode": "DUTY_CYCLED",
   "cycles": 172,
   "determination": "DEADLINE_REACHED",
   "failed_tasks": 0,
   "journal": {
    "head": "ccf8393c295167e6fec7c1ff21130249d1f1a9ed7c3998ce753108b09f6bd103",
    "n": 196,
    "ok": true
   },
   "scheduled_seconds": 3600.0,
   "state": "COMPLETED",
   "utilization": 0.02863,
   "verified_execute_tasks": 688,
   "verified_productive_hours": 0.0174,
   "worker_hours_process_alive": 0.6078
  },
  "memory": {
   "arms": {
    "A_no_memory": {
     "accuracy": 0.0,
     "correct": 0,
     "wilson95": [
      0.0,
      0.1611
     ]
    },
    "B_verified_memory": {
     "accuracy": 1.0,
     "correct": 20,
     "wilson95": [
      0.8389,
      1.0
     ]
    }
   },
   "authority_violations": 0,
   "injection": {
    "memory_id": "mem_4a6be99c106f4210",
    "quarantine_status": "QUARANTINED",
    "recalled": false
   },
   "model_arm": {
    "model": "smollm2-360m",
    "no_context": {
     "correct": 0,
     "rows": [
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 22.801,
       "question": "How low may the E72 adaptive scheduler's soft cap fall?",
       "text": "The soft cap of the E72 adaptive scheduler can be lowered to a maximum of 100% if necessary",
       "tokens_out": 24
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 4.481,
       "question": "Which field must a cross-field verifier come from?",
       "text": "A cross-field verifier must come from a field that is not only cross-functional but also has a strong foundation",
       "tokens_out": 24
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 1.483,
       "question": "What state does a task keep when resource admission refuses it?",
       "text": "A task keeps when resource admission refuses it.",
       "tokens_out": 10
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 1.65,
       "question": "How many times may a coordinator ticket be used?",
       "text": "A coordinator ticket can be used only once.",
       "tokens_out": 10
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 3.43,
       "question": "Can PBFT consensus authorize an action?",
       "text": "Yes, the Proof-of-Bounded-Transactions (PBFT) consensus algorithm can authorize an action.",
       "tokens_out": 24
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 4.081,
       "question": "Which signature scheme identifies a worker process?",
       "text": "The signature scheme identifies a worker process is the Digital Signature Algorithm (DSA).",
       "tokens_out": 18
      }
     ]
    },
    "questions": 6,
    "with_context": {
     "correct": 4,
     "rows": [
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": true,
       "latency_s": 3.625,
       "question": "How low may the E72 adaptive scheduler's soft cap fall?",
       "text": "The E72 adaptive scheduler may never compute a soft cap below one.",
       "tokens_out": 16
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": true,
       "latency_s": 2.783,
       "question": "Which field must a cross-field verifier come from?",
       "text": "Cross-field verifier must come from a different field than the producer.",
       "tokens_out": 16
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 1.93,
       "question": "What state does a task keep when resource admission refuses it?",
       "text": "A task keeps its state when resource admission refuses it.",
       "tokens_out": 12
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": false,
       "latency_s": 0.807,
       "question": "How many times may a coordinator ticket be used?",
       "text": "One time.",
       "tokens_out": 4
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": true,
       "latency_s": 2.18,
       "question": "Can PBFT consensus authorize an action?",
       "text": "No, PBFT consensus cannot authorize an action.",
       "tokens_out": 11
      },
      {
       "class": "REAL_LOCAL_MODEL",
       "correct": true,
       "latency_s": 4.626,
       "question": "Which signature scheme identifies a worker process?",
       "text": "A worker process identifies itself using the Ed25519 signature scheme.",
       "tokens_out": 17
      }
     ]
    }
   }
  },
  "missions": {
   "data_iris": {
    "acceptance": true,
    "source": "https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv"
   },
   "infra_reliability": {
    "acceptance": true,
    "source": null
   },
   "repo_software": {
    "acceptance": true,
    "source": "/root/cain-studio-platform"
   },
   "research_literature": {
    "acceptance": true,
    "source": null
   },
   "security_review": {
    "acceptance": true,
    "source": "isolated two-function Python target (no listener, no network, no production code)"
   }
  },
  "models": {
   "execution_classes": {
    "REAL_LOCAL_MODEL": 56
   },
   "heldout": {
    "qwen2.5-0.5b": {
     "accuracy": 0.5,
     "correct": 4,
     "n": 8
    },
    "smollm2-135m": {
     "accuracy": 0.0,
     "correct": 0,
     "n": 8
    },
    "smollm2-360m": {
     "accuracy": 0.125,
     "correct": 1,
     "n": 8
    }
   },
   "real_local_calls": 56,
   "real_remote_calls": 0,
   "routing": {
    "P0_fixed_best_model": {
     "accuracy": 0.5,
     "model": "qwen2.5-0.5b"
    },
    "P1_family_no_minimum": {
     "accuracy": 0.25
    },
    "P2_calibrated_min2": {
     "accuracy": 0.5,
     "rows": [
      {
       "basis": "family accuracy with >=2 samples",
       "correct": false,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "arithmetic",
       "latency_s": 16.2414,
       "model": "qwen2.5-0.5b",
       "task_id": "h1",
       "text": "92"
      },
      {
       "basis": "family accuracy with >=2 samples",
       "correct": false,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "arithmetic",
       "latency_s": 0.5849,
       "model": "qwen2.5-0.5b",
       "task_id": "h2",
       "text": "42"
      },
      {
       "basis": "family accuracy with >=2 samples",
       "correct": false,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "classification",
       "latency_s": 0.8638,
       "model": "qwen2.5-0.5b",
       "task_id": "h3",
       "text": "Anomalous Login"
      },
      {
       "basis": "family accuracy with >=2 samples",
       "correct": false,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "classification",
       "latency_s": 0.6398,
       "model": "qwen2.5-0.5b",
       "task_id": "h4",
       "text": "Security"
      },
      {
       "basis": "overall measured accuracy (family has too few samples)",
       "correct": true,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "extraction",
       "latency_s": 6.4394,
       "model": "qwen2.5-0.5b",
       "task_id": "h5",
       "text": "```json\n{\n  \"tools\": [\n    \"curl\",\n    \"awk\"\n  ]\n}\n```"
      },
      {
       "basis": "overall measured accuracy (family has too few samples)",
       "correct": true,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "reasoning",
       "latency_s": 2.1681,
       "model": "qwen2.5-0.5b",
       "task_id": "h6",
       "text": "False."
      },
      {
       "basis": "overall measured accuracy (family has too few samples)",
       "correct": true,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "instruction",
       "latency_s": 2.2054,
       "model": "qwen2.5-0.5b",
       "task_id": "h7",
       "text": "DONE"
      },
      {
       "basis": "overall measured accuracy (family has too few samples)",
       "correct": true,
       "execution_class": "REAL_LOCAL_MODEL",
       "family": "code",
       "latency_s": 2.3213,
       "model": "qwen2.5-0.5b",
       "task_id": "h8",
       "text": "4"
      }
     ],
     "wilson95": [
      0.2152,
      0.7848
     ]
    }
   }
  },
  "recovery": {
   "cold_restart": {
    "evidence_verified": true,
    "missions_after_restart": 1,
    "new_mission_state": "COMPLETED",
    "note": "coordinator process restarted on the same SQLite root; this is not a machine reboot",
    "prior_mission_state": "COMPLETED"
   },
   "database_lock": {
    "error": "database is locked",
    "failed_closed": true,
    "integrity_after_release": "ok"
   },
   "host_reboot_demonstrated_by_postgres_demo": true,
   "ssh_demo_host_reboot": "BLOCKED",
   "status": "PASS",
   "worker_kill": {
    "duplicate_results": 0,
    "evidence_verified": true,
    "killed": {
     "pid": 266729,
     "worker_id": "w-079c8afece"
    },
    "max_attempt": 2,
    "mission_state": "COMPLETED",
    "problems": [],
    "tasks_total": 96,
    "verified": 96
   }
  },
  "replica": {
   "at": 1794631673.7122824,
   "branch": "evolution-72",
   "clean_clone": true,
   "head_matches": true,
   "host": "root@45.76.169.191",
   "key_present": true,
   "local_head": "3fa6c6fb826e6a51fa74cb7af51caaaaefb6c7d3",
   "note": "The replica is a separate VM in the same administrative domain, not an independent trust anchor. No production service, DNS, Stripe or customer data was touched.",
   "ok": true,
   "remote_head": "3fa6c6fb826e6a51fa74cb7af51caaaaefb6c7d3",
   "remote_ref": "3fa6c6fb826e6a51fa74cb7af51caaaaefb6c7d3",
   "remote_tests_ok": true,
   "remote_verify_gate_counts": {
    "BLOCKED": 3,
    "FAIL": 0,
    "PARTIAL": 5,
    "PASS": 20
   },
   "remote_verify_ok": true,
   "schema": "cain72.replication.v1",
   "steps": {
    "clone": {
     "cmd": "ssh -i /root/.ssh/cain42_cluster -o IdentitiesOnly=yes -o BatchMode=yes -o StrictHostKeyChecking=accept-new -o ConnectTimeout=10 root@45.76.169.191 rm -rf /root/e72-verify-3fa6c6f && git clone --quiet --branch evolution-72 /root/cain-repo.git /root/e72-verify-3fa6c6f && cd /root/e72-verify-3fa6c6f && git rev-parse HEAD",
     "code": 0,
     "stderr": "",
     "stdout": "3fa6c6fb826e6a51fa74cb7af51caaaaefb6c7d3\n"
    },
    "ls_remote": {
     "cmd": "git ls-remote root@45.76.169.191:/root/cain-repo.git refs/heads/evolution-72",
     "code": 0,
     "stderr": "",
     "stdout": "3fa6c6fb826e6a51fa74cb7af51caaaaefb6c7d3\trefs/heads/evolution-72\n"
    },
    "push": {
     "cmd": "git push root@45.76.169.191:/root/cain-repo.git evolution-72:evolution-72",
     "code": 0,
     "stderr": "To 45.76.169.191:/root/cain-repo.git\n   486508ef2..3fa6c6fb8  evolution-72 -> evolution-72\n",
     "stdout": ""
    },
    "tests": {
     "cmd": "ssh -i /root/.ssh/cain42_cluster -o IdentitiesOnly=yes -o BatchMode=yes -o StrictHostKeyChecking=accept-new -o ConnectTimeout=10 root@45.76.169.191 cd /root/e72-verify-3fa6c6f && python3 -m pytest tests/test_e72_evolution.py -q --no-header",
     "code": 0,
     "stderr": "",
     "stdout": "============================= test session starts ==============================\ncollected 18 items\n\ntests/test_e72_evolution.py ..................                           [100%]\n\n============================== 18 passed in 1.69s ==============================\n"
    },
    "verifier": {
     "cmd": "ssh -i /root/.ssh/cain42_cluster -o IdentitiesOnly=yes -o BatchMode=yes -o StrictHostKeyChecking=accept-new -o ConnectTimeout=10 root@45.76.169.191 cd /root/e72-verify-3fa6c6f && python3 scripts/evolution72/verify_e72_offline.py artifacts/e72/bundle --json",
     "code": 0,
     "stderr": "",
     "stdout": "{\n \"ok\": true,\n \"schema\": \"cain72.evidence.bundle.v1\",\n \"bundle\": \"/root/e72-verify-3fa6c6f/artifacts/e72/bundle\",\n \"pinned_fingerprint\": \"fb1678b2f535fdc3811ca3544422bde6e1918ccffc36c403e54b893dec500843\",\n \"git_head\": \"486508ef2635d87972a94a05139697bcbcf8c132\",\n \"branch\": \"evolution-72\",\n \"checks\": [\n  \"manifest_signature\",\n  \"files_checked:551\",\n  \"gates_checked:28\",\n  \"scheduling_hard_cap_respected\",\n  \"scheduling_zero_lease_expiries\",\n  \"naive_lease_failure_preserved\",\n  \"scheduling_missions_completed\",\n  \"no_remote_model_call\",\n  \"local_execution_only\",\n  \"fallback_recovered\",\n  \"memory_no_authority\",\n  \"memory_injection_not_recalled\",\n  \"memory_heldout_improvement\",\n  \"cross_host_distinct\",\n  \"cross_host_both_executed\",\n  \"cross_host_no_duplicates\",\n  \"host_reboot_blocked\",\n  \"worker_kill_recovered\",\n  \"db_lock_fails_closed\",\n  \"productive_le_alive:4\",\n  \"productive_le_alive:8\",\n  \"long_horizon_productive_le_alive\",\n  \"no_multiday_claim\",\n  \"long_horizon_cycles_present\",\n  \"benchmark_denominator:arithmetic\",\n  \"benchmark_denominator:citation_extract\",\n  \"benchmark_denominator:data_analysis_deterministic\",\n  \"benchmark_denominator:data_analysis_model\",\n  \"benchmark_denominator:memory_retrieval\",\n  \"benchmark_denominator:planning\",\n  \"benchmark_denominator:repository_defect\",\n  \"benchmark_denominator:test_generation\",\n  \"e70_all_workers_recomputed\",\n  \"e70_delta_explained\",\n  \"experiment_history_chain\",\n  \"no_self_promotion\",\n  \"security_lab_all_blocked\",\n  \"crosshost_m1_independent\",\n  \"crosshost_m1_all_verified\",\n  \"crosshost_failover_completed\",\n  \"crosshost_fencing_monotonic\",\n  \"host_reboot_real\",\n  \"reboot_mission_completed\",\n  \"expired_authority_refused\",\n  \"crosshost_event_chain\",\n  \"crosshost_no_duplicate_results\",\n  \"stale_fencing_rejected\",\n  \"benchmark_three_models\",\n  \"benchmark_denominator:qwen2.5-0.5b\",\n  \"benchmark_denominator:smollm2-135m\",\n  \"benchmark_denominator:smollm2-360m\",\n  \"benchmark_tool_assisted_ci\",\n  \"memory_v2_improvement\",\n  \"memory_v2_no_authority\",\n  \"mission_bundle:data_iris\",\n  \"mission_bundle:infra_reliability\",\n  \"mission_bundle:repo_software\",\n  \"mission_bundle:research_literature\",\n  \"mission_bundle:security_review\",\n  \"mission_bundles:5\"\n ],\n \"problems\": [],\n \"files\": 551,\n \"gates\": 28,\n \"gate_counts\": {\n  \"PASS\": 20,\n  \"PARTIAL\": 5,\n  \"BLOCKED\": 3,\n  \"FAIL\": 0\n },\n \"note\": \"A self-generated key proves internal consistency and tamper detection, not a third-party identity. Pin the fingerprint out of band for that use.\"\n}\n"
    }
   }
  },
  "scheduling": {
   "levels": 6,
   "max_busy_64": 8,
   "naive_lease_failures": 36,
   "p95_controller_32": 0.544,
   "p95_naive_32": 3.478
  },
  "security": {
   "blocked": 103,
   "bypassed": 0,
   "scenarios": 103
  },
  "self_improvement": {
   "experiments": [
    {
     "approval": "AWAITING_OWNER_APPROVAL",
     "baseline": 0.25,
     "candidate": 0.5,
     "change": "CalibratedRouter(min_samples=2) replaces family-only routing",
     "delta": 0.25,
     "experiment_id": "exp-e6711e9ef7206a",
     "heldout_tasks": 8,
     "hypothesis": "Requiring at least two observations per task family before specialising, and otherwise falling back to overall measured accuracy, improves held-out accuracy over unspecialised family routing on small calibration sets.",
     "status": "MEASURED",
     "target": "model_routing"
    },
    {
     "approval": "AWAITING_OWNER_APPROVAL",
     "baseline": 0.3333333333333333,
     "candidate": 1.0,
     "change": "AdaptiveResourceController soft/hard caps + lease-aware refusal",
     "delta": 0.6666666666666667,
     "experiment_id": "exp-68f3d9a27e1369",
     "heldout_tasks": 3,
     "hypothesis": "Resource-aware admission sized from measured task cost prevents lease expiry and mission failure under a 1 s lease, without changing the definition of productive work.",
     "status": "MEASURED",
     "target": "scheduling_admission"
    }
   ],
   "history": {
    "experiments": 4,
    "ok": true,
    "problems": []
   },
   "refusals": [
    {
     "case": "protected_target",
     "error": "EvolutionError: target 'authorization' is a protected governance component",
     "refused": true
    },
    {
     "case": "no_heldout",
     "error": "EvolutionError: held-out tasks required",
     "refused": true
    },
    {
     "case": "evaluator_substitution",
     "error": "EvolutionError: evaluator changed; self-referential evaluation refused",
     "refused": true
    },
    {
     "case": "self_promotion",
     "error": "an external approver is required; agent identities cannot promote",
     "refused": true
    }
   ]
  },
  "verified_deliverables": 5
 },
 "generated_at": 1794631704.6585536,
 "head": "3fa6c6fb826e6a51fa74cb7af51caaaaefb6c7d3",
 "measured_vs_interpreted": {
  "future_targets": [
   "6/24/72/168 h missions",
   "500/1,000/5,000/10,000 verified productive hours",
   "network partition tolerance",
   "repeated reboot/soak recovery",
   "remote model providers",
   "third-party audit"
  ],
  "interpreted": [
   "the controller is more economical for CPU-bound work at the smallest capable width",
   "routing P2 is better than P1 but not better than the fixed strongest model",
   "the long-horizon campaign demonstrates persistence, not throughput"
  ],
  "measured": [
   "model calls and classifications",
   "mission acceptance from tests/artifacts",
   "cross-host task counts",
   "recovery outcomes",
   "benchmark denominators",
   "CPU seconds, PSS, utilization",
   "journal head",
   "gate statuses"
  ]
 },
 "not_claimed": [
  "production readiness",
  "third-party audit",
  "network partition tolerance",
  "remote model capability",
  "general intelligence",
  "10,000 productive hours",
  "hardware attestation",
  "multi-reboot soak"
 ],
 "schema": "cain72.final_audit.v1"
}