{
 "schema": "cain.e93.claim_register.v1",
 "generated_at": "2026-11-18T13:24:20Z",
 "commit": "c522b17215a564a47c66dd00bf6c7f9dbe7cb127",
 "designation": "Not VAGI. Current outcome: 3 (governed agentic intelligence infrastructure). Outcome 3 is E93's maximum under the pre-registered rule (owner decision R1).",
 "claims": [
  {
   "id": "E93-C01",
   "claim": "E92's procedure-synthesis result does not show generation: a no-search template-selection arm ties it 120/120, and `return True` passes staged_repair 40/40.",
   "status": "REPRODUCED",
   "evidence": [
    {
     "path": "artifacts/e93/audit/e92_reproduction/reproduction.json",
     "sha256": "65865a373f2b69d3525d20e39be20ad9745445c3110a4101cf9e218a2cd8a3a7"
    }
   ],
   "verifier_checks": null,
   "recomputed_by": "S66 (re-run of E92 code); S92 found the constant-True flaw",
   "limits": "E92 withdrawn publicly 2026-11-18 (5d3513ac2)"
  },
  {
   "id": "E93-C02",
   "claim": "Open, family-agnostic program search produced procedures that beat designer-supplied selection on 510 held-out tasks: G2 369 vs G 200 (seed 9301), 367 vs 223 (new families, seed 9302); Holm-significant vs D, F and G.",
   "status": "DEMONSTRATED_INTERNAL",
   "evidence": [
    {
     "path": "artifacts/e93/synthesis/main-seed-9301/rows.jsonl",
     "sha256": "6d61e15de20bc4110a1e28b046d5d414d53f19d2333cd4aa6982b0e379eb5a23"
    },
    {
     "path": "artifacts/e93/synthesis/main-seed-9301/results.json",
     "sha256": "c555121d434f7c24538e33ed0537696b5a4a78b248604f8fbd204702d7fb879a"
    },
    {
     "path": "artifacts/e93/synthesis/replication-seed-9302/rows.jsonl",
     "sha256": "deb5ed38e7d486fa07c28da963b3448773eaf31ff5b71adbacbcf316b07ae484"
    },
    {
     "path": "artifacts/e93/synthesis/replication-seed-9302/results.json",
     "sha256": "1deab9fe189e4e1e971521ae38b26379a79a113f62154dda37cc8963f897eb43"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'synthesis'",
   "recomputed_by": "S66 from raw rows",
   "limits": "internal benchmark; in-distribution targets; same-team authorship; no external review; ~20x selection's dev compute"
  },
  {
   "id": "E93-C03",
   "claim": "At matched dev compute, synthesis is no better than selection (G2m 208 vs G 200; 212 vs 223).",
   "status": "NEGATIVE_RESULT",
   "evidence": [
    {
     "path": "artifacts/e93/synthesis/main-seed-9301/results.json",
     "sha256": "c555121d434f7c24538e33ed0537696b5a4a78b248604f8fbd204702d7fb879a"
    },
    {
     "path": "artifacts/e93/synthesis/replication-seed-9302/results.json",
     "sha256": "1deab9fe189e4e1e971521ae38b26379a79a113f62154dda37cc8963f897eb43"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'synthesis'",
   "recomputed_by": "S66 from raw rows",
   "limits": "internal benchmark; in-distribution targets; same-team authorship; no external review"
  },
  {
   "id": "E93-C04",
   "claim": "12 + 12 procedures promoted; 12/12 syntactically new and 9 / 10 behaviourally new versus every repertoire instantiation.",
   "status": "DEMONSTRATED_INTERNAL",
   "evidence": [
    {
     "path": "artifacts/e93/synthesis/main-seed-9301/procedures.json",
     "sha256": "e2af2627d99fa441b13e2a7dedc1e3defb35f9af1e1787b736a1862bddf58a8b"
    },
    {
     "path": "artifacts/e93/synthesis/replication-seed-9302/procedures.json",
     "sha256": "c7facddf58d5e1d093a6218df7f113ad8f2a03bdb4fbf4db1ed66002b00494b4"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'synthesis'",
   "recomputed_by": "S66 from raw per-family flags (verifier)",
   "limits": "internal benchmark; in-distribution targets; same-team authorship; no external review; automated same-process approver (a different identity than the synthesizer)"
  },
  {
   "id": "E93-C05",
   "claim": "Held-out accuracy rises with verified experience: k=0 to k=60 verified dev tasks gives 87 to 369 and 114 to 367 of 510, monotone, McNemar p 2.6e-85 / 1.4e-76.",
   "status": "DEMONSTRATED_INTERNAL",
   "evidence": [
    {
     "path": "artifacts/e93/learning/curve-seed-9301/rows.jsonl",
     "sha256": "ac8702332425e616d58591cdcc234584ee62816fd32f06af921b48b54c7f9c41"
    },
    {
     "path": "artifacts/e93/learning/curve-seed-9302/rows.jsonl",
     "sha256": "7b8ac5e54711c9b8c82ecfdb4618e503ad95e65b5b8773f610d7415d83bd58f7"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'learning'",
   "recomputed_by": "S66 from raw rows",
   "limits": "internal benchmark; in-distribution targets; same-team authorship; no external review"
  },
  {
   "id": "E93-C06",
   "claim": "A pre-registered adaptive rule (selection first, synthesis only if needed) scores 390 and 418 of 510, beating synthesis alone (p 7.5e-4, 1.6e-11) at 87% / 99% of its dev cost.",
   "status": "DEMONSTRATED_INTERNAL",
   "evidence": [
    {
     "path": "artifacts/e93/synthesis/main-seed-9301/secondary.json",
     "sha256": "a1e568f8a1295d655c5cb195b1fa593ae556b2c33d223d754ca91f292dd0c28f"
    },
    {
     "path": "artifacts/e93/synthesis/replication-seed-9302/secondary.json",
     "sha256": "630edf4b02ea0d3759c5556e1c2fe4199e08d1cb0e680351aae66c492e311b8a"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'adaptive'",
   "recomputed_by": "S66 from raw rows",
   "limits": "internal benchmark; in-distribution targets; same-team authorship; no external review; selection always runs first and its cost always counts"
  },
  {
   "id": "E93-C07",
   "claim": "Cross-family transfer: 0/80 and 0/74 gates passed (no positive transfer); the gate blocked all 15 and 34 harmful transfers.",
   "status": "NEGATIVE_RESULT",
   "evidence": [
    {
     "path": "artifacts/e93/synthesis/main-seed-9301/secondary.json",
     "sha256": "a1e568f8a1295d655c5cb195b1fa593ae556b2c33d223d754ca91f292dd0c28f"
    },
    {
     "path": "artifacts/e93/synthesis/replication-seed-9302/secondary.json",
     "sha256": "630edf4b02ea0d3759c5556e1c2fe4199e08d1cb0e680351aae66c492e311b8a"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'transfer'",
   "recomputed_by": "S66 from detail rows",
   "limits": "within one synthetic domain; cross-domain transfer not run"
  },
  {
   "id": "E93-C08",
   "claim": "Local-model proposals: 34 Qwen2.5-1.5B calls, 4 parsed, 0 passed validation.",
   "status": "NEGATIVE_RESULT",
   "evidence": [
    {
     "path": "artifacts/e93/synthesis/model-proposals-9301/model_proposals.jsonl",
     "sha256": "4c65c3e9b9689bccfa264a8e5b3f670522587c3a880315acb7adee8d13f0513d"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'model proposals'",
   "recomputed_by": "S66 from raw rows",
   "limits": "one small local model; no paid models (owner decision)"
  },
  {
   "id": "E93-C09",
   "claim": "Software vertical: a learned repair ranking (L_30) fixes 16/30 held-out generated repositories vs 3/30 for the fixed baseline (McNemar 15 vs 2, p=0.00235, pre-registered); vs no-learning L_0 6/30 (p=0.00635, exploratory).",
   "status": "DEMONSTRATED_INTERNAL",
   "evidence": [
    {
     "path": "artifacts/e93/learning/software/rows.jsonl",
     "sha256": "04ace0419319099ca38c6772dbff9d67aa79cc072dacd02fba98ec8e92e1034e"
    },
    {
     "path": "artifacts/e93/verticals/manifest.json",
     "sha256": "03a398266fd99ca8f4272d6b1bf764cc6937ee699055b6a6172551c8edbe293a"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'software learning'",
   "recomputed_by": "S66 from raw rows",
   "limits": "53% absolute, below the 70%/60% thresholds; ranking learned over the baseline's own candidate generator; saturates by k=5; not procedure synthesis"
  },
  {
   "id": "E93-C10",
   "claim": "Hidden-test checkers cannot be satisfied for free: a bypass found by S92 was fixed before any arm saw software held-out (expected values never enter the sandbox; no imports/dunder in patches).",
   "status": "DEMONSTRATED_INTERNAL",
   "evidence": [
    {
     "path": "tests/test_e93v_checkers.py",
     "sha256": "3da9d16347678b89ffe4900a3bd9399c6f518fcddf02ef025116bb13910427ad"
    }
   ],
   "verifier_checks": null,
   "recomputed_by": "S92 (exploit 0/5 after fix); 8 regression tests",
   "limits": "tested threat model only"
  },
  {
   "id": "E93-C11",
   "claim": "Fixed baselines on the 90 frozen held-out vertical tasks: software 3/30, science 30/30, cyber 1/30 (floor).",
   "status": "MEASURED",
   "evidence": [
    {
     "path": "artifacts/e93/verticals/baseline-F-heldout/rows.jsonl",
     "sha256": "4147a4937fd3fbabc04ac3492a6a5adc9db168e305966d4d9c55aa92adb32283"
    },
    {
     "path": "artifacts/e93/verticals/manifest.json",
     "sha256": "03a398266fd99ca8f4272d6b1bf764cc6937ee699055b6a6172551c8edbe293a"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'verticals'",
   "recomputed_by": "S92 from raw rows",
   "limits": "science is solved by a designer rule (ceiling environment); cyber is the E91 environment reused"
  },
  {
   "id": "E93-C12",
   "claim": "Governance at the resource: 19/19 admission attacks refused, 10/10 tool-gate checks, and in 20/20 kill-switch trials with 4 worker processes 36,565 effects landed before the fence and 0 after (56,333 post-fence attempts refused; fence median 4.55 ms, max 55.7 ms).",
   "status": "DEMONSTRATED_INTERNAL",
   "evidence": [
    {
     "path": "artifacts/e93/governance/rows.jsonl",
     "sha256": "29c255a5d78f1e44ca619f3bc2d1805d1dbba3e43f5358fc731c12faa225ed29"
    },
    {
     "path": "artifacts/e93/governance/kill_trials.jsonl",
     "sha256": "db8f0b1f520a06c2ba6c0ba4d0611aa4183c0739dd22623659cc3068a1441f5b"
    },
    {
     "path": "artifacts/e93/governance/summary.json",
     "sha256": "75954f6130650ee5592efc443da939643f61b7e48df146ad02567b1cae2c51a7"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'governance'",
   "recomputed_by": "S92 from raw rows",
   "limits": "single host, SQLite resource; not distributed, not BFT"
  },
  {
   "id": "E93-C13",
   "claim": "A >= 6 h mission on fresh vertical tasks with 20 injected faults and independent post-run verification.",
   "status": "IN_PROGRESS",
   "evidence": [
    {
     "path": "artifacts/e93/mission/run_6h/prereg.json",
     "sha256": "d4d5d744b96b13db9dd43098cba6bf2feb3fe7ba2ece8941bc46f9568f5a72f4"
    }
   ],
   "verifier_checks": "verify_e93_offline.py section 'mission'",
   "recomputed_by": "pending",
   "limits": "work is the fixed strategy F, not a learned agent"
  }
 ],
 "not_claimed": [
  "CAIN-42 is not claimed to be AGI, VAGI, human-level or superintelligent.",
  "No external benchmark was run in E93 (ARC-AGI-2/3, MemoryArena, AMA-Bench, UltraHorizon, AgentLAB: NOT RUN).",
  "No external review or reproduction (BLOCKED: no reviewer).",
  "No new E93 defensive-cyber environment (E91 environment reused; blocker B09).",
  "No cross-domain transfer, no collaboration advantage, no memory experiment in E93.",
  "No paid or frontier model was used in E93 (owner decision: $0 further spend).",
  "Signatures in E93 bundles are self-generated keys; hashing shows integrity, not independent witnessing."
 ],
 "status_vocabulary": [
  "DEMONSTRATED_INTERNAL",
  "NEGATIVE_RESULT",
  "MEASURED",
  "REPRODUCED",
  "IN_PROGRESS",
  "EVIDENCE_MISSING"
 ]
}