{
  "schema": "kingy-document-independent-candidate-review/1.0.0",
  "reviewedAt": "2026-10-18T05:58:10.796012+00:00",
  "candidateVersion": "kingy-document-runner/1.5.6",
  "developmentBuild": 4,
  "runnerSha256": "3bcc03ffef818fa7fb6062ae17f20aa09d2fcc25fc9d57572c77ec35bc1ef02b",
  "sourceSha256": {
    "runner.py": "3bcc03ffef818fa7fb6062ae17f20aa09d2fcc25fc9d57572c77ec35bc1ef02b",
    "start_server.py": "7f54900199a61c495bf5f6753a5caa13b529f86061d56ec2fca5a63be3844b2a",
    "prepare_model.py": "2a66acb34c1bf136b087179dbc2c5f676156bb31d971ed33b82b05e77c12f576",
    "evaluate.py": "8daa63c6ceb93111fd898683d40712a414ebda0fdf08b6a165638250871e4152",
    "offline.sb": "e5462ef1daae3785012cb8e4fa05b89105bf4ac4d98be9580a8c357365a4f54f",
    "requirements.txt": "3ec8922905dc4e3a10c3b46bd7fba023c37ffc309d51c6bed2a13ec3e98177fd",
    "SCORING.md": "2d0b2201a894bb663f5f71a6c2464bcf7d53138b84b83de153ed9bf18f553259",
    "README.md": "c05ca8b1d7aefd859faf23f1a5d6ecd33be4e570a2060aa020a869155b93c0eb",
    "LICENSE.txt": "7b5272ed8db5e81567ea2a9624ab25c04db44dc168db692c760a61f4d392100c",
    "common_grade.py": "e9b95ce6b5a821c0ab703c60de388f7604d919d62cfe13ac544cfbcb95c14c5f"
  },
  "approvedForMethodFreezeAndFreshCorpusAuthorship": true,
  "releaseApproved": false,
  "qualificationGranted": false,
  "humanValidation": false,
  "secondMachineValidation": false,
  "futureBankAccess": false,
  "v20Rerun": false,
  "originalEvidenceChanged": false,
  "scorerChanged": false,
  "summary": "The general field-binding and selected-quote grounding repairs resolve the demonstrated development defects. The exact final source is suitable to freeze before a separate agent authors a fresh unseen qualification corpus. This is engineering approval for the next study, not release or comparative qualification.",
  "reviewScope": {
    "actualBuild3DevelopmentAttemptsReviewed": 33,
    "actualBuild3ModelRequestsReviewed": 91,
    "build4GuardReplayModelCalls": 0,
    "build4SourceAndReplayChecks": 221,
    "independentlyAuthoredBoundaryControls": 16,
    "actualBuild4PublicPdfSelftests": 4,
    "actualBuild4PublicPdfPasses": 4,
    "actualBuild4PublicModelRequests": 10,
    "newIndependentBuild4ModelAttempts": 4,
    "newIndependentBuild4ModelPasses": 4,
    "newIndependentBuild4ModelRequests": 13,
    "repairOwnerDeterministicControls": 35,
    "repairOwnerDeterministicControlsPassed": 35
  },
  "resolvedFindings": [
    {
      "id": "selected-passage-sufficiency-factual-heading-leak",
      "priorActualCase": "ind-single-interrogative-two-facts",
      "priorFailurePreserved": true,
      "resolution": "Single-interrogative conjunctions bind only at one unambiguous top-level split; full original context remains available. Sufficiency requests now require a verbatim excerpt from the selected quote, and a deterministic boundary guard refuses a supplied value absent from that quote. Headings are explicitly scope metadata, not answer evidence.",
      "verifiedBy": "Reviewed corrected actual build3 answer, re-evaluated all retained guard decisions under build4, and newly measured reversed-field compound question with different values/source and a warehouse distractor."
    },
    {
      "id": "partial-number-boundary",
      "resolution": "Pure guard rejects numeric suffixes after digits/comma and partial names, signs and decimal values. Outside-helper AST and other nine files are identical to build3.",
      "verifiedBy": "Independent boundary cases and exact request equality on actual final-source public PDF selftests."
    }
  ],
  "independentCases": [
    {
      "id": "b4-independent-single-question-reversed-fields",
      "passed": true
    },
    {
      "id": "b4-independent-absent-single-question",
      "passed": true
    },
    {
      "id": "b4-independent-quoted-title-shared-scope",
      "passed": true
    },
    {
      "id": "b4-independent-negative-policy-and-anchor",
      "passed": true
    }
  ],
  "reviewerValidatorCorrection": {
    "file": "parsed-context-review.json",
    "originalReviewerReceiptPreserved": true,
    "modelCallsMade": 0,
    "reason": "The original literal-string check falsely failed one quoted-source-title case because JSON escaped double quotes. Parsing the retained sufficiency payload confirms full original context is exact. The actual facts, source, quote and grounding checks all passed in the original run."
  },
  "nonHardcodedGeneralizationReview": {
    "passed": true,
    "details": "Runtime changes use generic question/field vocabulary, top-level clause boundaries and selected-source excerpts. No fixture names, expected answers, case IDs, development paths or new bank rules were introduced. Grading references remain outside ask()/ask_field(); scorer1.5.0 is byte-identical."
  },
  "preservation": {
    "priorIndependentSixOriginalsAndHoldPreserved": true,
    "old1_5_5SourceTenFilesPreserved": true,
    "v20StudyEditedOrRerunByReviewer": false,
    "finalSourceUnchangedAfterIndependentInference": true
  },
  "limits": [
    "This is same-Mac development validation with the installed local Qwen model; no human validation, second machine, unseen qualification or competitor superiority is established.",
    "Thirty-three actual development attempts ran on build3; their 91 request bodies and known guard decisions are replayed under build4 without inference. Build4 additionally has four actual public PDF tests and four newly authored actual extracted-page probes. These counts must not be described as a fresh build4 33-question model rerun.",
    "The new independent fixtures validate extracted-page logic rather than PDF parsing; the four public tests separately exercise real PDF loading.",
    "The verbatim guard establishes a selected-quote excerpt boundary, not semantic truth. The two model phases still judge target relevance, condition, version and completeness; retrieval remains bounded to eight pages/twelve candidate passages per field.",
    "Returned answers remain source passages and can include extra context or other conditional values. These tests do not prove optimal concision.",
    "Freeze the exact complete prospective method before a new unseen bank is authored or consumed; keep future measurement and launch eligibility separate from this engineering approval."
  ],
  "inputSha256": {
    "source-and-replay-review.json": "2129708b038d7da42b0c12f4982cc2e9a3d34cd6cf866b8778acded5e53f95e2",
    "independent-development-inputs.json": "eaf7ccd1fbfef4adde617e8d94f9c0a32c0ce442981126f936c921a8be607f66",
    "independent-development-freeze.json": "9d801b6234ce0a17b01c39b8c528b762cd08913e1c27394793d781b63e9f3313",
    "independent-model-results.jsonl": "e78da16125840513387961f1814f8f06f9eec10f8731531cef1e9ee8061d0fe1",
    "independent-run-completion.json": "9ce3344b7adca99ad31eb2beff5bc5a218675c452d4212ec08281dbc50d9bf84",
    "parsed-context-review.json": "fb8c34026c1b127ff5d3ba48c6ffed66c43d62e567849c20edf914a9f7b85eb0"
  },
  "externalEvidenceSha256": {
    "build4-source-freeze.json": "d907a99d68ea04f48126310514c5faa8c8c8e833633a3ec42e5c9ad29dba1a5f",
    "build3-completion.json": "ba41349f038b05ab285f9f5f0263b23045b390b3448658fe8f4ef2df3e1f5e4c",
    "build4-public-selftest.jsonl": "8436afe7fd519e6a048aee5503e392bf17aa5b9992628e075941aa450a629a71",
    "build4-public-request-equivalence.json": "bd63460a7a0ef22001f1e05625c14f32aca11430e52a1e00e7071e2a6b4ff299",
    "prior-independent-hold.json": "7a81d34f68c4acbf97aa932e9e739243d8e66dc4d2643057d38a7545f9cf50fc"
  },
  "modelCleanup": {
    "ownedServerStopped": true,
    "port18482ListeningAfter": false,
    "generatedKeyRemoved": true
  }
}