{
  "schema_version": 1,
  "date": "2026-09-20",
  "scope": "Local evaluator artifact-safety repair; synthetic fixtures with fitting intercepted",
  "source_commit": "b8d32393a5f37d20e7c4f98920e453e57c421d4b",
  "baseline_commit": "dd7b36122354da8984be73c228bc5dc47d814d28",
  "publication": "Local source checkpoint; no MMG source push or deployment",
  "source_identity": {
    "files": {
      "backend/research_v3_eval.py": "8ae8fd16de963f0ae10beb3a0447745da2572cdc12af194e46b74680daa53c65",
      "backend/tests/test_research_pit_eval.py": "ff09c349ebcbf956d45413450a730654dd1cdf31ca894962f0aa4f9643b877fd"
    },
    "algorithm": "SHA-256 of UTF-8 source after CRLF-to-LF normalization"
  },
  "contract": {
    "failure": "A failed run retains an incomplete reservation that cannot be reused; nonfinite results are rejected before JSON is written",
    "output_identity": [
      "training manifest and manifest hash",
      "training data hash",
      "transformation configuration hash",
      "evaluator settings and feature lists with configuration hash",
      "evaluator and shared-contract source hashes",
      "runtime dependency versions",
      "combined identity hash"
    ],
    "destination": "Explicit new path; existing files and final-path symlinks rejected; exclusive creation reserves the output before fitting",
    "input": "Explicit PIT training artifact; shared manifest, hash, schema, feature-order and label-time validation before fitting",
    "output_schema": "research_component_evaluation_v1",
    "additional_checks": "Binary evaluator labels, current eligibility and minimum interval count"
  },
  "verification": {
    "scope_note": "Overlapping runs, not additive totals. All nine focused cases use synthetic inputs and intercept model fitting.",
    "independent_source_review": "No actionable defect found within evaluator artifact-safety scope",
    "owner_backend_suite": {
      "skipped": 78,
      "warnings": 2,
      "passed": 203
    },
    "independent_focused_run": {
      "exit_code": 0,
      "seconds": 2.05,
      "passed": 9
    }
  },
  "limits": [
    "No actual component evaluation, real-data fitting, new model, metric recomputation or fresh holdout was performed.",
    "The historical component procedure does not evaluate the exact displayed serving rank; its statistical methodology has not been revalidated.",
    "No compatible real PIT dataset or model was produced. Historical metrics and artifacts remain unchanged.",
    "No MMG deployment or hosted acceptance follows from these checks.",
    "Exclusive output creation is not a power-loss durability guarantee or protection against concurrent mutation of input or source files."
  ]
}