{
  "schemaVersion": "b18.sft-greenlight-receipt.r1",
  "runId": "sha256:f81e74a1f17de90665b17015f457a215d2e20cbc62f80e3e696df2bf3ab43373",
  "suiteId": "B18-R1",
  "sourceCommit": "a8dc0ea51076e1ec77c41770d759f93bec751626",
  "sourceTree": "128bf0e195c1731aa6480eb0615df159f79d1c54",
  "startedAt": "2026-08-21T03:30:00.000Z",
  "finishedAt": "2026-08-21T03:30:00.000Z",
  "resultState": "VERIFIED_MECHANISM",
  "conclusion": "FAILED",
  "hardSafetyGates": {
    "sftRequiredEntityRecallIsComplete": true,
    "sftRequiredRelationRecallIsComplete": true,
    "sftRequiredMetricFactRecallIsComplete": true,
    "noFalseAuthoritativeExecution": true,
    "missingEvidenceControlsRefused": true,
    "sftSolverParityIsComplete": true,
    "sourceRevisionTraceabilityIsComplete": true
  },
  "efficiencyGate": {
    "targetMedianBytesRatioToFull": 0.6,
    "observedMedianBytesRatioToFull": 0.785827,
    "passed": false
  },
  "modelComparison": "NOT_RUN_EFFICIENCY_GATE_FAILED",
  "rawArtifacts": [
    "manifest.v1.json",
    "oracle.v1.json",
    "comparison-contexts.v1.json",
    "deterministic-results.v1.json",
    "raw-rows.v1.csv",
    "pinned-model-protocol.v1.json"
  ],
  "invalidRunChecks": {
    "fixtureCorpusChangedDuringTuning": false,
    "hiddenMetricFactsSupplied": false,
    "falseAuthoritativeExecutionObserved": false,
    "missingRowsRemoved": false,
    "modelHoldoutRunDespiteFailedEligibilityGate": false
  },
  "losses": [
    "Median SFT serialized context was 0.785827 of full scene; the predeclared maximum was 0.600000.",
    "The pinned-model holdout was not run because the deterministic efficiency gate failed."
  ],
  "limitations": [
    "The benchmark compares deterministic context construction on seven authored fixture cases, including one missing-evidence control; it is not a frontier-model answer-quality benchmark.",
    "Recall and irrelevant-context metrics use an independently authored entity, relation, and metric-fact oracle per case.",
    "LEXICAL_TOP_K is a transparent lexical baseline, not an embedding database or production Vector RAG implementation.",
    "GRAPH_NEIGHBORHOOD is a one-hop structural baseline, not Microsoft GraphRAG or another complete graph-retrieval system.",
    "Latency rows measure this local Node.js process and are descriptive, not a production performance claim.",
    "The predeclared median serialized-byte target was not met, so the frozen pinned-model holdout was not executed."
  ],
  "nextExperiment": "Do not tune this holdout. Redesign or simplify context construction on a new training corpus, freeze a new evaluation corpus, and rerun the deterministic gate before model spending.",
  "receiptDigest": "sha256:836386e6628e2083fe20e600b23e01eb54434563a083c0a014566cae33793648"
}