{
  "schema": "flamehaven_api_v1_vocabulary",
  "lanes": {
    "EQA": {
      "name": "Computational Proof Reproduction",
      "description": "Takes a published theorem or equation and implements it as executable, CI-tested code. Asks: does running this implementation reproduce the numerical results claimed in the paper, at the stated precision, under the stated conditions?",
      "portal": "https://flamehaven01.github.io/Flamehaven-Verification-Ledger/eqa.html"
    },
    "BAV": {
      "name": "Biomedical AI Pipeline Governance",
      "description": "Audits the full biomedical AI pipeline from molecule input to clinical decision output. Multi-model disagreement is treated as a signal to investigate, not noise to smooth over.",
      "portal": "https://flamehaven01.github.io/Flamehaven-Verification-Ledger/index.html"
    },
    "BSC": {
      "name": "Bioscience Repository Compliance",
      "description": "Static, zero-execution safety scans of public bioscience AI repositories. Maps findings to the MIT AI Risk Repository (AIRI) 4-tier evidence-readiness classification.",
      "portal": "https://flamehaven01.github.io/Flamehaven-Verification-Ledger/index.html"
    }
  },
  "verdicts": [
    {
      "code": "PASS",
      "label": "Pass",
      "meaning": "All checks passed; reproduction or classification succeeded within stated bounds."
    },
    {
      "code": "DEGRADED",
      "label": "Degraded pass",
      "meaning": "Partial pass: some checks passed but with scope caveats or boundary limitations."
    },
    {
      "code": "ACCEPT",
      "label": "Accept with bounds",
      "meaning": "Research-viable as a supporting construct within explicitly stated bounds; not suitable for core promotion."
    },
    {
      "code": "BLOCK",
      "label": "Block",
      "meaning": "Hard constraint violation triggered; dangerous false-pass risk identified. Advancement blocked."
    },
    {
      "code": "REJECTED",
      "label": "Gate rejected",
      "meaning": "Claim scope exceeds what the mathematical or technical construction supports."
    },
    {
      "code": "FAIL",
      "label": "Fail",
      "meaning": "Pipeline failure surface identified; PASS-eligible controls routed incorrectly."
    },
    {
      "code": "HOLD",
      "label": "Keep observer",
      "meaning": "All arms unverified; multi-model disagreement kept visible as signal rather than resolved by majority vote."
    },
    {
      "code": "ABSTAIN",
      "label": "Honest abstain",
      "meaning": "Calibration metrics are statistically sound on pilot data, but cross-domain consistency fails; system correctly declined to claim generalisation."
    },
    {
      "code": "NULL",
      "label": "Truthful null",
      "meaning": "Gate rejected all candidates quickly and correctly. Value is in what was NOT built: confirmed negative rather than a failure to find a positive."
    },
    {
      "code": "T1",
      "label": "High Risk",
      "meaning": "Significant deficiencies. Use only with explicit mitigations and documented limitations. (BSC lane)"
    },
    {
      "code": "T2",
      "label": "Caution",
      "meaning": "Partial compliance. Usable with acknowledged limitations and active monitoring. (BSC lane)"
    }
  ],
  "metrics": {
    "sr9": {
      "name": "Cross-domain consistency (SR9)",
      "description": "Can the pipeline reasoning transfer across domains it was not explicitly trained on?",
      "pipeline_guard": 0.7,
      "honesty_gate": 0.8,
      "direction": "higher_is_better",
      "provenance": "advisory_internal",
      "not_externally_validated": true
    },
    "di2": {
      "name": "Reasoning deviation (DI2)",
      "description": "How far does the model internal logic drift from its stated claims?",
      "pipeline_guard": 0.3,
      "honesty_gate": 0.2,
      "direction": "lower_is_better",
      "provenance": "advisory_internal",
      "not_externally_validated": true
    },
    "p_e2e": {
      "name": "End-to-end reliability (p_e2e)",
      "description": "Probability that a true positive survives the full pipeline: input capture x signal transfer x model correct x clinical decision correct.",
      "direction": "higher_is_better",
      "provenance": "advisory_internal",
      "not_externally_validated": true
    },
    "balanced_accuracy": {
      "name": "Balanced accuracy",
      "description": "Mean of PASS recall and BLOCK recall. 1.0 = perfect discrimination; 0.5 = random.",
      "direction": "higher_is_better",
      "provenance": "standard_ml"
    },
    "brier_score": {
      "name": "Brier score",
      "description": "Calibration quality: mean squared error between predicted probabilities and outcomes. 0.0 = perfect calibration.",
      "direction": "lower_is_better",
      "provenance": "standard_ml"
    },
    "final_score_bsc": {
      "name": "BSC compliance score",
      "description": "Weighted composite: 0.4 x README intent + 0.2 x consistency + 0.4 x code safety. Risk penalty subtracted. Capped at 69 for clinical-adjacent repositories without an explicit boundary statement.",
      "direction": "higher_is_better",
      "provenance": "advisory_internal",
      "not_externally_validated": true
    }
  },
  "disclaimer": "All entries are archival evidence records. They do not constitute certification, regulatory approval, or clinical recommendation. Internal pipeline metrics (SR9, DI2, p_e2e) are advisory Flamehaven-internal scores, not externally validated. Verification weight rests on external anchors: public MIT repositories, the Zenodo DOI, and standard third-party metrics such as AlphaFold pLDDT and PAE.",
  "generated_at_utc": "2026-06-18T04:04:39.065595+00:00"
}