{
  "format": "vercy-model-allocation-candidate/v1",
  "contourId": "EM-AI-03",
  "proposedName": "Reusable Evaluation Protocol",
  "modelId": null,
  "registryId": null,
  "allocationState": "unassigned",
  "decision": "NO NEW MODEL",
  "canonicalPublishable": false,
  "identityTest": {
    "stableIdentity": "A protocol is independently identifiable only when a governed evaluation specification is versioned, approved and reused across evaluation instances.",
    "versionIdentity": "Changes to task contract, metric set, thresholds, scorer requirements or validity scope create immutable revisions.",
    "independentLifecycle": [
      "draft",
      "approved",
      "effective",
      "suspended",
      "superseded",
      "retired"
    ],
    "mastership": null,
    "mastershipHold": "no frozen governance master exists for this candidate; mastership is unresolved and may not be inferred."
  },
  "boundary": {
    "owns": [
      "stable reusable protocol identity",
      "immutable protocol revisions",
      "protocol-level selection and binding of externally mastered task, metric, scorer and threshold definitions",
      "validity scope and transfer conditions",
      "effective periods and approval lineage"
    ],
    "references": [
      {
        "target": "WM-AI-003",
        "purpose": "Evaluation instances that apply a pinned protocol revision"
      },
      {
        "target": "WM-AI-009",
        "purpose": "Benchmark specification and release contract"
      },
      {
        "target": "WM-AI-008",
        "purpose": "Safety assessments consuming evaluation evidence"
      },
      {
        "target": "WM-SFT-004",
        "purpose": "Pinned evaluated model artifact"
      },
      {
        "target": "WM-DAT-001",
        "purpose": "Pinned evaluation dataset snapshot"
      },
      {
        "target": "WM-MAT-008",
        "purpose": "Result observations and uncertainty"
      }
    ],
    "excludes": [
      "evaluation-run execution identity",
      "benchmark and dataset identity",
      "model artifact identity",
      "safety judgement",
      "deployment authorization",
      "metric definition mastership",
      "benchmark task contract mastership",
      "threshold definition mastership",
      "metric definition mastership",
      "benchmark task contract mastership",
      "threshold definition mastership",
      "metric definition mastership",
      "benchmark task contract mastership",
      "threshold definition mastership"
    ]
  },
  "objects": {
    "EvaluationProtocol": {
      "identity": [
        "evaluationProtocolId"
      ],
      "required": [
        "name",
        "ownerRef",
        "status",
        "currentRevisionRef"
      ],
      "optional": [
        "successorRef"
      ],
      "lifecycle": [
        "draft",
        "approved",
        "effective",
        "suspended",
        "superseded",
        "retired"
      ]
    },
    "EvaluationProtocolRevision": {
      "identity": [
        "evaluationProtocolId",
        "revision"
      ],
      "required": [
        "taskContract",
        "metricRefs",
        "scorerRequirements",
        "thresholds",
        "validityScope",
        "contentDigest",
        "effectiveFrom",
        "approvalRef",
        "approvedAt",
        "approvedBy"
      ],
      "optional": [
        "effectiveTo",
        "supersedesRevision",
        "transferConditions"
      ]
    }
  },
  "invariants": [
    "Independent identity requires governed versioning and reuse across evaluations.",
    "Every applying evaluation pins one immutable protocol revision.",
    "Protocol and thresholds are registered before results are observed.",
    "Post-hoc changes require authorized disclosure and successor evaluation evidence.",
    "A protocol revision never owns an execution outcome.",
    "Benchmark, dataset, model artifact and metric definitions retain external mastership.",
    "Every result pins artifact, configuration, benchmark, dataset, metric, scorer and environment.",
    "Split labels alone never prove non-contamination.",
    "Evaluation validity is bounded by purpose, population and deployment context.",
    "Corrections and rescoring are additive and preserve originals.",
    "Measurement never implies safety or deployment permission.",
    "Evidence is revision-bound and becomes stale by declared rule.",
    "Protected benchmark items and confidential artifacts remain outside the public catalogue.",
    "Retired protocol identifiers and revisions remain resolvable and are never recycled.",
    "An unassigned candidate is not referenceable as an identified model and no identifier may be allocated, inferred or guessed.",
    "An unassigned candidate is not referenceable as an identified model and no identifier may be allocated, inferred or guessed.",
    "An unassigned candidate is not referenceable as an identified model and no identifier may be allocated, inferred or guessed."
  ],
  "holds": [
    "Independent Grok review and the single frozen semantic audit are complete; all material defects were remediated without rerun.",
    "The candidate has no present independent identity and remains identifier-unassigned.",
    "Grok's claim that this candidate can never require independent identity is rejected; absence is present-tense.",
    "No frozen governance master exists for this candidate.",
    "Registry allocation and canonical base publication remain unresolved.",
    "No installability or publication-readiness claim is made."
  ],
  "candidateRevision": 2,
  "status": "held",
  "installable": false,
  "presentIndependentIdentity": false,
  "conditionsUnmet": [
    "independent lifecycle and mastership distinct from the reused base models",
    "registry allocation"
  ],
  "decisionScope": "this checkpoint only",
  "conditional": true,
  "reopenCondition": "evidence of a lifecycle, approval and mastership independent of one evaluation occurrence or one issued decision record"
}