{
  "format": "vercy-enterprise-profile-candidate/v1",
  "contourId": "EM-AI-03",
  "name": "Enterprise AI Evaluation, Safety and Deployment Decision Binding",
  "decision": "PROFILE",
  "newRuntimeId": false,
  "bases": [
    "WM-AI-003",
    "WM-AI-009",
    "WM-AI-008",
    "WM-AI-010",
    "WM-SFT-004",
    "WM-DAT-001",
    "WM-MAT-008",
    "WM-KNW-010",
    "WM-ACT-024",
    "WM-REC-010"
  ],
  "constraints": [
    "WM-AI-003 owns one evaluation occurrence with distinct applied-plan, run and result roles; a run cannot rewrite its plan and a result requires a completed run.",
    "WM-AI-009 owns the benchmark contract and the benchmark dataset snapshot definition, including item set composition, item digests, sampling and hidden-test controls; WM-DAT-001 owns the source dataset master; WM-AI-003 pins release, snapshot digest, split designation and contamination-check result. Replacing the snapshot creates a new evaluation occurrence.",
    "Every accepted result pins model artifact, configuration, benchmark, dataset snapshot, metric, scorer, code and tool versions, environment, context, intended use, population, uncertainty and limitations.",
    "Capability measurement is not a safety judgement; WM-AI-008 consumes pinned evidence and does not authorize deployment.",
    "Deployment rationale, authorized act and issued record remain WM-KNW-010, WM-ACT-024 and WM-REC-010 roles; editing rationale after issue requires a new decision.",
    "Dataset, scorer, configuration, environment, context, intended-use or population change ends the transferability of prior evidence and removes prior authorization from effect for the changed scope only; prior evaluation, assessment and issued records remain immutable and remain valid within their own pinned scope.",
    "Contamination, leakage or context drift triggers reassessment and never an in-place score update.",
    "WM-AI-010 cites incident evidence mastered externally by WM-MAT-008 and WM-SFT-004 and never contains or replaces it.",
    "A high benchmark score is historical measurement only and cannot widen authorization scope or prove universal safety.",
    "Reusable Evaluation Protocol and Standing Deployment Authorization Instrument have no present independent identity and no assigned identifier.",
    "Within one WM-AI-003 occurrence, applied plan, run and result are append-only sub-records keyed by (occurrenceRef, role, sequence), each carrying its own recordedAt, actor and state; the plan is sealed at planSealedAt before any run begins; a run carries state in {started, completed, failed, aborted}; a result exists only for a run in state completed; a retry with byte-identical pins is a new run sequence inside the same occurrence; any pin delta requires a new occurrence.",
    "Rescoring the same sealed plan, run and pinned inputs with a changed scorer version produces an additive successor result inside the same occurrence, citing the superseded result; a change to any other pin requires a new occurrence.",
    "Benchmark and dataset rights, licence and redistribution terms are declared before use; protected benchmark items, hidden-test content, confidential incident content and personal data are excluded from published artifacts; disaggregated results require a declared minimum subgroup size.",
    "WM-AI-009 owns metric definitions and the benchmark-level protocol contract; WM-AI-003 owns only the applied protocol instance and may narrow but never redefine the contract or a metric definition.",
    "A WM-AI-003 gate decision, approval or waiver governs evaluation acceptance only; deployment authorization arises solely from WM-KNW-010 rationale, WM-ACT-024 act and WM-REC-010 instrument; waiver expiry is a reassessment trigger and never extends authorization.",
    "Within one WM-AI-003 occurrence, applied plan, run and result are append-only sub-records keyed by (occurrenceRef, role, sequence), each carrying its own recordedAt, actor and state; the plan is sealed at planSealedAt before any run begins; a run carries state in {started, completed, failed, aborted}; a result exists only for a run in state completed; a retry with byte-identical pins is a new run sequence inside the same occurrence; any pin delta requires a new occurrence.",
    "Rescoring the same sealed plan, run and pinned inputs with a changed scorer version produces an additive successor result inside the same occurrence, citing the superseded result; a change to any other pin requires a new occurrence.",
    "Benchmark and dataset rights, licence and redistribution terms are declared before use; protected benchmark items, hidden-test content, confidential incident content and personal data are excluded from published artifacts; disaggregated results require a declared minimum subgroup size.",
    "WM-AI-009 owns metric definitions and the benchmark-level protocol contract; WM-AI-003 owns only the applied protocol instance and may narrow but never redefine the contract or a metric definition.",
    "A WM-AI-003 gate decision, approval or waiver governs evaluation acceptance only; deployment authorization arises solely from WM-KNW-010 rationale, WM-ACT-024 act and WM-REC-010 instrument; waiver expiry is a reassessment trigger and never extends authorization.",
    "Within one WM-AI-003 occurrence, applied plan, run and result are append-only sub-records keyed by (occurrenceRef, role, sequence), each carrying its own recordedAt, actor and state; the plan is sealed at planSealedAt before any run begins; a run carries state in {started, completed, failed, aborted}; a result exists only for a run in state completed; a retry with byte-identical pins is a new run sequence inside the same occurrence; any pin delta requires a new occurrence.",
    "Rescoring the same sealed plan, run and pinned inputs with a changed scorer version produces an additive successor result inside the same occurrence, citing the superseded result; a change to any other pin requires a new occurrence.",
    "Benchmark and dataset rights, licence and redistribution terms are declared before use; protected benchmark items, hidden-test content, confidential incident content and personal data are excluded from published artifacts; disaggregated results require a declared minimum subgroup size.",
    "WM-AI-009 owns metric definitions and the benchmark-level protocol contract; WM-AI-003 owns only the applied protocol instance and may narrow but never redefine the contract or a metric definition.",
    "A WM-AI-003 gate decision, approval or waiver governs evaluation acceptance only; deployment authorization arises solely from WM-KNW-010 rationale, WM-ACT-024 act and WM-REC-010 instrument; waiver expiry is a reassessment trigger and never extends authorization."
  ],
  "holds": [
    "Independent Grok review and the single frozen semantic audit are complete; all 15 material defects were remediated without rerun.",
    "WM-AI-009→WM-DAT-001 relation is unapproved; snapshot-versus-source mastership is a publication blocker.",
    "Evaluation-occurrence master identity remains a gap; occurrenceRef is a declared descriptor, not an allocated identifier.",
    "Rights and confidentiality clearance for benchmark items, dataset snapshots and incident content is unresolved and is a publication blocker.",
    "No frozen deployment/endpoint or incident-occurrence master exists; context and incident pins are descriptors pending a master.",
    "WM-AI-003/WM-AI-009 protocol and metric overlap is arbitrated by constraint but the base drafts are unreconciled.",
    "WM-AI-010 relations and WM-AI-003 incident/risk siblings are unapproved or unregistered; incident evidence routing is unenforced.",
    "Base pins carry null version and digest; unversioned base pins block profile publication under code BASE_PIN_UNVERSIONED and are distinct from result-pin acceptance.",
    "Both conditional candidates remain identifier-unassigned; canonicalPublishable and installable remain false."
  ],
  "candidateRevision": 2,
  "status": "held",
  "canonicalPublishable": false,
  "installable": false,
  "constraintsAreNonExhaustive": false,
  "basePins": [
    {
      "modelId": "WM-AI-003",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-AI-009",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-AI-008",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-AI-010",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-SFT-004",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-DAT-001",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-MAT-008",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-KNW-010",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-ACT-024",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    },
    {
      "modelId": "WM-REC-010",
      "version": null,
      "digest": null,
      "verification": "unverified-current-draft"
    }
  ],
  "invariants": [
    "Plan, run and result are distinct roles of one evaluation occurrence.",
    "Result follows a completed run and neither run nor result rewrites the applied plan.",
    "Benchmark contract and dataset snapshot retain separate masters.",
    "A dataset snapshot change requires a new evaluation occurrence.",
    "Capability measurement is not a safety judgement.",
    "A safety assessment does not authorize deployment.",
    "Decision rationale, authorized act and issued record remain distinct.",
    "Issued rationale is immutable; correction requires a successor decision.",
    "Scores do not transfer across dataset, configuration, context, intended use or population except under a dated, authorized transfer justification that names the delta and its construct impact.",
    "Incident reports cite and never contain or replace incident evidence.",
    "A change in context, intended use or population places the changed scope outside prior authorization; silent carry-forward is rejected and the unchanged scope is unaffected.",
    "Uncertainty and limitations are mandatory on result and assessment.",
    "Contamination or drift triggers reassessment, never in-place score mutation.",
    "Missing or unversioned result pins block acceptance of that result.",
    "No conditional candidate has an assigned identifier or present independent identity.",
    "Split labels never prove non-contamination; non-overlap requires declared contamination evidence.",
    "Absence of a metric is not absence of risk.",
    "Evidence is revision-bound and becomes stale by a declared rule; stale evidence is marked stale and is not silently carried.",
    "Event, observation, ingestion, approval, publication and effective times are distinct RFC 3339 instants with explicit offset.",
    "Corrections and rescoring are additive; the original value, scorer version, actor, time and reason are preserved and never overwritten.",
    "Thresholds and the applied plan are registered before any result is observed; a later change requires an authorized, dated change record disclosing prior result visibility and does not amend the original result.",
    "Benchmark and dataset rights, licence and redistribution terms are declared before use; protected benchmark items, hidden-test content, confidential incident content and personal data are excluded from published artifacts; disaggregated results are reported only at or above a declared minimum subgroup size.",
    "Prior records remain immutable and valid within their pinned scope; change removes transferability, not validity.",
    "Deployment context, intended use, population, incident occurrence and evaluation occurrence are declared descriptors with no allocated identifier; they are never used as resolvable identifiers.",
    "A WM-AI-003 gate approval or waiver is not a deployment authorization.",
    "Split labels never prove non-contamination; non-overlap requires declared contamination evidence.",
    "Absence of a metric is not absence of risk.",
    "Evidence is revision-bound and becomes stale by a declared rule; stale evidence is marked stale and is not silently carried.",
    "Event, observation, ingestion, approval, publication and effective times are distinct RFC 3339 instants with explicit offset.",
    "Corrections and rescoring are additive; the original value, scorer version, actor, time and reason are preserved and never overwritten.",
    "Thresholds and the applied plan are registered before any result is observed; a later change requires an authorized, dated change record disclosing prior result visibility and does not amend the original result.",
    "Benchmark and dataset rights, licence and redistribution terms are declared before use; protected benchmark items, hidden-test content, confidential incident content and personal data are excluded from published artifacts; disaggregated results are reported only at or above a declared minimum subgroup size.",
    "Prior records remain immutable and valid within their pinned scope; change removes transferability, not validity.",
    "Deployment context, intended use, population, incident occurrence and evaluation occurrence are declared descriptors with no allocated identifier; they are never used as resolvable identifiers.",
    "A WM-AI-003 gate approval or waiver is not a deployment authorization.",
    "Split labels never prove non-contamination; non-overlap requires declared contamination evidence.",
    "Absence of a metric is not absence of risk.",
    "Evidence is revision-bound and becomes stale by a declared rule; stale evidence is marked stale and is not silently carried.",
    "Event, observation, ingestion, approval, publication and effective times are distinct RFC 3339 instants with explicit offset.",
    "Corrections and rescoring are additive; the original value, scorer version, actor, time and reason are preserved and never overwritten.",
    "Thresholds and the applied plan are registered before any result is observed; a later change requires an authorized, dated change record disclosing prior result visibility and does not amend the original result.",
    "Benchmark and dataset rights, licence and redistribution terms are declared before use; protected benchmark items, hidden-test content, confidential incident content and personal data are excluded from published artifacts; disaggregated results are reported only at or above a declared minimum subgroup size.",
    "Prior records remain immutable and valid within their pinned scope; change removes transferability, not validity.",
    "Deployment context, intended use, population, incident occurrence and evaluation occurrence are declared descriptors with no allocated identifier; they are never used as resolvable identifiers.",
    "A WM-AI-003 gate approval or waiver is not a deployment authorization."
  ]
}