{
  "boundary_failure_resolved": "unproven",
  "campaign_id": "glm-sampling-fix-20261025-r1",
  "candidate_inference_requests": 0,
  "candidate_weights_loaded": false,
  "comparison_evidence": null,
  "complete": true,
  "completion_scope": "Closed diagnostic campaign and verified readiness restoration; kernel gate failed and candidate inference was not run.",
  "decision": {
    "decision_labels": [
      "no-promotion"
    ],
    "evidence_labels": [
      "functional"
    ],
    "next_gate": "Capture the failing ordinary mask difference and probability mass against an independent reference before any proposed gate revision.",
    "promotion_authorized": false
  },
  "elapsed_seconds": 627.1426916122437,
  "evidence": [
    "kernel-log-excerpt.json",
    "kernel-cases.json",
    "candidate-status.json",
    "build.json",
    "plan.json",
    "identity.json",
    "source-freeze.json",
    "restoration.json",
    "projection-ledger.json",
    "closure.json"
  ],
  "execution": "stopped_at_kernel_gate",
  "failures": [
    "Ordinary-logit exact mask parity failed for vocabulary 154856, batch 64. Candidate exited 1."
  ],
  "forced_cases_passed": 6,
  "forced_cases_planned": 8,
  "forced_cases_unrun": 2,
  "gates": [
    {
      "evidence": "kernel-cases.json",
      "name": "forced-logit-regression",
      "status": "partial"
    },
    {
      "evidence": "kernel-log-excerpt.json",
      "name": "ordinary-exact-parity",
      "status": "failed"
    },
    {
      "name": "end-to-end-boundary",
      "status": "not_run"
    },
    {
      "evidence": "restoration.json",
      "name": "restoration",
      "status": "passed"
    }
  ],
  "headline_results": [
    {
      "evidence": "kernel-cases.json",
      "name": "forced_cases_passed",
      "planned": 8,
      "value": 6
    },
    {
      "evidence": "kernel-cases.json",
      "name": "ordinary_parity_passed",
      "planned": 4,
      "value": 3
    },
    {
      "evidence": "candidate-status.json",
      "name": "candidate_inference_requests",
      "value": 0
    }
  ],
  "identity": {
    "evidence": "identity.json",
    "hardware": "One RTX PRO 6000 Blackwell Max-Q GPU; two reserved for planned TP2 serve",
    "model": "nvidia/GLM-5.3-Flash-FP4-Local"
  },
  "ignis": "hold",
  "kernel_gate": "failed",
  "limitations": [
    "Counts reconstructed from ordered execution and failure site; no final success receipt.",
    "No mismatching-token count or probability-mass oracle retained; correctness of the ordinary difference is unknown.",
    "No candidate weights, serving-worker boundary trace, end-to-end inference or performance comparison.",
    "Peak tensor-memory assertion was not reached.",
    "Public bundle assembled after run from privately retained plan/source/managed logs.",
    "Restoration used authenticated readiness and exact identity checks; no generation smoke was sent."
  ],
  "measurement": {
    "path": "synthetic kernel regression before model weights",
    "sample_count": 10,
    "sample_count_definition": "Six completed forced cases plus four ordinary comparisons, including the failure. No inference requests.",
    "statistics": [],
    "warm_cold_state": "JIT/cache condition uncontrolled; no timing comparison"
  },
  "metrics_coverage": {
    "profile": "focused-functional-diagnostic",
    "reason": "Preweight kernel correctness gate; performance and full serving comparison out of scope.",
    "status": "not-applicable"
  },
  "ordinary_parity_failed": 1,
  "ordinary_parity_passed": 3,
  "ordinary_parity_planned": 4,
  "promoted": false,
  "promotion_authorized": false,
  "restoration": "verified",
  "schema": "anvil-serving.benchmark-decision-summary/v1",
  "workload": {
    "dtype": "float32",
    "evidence": "plan.json",
    "scope": "focused-functional-kernel-regression",
    "seed": 58231,
    "top_p": 0.95
  }
}