{
  "campaign_id": "2026-10-16-glm-natural-baseline",
  "coding_r1": {
    "aggregate_status": "R1 retained; final comparison is in coding-summary-final.json",
    "artifact": "native/cached-swe-r1.json",
    "attempted": 5,
    "graded": 5,
    "official_grader_complete": true,
    "promotion_scope": "This campaign proposes no promotion.",
    "resolve_rate": 0.8,
    "resolved": 4,
    "scope": "cached fixed five-task SWE stage"
  },
  "complete": true,
  "decision": {
    "decision_labels": [
      "current",
      "no-promotion"
    ],
    "evidence_labels": [
      "functional",
      "capacity"
    ],
    "next_gate": "No further measurement gate in this closed campaign; retain the negative result and no-promotion decision.",
    "coding_repeatability_established": false,
    "coding_summary": "coding-summary-final.json",
    "independent_r2_review": "independent-r2-review.json",
    "promotion_in_campaign": false
  },
  "evidence": [
    "timing-summary.json",
    "graph-manifest.json",
    "graphs/timing-graph-data.json",
    "graphs/timing-matrix.svg"
  ],
  "failures": [
    {
      "aggregate_throughput_eligible": false,
      "artifact": "native/natural-ctx8192-c1-r2.json",
      "class": "stream_validation_failed",
      "completion_tokens": 16384,
      "content_chunks": 0,
      "elapsed_ms": 213753.305,
      "error": "stream completed without visible content",
      "finish_reason": "length",
      "performance_eligible": false,
      "prompt_tokens": 6078,
      "reasoning_chunks": 15397,
      "request_index": 16
    }
  ],
  "gates": [
    {
      "claim": "stream compatibility only",
      "name": "routed preflight",
      "status": "passed"
    },
    {
      "claim": "observational timing only; no stable p99 or service-stability claim",
      "name": "timing matrix",
      "status": "complete-with-one-retained-failure"
    },
    {
      "claim": "R2 terminal incomplete; no pooled score or repeatability claim",
      "name": "cached five-task coding repeats",
      "status": "closed-negative"
    }
  ],
  "headline_results": [
    "Timing: 239/240 retained; one no-visible-output failure at the output cap.",
    "Coding: R1 4/5 fully graded; R2 terminal incomplete with 0/2 graded resolved and three step-limit failures."
  ],
  "identity": {
    "configuration": "configuration-identity.json"
  },
  "limitations": [
    "Variable natural-output lengths.",
    "Completion tokens may include reasoning tokens.",
    "Shared-router observation without traffic isolation.",
    "Successful timing rows lack finish reasons and semantic validation.",
    "Twenty requests per cell cannot establish stable p99 latency or service stability.",
    "One C1/8K request reached the output cap without visible content; C1/32K retains two long reasoning outliers.",
    "Coding uses five reused pinned tasks twice, not ten fresh held-out tasks."
    ,"Post-run free storage is below the required reserve; future trials are blocked pending separately owned storage work, without attributing SWE outcomes to storage."
  ],
  "measurement": {
    "path": "routed shared-router natural-output observation",
    "sample_count": 240,
    "statistics": [
      "12 timing artifacts: 239 completed requests and one retained no-visible-output failure; coding repeatability not established"
    ],
    "warm_cold_state": "unique prompt bytes; no cache flush or traffic isolation"
  },
  "schema": "anvil-serving.benchmark-decision-summary/v1",
  "workload": {
    "manifest": "workload-manifest.json"
  }
}