{
  "run_id": "2026-10-06-quick-greedy-nothink-gateway-json-schema-414e03",
  "suite": "typed-decisions",
  "tier": "quick",
  "suite_hash": "5995f56de30d9a5e5b8ab8d8342a24025e75a52464b4861e2c1d16ca1fc5c7b7",
  "subject": "qwen-3.8-27b/cerebras-2026-10/opaque/cerebras",
  "mode": "gateway:json-schema",
  "profile": "greedy-nothink",
  "status": "complete",
  "label": "provider-opaque",
  "items": {
    "planned": 100,
    "attempted": 100,
    "succeeded": 99,
    "failed": 1,
    "failed_by_status": {
      "502": 1
    }
  },
  "metrics": {
    "decisions": 500,
    "valid": 495,
    "accuracy": {
      "value": 0.722,
      "low": 0.681151,
      "high": 0.759463,
      "n": 500
    },
    "kl_from_gold": {
      "value": 3.930521,
      "low": 3.443078,
      "high": 4.457374,
      "n": 495
    },
    "brier": {
      "value": 0.246899,
      "low": 0.222453,
      "high": 0.271895,
      "n": 495
    },
    "ece": 0.144141,
    "accuracy_by_type": {
      "choice": {
        "value": 0.713333,
        "low": 0.636351,
        "high": 0.779662,
        "n": 150
      },
      "noul": {
        "value": 0.846667,
        "low": 0.78042,
        "high": 0.8956,
        "n": 150
      },
      "score": {
        "value": 0.635,
        "low": 0.566317,
        "high": 0.698595,
        "n": 200
      }
    },
    "accuracy_by_workflow": {
      "agent_trace_observability": {
        "value": 0.648,
        "low": 0.561003,
        "high": 0.726172,
        "n": 125
      },
      "customer_service": {
        "value": 0.68,
        "low": 0.593908,
        "high": 0.755359,
        "n": 125
      },
      "invoice_processing": {
        "value": 0.824,
        "low": 0.747877,
        "high": 0.880802,
        "n": 125
      },
      "security_incidents": {
        "value": 0.736,
        "low": 0.652526,
        "high": 0.805402,
        "n": 125
      }
    }
  },
  "invalid_reasons": {
    "HTTP 502 invalid_provider_output": 5
  },
  "notes": [
    "accuracy counts every dispatched decision; kl_from_gold, brier and ece use valid decisions only",
    "ece uses 10-bin top-label confidence; the dataset card does not define its ECE, so ece is not comparable with the card",
    "intervals treat decisions as independent; each case contributes five related decisions, so true uncertainty is somewhat wider"
  ],
  "latency_ms": {
    "p50": 572.197708,
    "p95": 1435.922875,
    "p99": 2680.427542,
    "note": "successful requests, nearest rank, measured from America/Chicago"
  },
  "usage": {
    "input_tokens": 290097,
    "output_tokens": 10579,
    "cost_usd": 0.302959
  }
}