{
  "run_id": "support-extraction-v1-20260914113427",
  "started_at": "2026-09-14T11:34:27.461Z",
  "mode": "paired_run_verified",
  "prompt_profile": "short",
  "prompt_version": "support-extraction-prompt-v1",
  "planned_cells": 180,
  "denominators": {
    "attempted": 180,
    "completed": 174,
    "completed_parsed": 173,
    "transport_failures_reported_separately": 6
  },
  "configurations": {
    "A": {
      "model_id": "anthropic/claude-sonnet-5",
      "attempted": 90,
      "completed": 87,
      "transport_failures": 3,
      "schema_pass": "86/87 (98.9%)",
      "task_acceptance": "84/90 (93.3%)",
      "schema_valid_task_fail": 2,
      "field_failures": {
        "urgency": 2
      },
      "failing_cases": {
        "case-20": [
          "urgency: expected \"normal\" got \"unknown\" (r1)",
          "urgency: expected \"normal\" got \"unknown\" (r2)"
        ]
      },
      "latency_ms": {
        "p50": 1433,
        "p95": 1628,
        "n": 87
      },
      "usage": {
        "prompt_tokens": 68055,
        "uncached_prompt_tokens": 68055,
        "cached_prompt_tokens": 0,
        "cache_write_tokens": 0,
        "completion_tokens": 6517,
        "reasoning_tokens_within_completion": 2468,
        "calls_with_cache_hit": "0/87",
        "finish_reasons": {
          "stop": 86,
          "length": 1
        }
      },
      "rates_usd_per_1m": {
        "input": 2,
        "output": 10,
        "cache_read": 0.2,
        "cache_write": 2.5
      },
      "calculated_list_cost_usd": 0.2013,
      "gateway_reported_charge_usd": 0.2013,
      "cost_per_acceptable_result_usd": 0.0024
    },
    "B": {
      "model_id": "openai/gpt-5.6-terra",
      "attempted": 90,
      "completed": 87,
      "transport_failures": 3,
      "schema_pass": "87/87 (100.0%)",
      "task_acceptance": "77/90 (85.6%)",
      "schema_valid_task_fail": 10,
      "field_failures": {
        "category": 4,
        "urgency": 3,
        "requested_action": 3
      },
      "failing_cases": {
        "case-11": [
          "category: expected \"unknown\" got \"feature\" (r1)"
        ],
        "case-20": [
          "urgency: expected \"normal\" got \"unknown\" (r1)",
          "urgency: expected \"normal\" got \"unknown\" (r2)",
          "urgency: expected \"normal\" got \"unknown\" (r3)"
        ],
        "case-21": [
          "requested_action: expected null got \"Please cancel my subscription.\" (r1)",
          "requested_action: expected null got \"Please cancel my subscription.\" (r2)",
          "requested_action: expected null got \"Please cancel my subscription.\" (r3)"
        ],
        "case-23": [
          "category: expected \"unknown\" got \"feature\" (r2)",
          "category: expected \"unknown\" got \"account\" (r3)"
        ],
        "case-24": [
          "category: expected \"billing\" got \"unknown\" (r2)"
        ]
      },
      "latency_ms": {
        "p50": 959,
        "p95": 1273,
        "n": 87
      },
      "usage": {
        "prompt_tokens": 26373,
        "uncached_prompt_tokens": 26373,
        "cached_prompt_tokens": 0,
        "cache_write_tokens": 0,
        "completion_tokens": 3251,
        "reasoning_tokens_within_completion": 97,
        "calls_with_cache_hit": "0/87",
        "finish_reasons": {
          "stop": 87
        }
      },
      "rates_usd_per_1m": {
        "input": 2,
        "output": 12,
        "cache_read": 0.2,
        "cache_write": 2.5
      },
      "calculated_list_cost_usd": 0.0918,
      "gateway_reported_charge_usd": 0.0918,
      "cost_per_acceptable_result_usd": 0.00119
    }
  },
  "case_agreement": {
    "both_pass_all_reps": 23,
    "a_only_pass_all_reps": 4,
    "b_only_pass_all_reps": 1,
    "neither_pass_all_reps": 2,
    "unstable_across_reps": [
      "case-11:B",
      "case-20:A",
      "case-23:B",
      "case-24:B",
      "case-29:A"
    ]
  },
  "label": "Paired-run provenance validated. Outputs are caller-supplied; their model origin is not independently verified. Results apply only to this synthetic fixture, not general quality or production safety."
}
