{
  "run": {
    "run_id": "gateway-matched-v2-pinned-40-luna",
    "provider": "Vercel AI Gateway",
    "simulated": false
  },
  "model_id": "openai/gpt-5.6-luna",
  "unique_cases": 40,
  "coverage": {
    "valid_cases": 39,
    "fraction": 0.975,
    "missing_case_ids": [],
    "failed_cases": {
      "call-040": "API error: {\"name\":\"GatewayResponseError\",\"status\":500,\"message\":\"Invalid error response format: Gateway request failed: The operation was aborted due to timeout\"}"
    }
  },
  "overall_accuracy": 0.95,
  "latency_ms": {
    "scope": "first observation per case, including timed failures; no imputation",
    "observed_cases": 40,
    "coverage": 1.0,
    "p50": 1803.6192499999997,
    "p95": 2461.2340500000005
  },
  "questions": {
    "appointment_newly_booked": {
      "accuracy": 0.95,
      "per_class": {
        "yes": {
          "precision": 1.0,
          "recall": 1.0,
          "f1": 1.0,
          "support": 9,
          "tp": 9,
          "fp": 0,
          "fn": 0
        },
        "no": {
          "precision": 0.9629629629629629,
          "recall": 0.9629629629629629,
          "f1": 0.9629629629629629,
          "support": 27,
          "tp": 26,
          "fp": 1,
          "fn": 1
        },
        "unclear": {
          "precision": 1.0,
          "recall": 0.75,
          "f1": 0.8571428571428571,
          "support": 4,
          "tp": 3,
          "fp": 0,
          "fn": 1
        }
      },
      "macro": {
        "precision": 0.9876543209876543,
        "recall": 0.9043209876543209,
        "f1": 0.9400352733686067
      },
      "confusion_expected_rows_predicted_columns": {
        "yes": {
          "yes": 9,
          "no": 0,
          "unclear": 0,
          "failure": 0
        },
        "no": {
          "yes": 0,
          "no": 26,
          "unclear": 0,
          "failure": 1
        },
        "unclear": {
          "yes": 0,
          "no": 1,
          "unclear": 3,
          "failure": 0
        }
      },
      "brier": {
        "mean": null,
        "cases": 0,
        "coverage": 0.0,
        "definition": "sum squared error over 3 classes; range 0..2"
      },
      "threshold": {
        "value": 0.8,
        "selected_cases": 0,
        "coverage_all_cases": 0.0,
        "accuracy_selected": null,
        "basis": "probability assigned to returned label; vendor confidence unused"
      }
    },
    "commercial_sales_discussion": {
      "accuracy": 0.975,
      "per_class": {
        "yes": {
          "precision": 1.0,
          "recall": 1.0,
          "f1": 1.0,
          "support": 16,
          "tp": 16,
          "fp": 0,
          "fn": 0
        },
        "no": {
          "precision": 1.0,
          "recall": 0.95,
          "f1": 0.9743589743589743,
          "support": 20,
          "tp": 19,
          "fp": 0,
          "fn": 1
        },
        "unclear": {
          "precision": 1.0,
          "recall": 1.0,
          "f1": 1.0,
          "support": 4,
          "tp": 4,
          "fp": 0,
          "fn": 0
        }
      },
      "macro": {
        "precision": 1.0,
        "recall": 0.9833333333333334,
        "f1": 0.9914529914529915
      },
      "confusion_expected_rows_predicted_columns": {
        "yes": {
          "yes": 16,
          "no": 0,
          "unclear": 0,
          "failure": 0
        },
        "no": {
          "yes": 0,
          "no": 19,
          "unclear": 0,
          "failure": 1
        },
        "unclear": {
          "yes": 0,
          "no": 0,
          "unclear": 4,
          "failure": 0
        }
      },
      "brier": {
        "mean": null,
        "cases": 0,
        "coverage": 0.0,
        "definition": "sum squared error over 3 classes; range 0..2"
      },
      "threshold": {
        "value": 0.8,
        "selected_cases": 0,
        "coverage_all_cases": 0.0,
        "accuracy_selected": null,
        "basis": "probability assigned to returned label; vendor confidence unused"
      }
    }
  },
  "false_appointment_booked_positives": {
    "count": 0,
    "cases": [],
    "by_expected_label": {
      "no": 0,
      "unclear": 0
    }
  },
  "repeats": {
    "policy": "first observation only for all primary metrics; repeats measure stability",
    "extra_rows": 0,
    "cases": {}
  },
  "freeze": {
    "algorithm": "sha256",
    "frozen_date": "2026-09-17",
    "fixture_version": "2026-09-17-v1",
    "files": {
      "cases.json": "fd1aa514f6d9fa3224542391637791121b521dcd5de1128b2dedfc9dcc8371a6",
      "rubric.md": "bce01e8bf6ef5e711ec379cd5c621abb3272f2799833aab7315c31c213cb1964",
      "prompt.txt": "08638a6a301c6041a3569eb251a6844533d8796f3c8d6e7328d9f2ccc5e9bf60"
    }
  }
}