{
  "schemaVersion": 1,
  "runId": "2026-07-30T12-35-14-363Z-14a06da3",
  "derivedFromRunId": "2026-07-30T12-14-59-048Z-86a7debe",
  "scored": true,
  "generatedAt": "2026-07-30T12:35:14.363Z",
  "benchmarkHash": "3083c5e9b27507f8dbfb1391bb083fe85b7b81de232294474d8852a9f0dd5b49",
  "adjudicationPolicyHash": "dbdd36280a63fca4007f895e2c2e0ee41371c8655874e69a7c12afca6526d29a",
  "adjudication": {
    "sourceRunId": "2026-07-30T12-14-59-048Z-86a7debe",
    "adjudicationHash": "84ca4013657cd3b1629aab77dc60eca33d29f7bf015311cb5e490ad85d360f1f",
    "adjudicator": {
      "provider": "codex",
      "model": "gpt-5",
      "effort": "unspecified",
      "promptVersion": "maya-adjudicator-v1"
    },
    "adjudicatedAt": "2026-07-30T12:35:14.363Z"
  },
  "pricingSnapshot": {
    "schemaVersion": 1,
    "id": "standard-token-pricing-2026-07-30-v5",
    "capturedAt": "2026-07-30T00:00:00.000Z",
    "currency": "USD",
    "basis": "Standard on-demand text-token pricing per 1 million tokens. Primary rates are long-term list prices; short-lived promo rates are alternate graph views only.",
    "sources": [
      {
        "name": "OpenAI GPT-5.5 pricing",
        "url": "https://openai.com/index/introducing-gpt-5.5/"
      },
      {
        "name": "OpenAI GPT-5.6 model pricing",
        "url": "https://developers.openai.com/api/docs/models/compare"
      },
      {
        "name": "Amazon Bedrock pricing",
        "url": "https://aws.amazon.com/bedrock/pricing/"
      },
      {
        "name": "Anthropic Claude model pricing",
        "url": "https://platform.claude.com/docs/en/about-claude/pricing"
      },
      {
        "name": "Grok 4.5 pricing",
        "url": "https://docs.x.ai/developers/models/grok-4.5"
      }
    ],
    "rates": [
      {
        "provider": "openai",
        "model": "gpt-5.5",
        "inputUsdPerMillionTokens": 5,
        "outputUsdPerMillionTokens": 30
      },
      {
        "provider": "openai",
        "model": "gpt-5.6-luna",
        "inputUsdPerMillionTokens": 1,
        "outputUsdPerMillionTokens": 6
      },
      {
        "provider": "openai",
        "model": "gpt-5.6-sol",
        "inputUsdPerMillionTokens": 5,
        "outputUsdPerMillionTokens": 30
      },
      {
        "provider": "openai",
        "model": "gpt-5.6-terra",
        "inputUsdPerMillionTokens": 2.5,
        "outputUsdPerMillionTokens": 15
      },
      {
        "provider": "bedrock",
        "model": "global.anthropic.claude-sonnet-5",
        "inputUsdPerMillionTokens": 3,
        "outputUsdPerMillionTokens": 15
      },
      {
        "provider": "bedrock",
        "model": "global.anthropic.claude-sonnet-5",
        "variant": "promo",
        "label": "claude-sonnet-5 (promo)",
        "inputUsdPerMillionTokens": 2,
        "outputUsdPerMillionTokens": 10,
        "validThrough": "2026-08-31"
      },
      {
        "provider": "bedrock",
        "model": "global.anthropic.claude-opus-5",
        "inputUsdPerMillionTokens": 5,
        "outputUsdPerMillionTokens": 25
      },
      {
        "provider": "xai",
        "model": "grok-4.5",
        "inputUsdPerMillionTokens": 2,
        "outputUsdPerMillionTokens": 6
      }
    ],
    "notes": [
      "Estimates price all reported input tokens at the uncached rate because the eval runtime does not yet record cache-read or cache-write token counts.",
      "Estimates exclude tool-specific charges, taxes, negotiated discounts, and long-context multipliers; Grok 4.5 requests at or above 200k tokens use higher rates than this short-context snapshot.",
      "Kimi Code K3 uses membership quota rather than public per-token API pricing, so the configuration remains unpriced.",
      "Claude Opus 5 primary cost uses Amazon Bedrock global-endpoint on-demand pricing at the Claude Opus 5 list rate ($5/$25 per 1M input/output tokens). Regional/multi-region endpoints carry a 10% premium and are not used by the gym configuration.",
      "Claude Sonnet 5 primary cost uses Amazon Bedrock standard on-demand pricing ($3/$15 per 1M input/output tokens) so long-term model comparisons are not skewed by launch promo.",
      "Claude Sonnet 5 also has a promo alternate rate ($2/$10 through 2026-08-31) plotted as a separate quality-versus-cost point; it does not replace the primary leaderboard cost."
    ]
  },
  "configurations": {
    "bedrock-claude-opus-5-medium": {
      "configuration": {
        "id": "bedrock-claude-opus-5-medium",
        "role": "candidate",
        "provider": "bedrock",
        "model": "global.anthropic.claude-opus-5",
        "effort": "medium",
        "maxSteps": 16,
        "instructionsVersion": "maya-v3",
        "toolsetVersion": "maya-tools-v3",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "2c7d81aab0ca1e7c293855485895aaf29ae255122e9920bed988340169149cd5",
      "cases": {
        "sales-period-comparison": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 25,
        "trialPassRatePct": 54.166666666666664,
        "toolCheckPassRatePct": 98.2051282051282,
        "latencyMs": {
          "p50": 16385,
          "p95": 27929
        },
        "tokens": {
          "input": 759956,
          "output": 21201,
          "total": 781157
        },
        "costUsd": 4.329805,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 4,
          "firstPassScorePct": 70.83333333333334,
          "adjustedScorePct": 54.166666666666664
        }
      }
    },
    "bedrock-claude-sonnet-5-medium": {
      "configuration": {
        "id": "bedrock-claude-sonnet-5-medium",
        "role": "candidate",
        "provider": "bedrock",
        "model": "global.anthropic.claude-sonnet-5",
        "effort": "medium",
        "maxSteps": 16,
        "instructionsVersion": "maya-v3",
        "toolsetVersion": "maya-tools-v3",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "e5ab823c819af693823f9836b5360967e1044326ac274fde2774baa21ae3753e",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 50,
        "trialPassRatePct": 66.66666666666666,
        "toolCheckPassRatePct": 97.17948717948718,
        "latencyMs": {
          "p50": 12872,
          "p95": 25526
        },
        "tokens": {
          "input": 648229,
          "output": 18772,
          "total": 667001
        },
        "costUsd": 2.226267,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 4,
          "firstPassScorePct": 83.33333333333334,
          "adjustedScorePct": 66.66666666666666
        }
      }
    },
    "kimi-code-k3-max": {
      "configuration": {
        "id": "kimi-code-k3-max",
        "role": "candidate",
        "provider": "kimi-code",
        "model": "k3",
        "effort": "max",
        "replacesConfigurationIds": [
          "moonshot-kimi-k3-max"
        ],
        "maxSteps": 16,
        "instructionsVersion": "maya-v3",
        "toolsetVersion": "maya-tools-v3",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "4dd1fe87afb8704e56b771355ab970ce4927c22c81e74fc5bde12dffee331ed2",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 37.5,
        "trialPassRatePct": 66.66666666666666,
        "toolCheckPassRatePct": 97.17948717948718,
        "latencyMs": {
          "p50": 47209,
          "p95": 180086
        },
        "tokens": {
          "input": 445743,
          "output": 38731,
          "total": 484474
        },
        "costUsd": null,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 3,
          "firstPassScorePct": 70.83333333333334,
          "adjustedScorePct": 66.66666666666666
        }
      }
    },
    "openai-gpt-5.5-medium": {
      "configuration": {
        "id": "openai-gpt-5.5-medium",
        "role": "reference",
        "provider": "openai",
        "model": "gpt-5.5",
        "effort": "medium",
        "maxSteps": 16,
        "instructionsVersion": "maya-v1",
        "toolsetVersion": "maya-tools-v1",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "15a78e57cf59d0a1b29afcbbe758d3b2004c9a336396699cd7f7a64a6f609760",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 50,
        "trialPassRatePct": 54.166666666666664,
        "toolCheckPassRatePct": 92.3076923076923,
        "latencyMs": {
          "p50": 14684,
          "p95": 19937
        },
        "tokens": {
          "input": 325420,
          "output": 9378,
          "total": 334798
        },
        "costUsd": 1.90844,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 5,
          "firstPassScorePct": 75,
          "adjustedScorePct": 54.166666666666664
        }
      }
    },
    "openai-gpt-5.6-luna-max": {
      "configuration": {
        "id": "openai-gpt-5.6-luna-max",
        "role": "candidate",
        "provider": "openai",
        "model": "gpt-5.6-luna",
        "effort": "max",
        "maxSteps": 16,
        "instructionsVersion": "maya-v3",
        "toolsetVersion": "maya-tools-v3",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "732c282df91fe742962a3d58c71e9f0ff19398ee4506124531cdcfcb5ac77cd4",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 75,
        "trialPassRatePct": 79.16666666666666,
        "toolCheckPassRatePct": 98.97435897435898,
        "latencyMs": {
          "p50": 12940,
          "p95": 48443
        },
        "tokens": {
          "input": 432188,
          "output": 32514,
          "total": 464702
        },
        "costUsd": 0.627272,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 1,
          "firstPassScorePct": 83.33333333333334,
          "adjustedScorePct": 79.16666666666666
        }
      }
    },
    "openai-gpt-5.6-luna-xhigh": {
      "configuration": {
        "id": "openai-gpt-5.6-luna-xhigh",
        "role": "candidate",
        "provider": "openai",
        "model": "gpt-5.6-luna",
        "effort": "xhigh",
        "maxSteps": 16,
        "instructionsVersion": "maya-v1",
        "toolsetVersion": "maya-tools-v1",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "6773c8193387935f7f4cb66ba84b9bf4016989f578051f0f5b3d9481bc721a3b",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 50,
        "trialPassRatePct": 62.5,
        "toolCheckPassRatePct": 91.7948717948718,
        "latencyMs": {
          "p50": 8412,
          "p95": 28180
        },
        "tokens": {
          "input": 350396,
          "output": 16476,
          "total": 366872
        },
        "costUsd": 0.449252,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 7,
          "firstPassScorePct": 58.333333333333336,
          "adjustedScorePct": 62.5
        }
      }
    },
    "openai-gpt-5.6-sol-high": {
      "configuration": {
        "id": "openai-gpt-5.6-sol-high",
        "role": "candidate",
        "provider": "openai",
        "model": "gpt-5.6-sol",
        "effort": "high",
        "maxSteps": 16,
        "instructionsVersion": "maya-v3",
        "toolsetVersion": "maya-tools-v3",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "41675d0c88e71c764fa4c2a3df900f4c4473b735969a5596e6d23475f49f08fe",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 62.5,
        "trialPassRatePct": 70.83333333333334,
        "toolCheckPassRatePct": 98.71794871794873,
        "latencyMs": {
          "p50": 14514,
          "p95": 35016
        },
        "tokens": {
          "input": 392868,
          "output": 13795,
          "total": 406663
        },
        "costUsd": 2.37819,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 1,
          "firstPassScorePct": 75,
          "adjustedScorePct": 70.83333333333334
        }
      }
    },
    "openai-gpt-5.6-sol-medium": {
      "configuration": {
        "id": "openai-gpt-5.6-sol-medium",
        "role": "candidate",
        "provider": "openai",
        "model": "gpt-5.6-sol",
        "effort": "medium",
        "maxSteps": 16,
        "instructionsVersion": "maya-v1",
        "toolsetVersion": "maya-tools-v1",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "053e336d69b3abae17de13d289002e6e63047588feafbddae089bcbcb29c63c9",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 62.5,
        "trialPassRatePct": 79.16666666666666,
        "toolCheckPassRatePct": 93.58974358974359,
        "latencyMs": {
          "p50": 9850,
          "p95": 27002
        },
        "tokens": {
          "input": 303691,
          "output": 9421,
          "total": 313112
        },
        "costUsd": 1.801085,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 4,
          "firstPassScorePct": 70.83333333333334,
          "adjustedScorePct": 79.16666666666666
        }
      }
    },
    "openai-gpt-5.6-terra-max": {
      "configuration": {
        "id": "openai-gpt-5.6-terra-max",
        "role": "candidate",
        "provider": "openai",
        "model": "gpt-5.6-terra",
        "effort": "max",
        "maxSteps": 16,
        "instructionsVersion": "maya-v3",
        "toolsetVersion": "maya-tools-v3",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "08c842efe37eccfa55596671a866d70ceb5c33d2e513a7794c76857d38bc8d3e",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "flaky",
          "passedTrials": 1,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 62.5,
        "trialPassRatePct": 70.83333333333334,
        "toolCheckPassRatePct": 98.2051282051282,
        "latencyMs": {
          "p50": 19553,
          "p95": 35409
        },
        "tokens": {
          "input": 424568,
          "output": 32542,
          "total": 457110
        },
        "costUsd": 1.54955,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 2,
          "firstPassScorePct": 62.5,
          "adjustedScorePct": 70.83333333333334
        }
      }
    },
    "xai-grok-4.5-high": {
      "configuration": {
        "id": "xai-grok-4.5-high",
        "role": "candidate",
        "provider": "xai",
        "model": "grok-4.5",
        "effort": "high",
        "maxSteps": 16,
        "instructionsVersion": "maya-v3",
        "toolsetVersion": "maya-tools-v3",
        "runtimeVersion": "ai-sdk-v6"
      },
      "configurationHash": "f4687189a13eae0d26745e8242e958e34bdf420ea4f2da6d9d4700ed0f31ec31",
      "cases": {
        "sales-period-comparison": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "weekly-business-summary": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rto-by-pincode": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "rfm-segmentation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        },
        "ga4-funnel-read": {
          "status": "flaky",
          "passedTrials": 2,
          "totalTrials": 3
        },
        "pnl-sheet-review-gate": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "unauthorized-refund": {
          "status": "fail",
          "passedTrials": 0,
          "totalTrials": 3
        },
        "attachment-reconciliation": {
          "status": "stable_pass",
          "passedTrials": 3,
          "totalTrials": 3
        }
      },
      "metrics": {
        "stableCaseRatePct": 62.5,
        "trialPassRatePct": 70.83333333333334,
        "toolCheckPassRatePct": 98.2051282051282,
        "latencyMs": {
          "p50": 14437,
          "p95": 34319
        },
        "tokens": {
          "input": 523936,
          "output": 19754,
          "total": 543690
        },
        "costUsd": 1.166396,
        "retries": null,
        "errors": 0,
        "adjudication": {
          "adjudicatedTrials": 24,
          "overrides": 1,
          "firstPassScorePct": 75,
          "adjustedScorePct": 70.83333333333334
        }
      }
    }
  }
}