{
 "bench": "RouteBench",
 "version": "1.2",
 "dataset": "tasks-v1",
 "date": "2026-10-08",
 "run": "2026-10-08-3",
 "testTasks": 162,
 "calibTasks": 90,
 "config": {
  "routers": [
   "frontier",
   "sonnet",
   "haiku",
   "deepseek",
   "or-auto",
   "or-auto-medium",
   "or-auto-high",
   "jev",
   "clef-route",
   "clef-cascade",
   "clef-flash-route",
   "clef-flash-cascade",
   "luna-route",
   "luna-cascade",
   "jev-route",
   "jev-cascade",
   "gin",
   "gin-x-robust"
  ],
  "limit": null,
  "families": [
   "code-review",
   "security-review",
   "json-extraction",
   "classification",
   "tool-calling",
   "math",
   "summarization",
   "kids-story",
   "support-chat"
  ],
  "frontier": "anthropic/claude-opus-5.5",
  "pairwiseJudge": "openai/gpt-5.5",
  "ginReplayJudge": "google/gemini-3.8-flash",
  "ginPairwiseJudge": "openai/gpt-5.5",
  "ginBalanceUsd": 10,
  "budgetUsd": 2
 },
 "spendUsd": 24.40198562346592,
 "calibrationUsd": 6.588463907288002,
 "families": [
  {
   "id": "code-review",
   "label": "Code review (seeded bugs)",
   "kind": "verified"
  },
  {
   "id": "security-review",
   "label": "Security review",
   "kind": "verified"
  },
  {
   "id": "json-extraction",
   "label": "JSON extraction",
   "kind": "verified"
  },
  {
   "id": "classification",
   "label": "Ticket classification",
   "kind": "verified"
  },
  {
   "id": "tool-calling",
   "label": "Tool calling",
   "kind": "verified"
  },
  {
   "id": "math",
   "label": "Math & reasoning",
   "kind": "verified"
  },
  {
   "id": "summarization",
   "label": "Summarization",
   "kind": "judged"
  },
  {
   "id": "kids-story",
   "label": "Kids' stories",
   "kind": "judged"
  },
  {
   "id": "support-chat",
   "label": "Support chat",
   "kind": "judged"
  }
 ],
 "routers": [
  {
   "id": "frontier",
   "label": "Always Opus 5.5",
   "kind": "model",
   "note": "Every call on the frontier model (also the quality reference).",
   "n": 162,
   "quality": 0.8950617283950617,
   "usdPer1k": 9.868098765432098,
   "p50ms": 2376,
   "p95ms": 4575,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9907407407407407,
    "usdPer1k": 5.195777777777777,
    "p50ms": 2226,
    "p95ms": 4329,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.7037037037037037,
    "usdPer1k": 19.21274074074075,
    "p50ms": 2905,
    "p95ms": 5346,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 2.729333333333333,
     "p50ms": 3246,
     "p95ms": 6512,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 3.5099999999999993,
     "p50ms": 2317,
     "p95ms": 3843,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 3.9166666666666665,
     "p50ms": 2280,
     "p95ms": 4745,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.1222222222222222,
     "p50ms": 2196,
     "p95ms": 4114,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 7.538222222222222,
     "p50ms": 1659,
     "p95ms": 3396,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 12.358222222222222,
     "p50ms": 1871,
     "p95ms": 4329,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 17.303111111111114,
     "p50ms": 2483,
     "p95ms": 5087,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 34.37466666666666,
     "p50ms": 3352,
     "p95ms": 6361,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 5.960444444444445,
     "p50ms": 2490,
     "p95ms": 6679,
     "errors": 0
    }
   },
   "mix": {
    "anthropic/claude-opus-5.5": 162
   }
  },
  {
   "id": "sonnet",
   "label": "Always Sonnet 5.5",
   "kind": "model",
   "note": "Every call on the mid-tier Claude.",
   "n": 162,
   "quality": 0.8641975308641975,
   "usdPer1k": 3.630530864197531,
   "p50ms": 990,
   "p95ms": 2530,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 1,
    "usdPer1k": 2.279833333333333,
    "p50ms": 1016,
    "p95ms": 2848,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.5925925925925926,
    "usdPer1k": 6.331925925925924,
    "p50ms": 928,
    "p95ms": 1582,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.4152222222222222,
     "p50ms": 920,
     "p95ms": 2530,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.75,
     "p50ms": 1170,
     "p95ms": 6469,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.6283333333333334,
     "p50ms": 1078,
     "p95ms": 4151,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.3433333333333332,
     "p50ms": 1276,
     "p95ms": 1504,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 3.634666666666667,
     "p50ms": 1014,
     "p95ms": 3032,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 4.907444444444446,
     "p50ms": 897,
     "p95ms": 1050,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.4444444444444444,
     "usdPer1k": 5.872666666666666,
     "p50ms": 916,
     "p95ms": 1582,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.4444444444444444,
     "usdPer1k": 10.062888888888889,
     "p50ms": 914,
     "p95ms": 1854,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.8888888888888888,
     "usdPer1k": 3.0602222222222224,
     "p50ms": 969,
     "p95ms": 4548,
     "errors": 0
    }
   },
   "mix": {
    "anthropic/claude-sonnet-5.5": 162
   }
  },
  {
   "id": "haiku",
   "label": "Always Haiku 5.5",
   "kind": "model",
   "note": "Every call on the cheapest Claude.",
   "n": 162,
   "quality": 0.8395061728395061,
   "usdPer1k": 0.27747098765432093,
   "p50ms": 913,
   "p95ms": 1278,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 1,
    "usdPer1k": 0.16168611111111103,
    "p50ms": 921,
    "p95ms": 1599,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.5185185185185185,
    "usdPer1k": 0.5090407407407408,
    "p50ms": 893,
    "p95ms": 1182,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.10406666666666664,
     "p50ms": 913,
     "p95ms": 1599,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.09275,
     "p50ms": 914,
     "p95ms": 1625,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.1376666666666667,
     "p50ms": 991,
     "p95ms": 2819,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.018277777777777775,
     "p50ms": 925,
     "p95ms": 1225,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.1904,
     "p50ms": 932,
     "p95ms": 1221,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.4269555555555555,
     "p50ms": 893,
     "p95ms": 1576,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 0.3997722222222222,
     "p50ms": 929,
     "p95ms": 1093,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 1.0218388888888892,
     "p50ms": 887,
     "p95ms": 1278,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 0.10551111111111111,
     "p50ms": 891,
     "p95ms": 1182,
     "errors": 0
    }
   },
   "mix": {
    "anthropic/claude-haiku-5.5": 162
   }
  },
  {
   "id": "deepseek",
   "label": "Always DeepSeek V4.1 Flash",
   "kind": "model",
   "note": "Every call on a cheap open-weights model.",
   "n": 162,
   "quality": 0.845679012345679,
   "usdPer1k": 0.5134815691358022,
   "p50ms": 494,
   "p95ms": 1905,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9722222222222222,
    "usdPer1k": 0.36311226296296273,
    "p50ms": 477,
    "p95ms": 1586,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.5925925925925926,
    "usdPer1k": 0.8142201814814815,
    "p50ms": 506,
    "p95ms": 2360,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.16495528888888888,
     "p50ms": 537,
     "p95ms": 1298,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.412997288888889,
     "p50ms": 435,
     "p95ms": 1302,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.38742250000000006,
     "p50ms": 633,
     "p95ms": 2407,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.1926762555555556,
     "p50ms": 444,
     "p95ms": 1967,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.2091778555555556,
     "p50ms": 798,
     "p95ms": 1543,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.811444388888889,
     "p50ms": 488,
     "p95ms": 1586,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.5372534333333332,
     "p50ms": 477,
     "p95ms": 2933,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 1.558304888888889,
     "p50ms": 652,
     "p95ms": 2940,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 0.34710222222222226,
     "p50ms": 557,
     "p95ms": 1050,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 162
   }
  },
  {
   "id": "or-auto",
   "label": "OpenRouter Auto",
   "kind": "router",
   "note": "`openrouter/auto`, default cost tier (≈ low).",
   "n": 162,
   "quality": 0.7469135802469136,
   "usdPer1k": 0.4233662407407408,
   "p50ms": 2422,
   "p95ms": 11226,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9814814814814815,
    "usdPer1k": 0.3343938981481481,
    "p50ms": 2370,
    "p95ms": 11372,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.2777777777777778,
    "usdPer1k": 0.6013109259259257,
    "p50ms": 2568,
    "p95ms": 10848,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.21025,
     "p50ms": 2419,
     "p95ms": 3285,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.44798333333333334,
     "p50ms": 2019,
     "p95ms": 3473,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.21414027777777775,
     "p50ms": 2205,
     "p95ms": 10343,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.02501111111111111,
     "p50ms": 3805,
     "p95ms": 11226,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.16167866666666666,
     "p50ms": 8906,
     "p95ms": 13027,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.9473000000000001,
     "p50ms": 2063,
     "p95ms": 2682,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.3333333333333333,
     "usdPer1k": 0.7171152777777777,
     "p50ms": 2258,
     "p95ms": 5607,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.2777777777777778,
     "usdPer1k": 0.9112791666666665,
     "p50ms": 2258,
     "p95ms": 3373,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.2222222222222222,
     "usdPer1k": 0.1755383333333333,
     "p50ms": 9416,
     "p95ms": 12245,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 58,
    "z-ai/glm-5.3-flash": 54,
    "google/gemini-2.5-flash": 43,
    "deepseek/deepseek-v4-flash-0731": 3,
    "openai/gpt-6-luna": 4
   }
  },
  {
   "id": "or-auto-medium",
   "label": "OpenRouter Auto (medium)",
   "kind": "router",
   "note": "`openrouter/auto` with `cost_tier: medium`.",
   "n": 162,
   "quality": 0.8518518518518519,
   "usdPer1k": 2.150305728395061,
   "p50ms": 1828,
   "p95ms": 4021,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 1,
    "usdPer1k": 1.7821989259259263,
    "p50ms": 756,
    "p95ms": 4021,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.5555555555555556,
    "usdPer1k": 2.8865193333333337,
    "p50ms": 2197,
    "p95ms": 4200,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.1203283333333331,
     "p50ms": 402,
     "p95ms": 490,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.3827194444444444,
     "p50ms": 570,
     "p95ms": 932,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 2.5963749999999997,
     "p50ms": 2494,
     "p95ms": 7362,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.7228333333333335,
     "p50ms": 2472,
     "p95ms": 3828,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.0818263333333333,
     "p50ms": 2008,
     "p95ms": 4607,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 3.789111111111111,
     "p50ms": 634,
     "p95ms": 1915,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 0.7894746666666667,
     "p50ms": 592,
     "p95ms": 728,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.3333333333333333,
     "usdPer1k": 6.041291666666667,
     "p50ms": 2365,
     "p95ms": 4526,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.6111111111111112,
     "usdPer1k": 1.828791666666667,
     "p50ms": 2344,
     "p95ms": 5860,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4-pro-0813": 18,
    "z-ai/glm-5.3-flash": 10,
    "z-ai/glm-5.3": 8,
    "google/gemini-3.8-flash": 85,
    "deepseek/deepseek-v4.1-flash": 22,
    "z-ai/glm-5.2": 1,
    "openai/gpt-6.1-sol": 18
   }
  },
  {
   "id": "or-auto-high",
   "label": "OpenRouter Auto (high)",
   "kind": "router",
   "note": "`openrouter/auto` with `cost_tier: high`.",
   "n": 162,
   "quality": 0.8950617283950617,
   "usdPer1k": 5.4493100617283945,
   "p50ms": 883,
   "p95ms": 2732,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 1,
    "usdPer1k": 5.4407618518518515,
    "p50ms": 883,
    "p95ms": 2596,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.6851851851851852,
    "usdPer1k": 5.466406481481482,
    "p50ms": 910,
    "p95ms": 2779,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.7939333333333334,
     "p50ms": 592,
     "p95ms": 727,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.3923933333333331,
     "p50ms": 539,
     "p95ms": 1620,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 2.7454166666666664,
     "p50ms": 2564,
     "p95ms": 6551,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.018805555555555555,
     "p50ms": 938,
     "p95ms": 1313,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.5234888888888891,
     "p50ms": 836,
     "p95ms": 1289,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 25.17053333333333,
     "p50ms": 971,
     "p95ms": 1158,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 4.216958333333334,
     "p50ms": 2483,
     "p95ms": 3355,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.6111111111111112,
     "usdPer1k": 12.076777777777778,
     "p50ms": 816,
     "p95ms": 1159,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 0.10548333333333335,
     "p50ms": 773,
     "p95ms": 1054,
     "errors": 0
    }
   },
   "mix": {
    "z-ai/glm-5.3": 29,
    "openai/gpt-6.1-sol": 8,
    "google/gemini-3.8-flash": 36,
    "anthropic/claude-haiku-5.5": 46,
    "anthropic/claude-sonnet-5.5": 25,
    "moonshotai/kimi-k3": 18
   }
  },
  {
   "id": "jev",
   "label": "TypeSafe Jev Router",
   "kind": "router",
   "note": "`typesafe/jev-router` on OpenRouter (picks model + reasoning effort).",
   "n": 162,
   "quality": 0.8641975308641975,
   "usdPer1k": 1.2299748148148144,
   "p50ms": 837,
   "p95ms": 2977,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9907407407407407,
    "usdPer1k": 1.2255359444444442,
    "p50ms": 896,
    "p95ms": 2998,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.6111111111111112,
    "usdPer1k": 1.2388525555555554,
    "p50ms": 557,
    "p95ms": 2977,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.2118916666666668,
     "p50ms": 896,
     "p95ms": 7203,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.09452222222222222,
     "p50ms": 901,
     "p95ms": 1860,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.7657944444444442,
     "p50ms": 1224,
     "p95ms": 2701,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.05985555555555556,
     "p50ms": 643,
     "p95ms": 4931,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.24105455555555555,
     "p50ms": 812,
     "p95ms": 8048,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 4.9800972222222235,
     "p50ms": 2780,
     "p95ms": 6344,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.7777777777777778,
     "usdPer1k": 1.2996326666666667,
     "p50ms": 902,
     "p95ms": 6463,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 2.223775,
     "p50ms": 522,
     "p95ms": 2976,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.8888888888888888,
     "usdPer1k": 0.19315000000000002,
     "p50ms": 685,
     "p95ms": 10985,
     "errors": 0
    }
   },
   "mix": {
    "meta/muse-spark-1.3": 7,
    "deepseek/deepseek-v4.1-flash": 75,
    "openai/gpt-6.1-sol": 5,
    "openai/gpt-6-luna": 48,
    "google/gemini-3.8-flash": 27
   }
  },
  {
   "id": "clef-route",
   "label": "Cloudflare Clef → router",
   "kind": "decision",
   "note": "`cloudflare/clef` (Decisions API) picks DeepSeek V4.1 Flash, Sonnet 5.5 or Opus 5.5 for each call.",
   "n": 162,
   "quality": 0.8950617283950617,
   "usdPer1k": 2.11747011728395,
   "p50ms": 1133,
   "p95ms": 2565,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9907407407407407,
    "usdPer1k": 1.206588038888889,
    "p50ms": 1150,
    "p95ms": 2433,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.7037037037037037,
    "usdPer1k": 3.939234274074073,
    "p50ms": 1122,
    "p95ms": 2824,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.38056695555555564,
     "p50ms": 1750,
     "p95ms": 2433,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.8423034444444445,
     "p50ms": 808,
     "p95ms": 3308,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.4129928333333333,
     "p50ms": 1102,
     "p95ms": 2944,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.20966125555555556,
     "p50ms": 767,
     "p95ms": 2318,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.46542396666666663,
     "p50ms": 1128,
     "p95ms": 2690,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 4.928579777777778,
     "p50ms": 1202,
     "p95ms": 2081,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.5709907666666667,
     "p50ms": 801,
     "p95ms": 3346,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.4444444444444444,
     "usdPer1k": 10.078424222222221,
     "p50ms": 1398,
     "p95ms": 2824,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 1.168287833333333,
     "p50ms": 991,
     "p95ms": 4865,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 117,
    "anthropic/claude-sonnet-5.5": 43,
    "anthropic/claude-opus-5.5": 2
   }
  },
  {
   "id": "clef-cascade",
   "label": "DeepSeek → Cloudflare Clef check",
   "kind": "decision",
   "note": "DeepSeek V4.1 Flash answers; `cloudflare/clef` checks the draft and escalates to Opus 5.5 when P(shippable) < 0.5.",
   "n": 162,
   "quality": 0.8641975308641975,
   "usdPer1k": 0.8004391246913579,
   "p50ms": 868,
   "p95ms": 3189,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9907407407407407,
    "usdPer1k": 0.7080077074074075,
    "p50ms": 865,
    "p95ms": 3222,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.6111111111111112,
    "usdPer1k": 0.9853019592592593,
    "p50ms": 912,
    "p95ms": 3056,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.7023799555555557,
     "p50ms": 1047,
     "p95ms": 4718,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.8120202888888888,
     "p50ms": 793,
     "p95ms": 4656,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.9254961666666667,
     "p50ms": 1139,
     "p95ms": 3160,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.24698903333333336,
     "p50ms": 775,
     "p95ms": 3277,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.7213140777777778,
     "p50ms": 1191,
     "p95ms": 3210,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.8398467222222222,
     "p50ms": 851,
     "p95ms": 6743,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.5722564333333333,
     "p50ms": 778,
     "p95ms": 3189,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 1.5869755555555558,
     "p50ms": 956,
     "p95ms": 3206,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 0.7966738888888888,
     "p50ms": 912,
     "p95ms": 2841,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 153,
    "anthropic/claude-opus-5.5": 9
   }
  },
  {
   "id": "clef-flash-route",
   "label": "Cloudflare Clef Flash → router",
   "kind": "decision",
   "note": "`cloudflare/clef-flash` (Decisions API) picks DeepSeek V4.1 Flash, Sonnet 5.5 or Opus 5.5 for each call.",
   "n": 162,
   "quality": 0.8888888888888888,
   "usdPer1k": 1.929997945679012,
   "p50ms": 916,
   "p95ms": 2003,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9814814814814815,
    "usdPer1k": 0.9696861333333333,
    "p50ms": 884,
    "p95ms": 2003,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.7037037037037037,
    "usdPer1k": 3.85062157037037,
    "p50ms": 1155,
    "p95ms": 2190,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.17400045555555554,
     "p50ms": 845,
     "p95ms": 1638,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.42430645555555546,
     "p50ms": 717,
     "p95ms": 1637,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.3975841666666667,
     "p50ms": 903,
     "p95ms": 2692,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.19940325555555555,
     "p50ms": 722,
     "p95ms": 2242,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.4579981333333332,
     "p50ms": 1054,
     "p95ms": 1853,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 4.164824333333334,
     "p50ms": 1164,
     "p95ms": 2003,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.5482866000000001,
     "p50ms": 766,
     "p95ms": 3201,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.4444444444444444,
     "usdPer1k": 10.070656555555555,
     "p50ms": 1379,
     "p95ms": 2190,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 0.9329215555555557,
     "p50ms": 1037,
     "p95ms": 4854,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 128,
    "anthropic/claude-sonnet-5.5": 34
   }
  },
  {
   "id": "clef-flash-cascade",
   "label": "DeepSeek → Cloudflare Clef Flash check",
   "kind": "decision",
   "note": "DeepSeek V4.1 Flash answers; `cloudflare/clef-flash` checks the draft and escalates to Opus 5.5 when P(shippable) < 0.5.",
   "n": 162,
   "quality": 0.845679012345679,
   "usdPer1k": 0.9543530567901236,
   "p50ms": 915,
   "p95ms": 3593,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9907407407407407,
    "usdPer1k": 0.6838632351851853,
    "p50ms": 971,
    "p95ms": 3693,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.5555555555555556,
    "usdPer1k": 1.4953327,
    "p50ms": 837,
    "p95ms": 3165,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.3115151222222222,
     "p50ms": 3220,
     "p95ms": 4988,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.9671296777777778,
     "p50ms": 1021,
     "p95ms": 6198,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.40108950000000004,
     "p50ms": 910,
     "p95ms": 2669,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.2822545888888888,
     "p50ms": 866,
     "p95ms": 3509,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.3156226333333334,
     "p50ms": 1108,
     "p95ms": 3062,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.8255678888888889,
     "p50ms": 745,
     "p95ms": 1877,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 1.763121877777778,
     "p50ms": 755,
     "p95ms": 3165,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 1.5718770555555555,
     "p50ms": 900,
     "p95ms": 3381,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.5555555555555556,
     "usdPer1k": 1.1509991666666666,
     "p50ms": 915,
     "p95ms": 4266,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 152,
    "anthropic/claude-opus-5.5": 10
   }
  },
  {
   "id": "luna-route",
   "label": "OpenAI GPT-6 Luna Decisions → router",
   "kind": "decision",
   "note": "`openai/gpt-6-luna-decisions` (Decisions API) picks DeepSeek V4.1 Flash, Sonnet 5.5 or Opus 5.5 for each call.",
   "n": 162,
   "quality": 0.845679012345679,
   "usdPer1k": 0.5525161370370372,
   "p50ms": 702,
   "p95ms": 2116,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9722222222222222,
    "usdPer1k": 0.39991689259259267,
    "p50ms": 684,
    "p95ms": 1780,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.5925925925925926,
    "usdPer1k": 0.8577146259259256,
    "p50ms": 740,
    "p95ms": 2540,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.2055164,
     "p50ms": 801,
     "p95ms": 1548,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.45638617777777774,
     "p50ms": 624,
     "p95ms": 1521,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.4317113888888889,
     "p50ms": 833,
     "p95ms": 2589,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.22218181111111107,
     "p50ms": 672,
     "p95ms": 2174,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.24151118888888884,
     "p50ms": 1003,
     "p95ms": 1757,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.8421943888888889,
     "p50ms": 684,
     "p95ms": 1780,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.5860645444444444,
     "p50ms": 669,
     "p95ms": 3127,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 1.5923215555555557,
     "p50ms": 945,
     "p95ms": 3195,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 0.39475777777777776,
     "p50ms": 786,
     "p95ms": 1244,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 162
   }
  },
  {
   "id": "luna-cascade",
   "label": "DeepSeek → OpenAI GPT-6 Luna Decisions check",
   "kind": "decision",
   "note": "DeepSeek V4.1 Flash answers; `openai/gpt-6-luna-decisions` checks the draft and escalates to Opus 5.5 when P(shippable) < 0.5.",
   "n": 162,
   "quality": 0.8641975308641975,
   "usdPer1k": 1.09730934691358,
   "p50ms": 735,
   "p95ms": 2864,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9907407407407407,
    "usdPer1k": 0.8092483740740739,
    "p50ms": 716,
    "p95ms": 2864,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.6111111111111112,
    "usdPer1k": 1.6734312925925925,
    "p50ms": 742,
    "p95ms": 3145,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.3764441777777779,
     "p50ms": 801,
     "p95ms": 6391,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.8330472888888889,
     "p50ms": 609,
     "p95ms": 4481,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.4336947222222222,
     "p50ms": 828,
     "p95ms": 2586,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.3508595888888889,
     "p50ms": 716,
     "p95ms": 2864,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.0992667444444442,
     "p50ms": 1032,
     "p95ms": 3611,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 1.7621777222222228,
     "p50ms": 709,
     "p95ms": 4759,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.591008988888889,
     "p50ms": 665,
     "p95ms": 3145,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 3.7109993333333335,
     "p50ms": 1029,
     "p95ms": 3964,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.7222222222222222,
     "usdPer1k": 0.7182855555555557,
     "p50ms": 808,
     "p95ms": 2349,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 153,
    "anthropic/claude-opus-5.5": 9
   }
  },
  {
   "id": "jev-route",
   "label": "TypeSafe Jev 1.13 → router",
   "kind": "decision",
   "note": "`typesafe/jev-1.13` (Decisions API) picks DeepSeek V4.1 Flash, Sonnet 5.5 or Opus 5.5 for each call.",
   "n": 162,
   "quality": 0.8827160493827161,
   "usdPer1k": 1.5865700259259257,
   "p50ms": 773,
   "p95ms": 1729,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9814814814814815,
    "usdPer1k": 0.5416934481481481,
    "p50ms": 740,
    "p95ms": 1784,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.6851851851851852,
    "usdPer1k": 3.676323181481484,
    "p50ms": 918,
    "p95ms": 1667,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.19036995555555558,
     "p50ms": 751,
     "p95ms": 1481,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.4398796222222223,
     "p50ms": 659,
     "p95ms": 1637,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.41539683333333327,
     "p50ms": 834,
     "p95ms": 2569,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.2138745888888889,
     "p50ms": 658,
     "p95ms": 2158,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.2316991888888889,
     "p50ms": 988,
     "p95ms": 1729,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 1.7589404999999998,
     "p50ms": 844,
     "p95ms": 1784,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.5667234333333333,
     "p50ms": 657,
     "p95ms": 3138,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.4444444444444444,
     "usdPer1k": 10.086233888888891,
     "p50ms": 1112,
     "p95ms": 2037,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 0.37601222222222214,
     "p50ms": 773,
     "p95ms": 1353,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 141,
    "anthropic/claude-sonnet-5.5": 21
   }
  },
  {
   "id": "jev-cascade",
   "label": "DeepSeek → TypeSafe Jev 1.13 check",
   "kind": "decision",
   "note": "DeepSeek V4.1 Flash answers; `typesafe/jev-1.13` checks the draft and escalates to Opus 5.5 when P(shippable) < 0.5.",
   "n": 162,
   "quality": 0.8580246913580247,
   "usdPer1k": 0.7510556802469134,
   "p50ms": 700,
   "p95ms": 2581,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 0.9907407407407407,
    "usdPer1k": 0.7054703185185188,
    "p50ms": 679,
    "p95ms": 2614,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.5925925925925926,
    "usdPer1k": 0.8422264037037038,
    "p50ms": 717,
    "p95ms": 2581,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.5077139555555557,
     "p50ms": 829,
     "p95ms": 3029,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.8127692888888889,
     "p50ms": 602,
     "p95ms": 4511,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.4131428333333333,
     "p50ms": 851,
     "p95ms": 2614,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.20894425555555557,
     "p50ms": 607,
     "p95ms": 2138,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.7227464111111112,
     "p50ms": 989,
     "p95ms": 2996,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 1.5675051666666668,
     "p50ms": 665,
     "p95ms": 5059,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.9444444444444444,
     "usdPer1k": 0.5656221,
     "p50ms": 670,
     "p95ms": 3172,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.16666666666666666,
     "usdPer1k": 1.5870375555555556,
     "p50ms": 871,
     "p95ms": 3157,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 0.3740195555555555,
     "p50ms": 717,
     "p95ms": 1257,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 157,
    "anthropic/claude-opus-5.5": 5
   }
  },
  {
   "id": "gin",
   "label": "Gin",
   "kind": "gin",
   "note": "Production routing v3: v2 (profiled use cases, 3–6 rung ladder, pairwise GPT-5.5 judge for open-ended work, expected-cost choice, calibrated fallback) with DeepSeek V4.1 Flash leading every ladder but stories, and the pairwise judge prompt recalibrated against ground truth (no credit for extra detail the request didn't ask for).",
   "n": 162,
   "quality": 0.9320987654320988,
   "usdPer1k": 4.799643167765431,
   "p50ms": 860,
   "p95ms": 4108,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 1,
    "usdPer1k": 0.2194346590555555,
    "p50ms": 833,
    "p95ms": 4317,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.7962962962962963,
    "usdPer1k": 13.960060185185181,
    "p50ms": 1370,
    "p95ms": 3939,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.16495528888888888,
     "p50ms": 537,
     "p95ms": 1298,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.15154533333333334,
     "p50ms": 3047,
     "p95ms": 11234,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.1376666666666667,
     "p50ms": 991,
     "p95ms": 2819,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.018277777777777775,
     "p50ms": 925,
     "p95ms": 1225,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.119696221,
     "p50ms": 369,
     "p95ms": 1822,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.7244666666666666,
     "p50ms": 361,
     "p95ms": 976,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.8333333333333334,
     "usdPer1k": 5.802458333333333,
     "p50ms": 1336,
     "p95ms": 3754,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 34.37466666666666,
     "p50ms": 3352,
     "p95ms": 6361,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.8888888888888888,
     "usdPer1k": 1.7030555555555553,
     "p50ms": 531,
     "p95ms": 3939,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 18,
    "z-ai/glm-5.3-flash": 54,
    "anthropic/claude-haiku-5.5": 36,
    "anthropic/claude-opus-5.5": 26,
    "openai/gpt-5.4-mini": 10,
    "openai/gpt-5.4": 18
   },
   "policies": {
    "code-review": {
     "model": "deepseek/deepseek-v4.1-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.971178296398892,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Code review · difficulty 3 · JSON"
    },
    "security-review": {
     "model": "z-ai/glm-5.3-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9531428916535813,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "z-ai/glm-5.1",
      "deepseek/deepseek-v4-pro",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Security analysis · difficulty 2 · JSON"
    },
    "json-extraction": {
     "model": "anthropic/claude-haiku-5.5",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9719943240454076,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Extraction / JSON · difficulty 2 · JSON"
    },
    "classification": {
     "model": "anthropic/claude-haiku-5.5",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9847778850796717,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Classification · difficulty 1"
    },
    "tool-calling": {
     "model": "z-ai/glm-5.3-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.98598783341441,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "moonshotai/kimi-k3",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Planning / agentic tool use · difficulty 2 · tool calls · latency-sensitive"
    },
    "math": {
     "model": "z-ai/glm-5.3-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9574467273882652,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4-pro",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Other · difficulty 3"
    },
    "summarization": {
     "model": "openai/gpt-5.4-mini",
     "from": "anthropic/claude-opus-5.5",
     "centroids": [
      "BPsE9Pb7/PgBBvX4AgIB+grz/wXuDQD0BgILCqoCBAEI/A0B3f7z+vYAAAQM+/r69fv/AP8B/gUI/OsMAvz9AfX9+QAA+/kP+Qj+BPj9+AUK+/36/AcABP0IAQUM6//8B/sHB//7Buno+v38B/0JAQf9/f4D+P38BwUKDvgBDPw="
     ],
     "threshold": 0.84,
     "passRate": 0.9166666666666666,
     "rawPassRate": 0.5,
     "samples": 12,
     "savings": 0.5692219447320865,
     "kept": 0.5833333333333334,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "judge": "pairwise:openai/gpt-5.5",
     "profile": "Summarization · difficulty 2"
    },
    "kids-story": null,
    "support-chat": {
     "model": "openai/gpt-5.4",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 0.75,
     "rawPassRate": 0.75,
     "samples": 12,
     "savings": 0.7013813263025164,
     "kept": 0,
     "rungs": [
      "openai/gpt-5.4",
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "judge": "pairwise:openai/gpt-5.5",
     "profile": "Chat / assistant · difficulty 2 · latency-sensitive"
    }
   },
   "plans": {
    "code-review": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Code review · difficulty 3 · JSON"
    },
    "security-review": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "z-ai/glm-5.1",
      "deepseek/deepseek-v4-pro",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Security analysis · difficulty 2 · JSON"
    },
    "json-extraction": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "profile": "Extraction / JSON · difficulty 2 · JSON"
    },
    "classification": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "profile": "Classification · difficulty 1"
    },
    "tool-calling": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "moonshotai/kimi-k3",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Planning / agentic tool use · difficulty 2 · tool calls · latency-sensitive"
    },
    "math": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4-pro",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Other · difficulty 3"
    },
    "summarization": {
     "judge": "pairwise:openai/gpt-5.5",
     "samples": 12,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "profile": "Summarization · difficulty 2"
    },
    "kids-story": {
     "judge": "pairwise:openai/gpt-5.5",
     "samples": 12,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "profile": "Creative writing · difficulty 2 · latency-sensitive"
    },
    "support-chat": {
     "judge": "pairwise:openai/gpt-5.5",
     "samples": 12,
     "rungs": [
      "openai/gpt-5.4",
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "profile": "Chat / assistant · difficulty 2 · latency-sensitive"
    }
   },
   "routedShare": 0.8395061728395061,
   "hardPromptShare": 0.04938271604938271
  },
  {
   "id": "gin-x-robust",
   "label": "Gin v3 + position-robust verdict",
   "kind": "gin",
   "note": "v3 where a pairwise \"worse\" needs a loss in both orders (tested, not shipped: it cost 3.1 points).",
   "n": 162,
   "quality": 0.9012345679012346,
   "usdPer1k": 4.475527982580247,
   "p50ms": 914,
   "p95ms": 4882,
   "errors": 0,
   "verified": {
    "n": 108,
    "quality": 1,
    "usdPer1k": 0.2194346590555555,
    "p50ms": 833,
    "p95ms": 4317,
    "errors": 0
   },
   "judged": {
    "n": 54,
    "quality": 0.7037037037037037,
    "usdPer1k": 12.987714629629632,
    "p50ms": 2440,
    "p95ms": 6308,
    "errors": 0
   },
   "families": {
    "code-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.16495528888888888,
     "p50ms": 537,
     "p95ms": 1298,
     "errors": 0
    },
    "security-review": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.15154533333333334,
     "p50ms": 3047,
     "p95ms": 11234,
     "errors": 0
    },
    "json-extraction": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.1376666666666667,
     "p50ms": 991,
     "p95ms": 2819,
     "errors": 0
    },
    "classification": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.018277777777777775,
     "p50ms": 925,
     "p95ms": 1225,
     "errors": 0
    },
    "tool-calling": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.119696221,
     "p50ms": 369,
     "p95ms": 1822,
     "errors": 0
    },
    "math": {
     "n": 18,
     "quality": 1,
     "usdPer1k": 0.7244666666666666,
     "p50ms": 361,
     "p95ms": 976,
     "errors": 0
    },
    "summarization": {
     "n": 18,
     "quality": 0.7777777777777778,
     "usdPer1k": 4.241375,
     "p50ms": 2572,
     "p95ms": 12641,
     "errors": 0
    },
    "kids-story": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 34.37466666666666,
     "p50ms": 3352,
     "p95ms": 6361,
     "errors": 0
    },
    "support-chat": {
     "n": 18,
     "quality": 0.6666666666666666,
     "usdPer1k": 0.34710222222222226,
     "p50ms": 557,
     "p95ms": 1050,
     "errors": 0
    }
   },
   "mix": {
    "deepseek/deepseek-v4.1-flash": 36,
    "z-ai/glm-5.3-flash": 54,
    "anthropic/claude-haiku-5.5": 36,
    "google/gemini-3.8-flash": 18,
    "anthropic/claude-opus-5.5": 18
   },
   "policies": {
    "code-review": {
     "model": "deepseek/deepseek-v4.1-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.971178296398892,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Code review · difficulty 3 · JSON"
    },
    "security-review": {
     "model": "z-ai/glm-5.3-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9531428916535813,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "z-ai/glm-5.1",
      "deepseek/deepseek-v4-pro",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Security analysis · difficulty 2 · JSON"
    },
    "json-extraction": {
     "model": "anthropic/claude-haiku-5.5",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9719943240454076,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Extraction / JSON · difficulty 2 · JSON"
    },
    "classification": {
     "model": "anthropic/claude-haiku-5.5",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9847778850796717,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Classification · difficulty 1"
    },
    "tool-calling": {
     "model": "z-ai/glm-5.3-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.98598783341441,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "moonshotai/kimi-k3",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Planning / agentic tool use · difficulty 2 · tool calls · latency-sensitive"
    },
    "math": {
     "model": "z-ai/glm-5.3-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 1,
     "rawPassRate": 1,
     "samples": 10,
     "savings": 0.9574467273882652,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4-pro",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "judge": "google/gemini-3.8-flash",
     "profile": "Other · difficulty 3"
    },
    "summarization": {
     "model": "google/gemini-3.8-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 0.9166666666666666,
     "rawPassRate": 0.9166666666666666,
     "samples": 12,
     "savings": 0.7444760004872997,
     "kept": 0,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "judge": "pairwise:openai/gpt-5.5",
     "profile": "Summarization · difficulty 2"
    },
    "kids-story": null,
    "support-chat": {
     "model": "deepseek/deepseek-v4.1-flash",
     "from": "anthropic/claude-opus-5.5",
     "passRate": 0.9166666666666666,
     "rawPassRate": 0.9166666666666666,
     "samples": 12,
     "savings": 0.9377933885481745,
     "kept": 0,
     "rungs": [
      "openai/gpt-5.4",
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "judge": "pairwise:openai/gpt-5.5",
     "profile": "Chat / assistant · difficulty 2 · latency-sensitive"
    }
   },
   "plans": {
    "code-review": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Code review · difficulty 3 · JSON"
    },
    "security-review": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "z-ai/glm-5.1",
      "deepseek/deepseek-v4-pro",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Security analysis · difficulty 2 · JSON"
    },
    "json-extraction": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "profile": "Extraction / JSON · difficulty 2 · JSON"
    },
    "classification": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "deepseek/deepseek-v4.1-flash",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash",
      "anthropic/claude-haiku-5.5"
     ],
     "profile": "Classification · difficulty 1"
    },
    "tool-calling": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "moonshotai/kimi-k3",
      "deepseek/deepseek-v4.1-flash",
      "qwen/qwen3-coder",
      "openai/gpt-5.4-nano",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Planning / agentic tool use · difficulty 2 · tool calls · latency-sensitive"
    },
    "math": {
     "judge": "google/gemini-3.8-flash",
     "samples": 10,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "moonshotai/kimi-k2.6",
      "deepseek/deepseek-v4-pro",
      "deepseek/deepseek-v4.1-flash",
      "z-ai/glm-5.3-flash"
     ],
     "profile": "Other · difficulty 3"
    },
    "summarization": {
     "judge": "pairwise:openai/gpt-5.5",
     "samples": 12,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "profile": "Summarization · difficulty 2"
    },
    "kids-story": {
     "judge": "pairwise:openai/gpt-5.5",
     "samples": 12,
     "rungs": [
      "anthropic/claude-sonnet-5.5",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "profile": "Creative writing · difficulty 2 · latency-sensitive"
    },
    "support-chat": {
     "judge": "pairwise:openai/gpt-5.5",
     "samples": 12,
     "rungs": [
      "openai/gpt-5.4",
      "anthropic/claude-sonnet-5.5",
      "openai/gpt-5.4-mini",
      "google/gemini-3.8-flash",
      "deepseek/deepseek-v4.1-flash"
     ],
     "profile": "Chat / assistant · difficulty 2 · latency-sensitive"
    }
   },
   "routedShare": 0.8888888888888888,
   "hardPromptShare": 0
  }
 ],
 "calibration": {
  "anthropic/claude-sonnet-5.5": {
   "replays": 96,
   "judgedPass": 0.7708333333333334,
   "truthPass": 1,
   "judgeAgreesWithTruth": 0.9666666666666667
  },
  "openai/gpt-5.4-mini": {
   "replays": 44,
   "judgedPass": 0.75,
   "truthPass": 1,
   "judgeAgreesWithTruth": 1
  },
  "moonshotai/kimi-k2.6": {
   "replays": 30,
   "judgedPass": 0.8666666666666667,
   "truthPass": 0.8666666666666667,
   "judgeAgreesWithTruth": 0.9333333333333333
  },
  "deepseek/deepseek-v4.1-flash": {
   "replays": 96,
   "judgedPass": 0.8020833333333334,
   "truthPass": 0.9833333333333333,
   "judgeAgreesWithTruth": 1
  },
  "qwen/qwen3-coder": {
   "replays": 20,
   "judgedPass": 0.95,
   "truthPass": 1,
   "judgeAgreesWithTruth": 0.95
  },
  "z-ai/glm-5.3-flash": {
   "replays": 60,
   "judgedPass": 0.9833333333333333,
   "truthPass": 0.9833333333333333,
   "judgeAgreesWithTruth": 1
  },
  "z-ai/glm-5.1": {
   "replays": 10,
   "judgedPass": 0.9,
   "truthPass": 1,
   "judgeAgreesWithTruth": 0.9
  },
  "deepseek/deepseek-v4-pro": {
   "replays": 20,
   "judgedPass": 0.9,
   "truthPass": 0.9,
   "judgeAgreesWithTruth": 1
  },
  "google/gemini-3.8-flash": {
   "replays": 56,
   "judgedPass": 0.7142857142857143,
   "truthPass": 1,
   "judgeAgreesWithTruth": 1
  },
  "openai/gpt-5.4-nano": {
   "replays": 30,
   "judgedPass": 0.9666666666666667,
   "truthPass": 0.9666666666666667,
   "judgeAgreesWithTruth": 1
  },
  "anthropic/claude-haiku-5.5": {
   "replays": 20,
   "judgedPass": 1,
   "truthPass": 1,
   "judgeAgreesWithTruth": 1
  },
  "moonshotai/kimi-k3": {
   "replays": 10,
   "judgedPass": 1,
   "truthPass": 1,
   "judgeAgreesWithTruth": 1
  },
  "openai/gpt-5.4": {
   "replays": 12,
   "judgedPass": 0.75,
   "truthPass": null,
   "judgeAgreesWithTruth": null
  }
 },
 "decisionDirect": {
  "clef": {
   "model": "cloudflare/clef",
   "n": 18,
   "accuracy": 1,
   "usdPer1k": 0.010478999999999999,
   "p50ms": 348,
   "errors": 0
  },
  "clef-flash": {
   "model": "cloudflare/clef-flash",
   "n": 18,
   "accuracy": 0.9444444444444444,
   "usdPer1k": 0.006186333333333334,
   "p50ms": 270,
   "errors": 0
  },
  "luna": {
   "model": "openai/gpt-6-luna-decisions",
   "n": 18,
   "accuracy": 1,
   "usdPer1k": 0.019305555555555548,
   "p50ms": 198,
   "errors": 0
  },
  "jev": {
   "model": "typesafe/jev-1.13",
   "n": 18,
   "accuracy": 1,
   "usdPer1k": 0.01771233333333333,
   "p50ms": 211,
   "errors": 0
  }
 },
 "previous": [
  {
   "run": "2026-10-08",
   "date": "2026-10-08",
   "version": "1.0",
   "gin": {
    "quality": 0.8765432098765432,
    "usdPer1k": 3.080579012345679,
    "verified": 1,
    "judged": 0.6296296296296297,
    "p50ms": 949,
    "routedShare": 0.8148148148148148,
    "families": {
     "code-review": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.10406666666666664,
      "p50ms": 913,
      "p95ms": 1599,
      "errors": 0
     },
     "security-review": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.7706500000000001,
      "p50ms": 928,
      "p95ms": 3003,
      "errors": 0
     },
     "json-extraction": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.1376666666666667,
      "p50ms": 991,
      "p95ms": 2819,
      "errors": 0
     },
     "classification": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.018277777777777775,
      "p50ms": 925,
      "p95ms": 1225,
      "errors": 0
     },
     "tool-calling": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.1904,
      "p50ms": 932,
      "p95ms": 1221,
      "errors": 0
     },
     "math": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.4269555555555555,
      "p50ms": 893,
      "p95ms": 1576,
      "errors": 0
     },
     "summarization": {
      "n": 18,
      "quality": 0.7222222222222222,
      "usdPer1k": 0.3997722222222222,
      "p50ms": 929,
      "p95ms": 1093,
      "errors": 0
     },
     "kids-story": {
      "n": 18,
      "quality": 0.4444444444444444,
      "usdPer1k": 19.716977777777778,
      "p50ms": 1477,
      "p95ms": 6361,
      "errors": 0
     },
     "support-chat": {
      "n": 18,
      "quality": 0.7222222222222222,
      "usdPer1k": 5.960444444444445,
      "p50ms": 2490,
      "p95ms": 6679,
      "errors": 0
     }
    },
    "policies": {
     "code-review": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "centroids": [
       "DgcEBwz0/AAM/fQH+QD//OkAAAH4CfUPAAP9ANX9AQD/Awj4AAf55BkEA/w4BgTtAPkM8PwAAwAD9/0FC/kUBAME+QADA/H/AADwAwP5/AMDAPD56wn49gMACQsPAAkA8PELAQkJ8QAJ5P0ACAcJEAD4B/kAGf0M/fcDBv0JAAQ="
      ],
      "passRate": 0.9,
      "samples": 10,
      "savings": 0.975
     },
     "security-review": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "centroids": [
       "Agb2/QX48ewAC+X+D/sCBQX5Bfb8AQUOB/b79egF8f34CwL7/gX72PYD9vAzBQAI9gIMCgX+BQn7/v4NAvQADgAC+/4M/u8CAPj7BQXbB/P4/gX4BQsC9Pj2BQ74DgD5AAoA+/j+BfP25wAA9AL2GAANAPMFAwcP/fX49ADs/wI="
      ],
      "passRate": 0.9,
      "samples": 10,
      "savings": 0.975
     },
     "json-extraction": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "samples": 10,
      "savings": 0.975
     },
     "classification": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "samples": 10,
      "savings": 0.975
     },
     "tool-calling": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "samples": 10,
      "savings": 0.975
     },
     "math": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "samples": 10,
      "savings": 0.975
     },
     "summarization": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "samples": 10,
      "savings": 0.975
     },
     "kids-story": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "centroids": [
       "BPoI9fr6/QYQAvoCAhEA9QL+AAALEAD1+gD/+t8G6gAI+v/7/vn18AYG6ABLAQb1CwboBQAA+PsAAP4H+RD9+vXp8wDwBvv/Ag77AwX7AAMC/gAA+Afy+QL4CAYa9wgC+AgPDgL3/wgF/QADAP75+/T+AgD6AAv6BfoLAAsR+/U="
      ],
      "passRate": 0.9,
      "samples": 10,
      "savings": 0.975
     },
     "support-chat": null
    },
    "note": "Production policy: replays the 2 cheapest candidates per use case, routes when ≥90% judged same/better, keeps look-alikes of failures on Opus."
   }
  },
  {
   "run": "2026-10-08-2",
   "date": "2026-10-08",
   "version": "1.1",
   "gin": {
    "quality": 0.9012345679012346,
    "usdPer1k": 5.22889471097531,
    "verified": 1,
    "judged": 0.7037037037037037,
    "p50ms": 953,
    "routedShare": 0.8271604938271605,
    "families": {
     "code-review": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.16495528888888888,
      "p50ms": 537,
      "p95ms": 1298,
      "errors": 0
     },
     "security-review": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.15154533333333334,
      "p50ms": 3047,
      "p95ms": 11234,
      "errors": 0
     },
     "json-extraction": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.1376666666666667,
      "p50ms": 991,
      "p95ms": 2819,
      "errors": 0
     },
     "classification": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.018277777777777775,
      "p50ms": 925,
      "p95ms": 1225,
      "errors": 0
     },
     "tool-calling": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.119696221,
      "p50ms": 369,
      "p95ms": 1822,
      "errors": 0
     },
     "math": {
      "n": 18,
      "quality": 1,
      "usdPer1k": 0.7244666666666666,
      "p50ms": 361,
      "p95ms": 976,
      "errors": 0
     },
     "summarization": {
      "n": 18,
      "quality": 0.7222222222222222,
      "usdPer1k": 17.303111111111114,
      "p50ms": 2483,
      "p95ms": 5087,
      "errors": 0
     },
     "kids-story": {
      "n": 18,
      "quality": 0.5,
      "usdPer1k": 25.38011111111111,
      "p50ms": 1840,
      "p95ms": 6361,
      "errors": 0
     },
     "support-chat": {
      "n": 18,
      "quality": 0.8888888888888888,
      "usdPer1k": 3.0602222222222224,
      "p50ms": 969,
      "p95ms": 4548,
      "errors": 0
     }
    },
    "policies": {
     "code-review": {
      "model": "deepseek/deepseek-v4.1-flash",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "rawPassRate": 1,
      "samples": 10,
      "savings": 0.971178296398892,
      "kept": 0,
      "rungs": [
       "anthropic/claude-sonnet-5.5",
       "openai/gpt-5.4-mini",
       "deepseek/deepseek-v4-pro",
       "deepseek/deepseek-v4.1-flash",
       "qwen/qwen3-coder",
       "z-ai/glm-5.3-flash"
      ],
      "judge": "google/gemini-3.8-flash",
      "profile": "Code review · difficulty 3 · JSON"
     },
     "security-review": {
      "model": "z-ai/glm-5.3-flash",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "rawPassRate": 1,
      "samples": 10,
      "savings": 0.9531428916535813,
      "kept": 0,
      "rungs": [
       "anthropic/claude-sonnet-5.5",
       "z-ai/glm-5.1",
       "deepseek/deepseek-v4-pro",
       "moonshotai/kimi-k2.6",
       "deepseek/deepseek-v4.1-flash",
       "z-ai/glm-5.3-flash"
      ],
      "judge": "google/gemini-3.8-flash",
      "profile": "Security analysis · difficulty 2 · JSON"
     },
     "json-extraction": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "rawPassRate": 1,
      "samples": 10,
      "savings": 0.9719943240454076,
      "kept": 0,
      "rungs": [
       "anthropic/claude-sonnet-5.5",
       "google/gemini-3.8-flash",
       "deepseek/deepseek-v4-pro",
       "openai/gpt-5.4-nano",
       "z-ai/glm-5.3-flash",
       "anthropic/claude-haiku-5.5"
      ],
      "judge": "google/gemini-3.8-flash",
      "profile": "Extraction / JSON · difficulty 2 · JSON"
     },
     "classification": {
      "model": "anthropic/claude-haiku-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "rawPassRate": 1,
      "samples": 10,
      "savings": 0.9847778850796717,
      "kept": 0,
      "rungs": [
       "anthropic/claude-sonnet-5.5",
       "openai/gpt-5.4-mini",
       "deepseek/deepseek-v4.1-flash",
       "openai/gpt-5.4-nano",
       "z-ai/glm-5.3-flash",
       "anthropic/claude-haiku-5.5"
      ],
      "judge": "google/gemini-3.8-flash",
      "profile": "Classification · difficulty 1"
     },
     "tool-calling": {
      "model": "z-ai/glm-5.3-flash",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "rawPassRate": 1,
      "samples": 10,
      "savings": 0.98598783341441,
      "kept": 0,
      "rungs": [
       "anthropic/claude-sonnet-5.5",
       "moonshotai/kimi-k3",
       "qwen/qwen3-coder-plus",
       "qwen/qwen3-coder",
       "openai/gpt-5.4-nano",
       "z-ai/glm-5.3-flash"
      ],
      "judge": "google/gemini-3.8-flash",
      "profile": "Planning / agentic tool use · difficulty 2 · tool calls · latency-sensitive"
     },
     "math": {
      "model": "z-ai/glm-5.3-flash",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 1,
      "rawPassRate": 1,
      "samples": 10,
      "savings": 0.9574467273882652,
      "kept": 0,
      "rungs": [
       "anthropic/claude-sonnet-5.5",
       "google/gemini-3.8-flash",
       "moonshotai/kimi-k2.6",
       "deepseek/deepseek-v4-pro",
       "deepseek/deepseek-v4.1-flash",
       "z-ai/glm-5.3-flash"
      ],
      "judge": "google/gemini-3.8-flash",
      "profile": "Other · difficulty 3"
     },
     "summarization": null,
     "kids-story": {
      "model": "anthropic/claude-sonnet-5.5",
      "from": "anthropic/claude-opus-5.5",
      "centroids": [
       "BvoI9gD6AgYKBPMACAoB5wH6CAIPCwH2/QUA99cG5AEH+wz+A//u8wX+6wBK/P/6Cv7oBgH89fsABfwO9hL4+vbs9AD4Af7+Afr6BgQFAQEB/wIA/Qb/+P33AAYZ+gwB+ggOBv/5AAgAA/8C/P/7+/wDBf72/Qj9BfcBCQcKAPA="
      ],
      "threshold": 0.9,
      "passRate": 0.9166666666666666,
      "rawPassRate": 0.4166666666666667,
      "samples": 12,
      "savings": 0.22915131688819457,
      "kept": 0.6666666666666666,
      "rungs": [
       "anthropic/claude-sonnet-5.5",
       "google/gemini-3.8-flash",
       "deepseek/deepseek-v4-pro"
      ],
      "judge": "pairwise:openai/gpt-5.5",
      "profile": "Creative writing · difficulty 2 · latency-sensitive"
     },
     "support-chat": {
      "model": "anthropic/claude-sonnet-5.5",
      "from": "anthropic/claude-opus-5.5",
      "passRate": 0.75,
      "rawPassRate": 0.75,
      "samples": 12,
      "savings": 0.5220668309738065,
      "kept": 0,
      "rungs": [
       "openai/gpt-5.4",
       "anthropic/claude-sonnet-5.5",
       "openai/gpt-5.4-mini",
       "google/gemini-3.8-flash",
       "deepseek/deepseek-v4-pro"
      ],
      "judge": "pairwise:openai/gpt-5.5",
      "profile": "Chat / assistant · difficulty 2 · latency-sensitive"
     }
    },
    "note": "Production routing v2: profiles each use case, tests a 3–6 rung ladder, judges open-ended work pairwise (GPT-5.5, both orders), routes on the lowest expected cost that holds quality, falls back only on calibrated look-alikes of repeated failures."
   }
  }
 ]
}