{
  "schema": "cortex.benchmark.claims.v1",
  "slug": "tau2-gpt-router",
  "benchmark": "tau2-bench",
  "split": "base",
  "as_of": "2026-08-02",
  "estimand": "paired common-task point comparison",
  "comparison": {
    "router": {
      "tasks": 100,
      "passed": 66,
      "quality": 0.66,
      "cost": 0.6254493824999999
    },
    "baseline": {
      "tasks": 100,
      "passed": 65,
      "quality": 0.65,
      "cost": 1.3064008199999992
    },
    "quality_delta_percentage_points": 1.0000000000000009,
    "cost_reduction_percent": 52.124235309344016
  },
  "full_arm_metrics": [
    {
      "id": "luna",
      "label": "GPT-5.6 Luna",
      "tasks": 100,
      "passed": 46,
      "quality": 0.46,
      "cost": 0.2765724530000001
    },
    {
      "id": "terra",
      "label": "GPT-5.6 Terra",
      "tasks": 100,
      "passed": 57,
      "quality": 0.57,
      "cost": 0.7311155719999998
    },
    {
      "id": "sol",
      "label": "GPT-5.6 Sol",
      "tasks": 100,
      "passed": 65,
      "quality": 0.65,
      "cost": 1.3064008199999992
    },
    {
      "id": "router",
      "label": "Cortex task-level router",
      "tasks": 100,
      "passed": 66,
      "quality": 0.66,
      "cost": 0.6254493824999999
    }
  ],
  "route_counts": {
    "luna": 35,
    "terra": 34,
    "sol": 31
  },
  "published_rows_sha256": "815437e87efba6a9b71701c43cca63bbaac9b21941a9aa2254514efe53d0fe7a",
  "provenance": {
    "source_artifact": "tau2_codex_2arm_100",
    "source_artifact_sha256": "not recorded in supplied artifact",
    "router_build_id": "not recorded in supplied artifact",
    "route_score_provenance": "selected_arm imported from sticky_router_per_task.csv; published arm labels follow sticky summary_report series rename (router occupies former mid-tier series; mid-tier carries former routed aggregate). Decision mix still uses sticky selected_arm.",
    "benchmark_version": "not recorded in supplied artifact",
    "reward_implementation_version": "not recorded in supplied artifact",
    "pricing_source": "Costs use agent_cost recorded by the benchmark harness; the underlying price table was not included in the supplied artifact.",
    "pricing_as_of": "2026-08-02",
    "held_out_status": "The supplied artifact does not establish whether this cohort was held out from router or gate tuning."
  }
}