{
  "measurement": "Median elapsed backend seconds for one scripted caller turn including tool round trips.",
  "source": "../../delivery/results.json",
  "method": "Selected the 136 reviewed traces by SHA-256. Pooled turn_ms within each model, excluding the pilot. Counted only events of type tool.",
  "suites": [
    "matrix-v2",
    "new-phrases-v3"
  ],
  "models": [
    {
      "label": "Terra",
      "model": "gpt-6.1-terra",
      "seconds": 2.882576685049571,
      "turns": 129,
      "runs": 68,
      "passed": 68,
      "tool_calls": 157,
      "by_tool": {
        "get_order": 68,
        "check_credit_eligibility": 27,
        "register_credit": 11,
        "check_replacement_availability": 33,
        "create_replacement": 18
      }
    },
    {
      "label": "Luna",
      "model": "gpt-6-luna",
      "seconds": 3.182167283957824,
      "turns": 129,
      "runs": 68,
      "passed": 66,
      "tool_calls": 172,
      "by_tool": {
        "get_order": 80,
        "check_credit_eligibility": 29,
        "register_credit": 12,
        "check_replacement_availability": 33,
        "create_replacement": 18
      }
    }
  ]
}
