{
  "measured_at_start": "2026-10-20T19:49:30.431652+00:00",
  "measured_at_end": "2026-10-20T19:57:35.088155+00:00",
  "build": "teacher-objectives-main-6d4954bb72cae634",
  "model": "OpenVINO/Qwen3.5-4B-int4-ov",
  "submitted": 12,
  "completed": 12,
  "failed": 0,
  "busy_rejected": 0,
  "truncated": 0,
  "median_seconds": 36.013999999999996,
  "min_seconds": 35.638,
  "max_seconds": 49.802,
  "scenarios": [
    {
      "scenario": "introduce-one-half",
      "label": "Introduce one half",
      "requests": 4,
      "completed": 4,
      "median_seconds": 35.8575,
      "min_seconds": 35.828,
      "max_seconds": 36.694
    },
    {
      "scenario": "compare-unit-fractions",
      "label": "Compare thirds and fourths",
      "requests": 4,
      "completed": 4,
      "median_seconds": 49.0695,
      "min_seconds": 49.042,
      "max_seconds": 49.802
    },
    {
      "scenario": "unequal-intervals",
      "label": "Correct unequal intervals",
      "requests": 4,
      "completed": 4,
      "median_seconds": 35.831500000000005,
      "min_seconds": 35.638,
      "max_seconds": 36.105
    }
  ],
  "all_requested_headings_present": true,
  "protected_services_unchanged": true,
  "cloudflare_unchanged": true,
  "temporary_harness_removed": true,
  "min_available_host_gib": 14.785568237304688,
  "method": {
    "design": "12 serial requests: four interleaved cycles of three fictional Grade 3 number-line lesson scenarios, same captured 3.NF.2 source. One shared LiveAssistant, unchanged active-request guard. No retries, model changes or production saves. Requests are repeated synthetic scenarios, not 12 distinct curricula.",
    "controls": {
      "temperature": 0,
      "max_tokens": 2048,
      "stream": false,
      "thinking": false,
      "model_tools": false
    },
    "timing": "Application preparation, source selection validation, saved-assistant checks, local inference and response validation. Excludes HTTP authentication, browser/network/rendering, and saving.",
    "spot_check": "The first response in each scenario still suggested unavailable paper, notebooks or manipulatives despite the whiteboard-only instruction. This was a three-output spot-check, not a full educator evaluation.",
    "quality": "Response completion and heading checks are separate from educational correctness, source fidelity, teacher-constraint adherence and educator acceptance.",
    "statistic": "Median over all 12 successful requests; scenario mix affects the aggregate. Four observations per scenario. Single run, no p95 or sustained capacity claim."
  }
}