{
  "schema": "anvil-serving.quality-result-summary/v1",
  "evidence": "native/mtp-uva3/quality.json",
  "status": "passed",
  "questions": 10,
  "repetitions": 3,
  "attempts": 30,
  "passed": 30,
  "finish_reasons": [
    "stop"
  ],
  "failures": [],
  "protocol": {
    "budget_semantics": "visible_answer_tokens plus reasoning_headroom_tokens are sent as one max_tokens cap; the endpoint does not hard-partition the channels",
    "minimum_comparable_repetitions": 3,
    "minimum_pass_rate": 1.0,
    "reasoning_headroom_tokens": 8192,
    "records_finish_reason": true,
    "records_full_visible_answer": true,
    "records_reasoning_channel_metadata": true,
    "repetitions": 3,
    "version": 3,
    "visible_answer_tokens": 2048
  },
  "identity_binding": "identity-mtp-uva3.json",
  "scope": "Small ten-question repeated diagnostic; not full MMLU-Pro, independent samples, a broad quality ranking, or proof that MTP caused the different result."
}