A — 0: Success is a subjective demo reaction or an unexplained model score. B — 1: A rubric exists, but human and automated scoring have not been calibrated. C — 2: Deterministic checks and adjudicated human examples define acceptance; any model judge has a documented error review.
Evidence to attach: rubric and grader-calibration record. If weak: start with a small reviewed diagnostic set. Human calibration matters before automated grades influence a procurement decision.