Question 520
A model quality dashboard compares pre-release and post-release scores but fails to record which evaluator prompt or rubric version produced each measurement. What provenance field is essential?
Explanation
Correct Answer
Evaluation scores can change because the measurement instrument changed; rubric and dataset identifiers permit valid comparisons.
Incorrect Answers
A moving label destroys the history needed to separate evaluator drift from model drift.
Response length does not make scores from different evaluators equivalent.
Dashboard inventory is unrelated to metric semantics.