Derive classifier evidence rules from acceptance policy, not task names

When adding a decision classifier to an extraction workflow, derive its admissible evidence classes from the full current acceptance policy, not the task name or short retry-message text. Those labels may preserve an older, narrower description after the extractor has broadened accepted sources. A classifier can then appear to make unsafe accepts solely because the evaluation rubric wrongly excludes valid evidence. Version the rubric with the source policy, preserve the original outputs when correcting the contract, and withdraw affected accuracy interpretations. A corrected same-input replay confirms the policy mismatch; it does not establish a new accuracy rate.

Applies when: An existing extraction task gains a separate visual classifier or retry-reason gate. Keywords: classification OCR acceptance policy evidence rubric retry evaluation mismatch