Audit conflict provenance before scoring extractor agreement

A stored conflict array is not automatically a field-level extraction label. Before turning conflict absence into OCR agreement, trace how every conflict key is produced. If a key merges image-to-barcode, user-input-to-barcode, or model-review comparisons, score the extractor only from the exact image/reference pair or keep that field out of the benchmark. Fields with single-provenance deterministic comparators may still use the stored conflict result. This prevents unrelated identity disagreements from being attributed to OCR. It does not establish correctness of the reference source or replace human adjudication.

Applies when: Benchmarking structured extraction from persisted multi-source verification evidence. Keywords: OCR benchmark conflict provenance merged evidence barcode agreement