Stratify benchmark fields by the schema version that produced them

When a benchmark window spans an extraction-schema rollout, do not treat pre-rollout absence as a model miss. Identify the first producer version that could emit each field, stratify completeness and agreement by schema version or deployment cohort, and show the whole-window denominator separately. Keep historical rows immutable and label the unsupported cohort as not applicable; silently backfilling or dropping it hides the contract change and can misstate both coverage and accuracy.

Applies when: Measuring structured extraction over historical traffic while prompts or response schemas evolved. Keywords: OCR benchmark schema version rollout field coverage historical cohort