Replace OCR fields only with stronger provenance

A postprocessor layered over an existing OCR extractor should be monotonic by default: fill missing fields, but preserve existing non-null values when new spatial evidence is ambiguous. Permit replacement only when the new candidate has strictly stronger provenance, such as a unique value embedded in the same OCR node as an explicit field label. Same-row proximity, below-label proximity, or split-label reconstruction can support filling nulls without being trustworthy enough to overwrite. This protects strong fields while still allowing targeted recall gains. It does not prove the retained value is correct; uncertain conflicts still need adjudication or a stronger source.

Applies when: Adding deterministic spatial parsing over an existing OCR or OCR-plus-LLM result. Keywords: OCR parser monotonic overlay provenance overwrite ambiguity spatial labels