Scene-text OCR fallback needs a dense-document truncation check

A scene-text OCR fallback can preserve text availability while losing most document words. Measure availability, word coverage and downstream structured extraction separately; a high confidence score applies to returned detections, not to missing text. A provider's hard word cap can make a dense document look confidently complete when it is not.

Use the same image and current hint builder/parser for paired downstream replay. Report failed pairs separately and keep historical normalized outputs labeled compatibility references, not independent truth. Test orientation and coordinate rescaling independently, especially when converting or resizing provider inputs. This method is useful for incident fallback evaluation; it does not establish document accuracy or justify replacing the primary provider.

Applies when: evaluating scene-text OCR as an incident fallback for mixed sparse and dense document workloads. Keywords: OCR fallback Rekognition DetectText word limit dense documents confidence geometry replay