An OCR extraction miss is not proof that a field is absent

When a candidate-based OCR service maps every missing normalized identifier to required_field_missing, that reason does not establish visual absence. It can also represent recognition, label association, candidate selection, or normalization failure. If an independent image-policy check passes but extraction requests a retake, preserve the disagreement and inspect extractor evidence before blaming photo quality. A coordinator that falls back only on technical errors will treat this valid negative response as final, so repeated uploads can repeat the same rejection. Do not infer empty upstream OCR from an unpopulated coordinator diagnostic field.

Applies when: combining candidate-based OCR extraction with a separate image-policy gate. Keywords: OCR PaddleOCR Qwen Gemini extraction missing field fallback diagnostics