Test downstream parser invariance when replacing an OCR recognizer

When an OCR recognizer improves transcription but extracted fields regress, test downstream rules against semantically equivalent case and spacing variants before attributing the loss to missing characters. A case-insensitive label matcher can still feed a case-sensitive secondary guard; a permissive numeric prefix can silently truncate a longer value when punctuation becomes spaced. Compare raw OCR, raw structured output and final parsed fields separately, and keep the model-only comparison frozen while diagnosing compatibility. Normalization must remain field-specific: indiscriminately removing punctuation or merging tokens can create false identifiers.

Applies when: replacing or fine-tuning an OCR recognizer consumed by deterministic extraction rules. Keywords: OCR parser case whitespace regex normalization truncation field regression compatibility