Suppress identity false positives with provenance and bounded canonicalization

General lesson: when comparing a person’s name across form input, document OCR, and machine-readable barcode data, retain a separate result for every source pair and make only the requested pair authoritative. Canonicalize representation—not identity—before comparing: preserve ordered given-name tokens, normalize recognized suffix placement and harmless compound-name spacing, and allow only documented layout permutations. Treat missing or invalid extraction as unknown/evidence-error, never mismatch. Keep character edits, dropped non-suffix tokens, aliases, and nicknames reviewable unless an independently validated policy supports them. A model may retry extraction or classify evidence quality, but should not union holistic conflicts into deterministic field conflicts.

Applicability and limits: useful for multi-source identity verification where parsers split or order the same printed name differently. It does not justify unordered token-set matching or broad fuzzy acceptance, which can merge different people.

Applies when: multiple identity sources encode the same human name with different field boundaries, order, suffix placement, or evidence quality. Keywords: identity matching false positives barcode OCR name normalization provenance