OCR text is imperfect, and people do not always remember punctuation or spelling exactly. Phase 6 improved tolerance while keeping retrieval deterministic.
The project added normalization for case, whitespace, punctuation, and optional diacritics. A normalized-to-original range map preserves the exact source text for snippets and highlighting, while conservative edit-distance matching handles small spelling differences only when enabled.
Stronger evidence remains above weaker evidence. Exact, phrase, case-insensitive, and normalized matches rank above typo matches, short queries are protected, duplicate ranges are suppressed, and typo candidates are capped.
Readable labels explain when normalization or edit distance contributed. The system still does not expand synonyms or infer related concepts; tolerance changes comparison, not assumed meaning.
← Back to SnipDig