Fuzzy Matching

Specification v3.4.0, section V.E

1. Fuzzy matching may only be used on approved demographic combinations where the demographic is explicitly marked for fuzzy matching.
2. Fuzzy matching should be implemented as least edit distance tolerating 1 insertion, 1 deletion, 1 substitution, or a single transposition. This set of edits can best be described as a Damerau-Levenshtein distance of 1 (as opposed to a standard Levenshtein which would classify the transposition as a distance of 2). Jaro-Winkler may also be considered where available.
3. Fuzzy matching SHALL NOT be applied to query strings that are less than 5 characters long.
4. Soundex and other phonetic matching are prohibited; except we permit phonetic comparison on first and last name where the implementer can validate the technique against the approved test datasets at or below the framework’s false-positive threshold.