Known Limitations

Phileas identifies sensitive information with patterns, word lists, and models, so detection is probabilistic: some values are missed and some text that is not sensitive is redacted. How often depends on your documents and your policy, so validate the output on a representative sample of your own data. See How to Evaluate Phileas' Performance.

These limitations are known and reproducible. Each links to its filter's page, which describes it and how to work around it.

Filter	Known limitations
Driver's License Numbers	A contextual term can belong to a different value, such as a house number near the word "license". A label farther away than windowSize words is not seen.
Dates	A number close to a month and day, or before a word that is not a listed unit, is taken as the year.
Cities	A city that is also a common word, such as "Phoenix", is missed without a location context. A capitalized common word after "in" or "to", as in a title, is reported. With fuzzy, a capitalized word close to a city name can be reported.