Code that won’t compile, a password that “is definitely right”, a search that finds nothing, a spreadsheet match that fails: the culprit is often a character you cannot see. Reveals zero-width spaces, non-breaking spaces, bidi overrides, soft hyphens, Cyrillic or Greek lookalikes and smart quotes, then cleans them.
Lookalike detection covers the common Cyrillic and Greek letters used in phishing domains and copy-paste accidents; it is not a complete Unicode confusables database. Nothing is sent anywhere.