Unicode normalization form D (NFD) decomposes a single accented code point like 'é' (U+00E9) into a base letter 'e' (U+0065) followed by a combining acute accent (U+0301). Once the accent is its own code point, a regex strip of the combining-marks block (U+0300 through U+036F) removes it, leaving the base letter intact. This is the ASCII-folding trick used by every search engine and slug generator: cheap (one allocation), correct for Latin and Greek, and fully built into the language since ES2015. It does NOT handle scripts that have a single code point per character with no decomposition, like Arabic, Hebrew, or CJK, where the input passes through unchanged.

Search-as-you-type fields, autocomplete, and table filters need to match 'cafe' against 'Café' regardless of case or accent. Folding the haystack and needle through the same NFD + strip + lowercase pipeline turns the comparison into a vanilla String.prototype.includes, which V8 specialises into a fast indexOf. Doing this on the haystack at search time is fine for short text; for large catalogs, precompute the folded form once at index time and search against that to keep latency flat. Note that this is intentionally locale-naive: Turkish dotted/dotless I and German ß both have locale-specific case rules that this version ignores.

Stripping accents is the right move for indexing and slugs, but for user-facing sorting and equality you want Intl.Collator, which respects per-language rules without you hand-coding them. sensitivity: 'base' treats 'café' and 'cafe' as equal, while 'accent' keeps accent differences but ignores case, and 'variant' distinguishes everything. Pass the collator's compare method directly to Array.prototype.sort and you get locale-correct ordering for free, including correct handling of the Swedish å/ä/ö after z, the German ß, and Czech ch. Use the strip-accents helper for keys you store, but use a collator for anything the user sees.