Strings are stored as UTF-8, with lone surrogates kept as their own code units.
Length, string methods and relational comparisons such as `<` and `>` use UTF-16 semantics, including for lone surrogates.
`split` with a string separator and `replace` or `replaceAll` with a string pattern don't match a lone surrogate against half of a surrogate pair; regular-expression patterns do.
Output, file contents, and other encoded text write each lone surrogate as U+FFFD, as Node does.
File system paths and network host names that contain a lone surrogate reach the operating system unconverted.

One-argument `localeCompare` follows Node's default root-locale collation for Latin-script text (ASCII, Latin-1 Supplement, Latin Extended-A, combining diacritical marks, and U+2044 FRACTION SLASH): letters order before case and accents, so `"a".localeCompare("B")` is `-1`.
Strings containing other characters compare by code point, which can differ from Node's ICU collation; `SC6003` reports such string literals.
The locales and options arguments are unsupported.

`sort` and `toSorted` use stable natural merge sort.
Default ordering supports string, number, and boolean arrays, comparing their string forms in UTF-16 order; for example, `[2, 10, 1].sort()` produces `[1, 10, 2]`.
`sort` preserves holes at the end, while `toSorted` produces a fresh array with holes filled by `undefined`.
With an explicit comparator, consistent comparators produce the same sorted results as Node, but callback order and count can differ.

Typed-array `sort` and `toSorted` order numbers by default, with `-0` before `0` and `NaN` last.
With a comparator, they call it in the same order as Node and write the sorted values back after the last call.