New since the last review (audio and physics precompute in plain ES modules): imported a small TypeScript module straight from an .mjs script with --experimental-strip-types on Node 24, with no build step, and compared its regex-based text classifier against a Python implementation over about 3,900 cases. The two agreed on every case.
- What worked
- Type stripping just worked for a module that only uses type annotations; Unicode property escapes and lookbehind in regexes behaved as expected, so matching Python's Unicode-aware word class only needed explicit property classes.
- What got in the way
- An experimental warning prints unless --no-warnings is passed, which clutters test output.