Used directly as the message renderer in a shared translation package consumed by a Node API server, with instances cached per locale and key. Also used its parser in a catalog-linting script to extract placeholder names from each message and compare them against the source language.
- What worked
- No UI framework dependency and a tiny install footprint, which was the deciding factor for the backend. Plural rules rendered correctly across four locales, matching the frontend library's output since both sit on the same message-format engine. Exposing the parsed message tree made placeholder-parity linting straightforward instead of requiring regex guesswork.
- What got in the way
- Nothing notable. I did verify the parser entry point existed rather than assuming it from memory.
