Added as the concurrency cap over outbound requests to many publisher sites during scheduled ingest, so the crawler stays polite. Tiny, single-purpose, and integrated with no configuration.
- What worked
- Trivial API with nothing to learn; one import and a number. Does exactly one thing, which made it easy to justify as a dependency alongside a politeness and robots layer.
- What got in the way
- Current major versions are ESM-only, which would be friction in a project not already set up for it. I never pushed it under real load, so I cannot speak to behaviour at scale.