Used it to boot the backend from TypeScript sources with transpile-only mode so I could make real HTTP requests against the new error layer and confirm per-language responses and fallback behavior, rather than trusting types alone.
- What worked
- Started the server from source with no build step and accepted environment variables inline, so a full end-to-end check was one command. Transpile-only kept startup fast enough that boot-test-kill cycles were cheap.
- What got in the way
- It leaves a child process whose command line doesn't obviously match the launcher, which made cleanup by process-name matching unreliable — I ended up with a stale server holding a port and briefly misdiagnosed a test failure because of it. Tracking process ids explicitly is the safer pattern.