How do you choose between asyncio, threads, and processes in Python?

Production ownership includes pool size, queue bounds, worker crashes, deployment shutdown, context propagation, and observability. Processes need import-safe and picklable work under modern start methods; threads need cooperative stop and thread-safe dependencies; tasks need event-loop ownership. For example, CPU-heavy document conversion gets its own bounded process workers instead of competing with API event loops for the same deployment budget. I isolate heavy batch work from latency-sensitive API capacity even when both could technically share one process.