Can asyncio code have race conditions even when one event loop uses one thread?

I identify every suspension point inside the proposed critical sequence, including hidden awaits in helpers and async context managers. A lock that covers network I/O may preserve correctness but destroy latency, so redesign can be better than serialization. For a cache fill, I might coordinate only publication per key while allowing the slow fetch to run without one global lock. Stress tests control scheduling with events or barriers and verify durable outcomes, not merely the final in-memory value.