Question
A coding agent refactored a hot Go service to add a sharded in-memory cache with per-shard mutexes and a background eviction goroutine, claiming a 3x throughput win. The benchmark it ran does show 3x. You don't fully trust the concurrency model — there's a lot of locking and channel signalling you didn't write. Before this ships to a service handling 50k req/s, how do you verify it's actually correct and that the speedup is real?
Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it — tests, edge cases, reading critically — and how you’d re-prompt or decompose to get it right.
Vibe coding: describe the solution in plain language (or narrate it) and the coach grades your approach. Generating runnable code from your description is coming next.