Meet Alice. Alice is impatient.

The inspection paradox explains why customers experience much higher latency and longer outages than service metrics suggest. When you measure latency per-request, a long request counts as just one data point — but customers spend most of their waiting time on slow requests, so their perceived mean is dominated by the tail. A simulation shows how a service reporting 88ms mean latency can feel like 220ms to users. This is why tail latency and recovery time distributions matter far more than means or trimmed metrics, and why hiding tail latency with retries only partially helps.