Your service metrics tell one story. Your customers tell another. According to Marc Brooker, an engineer at Amazon Web Services, this contradiction resolves through what's known as the inspection paradox: customers don't experience your latency distribution—they experience a time-weighted version of it. When a request takes longer than average or an outage runs longer than expected, that counts more heavily in a customer's experience. Brooker illustrates the gap with a concrete example: if your median outage recovery time is thirty minutes, but your ninety-ninth percentile stretches to ten hours, your systems might report an average recovery time of just over one hour. But customers experience an average recovery time of around six hours. Both numbers are right—they're just measuring different things.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton