Measuring Delivery That Lasts, Not Just Ships

Ship rate is a vanity metric

Counting merged pull requests tells you a team is busy. It says nothing about whether the thing you shipped still works under load six months later — which is the only outcome a client actually pays for.

We track change-failure rate, time-to-recovery, and the age of the oldest unaddressed defect, because those numbers move with durability rather than motion.

Instrument the boring parts

The signals that predict trouble are unglamorous: retry storms, slow queries creeping past a threshold, a queue that never quite drains. Instrumenting them early is how a problem surfaces as a graph instead of an outage.