Production Verification
The claims a deployment can and cannot make about itself, and what closes the gap.
Overview
Verification is a sequence of increasingly expensive claims: the process is alive, its dependencies answer, a real user journey completes, and nothing regressed under real traffic. Each proves strictly more than the last, and each is routinely mistaken for the one above it.
Why This Exists
The failure this prevents is specific and common: a deployment reports success, every check is green, and the product is broken for users. It happens because the cheap checks answer a different question than the one anyone cares about. A process that started has demonstrated that it started. A health endpoint returning affirmative has demonstrated that an endpoint returns affirmative. Neither says whether someone can complete the task they came to do, and the distance between those claims is where releases fail.
Four claims, in ascending order of cost and value
Smoke tests: few, real, and end to end
The checks a browser has to make
Watching, and what to watch
Deciding to keep or roll back
Related Principles· 2
Related Documents· 4
Referenced By· 4
Version History
- v0.1.0
Initial structure and metadata established.
- v0.2.0
Written from real experience. Separates the four claims a deployment can make in ascending order of cost, scopes smoke tests to few real end-to-end journeys, names the failures only a browser can see, defines the watching window and its signals, and ties the keep-or-roll-back decision to the answer recorded during review.