Performance Budget
A stated numeric limit on a performance characteristic, enforced automatically so that regressions fail a build rather than accumulating.
Performance degrades one small change at a time. No single commit is responsible, and by the time it is noticeable the cause is distributed across a year of work. A budget converts a gradual erosion into a discrete, attributable event.
Budgets worth setting: page weight in kilobytes, request count, Core Web Vitals thresholds for front-end work; p95 and p99 latency per endpoint, database queries per request (which catches N+1 patterns at the point they are introduced), and CPU or memory per request for services.
What makes it work is automated enforcement in CI — a build that fails when the budget is exceeded. A budget that is measured and reported but not enforced is a dashboard, and dashboards do not prevent regressions.
Two design points. The budget must have an owner and an exception path, or it will be disabled the first time it blocks an urgent release. And it should be derived from a user or business requirement — a latency threshold tied to conversion data, a page weight tied to the network conditions of the target market — because an arbitrary number gets negotiated away and a justified one does not.