Performance tests

Catch a regression in cost without a check that fails on machine noise. Success is: every performance risk has a check that failed on a seeded regression and passes on the fix. The check runs where its noise allows: deterministic counts and ratios on every change, benchmarks and load on dedicated or scheduled runs. Thresholds come from measured baselines or SLOs.

Process

1. Name the risk, its unit and its cadence before choosing a tool: