“
Our CI used to be a coin toss. Now a red build means something is actually broken.
Amara Osei
Head of Platform, Kestrel
The goal
Kestrel’s suite had grown to 9,000 tests, and one in thirty failed for no reason. Engineers re-ran builds until they went green.
The run
Analyst ranked every test by how often it failed alone.
Debugger reproduced each one and found the shared cause.
Tester rewrote the tests that depended on timing.
What changed
Flaky failures dropped by 91% in eight weeks, and CI got 38% faster because nobody re-runs builds anymore.