the evidence
A pass is
not a reason.
A flaky test is an accusation without proof. Flakerad changes one environmental variable at a time, making the path from failure to pass legible.
read the method →the four controls
Change one thing. Watch the failure move.
01
fixed seed
same random inputs
−75%
failure delta / controlled rerun
the distribution
Confidence is
a shape.
baseline
failure delta · 0.00
fixed seed
failure delta · -0.18
frozen clock
failure delta · -0.36
isolated order
failure delta · -0.54
the reading
More reruns.
Less mythology.
reruns per condition
8
58%precision
75%recall
+11%net signal
the source
Open code.
Reproducible bugs.
To see the test runner source, explore benchmark fixtures, or to find bugs, visit the repository on GitHub. Built and maintained by Ayush Dubey.