Falsky watches every CI run, spots tests flipping between pass and fail, and quarantines them — so real bugs block merges and phantom failures don't.
Every run is recorded — result, retry count and timing for each test. Falsky's GitHub Action reads your JUnit XML output automatically.
A Bayesian trust score weighs pass-rate, retries and timing variance to separate flaky from genuinely broken. Better than naive pass/total.
Flaky tests leave the merge path with an issue attached. Real bugs stay blocking. Your pipeline stays green.
48 builds · each tile is one run — green stable, amber flaky, red real failure
of red builds are actually flaky
engineer time lost per false red
agents or code changes needed
Five minutes to wire in. Your next red build might not actually be red.