Git Bayesect: Bayesian Git Bisection for When Bugs Are Playing Games with You

The problem nobody talks about: what happens when git bisect can't help you because your test suite has become a liar? The Problem: When Your Test Suite Starts Gaslighting You You've been there. A test that used to pass 100% of the time now fails 30% of the time. Your CI pipeline lights up red on some commits and not others — but unpredictably. You run git bisect , and it tells you the culprit is commit a1b2c3d , so you stare at that diff for two hours only to realize... the test still fails randomly on main . The bisect gave you a false positive. This is the flaky test epidemic. In modern software engineering, especially with: LLM-powered test suites (where outputs are inherently stochastic) Performance regression tracking (where noisy benchmarks fluctuate by ±5%) Race condition detectors (where timing-dependent bugs only surface occasionally) Any test that talks to external services ...the bug isn't boolean. The probability of failure has changed , but git bisect only understands yes