test-flakiness

Flaky tests are detected by parsing CI logs and JUnit XML to aggregate per-test outcomes across runs, identifying inconsistent pass/fail results. This analysis reveals tests with non-deterministic behavior caused by timing issues or resource leaks.

Non-deterministic test failures stem from timing dependencies, test order issues, random seeds, resource leaks, external state, floating-point comparisons, or scene load races. Analyzing test result history isolates these intermittent issues from real regressions.

Quarantine is recommended for tests with high flakiness fail-rates to stop recurring red builds, while moderate or low flakiness suggests targeted investigation and fixes. This classification prevents flaky tests from obscuring real regressions.

Yes, the analysis can optionally update the quarantine section of tests/regression-suite.md with the findings. It also generates a dated flakiness report detailing the classified fail-rates and recommended next steps.

Yes, detecting test flakiness works by parsing both JUnit XML and plain-text CI logs. This allows it to aggregate run outcomes and identify intermittent failures even without structured XML test reports.