Your first insight

Once a few runs have landed, the dashboard turns them into answers. Here is the path from raw runs to a decision.

1. The flaky board

The flaky board ranks every test by its flaky score, worst first. The score is a Beta-Binomial estimate with exponential time-decay — recent behaviour counts for more than last month's. A test only appears once it has enough samples to be meaningful, so a single red run does not brand a test as flaky.

2. Reason codes

Open a test and the score breaks into reason codes — the human-readable evidence behind the number:

• same commit, different result — the strongest flaky signal: the code did not change but the outcome did.
• pass-on-rerun — failed, then passed on retry with no code change.
• intermittent across branches — fails on some branches and not others without a code reason.

Because every score ships with its reasons, you can trust it — or argue with it — instead of taking a black-box number on faith. See Flaky scoring for the model.

3. Test history that survives refactors

A test's history does not reset when you move or rename its file. The test identity engine stitches executions across refactors, so the flaky score reflects the test's real long-run behaviour rather than the age of its current file path.

4. Root-cause on failures

For genuinely new failure signatures, the AI RCA panel offers a structured "likely cause + suggested action". Failures are normalized and clustered first, so only new signatures reach the model — root-cause without a per-failure LLM bill.

5. Guard your pull requests

Wire the GitHub Action in and every PR gets a sticky comment plus a quality gate that distinguishes new failures this change introduced from tests that already flake on the base branch — blocking only what the change actually broke.

6. Track the trend

The Test health page answers the executive question: is flakiness improving or degrading? It reports flaky MTTR (how long a test stays flaky before it stabilizes), flaky-introduced vs -resolved per week, the current quarantine backlog, and the overall reliability trend — over a time range up to a quarter. A net-positive week (more introduced than resolved) is the early warning that flakiness is winning.