Reading results

Verdict

Each CI run is compared with a baseline: the last 30 comparable runs on the default branch.

A flow is slower when its median is at least 10% above the baseline and outside its normal run-to-run variation. With fewer than 10 baseline runs it only warns. With none, the result is Not compared.

Results on a pull request: N% slower, Unchanged, Failed, Runner noise, Not compared.

Steps

Where the time went lists each step’s median latency and its change from the baseline, so you see which request got slower.

Runner noise

When the CI machine was slow rather than your code, the result is Runner noise, re-run. It never fails the check. Re-run the job.

On the default branch

An alert opens after 2 slower runs in a row (3 on GitHub-hosted runners), in the app and in Slack.