Pipeline stability

How reliably each flow runs, over completed runs only. Two questions, kept apart because they answer to different things: did it work is the conclusion of every run, and did it work without help counts only runs that succeeded on the first attempt. The second is reported for retrieve, transform and match alone — those three have no auto-rerun of their own, so a second attempt can only mean a person clicked re-run. The rest carry a rerun-on-failure workflow or cancel-in-progress, where a later attempt says nothing about whether anyone was involved.

Last 30 days, collected 13 Sept, 23:58. Durations are execution time, measured across the jobs that actually ran, and average first-attempt successes only — GitHub rewrites a run's start time when it is re-run, so a retried run's elapsed time describes nothing real. Time spent queued for a runner is reported separately rather than folded in: one diff run waited 53 minutes and then built in 2m20s, and calling that a 55-minute build says something false about the build.