Run volume is the easiest automation metric to collect and the least useful for release decisions. A pipeline that retries three times and finally exits still often lands as a single success in the weekly pack.
We ask teams to name the business or engineering outcome each automation is meant to produce—ticket closed, artefact published, environment provisioned—then instrument that outcome separately from the run start event.
Once outcomes are distinct, retries become visible as cost rather than noise. The board pack shrinks, and the arguments about 'green dashboards' get shorter.