Whether an entry describes what the release is rather than how fast it was.
The benchmark action compares every entry it uploads against the release before and calls a value that grew
a regression. That is right for a runtime and wrong for a counter: a release with ten more linting rules or a
larger signature database is not slower. Those entries are written to infoGraphPath instead and are
uploaded under a suite of their own, where nothing alerts on them. The page merges the two back together.
The failures are counters too, but a run that fails to re-parse more slices than the one before is exactly
what an alert is for, so they stay with the measurements.
Whether an entry describes what the release is rather than how fast it was.
The benchmark action compares every entry it uploads against the release before and calls a value that grew a regression. That is right for a runtime and wrong for a counter: a release with ten more linting rules or a larger signature database is not slower. Those entries are written to infoGraphPath instead and are uploaded under a suite of their own, where nothing alerts on them. The page merges the two back together.
The failures are counters too, but a run that fails to re-parse more slices than the one before is exactly what an alert is for, so they stay with the measurements.