Individual runs that executed and aborted. A row-level count: a shard that fails before producing any rows leaves this at 0, which is why failedShards exists alongside it.
OptionaletaRough seconds remaining, extrapolated from the rate observed so far and assuming nothing else competes for workers. Runs conservative in practice — it has measured 2–5× long when a sweep spent part of its life waiting to be retried, since that wait dilutes the observed rate. Omitted, never zero, when it cannot be computed: a sweep with nothing finished yet has no rate to extrapolate from, and a zero would read as "about to finish". Excludes queue wait entirely; retrying and stalledSeconds are where that shows up.
Shards (or folds) that failed and will not be retried. Distinct from aborted: this counts whole units that never reported, not runs that ran badly.
Units that have not reported anything yet. Covers both work still queued behind other work and work claimed by a worker that stopped before it began, which is why a sweep with a persistent value here and a rising stalledSeconds is worth looking at.
Units whose last attempt failed on something transient — an I/O error, a worker that died mid-read — and which are queued to be attempted again. Not counted as failures, because they have not failed yet; a sweep with a non-zero value here is still expected to complete.
OptionalstalledSeconds since anything last advanced. Omitted on a finished sweep, where it would only measure how long ago it finished, and on sweeps submitted before this field existed.
How far along a sweep is, and — when the sweep is still running — enough to tell a healthy one from a stuck one. The counts partition the shards (or, for a walk-forward sweep, the folds): every unit is either finished, failed, waiting to be retried, or not yet started.