1.3.19 (2026-07-24)
Fixes
- Background health checks can no longer be held up by a single file. A file that could not be scheduled, whose first article was already gone, or whose check ended without a verdict is now given a verdict or a retry time instead of being retried continuously ahead of everything else.
- File checks no longer pause near completion when a provider stops responding part-way through a batch. The stalled connection is dropped after two seconds and only its unanswered articles move to another provider, instead of waiting for that provider to close the connection.
- A provider that loses a single connection is no longer set aside for the rest of a check. Only a provider that never answers is, so its remaining connections keep doing useful work.
- Large releases now get verification time proportional to their size. Previously a fixed budget could expire on every attempt, leaving those files reported as unverified after every retry.
Improvements
- Health checks no longer pause connection upkeep on providers they are not using, so playback and preparation keep their ready connections while a check runs.