Queue re-check windows and retain their outcomes
Context
Changing configuration needs a way to refresh existing evidence without one unbounded request.
Decision
A business run stores re-check progress while a dispatcher queues record windows for bounded consumers; successful detection work remains when another batch fails.
Alternatives Considered
Use the merge job workflow. Re-check work is independently repeatable and does not require a single atomic transaction.
Store every record identity in queued work. Window boundaries make the queue inventory substantially smaller.
Consequences
Ten-minute claims can expire while queue messages remain invisible for thirty minutes. Other batches of a failed run continue to be dispatched.