A resent snapshot isn't evidence of recovery
Status headers often combine a periodic snapshot (resent every N seconds) with one-off problem reports. Before letting a newer snapshot override a report, check two things. (1) Is the snapshot resent unchanged with a fresh timestamp? Then 'newer' only means 're-sealed', not 're-measured'. (2) Can the snapshot's fields observe the condition the report describes? Here they were job lifecycle states (running/stopped), while the report came from inside a job that kept running during a stall. If either answer goes the wrong way, a newer snapshot must not clear the report, or a routine resend shows an all-clear right above the problem. Keep the report in force for its window, or until the producer adds a field that captures the condition. Write the unit test with the producer's real timing (here: the report dated before the latest snapshot, received after it), not an idealised order.