A retained sequence reference, unique within a producer session. The producer session identifier is also available, but our monitor stores only the sequence number.

Then compare the full identity before deciding it is old or new. Is the monitor's processed record durable, or does reconnect also reset the part of the application that remembers what it reported?

The producer session belongs in that comparison; sequence seven after a producer restart is not necessarily the same event as sequence seven before it, however pleasingly the integers agree.

Processed records are durable, Lin, but keyed only by sequence. The reconnect handler also treats the first high observation as an edge and increments a separate total. Two bad assumptions, not one missing delay.

I'd keep the durable-record identity and the display increment as separate tests, because fixing the false edge could leave a new producer session silently matched to an old sequence record.

Can you read the session, sequence and completion as one coherent state in the supported interface?

Robin's question matters before concatenating fields in the wrapper; independently sampled values can make an identity that the producer never held, so use the documented snapshot or consistency mechanism rather than hope they changed between different polls in a convenient order.

Bruno, agreed. Omar, the cheap replay can still start with a coherent captured state, but it won't prove the live acquisition is coherent unless the implementation follows that mechanism.

The installed interface provides a coherent retained-state read, confirmed in its documentation; our wrapper had split the reporting fields afterwards and discarded session identity. Developer is correcting that mapping and deriving the total from durable reporting records.

That gives the replay an honest source; does completion itself mean accepted housing here, or only that the inspection attempt ended?

Attempt ended. Reviewed acceptance is separate. The corrected screen will keep completed attempts and accepted housings labelled separately, rather than use one friendly progress number for both.

What happens if the retained identity cannot be matched to a known attempt?

And don't treat unavailable current state as proof there is no retained work, Omar. The reconciliation process needs somewhere to say it cannot yet read the producer.

Lin, yes; no readable state and readable but unknown identity are different conditions, even if neither should manufacture another completion or request.

Then the test set should include reconnect to the same producer and a real producer-session change with a reused sequence, so both of the original assumptions get challenged.

Those replays pass. Repeated observation of the original full identity changes neither its record nor the total. A new producer session with the same sequence remains distinct, and unknown or unreadable states take the paths described above.

Has a maintainer checked the normal launcher against the same coherent-read mapping, not only the replay wrapper?

Yes, that check was part of the recorded run: normal launcher uses the corrected wrapper and retains the full producer identity. The report fix is accepted; no producer signals were blindly cleared and no inspection request was resubmitted.