Same attempt identity, Chen, with its result durably saved, Noah; the reconnect handler treats seeing the retained state as a fresh event regardless of that history

Reconcile the retained identity with saved attempts before applying any change. A known matching completion should leave state and counts unchanged. Keep an unknown identity separate for review rather than assigning it to whichever housing is currently selected.

What will the operator see during that review? I'd rather see one named housing awaiting reconciliation than a counter that briefly says the batch is finished while the program sorts itself out.

What if the same identity returns a different result?

Treat that as a conflict requiring the agreed review, not as permission to overwrite the saved result. Preserve both observations and their sources so the responsible person can investigate.

Developer has matching, unknown and conflicting cases in the replay now; Priya, the proposed view identifies the held housing and review reason without moving it into the accepted total

Does another reconnect preserve that unresolved reason?

Noah's question matters for the next shift too. They shouldn't inherit an unexplained pending row after the first person saw a useful conflict message and went home.

It initially lost the reason on a second reconnect; fixed persistence now keeps the unknown or conflicting observation visible with the original attempt history

And no automatic inspection request from that path?

None; replay asserts no request is sent, known completion leaves the total unchanged, and unknown or conflicting records remain awaiting the authorised decision

Include a disconnect while the reconciliation result is being saved. A normal reconnect test can miss the boundary where one part of the record is written and another is not.

Interrupted-save cases pass with the record either committed completely or still unresolved on reopening; the normal installed monitor has also reproduced the saved-completion case without recounting it

Thanks for checking the normal monitor. Did the exported batch total follow it, or does the supervisor still receive the inflated version from another reporting path?

Export matches the reconciled housing dispositions and retains the attempts beneath them; maintainer checked both after reconnect, and we've closed the duplicate-completion counting fault