Keep those as three levels: the physical housing, its inspection attempts, and delivery of each result. Removing repeated messages should not remove a genuine reviewed reinspection, but neither should a reinspection manufacture another housing.

Housing and attempt identities survive restart. Display says 164; we are checking deliveries against the stored attempts.

Who owns the current disposition when a housing has more than one attempt? I would want quality's rule before a developer chooses first pass, last arrival or any other convenient interpretation.

And leave the failed attempt readable. Someone will eventually ask why that accepted housing was reinspected.

Reconciliation found 158 distinct attempts on 151 housings. Seven reinspections, plus six repeated result deliveries in the displayed 164.

So 151 inspected is not yet your accepted count.

Exactly. Ask quality to classify those 151 by the authoritative disposition records. The remaining sixteen planned housings have no inspection attempt in this reconciliation; don't silently classify them as failed.

Quality reports 143 accepted, five held and three rejected among the 151. Sixteen uninspected remains separate.

Does the accepted list include the housings whose reviewed retry passed? Otherwise the new total could be unique but still ignore the permitted disposition changes you wanted to preserve.

Yes. Quality linked each reviewed retry to its housing and retained the original failed result. No attempt erased.

What will the next shift see instead of nearly done?

Proposed screen shows 143 accepted of 167 planned, with held, rejected and uninspected counts alongside; developer is rebuilding from disposition records.

Test delayed delivery of an older failed attempt after its reviewed successful retry. Arrival order should not reverse an authorised disposition merely because an old message came in last. Also test genuinely missing disposition evidence as unknown, not accepted.

What about a new failed check after acceptance?

Quality defined that as review-required hold, not automatic continued acceptance. Tests now include both older delayed failure and genuinely later failure.

Has the corrected screen survived reopening from the saved records? A display adjusted by subtracting six repeats could look right once and still return the old total on launch.

Reopen and replay checks passed. Reviewed report now rebuilds as 143 accepted, five held, three rejected, sixteen uninspected. Shift lead has the corrected total and original inflated report.

Thanks for the breakdown. The twenty-four not accepted are now visible without calling them twenty-four failed housings.