Yes, housing labels persist. The dashboard increments from completed inspection records, including retries.

Separate item identity, attempt identity and the quality decision. Completion alone isn't acceptance.

Derive the displayed total from distinct batch-item identities meeting the agreed acceptance rule, while retaining the attempt records; merely suppressing repeated screen increments will not make restart reconstruction reliable.

Who decides which reviewed retry changes the housing's acceptance, rather than just adds another test result?

Quality approves the retry. We haven't written how that decision affects the summary after a later failure.

Resolve that rule with quality. Neither latest result wins nor any past pass wins is automatic.

Can the dashboard show pending review while that item's disposition is undecided?

Yes, as a defined disposition separate from accepted count; retain the sequence and evidence so an uncertain item is not forced into pass or fail for the convenience of the total.

I've drafted separate accepted, pending-review and other dispositions. Attempt history stays visible beneath each housing.

And make sure one housing can't acquire a second batch-item identity simply because someone reloads the job.

Keep batches distinct too. Reused label text across batches mustn't merge different physical housings.

The key is now batch plus housing label. Attempt IDs are separate; retries don't create new items.

Test reconstruction from the stored records as well as live updates, including duplicate event delivery and a crash before the summary is saved; the expected total should come from the same rule in both paths.

Are any older attempts missing their housing identity?

Two older records lack a reliable label. They remain unassigned; I haven't guessed from neighbouring attempts.

Those need to stay visible as unresolved records even if the labelled housing total looks tidy.

Has quality approved the later-failure rule yet? The software tests cannot settle that decision.

Not yet. Duplicate and restart tests pass against the draft rule, but production summary release waits for quality and the two unassigned records.

Retain the draft-rule version with those results, so approval or revision of the policy leads to an explicit rerun rather than a claim that tests under different assumptions already cover it.