the dashboard counts retries as extra plates

Made a tiny example. One failure, approved retry, pass, then the same pass delivered again after restart. Their query counts two passes. At least we can reproduce it without the robot now.

Give that replay the same attempt identity as its original result, or the example will accidentally describe a new inspection rather than a duplicate delivery.

Does a later failed check ever withdraw a plate's earlier acceptance? You need that rule too. Counting anything that has ever passed sounds simple until someone reviews an accepted plate again.

Have quality define the current item disposition, including a reopened review, before making the report infer it from whichever result has the latest timestamp.

Quality says reopened review removes it from accepted until a new disposition is authorised. Programmer now has separate item, attempt and received-message examples. The original failed check stays visible.

And keep that authorised disposition linked to its supporting attempt and review, so a trainer can explain why the plate counts without pretending its earlier failure disappeared.

What about plates with no readable label? There always seems to be one that somebody wants included because everyone knows which it is.

We had exactly that argument over a smudged label. The report wasn't the place to resolve it. Put unmatched work somewhere visible for review, otherwise the nice total becomes an invitation to guess.

Unmatched records now stay outside the accepted item total with a review count beside it. No automatic match by time. Added a same-label-in-another-batch example as well, which exposed another shortcut.

Does the item identity include the batch context where labels can be reused?

And has replaying the full saved history produced the same total as receiving those messages one at a time? I'd like to hear that result.

Priya, yes, labels are scoped to the batch. Ben, full replay and incremental processing now agree across our saved cases, duplicates included. Quality checked the resulting item list against the batch records. Revised report accepted.

Thanks, Victor. One plate with a messy history explained this much better than my original paragraph about counters.

Keep that messy plate in the lesson. The spotless example teaches everyone the one case that never needed explaining.