Job counter and plate label survive with the result. Our seen list belongs to the connection and starts empty.

Move that check to retained history, but confirm the job counter's scope first. Does it start again when its producer restarts?

Yes. Producer restart resets the counter, so counter plus plate label could match an earlier run too.

Have the author provide a retained run identity with the attempt reference. Making the laptop remember more won't fix a producer identity that gets reused.

Keep the old ambiguous entries unresolved where the run cannot be recovered. Do not invent a run link merely to make the new key fit existing history.

Has the author settled that run identity, Zara?

Yes. Revised records include producer run and attempt identities. Reconnect now reconciles known results through saved history; old entries without a recoverable run link stay marked for review.

Test a repeated delivery within one run and the same counter in a new run. The first should be one record; the second must not disappear as a duplicate.

And assert that reconnect sends no new inspection request. The bookkeeping repair should not acquire a helpful retry on its way through the code.

Both run-scope cases passed, including application restart. The recorded commissioning replay also sends no new request on reconnect. Saved history and displayed check totals agree; unmatched old entries remain visible for review.