Only the previous signal state, in memory. It starts false on reconnect. The stored inspection row already contains the attempt identity.

Use that stored identity to reconcile the result, with the reporting owner defining how repeat delivery is handled. A reconnect should not invent a new inspection because the application's local memory started empty.

What will maintenance see if the retained identity is unknown?

We found a replay fixed in the display but still duplicated in the exported total. Check the stored row and export as well. Reconnection is only one way of presenting the same completion twice.

Kai, unknown identity goes to reconciliation, with no count added. Offline replay of an already stored attempt now leaves both row and total unchanged. We're checking restart and export next.

Also test the interruption between storing the row and updating the total. Neither finishing that update later nor receiving the result again should count the same accepted coupon twice.

That interruption reproduced a second bug. Restart rebuilt the count correctly, then the replay added it again through a separate export path. Reporting author is changing that path to use the same stored disposition.

And what happens when the same coupon has a later attempt, with a different acceptance disposition?

Did the corrected path survive both those cases? The original duplicate and a genuinely later result should not collapse into the same test.

Yes. Replayed attempts stay single; a later authorised disposition updates that coupon without adding another coupon. Restart during the stored-row update gives the same exported result. Reporting owner also checked the installed read-only capture replay. Closed this counting fault; unknown identities remain visible for reconciliation.