I cannot test our coupon result parser with an endlessly agreeable mock

No. We added that sequence and found the old pass stays on screen with no age or failure indication. New tests already earning their keep.

Check the exported result too. A correct screen won't help the next person reading an old pass from a report.

Keep the legitimate negative result distinct from request failure. I've seen both labelled failed, which left the operator unable to tell whether the coupon was rejected or never read. You can test that distinction without the cell.

And startup before any valid result.

Parser now distinguishes a valid negative inspection result from no valid response. Missing fields, timeout and unexpected shapes show unavailable. Startup has no retained result; after a failed read, any old result is separately labelled with its age.

Try a valid result for the wrong coupon identity too. A well-shaped response can still belong somewhere else.

Robin, did the export problem exist as well?

Yes, export copied the stale pass. It now uses the same validated result state as the screen. Wrong-coupon reply remains unmatched and cannot populate the selected coupon.

Try recovery to a matching valid reply after those failures. The screen should not stay unavailable forever just because the new tests taught it some caution.

Recovery passes, including after reopening. Maintainer compared screen and export across the saved cases; both agree on result, identity and age. We removed the always-true fake from this suite. Thanks, Noah, the two meanings of failed were genuinely confusing.