The FR5 test screen stays green after I remove the read

It only checks colour. Initial value is accepted, so no read is needed to satisfy it.

Start unknown. No observation should not look accepted.

Check the status and its source together. For our equipment screens I ask for the sample identity and observation time beside the result, because a correct-looking value can still belong to the previous sample. Your offline test can exercise that distinction without a controller.

Mina, yes, though Ada should first make the missing read fail. Otherwise a larger test can hide the same broken assertion under more setup.

Missing-read case fails now. Fake records calls, and display starts unknown. Next: return shapes, not more green.

Use the documented interface your adapter actually wraps. Don't invent a convenient response shape.

Does the fake also reject a method name your real interface doesn't have, or could a typo still get a cheerful answer?

Typos raise now. Unsupported arguments too. Recorded package documentation supplies the boundary cases; no catch-all true.

Try failure after a valid reading. That's the one I'd worry about as an operator: I have seen it work, then the connection fails, and the old good value sits there looking reassuring.

And distinguish a communication failure from an actual rejected housing. Both may need attention, but the person handling the housing needs to know which one occurred. Did your previous display make that difference visible?

No, Mina. Same red state for both. I've split unavailable from inspected reject and retained the reason.

Has the good-then-failed sequence caught anything else, Ada?

It left the old sample name beside unavailable. Result cleared, identity didn't. Fixed and added that assertion.

Nice catch. What does an unreadable result do?

Kai means a response your adapter can't interpret, I think, rather than a housing the inspection has rejected. Worth giving that its own test input.

Yes, malformed response. My wording was vague.

Thanks for spelling that out, Kai. Include the diagnostic reason without showing the housing as rejected; otherwise the operator may send a perfectly unassessed sample down the reject route.

Malformed response stays unavailable with a reason. Good/failure/reject, wrong identity and missing-read regressions all pass. Offline display bug closed.

Useful result. Give the bench maintainer those sequences and the package reference for the installed comparison, so they know exactly which behaviours you exercised rather than receiving a screenshot of a happy screen.