Thanks Aisha. I can get the installed version and recorded query returns. We only need this to test the parser and screen, not to imitate the physical cell.

Add explicit expected screen states for those recorded returns. Otherwise the mock may become more realistic while the test still checks only that the page renders.

And check what happens after a valid result followed by a failed request. I mean the screen shouldn't quietly keep the old result looking current just because the failure case itself no longer crashes

That case caught another gap. The old result stays visible with no age or failure indication. I've added a separate expected unavailable state to the screen review.

Include startup too, before a valid result exists. That case cannot borrow an old reading (unless the mock invents one for it).

Keep the original typo as a failing interface test after the replacement is written. It is a useful small check that the new substitute hasn't accidentally become permissive again