I want the failure shown honestly on our stores-side diagnostic screen. The newer SDK returns a container where our old wrapper expected a status value, and a failed read now looks plausible. I've asked the maintainer to compare the documented response formats, but what should we check on the screen as well?
Try a good read followed by a failure and watch the old value. It needs an obvious stale indication, not just an error tucked under a healthy-looking number.
The old value stays, and the page updates its time on every refresh. So even when it fails, the old reading gets a new-looking timestamp. Two problems for the maintainer.
Ask for the observation time separately from the refresh attempt; actually, label both if both are useful, because one unlabeled time is how we ended up reassuring ourselves with old data.
We have test cases for both documented versions, reported failure and malformed data. The revised screen shows the last valid observation time and a separate read failure. Still to be tried by the workshop users.
Let them explain what they'd do after the failure without prompting. That will tell you whether the wording helps or merely satisfies everyone who wrote it.