Either matters, Thomas, though the fixes differ. I'd first keep an example of the raw response with the installed environment details and the wrapper's output; that makes it possible to distinguish extraction from display persistence without guessing the SDK's structure.

It creates a fresh reading from part of the failed response; the old observation isn't involved in this case.

Our rebuilt laptop recently selected a different library copy than expected. I'd verify the loaded package and matching documentation here too. A neat adapter for the version written in the handover can still be wrong for the version actually imported.

Loaded version is confirmed, and its documented request result is not the status value our wrapper currently extracts.

Then make that request result the first gate. Only a documented successful response with the required data should become an observation; a failed or unfamiliar response should carry an explicit reason instead. Keep that interpretation in one place so the display doesn't improvise its own second conversion.

What will I see on the screen while that gate rejects an answer?

Thomas is asking about the person using it, not another return type. The display needs an unavailable current result and enough context to distinguish a failed read from no read attempted. Earlier observations can remain separately identified, if that is useful here.

I will put the failure reason beside the attempted-read time; any previous value will stay under a separate historical heading.

The adapter now rejects our captured failure and unfamiliar-response cases, but the screen still stamps the last good value with the latest attempt time; that separate display defect remains open.

That timestamp would still tell a misleading story, even with the value under a historical heading. Keep the screen out of reliance until the observation time survives the failure path, and retain this exact sequence as a test rather than closing the work at the adapter.