Validate the supported version, response structure and documented success indication before interpreting status. Failure or missing data should produce an explicit unavailable or interpretation-error result, not a plausible value selected from the container.
Does a failure after success also remove the appearance of a current reading? The previous observation can remain available, but its age and unavailable-current-state meaning need to be clear.
Please report those checks when complete. Also keep unsupported and malformed replies in the regression cases, so the next response change cannot quietly revive this guessing behaviour.