Our updated SDK returns more than the status value now. The dashboard wrapper treats the whole reply as a reading, and failed queries can look available. Which cases should the maintainer capture?
Capture the documented successful reply with its value, a failed query, a missing value and an unexpected response shape for the exact SDK version. Feed them through the same parser the dashboard uses. A non-empty container can look wonderfully reassuring while containing an error.
Include a genuine negative status and a legitimate zero where the contract allows it. Otherwise the repair might just trade always available for always unavailable. What happens before any successful reading?
Startup now shows no current reading. Documented positive, negative and zero values pass; failed, malformed and incomplete replies show unavailable. The installed dashboard passed the captured-response checks, including a valid read after failure.
Good, the valid read after failure matters. Keep those captures with the installed version so the next SDK change has something less optimistic than a non-empty test to beat.