I've checked our saved examples. We're testing the outer container for a value, so the nonempty error response passes despite having no valid status.

@PavelBaker0440 Start a narrow parser for this version and replay your two examples offline. Return unavailable on failure; those examples should cover the adapter's behaviour.

That covers this bug. What about a transport exception or missing field? Neither has to look like your saved error

@RobinBaker0512 Fair point. Those two are a regression case, not full coverage. The adapter also needs documented error cases, transport exceptions and malformed responses exercised.

Can our screen just say unavailable, but retain the reason?

Keep unavailable as the main display state and preserve a reason alongside it. Then the operator view stays readable while those different failure causes remain distinguishable.

My screen kept the last good value after disconnect. People read it as current until we added an explicit unavailable label

@IsabelBennett0735 Ours does that too. No old-observation label. Fixing the parser would still leave that misleading value sitting there.

@PavelBaker0440 Why keep it? Wouldn't clearing an old mode be less confusing?

@JoAli0244 You can clear it. Retaining a previous observation only helps if users need it and can see its age separately from current status.

@LouisBell0659 I'll make our main field unavailable and put any previous observation in labelled diagnostic detail. That seems clearer for this screen.

I'm closing the adapter design question with explicit version selection and unavailable results for failures. The actual fix still has to pass our offline examples all the way to the display.

@PavelBaker0440 Makes sense. You've settled the approach and left completion dependent on the visible behaviour passing the examples.