An update that could not check itself no longer reports success

Every platform update recompiles the code behind your content types against the new version. Because that can break something, the new instance builds everything before it takes over: if a type that used to work no longer compiles, it declines to go live, the update stops, and the version you were already running keeps serving. You see a paused update instead of broken pages.

That check had one gap. It reported a pass in two quite different situations:

• It looked, and found nothing wrong. The genuine all-clear.
• It could not look at all. If the step that lists your content types failed — it could not reach the store, or the request never came back — the failure was quietly discarded and the result came out as "nothing to report", which is indistinguishable from a clean pass.

So an instance that had verified nothing could report itself ready and take over. Nothing was broken by this on its own, but it was the one way an update could get past the very safeguard meant to catch it.

Now the two are told apart. A check that could not run is recorded as unanswered rather than passed: the new instance does not take over, the update waits, and the version you are on carries on serving. The status message says so in plain terms instead of showing a green tick nobody earned. If the cause was temporary, the next attempt simply runs the check properly and the update continues on its own.

Finding nothing is still a perfectly good answer. An instance with no content types of its own verifies successfully and goes live as normal — the change is only about the difference between "nothing is wrong" and "I could not find out".