A page caught mid-recycle recovers by itself

Nodes restart routinely — you save an edit that recompiles a view, an administrator recycles a space, a background release rolls a node forward. Restarting takes a moment, and any request that arrives inside that moment cannot be served yet.

The platform has always known the difference between "this address is coming back in a second" and "this address is gone", and everything that reads live data is built to sit out the first and give up only on the second. What went wrong is that the verdict was being lost in transit: the node correctly said "I am restarting", and the layer that carried the answer back to the page rewrote it as a permanent failure. The page then stopped waiting and showed an error for something that had already fixed itself.

Now the original verdict travels intact. In practice: a view you open (or that is already open) while its node is restarting waits the extra moment and renders, instead of showing an error you had to reload past. Genuinely missing content is unaffected — that still reports as missing straight away, with no retrying.