A create that cannot be saved now says so

Creating a node asks the storage layer to write it. Storage is a chain of backends, and each one answers "yes, that path is mine" or "no, not mine" so the next one gets a turn. When nobody claimed the path, one of the two ways that chain can be assembled reported a clean failure — and the other simply went quiet. Nothing was written, nothing was said, and whoever asked for the node waited indefinitely: no error, no timeout, no clue.

Which of the two you got depended entirely on how the portal's storage happened to be wired underneath, which is not something the caller can see or control. That made it look random and made it very hard to reproduce — the same action would fail politely on one deployment and hang on another.

Both paths now answer the same way, and the failure names the backend that declined so the misconfiguration can be found instead of guessed at. More generally, the create pipeline now has a backstop: if it ever finishes without producing a node and without explaining why, it replies with that fact rather than leaving the request unanswered. Waiting forever is never a legitimate outcome — an answer you can act on always is.