A blip reaching a hub no longer drops the message

Servers roll over routinely — a new version ships, a pod restarts, a connection briefly drops. Most of the platform already treats that as ordinary: a message aimed at a hub going through that moment is retried a few times, and once the hub is reachable again the message lands as if nothing happened.

One delivery path did not get that treatment. Messages routed to a hub that lives in its own process — a portal session, a cache, a background sync — were given exactly one attempt. If that attempt landed during the brief window a server was reconnecting or finishing its own restart, the sender was told immediately that the message had failed, even though a moment later the very same delivery would have succeeded.

Now that path retries too, the same way every other delivery in the platform does: a few attempts with a short, increasing pause between them, before giving up and reporting a real failure. A passing hiccup is absorbed instead of surfacing as a dropped message, and only a destination that is genuinely gone still gets a prompt, honest failure.