[indent]

After a restart that couldn't reuse the cached session, two logins got HTTP 503: 130: Over capacity from inbox_initial_state.json. Since then every user_updates.json poll has failed the same way (all 6 attempts, then backoff, ~20 give-ups/hour per login for 9+ hours). The polled URL has no cursor parameter. Before the restart, both logins had been polling without errors from a cached session's cursor.

Cause: syncUntrustedChannels is the only place that seeds PollingCursor. When GetInitialInboxState fails it just logs and returns, and connect starts polling anyway with an empty cursor. Nothing ever fetches the initial state again, so message requests and legacy group DMs stop arriving until the next reconnect.

Change: syncUntrustedChannels returns the fetch error, and the polling goroutine retries it with backoff (1 min, doubling up to 30 min) before calling StartPolling. Auth errors still go straight to polling, which already handles them.

Tag @indent to continue the conversation here.