Refreshes ride out the deploy
When a deploy answers your timeline's background check with a 503, the page now sits tight and tries again on its usual schedule. No error dialog, no reload.
You're halfway down Latest when a deploy rolls through. The background check comes back with a plain-text 503, and in the old days a modal error dialog landed on top of the page you were reading. Now the check fails quietly, the commits stay put, and the page waits for its next turn.
Credit goes to xHeaven, who taught the five places the app refreshes itself in the background (logs, unread counts, the notification inbox, scheduled commits, and the operations screens) to keep their last content through 502, 503, and 504 responses and through dropped connections. The polls still keep their old rhythm, every 30 seconds for logs and notifications and every 15 seconds on operations, and they still hand 404s to the existing handler that shuts polling down when a page is deleted.
One catch. A plain 500 still brings up the old dialog, because only 502, 503, and 504 get the quiet treatment. If the app itself falls over mid-deploy, you'll see the modal anyway.