I think decoupling retry logic from the “there’s something wrong” UI ends up being a better experience than tieing the UI state to the details of network retries. (For one thing, it gives you a chance to fix the “everything is broken” UI without any action on the user’s part.
