- Fail outbound sends while the app is rebuilding instead of returning quietly, so ChannelManager retries and logs the failure instead of counting the message as delivered
- Close the bot's HTTPX pools during teardown: Application.shutdown() skips them when initialize() never finished, leaking one pool per startup retry
- Propagate terminal startup errors (rejected token, bad proxy, bound webhook port) and clear the running flag instead of retrying forever while the channel still reports itself healthy
- Restrict startup retries to NetworkError/TimedOut, the exceptions HTTPXRequest wraps every httpx failure into
- Scrub the rejected-token failure so PTB's token-bearing message never reaches the log
- Wrap the getUpdates request pool to record completed round trips, since a healthy long poll completes one every ~10s even with no traffic
- Replace the idle keep-alive loop with a supervisor that tears down and rebuilds the application (including its HTTPX pools) when no round trip completes for 120s
- Retry failed startups with exponential backoff so the bot self-heals once the network path recovers
- Give up immediately on InvalidToken since a rejected token is a config error that retries cannot fix
- Tear down the freshly built app when stop() lands mid-startup so the supervisor never leaks a running application
- Bridge python-telegram-bot and httpx stdlib logging into loguru so polling failures are visible in nanobot logs
QQQ300kuaiandchengyongru