autostart was false on both hosts, so neither bridge survived a reboot — the standing link between favela and the existing bsky/fediverse identities would have quietly stopped at the next restart with nothing to indicate it. The watch loop already catches every cycle error (a peer restarting, a sink respawning on OTA, a socket that vanished are all ordinary here), so a boot race against favela-peer or hyper-proto-peer is survivable rather than fatal. That is what makes autostart safe. And the flag finding, which cost real time today. `--watch` calls cycle(true) unconditionally; SEND is read only by the one-shot path. So --send has no bearing in watch mode, and the deployed pgh args are ["--watch"] with no --send — which is why ashatron's daemon logs "mode SEND" while /proc/<pid>/cmdline shows no such flag. I read the cmdline, read `const SEND = argv.includes('--send')`, and concluded the two contradicted each other, when the loop simply does not consult SEND. Deliberately NOT resolved by making watch honour SEND: both deployed daemons depend on this and would silently stop publishing. Documented instead, the startup line now says SENDS on every cycle, and --limit 0 is the way to run a watch that publishes nothing. 17/17 pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>