# OL-4257 — Update rolled back (supervisor caught restart loop)
> The new binary crashed-on-startup three times in 60 seconds. The
> supervisor (systemd, launchd, etc.) restarted us each time; the
> third start observed the sentinel + `<exe>.bak` and rolled back to
> the previous version.
## When this fires
- A release shipped a crashing binary on your platform/arch.
- An on-disk corruption between the apply and the third restart.
- Hard external dependency change between the apply and restart
(filesystem layout, network config, etc.).
## How to fix it
This event is a fail-safe firing successfully — no action required.
The previous binary is back in place at `current_exe()`, the sentinel
+ `<exe>.bak` are cleared, and the restart-tracker is reset.
If you observe `OL-4257` repeatedly:
1. Capture the panic backtrace from the new binary (it's emitted to
stderr / journald before the supervisor catches the restart).
2. File an issue with the panic + your platform / arch — a release
that crashes on first run is the kind of bug the rollback exists to
buy time to fix.
## Related
- `.claude/rules/auto-update.md` — supervisor-restart-loop section.
- Telemetry: surfaced as `update_completed { rolled_back: true }`.