---
type: machine-down
coverage: 16
source: COMPLAINT_REGISTRY.jsonl via taxonomy.json _meta (2026-08-08)
absorbs: [system-crash, machine-lockout, graphics-crash]
---

# KB — machine-down

audience: AI coding agents first. Type-level patterns; per-incident entries land separately.

## Coverage

16 records. Sub-counts: `system-crash` 13, `graphics-crash` 2, `machine-lockout` remainder. Often downstream of overload on shared kernel — OOM/IO takes compositor with it.

## Recurring failure patterns

1. **Graphics/compositor death** — black screen, Cinnamon/Xorg restart loop; SSH may still work while VT7 dead.
2. **No evidence before reboot** — journal lost; repeat incident undiagnosed.
3. **SSH lockout on buildbox** — past root cause: sshd config edited by agent; owner locked out remotely.
4. **Watchdog reboot without parked runs** — harness auto-resume after unclean boot re-enters crash loop.
5. **Assumed total death** — active VT on tty1-6 ignored; `/sys/class/tty/tty0/active` not checked.

## Fix vectors that worked

- Capture `journalctl -b` + `dmesg` before power cycle.
- Switch VT, restart session service only when journal points to compositor.
- Buildbox wedge → reboot box; laptop stays up (placement isolation).
- Boot-to-drain: harness parked after watchdog reboot (I5 execution-plane invariant).

## Related skill

`od-recovery`