Find the right recovery recipe
Begin with read-only status and identify the affected machine/task. Shared services require coordination; do not repair a different task because its ports resemble an old example.
| Symptom | Start here |
|---|---|
| Shipyard stays off after power returns | Power recovery |
| SSH or Orca remote access fails | Connectivity |
| API setup stalls or health fails | API startup |
| Dashboard targets the wrong API or shows an error boundary | Dashboard |
| Browser viewer or Playwright is unavailable | Browser |
| Staging restore or permissions fail | Database and authorization |
| Archive fails or retained edits need recovery | Archive and preservation |
| Search-engine build or tests fail | Appliance |
| Package/cloud/GitHub access fails | Access |
| Memory/disk usage grows | Resources |
| Simulated cameras, webhook receiver, build server address or appliance cloud link fail | Home test bed |
Runbooks describe their tested limits. A command that changes data is not authorized merely by being listed here. For unknown behavior record the symptom, redacted error, target, relevant versions and attempted steps. Add a validated recipe after the cause is understood.