Name the state you intend to recover
“Restart the stage” is too broad to test. Define a known show state with a content revision, map or level, nDisplay/configuration revision, active camera and lens assumptions, tracking source mapping, color/view settings, playback cue, and a reference frame. Then name what is outside that state: external devices, signal routing, console variables, media servers, physical lens settings, and any manual cue. Recovery begins by distinguishing a saved scene from the wider live system.
Level Snapshots can save and restore actor layouts and selected properties. Epic’s current documentation calls them especially useful for resetting virtual environments between takes, but also states that asset changes are not recorded and lists exclusions including USD actors and console variables. A snapshot is therefore a component of recovery, not a substitute for a release record.
Make recovery steps observable
Write the runbook in order: identify the incident; place recording or camera in a safe state; restore the named content baseline; establish display/configuration; reconnect and audit tracking; restore the approved snapshot/cue; verify through the production camera; obtain a named go/no-go decision. Every step should produce a visible signal or logged fact. “Check everything” is not an operator instruction.
| Recovery layer | Named evidence | Common gap |
|---|---|---|
| Project content | Engine and revision ID | Unsaved local edits. |
| Scene state | Snapshot/cue ID and reference frame | Assuming it includes material or console changes. |
| External inputs | Tracker, lens, time reference status | Connection indicator mistaken for data audit. |
| Physical output | Correct wall/monitor image through camera | Checking only an operator UI. |
Run a real interruption drill
During a technical rehearsal, stop one planned component under controlled conditions: restart a render node, disconnect a non-critical source, or restore after a deliberately altered light. Time the drill, but do not make speed the only score. Log which state came back automatically, which steps required judgment, and whether the camera reference matched. Also log the exact person who called the recovered image acceptable; recovery has a production decision as well as a technical sequence. If the intended failure is unsafe or would disrupt other work, use an agreed simulation and state that limitation.
Illustrative example: after a node restart, the scene opens but the wall uses a prior color/view configuration. The team should record “content recovery pass; image verification fail; restored after display preset X” rather than announcing a complete recovery just because Unreal launched.
Improve the runbook after each drill
- Keep the shortest safe procedure, but include decision ownership and escalation contacts/roles.
- Update the known-state record when approved revisions change.
- Store reference frames and snapshot labels where the relief operator can reach them.
- Repeat the rehearsal after material changes to topology, mapping, or content release.
This Atlas routine is operational rehearsal, not a claim of automatic disaster recovery. Its value is exposing what the team cannot yet restore.
Sources & evidence
Primary documentation establishes tool behavior or a reported production. Checklists and illustrative examples are Atlas editorial guidance, not results of independent testing. Match documentation to your installed version.
What snapshots save/restore, virtual-production use, and stated limitations.
Source published: Not established · Retrieved 19 September 2026
Stage-machine configuration and control context for a wider recovery record.
Source published: Not established · Retrieved 19 September 2026