Restore lifecycle

Review draft. This source page has not completed website editorial and release-specific verification.

Load validates the request/container, reads canonical bytes, checks versions/integrity, normalizes supported legacy Actor record v1 data, and validates the current map policy before restore.

Pipeline

1. Document preflight: magic, versions, byte count, CRCs, record structure, request and map policy.
2. Contributor preflight: match IDs; enforce missing required and unknown-required failures, skip unknown optional with warning, and require exact schema equality. Then validate known payloads.
3. Before-Actor contributor dispatch: dispatch BeforeActorRestore contributors before the Actor restore manager starts.
4. Actor discovery and creation: match persistent IDs, resolve currently available placed/runtime targets, and provisionally spawn and register eligible missing runtime Actors. Missing placed Actors may be deferred under the existing policy.
5. Post-creation contributor dispatch: dispatch AfterActorCreationBeforeProperties contributors after eligible runtime Actor creation/registration and before saved Actor state is applied.
6. Actor property restore: apply supported Actor/component/subobject state and tombstones.
7. Post-property contributor dispatch: dispatch AfterActorPropertyRestore and then FinalPostRestore contributors.
8. Completion: normalize the final result and deliver operation completion notifications.

Each contributor selects one phase. Priorities order contributors within their phase (lower first, deterministic tie ordering). At AfterActorCreationBeforeProperties, eligible Actors have been discovered/created and runtime Actors are registered, while saved reflected properties and component/subobject state have not yet been applied. A placed Actor unavailable in the current world can remain deferred under the existing policy.

Failure and rollback limits

Required-record/schema validation is preflight: such failures happen before contributor restore callbacks and before Actor/world restore. If later pipeline work fails, tracked provisional spawns can be rolled back. State already applied to pre-existing Actors, components, subobjects, or external services is not generally transactionally reversed. SaveGame Director does not promise whole-world rollback. Result status/severity/messages/warnings report failures and partial restore.

Placed records can be deferred while their target registration/level is unavailable through supported lifecycle paths. V1 does not orchestrate map travel or guarantee all streamed-level layouts. The game controls map and level availability.