RID playback and rewind

The .scenerid stores recorded scene animation data. The .scene decides which recorded channel plays, which performer or camera receives it, where it is placed, and when it appears on a rewindable timeline. Neither resource is self-sufficient.

Research note: the field shapes and focused counts from sq012_braindance__part_a.scenerid and sq012_02a_braindance.scene are Observed in vanilla on Cyberpunk 2077 2.31a, re-extracted and inspected with WolvenKit 8.19.0 on 2026-08-09. The retained mod-owned RID/scene pair round-trips with WolvenKit 8.19.0 and is Structurally validated. Custom playback, camera, seek, rewind, facial/cyberware channels, and interruption remain Experimental.

Test from an untouched pre-install save. Scene playback, playhead position, clue state, camera ownership, and quest facts can survive in saves or the current session. A save already inside the braindance is not a clean replay.

Two resources, three reference levels

The complete recorded-animation lookup is:

scnRidResource (.scenerid)
  actor/camera scnRidTag(signature, serialNumber)
  animation scnRidSerialNumber + buffer
          |
          | selected by resource ID + animation serial
          v
scnSceneResource.resouresReferences
  ridAnimations[] / ridFacialAnimSets[] / ridCyberwareAnimSets[]
  ridCameraAnimations[]
          |
          | selected by scene-local reference-array ID
          v
scnRewindableSectionNode event
  performer + animResRefId
  or cameraRef + animSRRefId

The misspelled serialized property name resouresReferences is the native field exposed by these resources. Preserve the field name WolvenKit presents; do not “correct” a serialized property based on prose.

RID resource identity

An scnRidResourceHandler in the scene binds:

• a typed scnRidResourceId; and
• an exact .scenerid ResourcePath.

The resource ID is scene data, not the depot-path hash and not an array index. A scene reference such as scnRidAnimationSRRef then combines that resource ID with an animation scnRidSerialNumber from the selected RID.

Scene-local reference identity

RID animation events do not carry the resource path and serial directly. A body event’s scnRidAnimationSRRefId selects an entry in the scene’s RID animation-reference arrays. A camera event’s scnRidCameraAnimationSRRefId selects a camera-reference entry. Array membership and typed ID must be re-audited whenever resources or channels are added, removed, or reordered.

Performer and camera identity

scnPlayRidAnimEvent.performer is an scnPerformerId; it is not the RID signature. The scene’s actor/player/prop definitions establish the performer map. A camera event instead carries a world cameraRef NodeRef, a camera RID reference, and an origin marker.

This produces two independent joins:

RID actor signature -> RID animation serial -> scene RID ref -> performer
RID camera signature -> RID camera serial -> scene camera ref -> camera NodeRef

Do not force actor and camera identities into one shared counter scheme.

What version-5 .scenerid owns

The focused SQ012 part-A resource has root type scnRidResource, version: 5, and these principal fields:

In this exact vanilla part, nine actor entries and one camera entry are present. Seven actor entries have one body and one facial animation; one of those also has a cyberware animation. Two prop-like RID actors have no animation arrays. This observation demonstrates optional channel sets; it does not define a required custom layout.

Animation buffer invariants

An animation entry ultimately owns an animAnimation and an animation buffer such as animAnimationBufferCompressed. Focused properties include:

• animation and buffer duration;
• frame count;
• joint count and track count;
• raw/constant key counts;
• extra joint/track counts;
• compressed or deferred buffer bytes;
• optional motion extraction;
• camera LOD samples for camera animation.

Those numbers must describe one compatible rig/channel layout. A binary can serialize while still containing the wrong joint order, wrong track mapping, invalid root-motion basis, mismatched duration, or incompatible facial/ cyberware cardinality. WolvenKit round-trip success is therefore necessary but not sufficient.

For every custom channel, record a small manifest like this before linking the scene:

Do not publish a vanilla .scenerid, its full serialized JSON, or its compressed buffer bytes. The table is a focused provenance record, not a redistribution mechanism.

Recorded placement is another contract

A body event observed in the SQ012 scene includes:

• performer (scnPerformerId);
• animResRefId (scnRidAnimationSRRefId);
• animOriginMarker with a global scene-marker NodeRef;
• actorPlacement: SceneOrigin;
• startTime and duration;
• blend, collision, component, camera-parallax, and trajectory settings.

The first event in the inspected section uses the global origin #sq012_02a_sm_braindance, begins at 0, and lasts 33333 scene-time units. That exact value is Observed in vanilla, not a universal conversion rule for seconds or frames.

If an actor’s recorded root is authored in a different coordinate basis than the scene marker, every otherwise valid pose can appear offset or rotated. Audit these as one transform equation:

world scene-marker transform
  * event placement/origin policy
  * RID actor root / motion extraction
  = expected world-space performance

Test translation, yaw, elevation, and return-to-origin separately. A single actor standing at (0, 0, 0) cannot expose every basis error.

Recorded-perspective camera

The inspected scneventsPlayRidCameraAnimEvent contains:

The SQ012 scene’s camera NodeRef, the camera prop/entity, the RID camera reference, the origin marker, and the event are five separate checks. Seeing a camera in the streamed world does not prove the RID camera channel is active.

engine\scenesystem\camera.ent exists in the installed 2.31a archives and is a useful type/shape comparison for a placed scene camera. Its existence is not permission to reuse arbitrary vanilla placement or quest-local IDs.

Rewindable section anatomy

The native timeline container is scnRewindableSectionNode, not an ordinary scnSectionNode with a rewind flag added elsewhere. In the focused SQ012 scene it owns:

• sectionDuration.stu: 89300;
• ffStrategy: automatic;
• forward slow/fast/very-fast modifiers 0.5, 3, and 6;
• backward slow/fast/very-fast modifiers 0.5, 3, and 6;
• 115 timed events;
• output sockets for ordinary progress and side routes.

Its event inventory includes dialog/audio, skeletal animation, RID actor animation, RID camera animation, clue windows, BD visibility, VFX, prop attachment, and socket events. This is Observed in vanilla for the named scene. It demonstrates that rewindability coordinates multiple event types; it does not require every custom recording to contain all of them.

Rewind safety questions

For each event, answer:

1. Is it a pure function of current timeline time, or does it cause a durable side effect?
2. What state must be reconstructed when the playhead seeks into its middle?
3. What is undone when the playhead moves backward across its start?
4. Does its end state differ when playback jumps past the whole interval?
5. Does it target an actor, prop, camera, clue, VFX, UI state, fact, or socket?
6. Is the side effect allowed to happen more than once?

Facts, journal mutations, item operations, spawning, and quest sockets deserve special scrutiny. A timeline may revisit their event time, while the quest side effect may be intentionally one-shot. Make that policy explicit instead of assuming rewind will undo persistent state.

Inspect the RID link in WolvenKit

Use a disposable project and extract:

base\quest\side_quests\sq012\scenes\sq012_02a_braindance.scene
base\animations\quest\side_quests\sq012\sq012_braindance\rid\sq012_braindance__part_a.scenerid

Then record this focused cross-reference without copying the resources:

1. In the scene, list each ridResources[] handler’s resource ID and depot path.
2. In resouresReferences, list each RID animation/camera reference’s resource ID and serial number.
3. In the RID, find the corresponding actor/camera tag and animation serial.
4. In the rewindable section, select one body event and one camera event; record their local reference IDs, performer/camera, origin, start, and duration.
5. Verify that the event’s local reference resolves to the expected RID resource/serial pair.
6. Compare the event interval with the animation duration and the section duration.

Do not rely on graph-screen position. The typed IDs and property values are the evidence.

Manual composition order for a mod-owned candidate

1. Finish the ordinary scene actor/prop acquisition and named entry/exit contract before introducing RID playback.
2. Add the mod-owned RID resource and inventory all tags, serials, channels, durations, joint counts, and track counts.
3. Allocate a unique scene scnRidResourceId and add its handler/path.
4. Add scene-level body/facial/cyberware/camera references, pairing the exact resource ID with exact RID animation serials.
5. Add one short body event and verify its performer, origin, placement, duration, and local reference ID structurally.
6. Add one short camera event and verify its camera entity and marker chain.
7. Only then add the full rewindable section and the remaining channels.
8. Re-inspect every reference ID after array edits; do not assume earlier indexes stayed stable.
9. Serialize to CR2W, serialize the binary back to JSON, and compare root type, RID handlers, serial mappings, event references, durations, and cardinalities.
10. Keep the result Experimental until all runtime cases pass on the exact installed build.

This sequence localizes failures. It is not a claim that arbitrary compressed animation buffers can be authored correctly by filling WolvenKit fields by hand. A buffer-production tool may be needed, but it remains an author tool, not evidence that the native runtime contract passed.

Structural acceptance checklist

• .scenerid root type/version and each actor/camera tag are recorded.
• Every tag and animation serial is unique in its required scope.
• Every buffer has expected duration, frame, joint, and track counts.
• Every scene RID handler resolves an exact mod-owned depot path.
• Every scene RID reference resolves an existing resource/serial pair.
• Every event’s local reference ID selects the intended scene reference.
• Performer/camera, origin marker, placement, start, and duration are explicit.
• The rewindable section contains no unexplained durable side effect.
• WolvenKit binary-to-JSON-to-binary conversion succeeds.
• The archive contains the intended depot paths before runtime testing.

Passing this checklist supports Structurally validated, not Runtime-proven.

Failure routing

Next: Clue layers, cleanup, and acceptance.