performance

Reel separates immutable setup data from mutable runtime data so a large group of matching characters can avoid repeated loading and binding work.

Data ownership

Data	Recommended ownership
AnimationAsset	Shared per animation
RigDesc	Shared per matching skeleton
AnimationBinding	Shared per asset/descriptor pair
RigState	One per character and worker
AnimationState / DeltaState	One per playing track and character
Prepared native/delta lists	Owned by the character or worker that mutates them

Assets, descriptors, and bindings are frozen. Do not share mutable rig state, track state, or prepared lists across Actors.

Parallel Luau shape

Partition characters by Actor or small batches of characters.

Within each worker:

1. Update local track states
2. Call solve_prepared()
3. Optionally apply delta layers and reconstruct world transforms
4. Call task.synchronize() before assigning Motor6D.Transform, AnimationConstraint.Transform, or Bone.Transform

Keep one descriptor and one binding per compatible asset/rig layout in that worker. Avoid splitting the solve for a single small rig across workers, the coordination cost usually dominates.

Allocation rules

• Load KeyframeSequence assets and rigs during setup, never per frame
• Reuse prepared lists while track membership and priority are stable
• Allocate native blend scratch once for rigs that can blend two effective tracks
• Pass a preallocated world output array in hot loops
• Skip delta processing and world reconstruction when the caller does not need them

Profile solving, world reconstruction, and Instance commits separately for parallel luau.