Volumetric video.
Ship the cloud's motion, not the light-field.
The frontier of "video": a scene that is volumetric and moves. A
wai.splat.4d carries a moving Gaussian cloud as per-Gaussian keyframed
tracks — position, scale, rotation, color, opacity over time — not a recorded
light-field. Your browser interpolates the cloud at any frame with the same deterministic
Rust engine (compiled to WASM) that made the conformance corpus, and re-derives every
frame's Gaussian-set hash: frame-state-equivalence, the temporal counterpart of
the static splat. Scrub the frame slider — the cloud is reconstructed, Gaussian by Gaussian,
in the fixed-point floor.
Motion as instructions
A frame is the cloud interpolated from keyframed tracks — the wire carries the trajectories, the sink reconstructs the Gaussians. The render of that set — EWA splatting, depth sort, alpha blend — is the sink's; here it's a simple 2D projection, but the verified object is the cloud, not the pixels.
Frame-state-equivalence
Reconstruction is bit-exact: the wai.det.fixed64 floor, linear interpolation
of each Gaussian's keyframes. The Gaussian set at frame N hashes to one BLAKE3 on every
machine — the temporal counterpart of the static splat, with no tolerance parameter.
Honest about size
Volumetric is heavy — a real cloud is megabytes. The win is the same as the static splat: content-addressing, dedup, one signed envelope, and a verifiable per-frame state — not payload reduction. We say that plainly.