open-standards/wai/volumetric-conformance

Volumetric / 4D Splat Conformance Corpus

Vectors for the volumetric extension (wai.splat.4d). The criterion is frame-state-equivalence: given the same temporal-splat clip, a conforming sink MUST reconstruct the identical Gaussian set at every frame, hence the identical BLAKE3 — on every machine, no tolerance parameter. The temporal counterpart of the static splat, with the single cloud replaced by per-Gaussian keyframed tracks.

Provenance. Generated here: wai_splat4d_vectors gen writes every vector from clouds and tracks authored in code. No capture.

Each Gaussian carries a track of keyframes (frame, position, scale, rotation quaternion, color, opacity, all wai.det.fixed64); the sink interpolates linearly in the Fx floor at any frame. The frame hash is BLAKE3(domain || frame || n_gaussians | per-Gaussian 14×i64). The render of that set (EWA splatting, sort, blend) is the sink's presentation, never in the hash.

Clips are self-contained, so each vector is a .w4ds container + its per-frame expected hashes.

Layout

• vectors/<name>.w4ds — the W4DS container (contract + Gaussian tracks).
• expected_hashes.json — per clip: n_gaussians, n_frames, clip_hash, and frames[] = {frame, hash} for every frame.

Vectors:

• orbit — two Gaussians swap x positions across 12 frames.
• pulse — one Gaussian scales up and fades out across 10 frames.
• morph — one Gaussian shifts color red → cyan across 8 frames.

Running

cd ../wai-rs
cargo run --features splat4d --bin wai_splat4d_vectors -- verify ../volumetric-conformance
cargo run --features splat4d --bin wai_splat4d_vectors -- gen    ../volumetric-conformance   # regenerate

Pass = clip-hash and every frame-hash equality.

Integer neural fields

fields/ holds the corpus of wai.field.int_grid (neural-fields): payloads, the values and frames every conforming sink computes from them, and fault payloads with their refusal codes. Its README.md says how it is made and checked; field_verify.py is its second implementation.