Interface ViewState

Which dimensions to display (max 3, indices into nD space)

Current position in nD space (one value per dimension)

Tolerance for slicing in each dimension (radius in non-displayed dims, 0 for displayed)

Dimension metadata for the dataset.

IMPORTANT: This field is REQUIRED for the extend_to_all feature to work. If extend_to_all is configured on a node but dimensions is undefined, the optimization will silently be skipped.

Always provide dimensions when using nD datasets with extend_to_all.

Same shape as LinesViewState.dimensions, GSplatsViewState.dimensions, PointsViewState.dimensions, and BaseViewState.dimensions. Callers that need richer dimension state (selected dim, animation state, etc.) read it off sceneDimsManager directly rather than reaching for a different shape on this field.

Per-tick LOD time budget (ms) for progressive loaders during dimension ANIMATION playback, or absent for normal full-refinement behavior.

PER-PASS DIRECTIVE, not state. This field rides only the updateView(partial) call: the scene loader destructures it OUT before merging into its persistent view state (a stale budget would leave loaders capped after playback ends), threads it through the per-type handler ctxs, and injects it into the DERIVED per-node view state right before loader.updateView — so refinement and retry passes (which derive independently) are budget-free by construction. It must never enter viewStatesEqual (would reset ladders on play/pause) nor the SliceCache key (buildSliceViewSig).

Pinned additive-ladder depth for progressive loaders during dimension playback and scrubbing — the "playback detail" setting. A number makes a pass load EXACTLY min(ladderDepth, nLods) rungs, cold or not, ignoring the time budget, so every frame of a time-lapse is drawn at the same rung and the tick waits for the data instead of showing whatever happened to be resident. 'auto' lets each loader resolve its own depth from its energy stamps (resolveLadderDepth); a ladder without stamps stays time-budgeted. Absent = the time-budgeted behaviour.

Like frameBudgetMs this is a PER-PASS DIRECTIVE, not state: the scene loader strips it before persisting the view state, threads it through the handler ctxs into the DERIVED per-node view state, and the SlicePrefetcher hands the same value to its shadow passes so the t+1 S-cache entry carries the pinned prefix. It must never enter viewStatesEqual nor the SliceCache key.

Set only on the SlicePrefetcher's shadow pass: marks stores as PREFETCH so the loader pins the cached ladder until the foreground tick restores it (see SliceCache.set({ pin })). Without the pin, the just-stored t+1 is the MRU entry and the FIRST victim of a subsequent scan-eviction under budget pressure — defeating the prefetch exactly in the thrash regime. Like frameBudgetMs this is a per-pass directive: it must never enter viewStatesEqual nor the SliceCache key (buildSliceViewSig).

Set by deriveNodeViewState when this node's composed nd_transform maps the current WORLD slice to a position that no local value can occupy — a DISCRETE dimension whose inverse image falls between grid points (e.g. scale: 2 at an odd world frame). The node must then render NOTHING.

DERIVED PER-NODE, not global state. It is computed from the node's own transform in invertNdTransformForQuery (which documents the rule), so it only ever appears on the derived per-node view state, never on the scene loader's shared one. Each geometry's range query honours it by returning an empty range list, whose existing "no visible elements" path clears the geometry. It must never enter viewStatesEqual nor the SliceCache key — both already vary with slicePosition, from which this is a pure function.