Class DirectionalLightShadow

How a DirectionalLight casts (§69) — the shadow-map resolution, the two bias controls, and the orthographic volume the map covers.

const sun = new DirectionalLight({ intensity: 3, castShadow: true });
sun.shadow.mapSize = 2048;
sun.shadow.extent = 12;      // a 24 m × 24 m box about the light's position
sun.shadow.bias = 0.002;
Copy

Every property is a validated accessor (the F14 policy, 2026-08-07): the constructor's §85 check runs again on every write, so a NaN map size is a RangeError at the assignment that caused it rather than a black frame three systems later. Nothing here is clamped or reordered — WP-3.3's no-silent-rewrites rule — so an unusable value is refused, never quietly repaired.

The volume, and why it is authored rather than fitted

A directional light has no position as far as lighting is concerned (§68 — see Node.getWorldDirection), but a shadow map is a rendering of a bounded volume and something has to bound it. This tier takes the volume from the light's own node: the map is rendered from the light's world position, looking along its −Z axis, through an orthographic box 2 × extent wide and tall with the near and far planes below. Moving or aiming the light node therefore moves the shadow volume, which is the one control an author already has in their hands.

The alternative — fitting the box to the scene's bounds every frame — is not this tier's, and deliberately: it needs a world-space bounds pass this engine does not yet run (§87's culling packet owns it), and a volume that silently resizes makes shadow texel density frame-dependent, which is the shimmer §33's determinism language exists to keep out of a recorded run.

Staged, with the date (2026-08-09, R-18)

§69 lists ten features; this settings object is sized for the one tier that ships. Named here rather than sketched, because each is public shape:

• cascaded shadow maps — one map, one volume. Cascades need a split scheme, a map per split, and a per-fragment cascade selection, and they are what make a large outdoor volume workable; extent is the honest single-cascade knob until then.
• point-light cubemap shadows and spot-light shadows — neither PointLight nor SpotLight carries castShadow at all (see the module header). A cube map needs six passes and a samplerCube; a spot needs a perspective shadow matrix and a second map slot, which is the shadow-atlas item below.
• shadow atlas management — one map, one caster light. An atlas is what turns "a per-light shadow index" into something a bounded uniform set can address.
• transparent shadow masks — the depth-only pass writes geometry, not alpha, so a sprite or a cut-out texture casts nothing here (§55 quads are skipped by the caster pass entirely rather than casting a hard rectangle).
• contact shadows — a screen-space effect, and so §70's business rather than this record's.