Shadow
View-depth extent of shadow-relevant scene content, accumulated from drawable bounding spheres as they are offered and consumed by ShadowLayer on the next frame (the shadow layer renders before content layers). Fitting cascades to actual content — instead of a terrain-ray probe — is what keeps street-level views at centimetre texels.
Properties
More casters contributed than casterSpheres holds; maxCasterLightZ kicks in.
World bounding spheres (x, y, z, radius) of ELEVATED casters (camera-relative light-axis top above ELEVATED_CASTER_THRESHOLD), consumed by ShadowLayer to lift only the cascades whose light-space window each sphere overlaps — one model at altitude must not inflate street-level cascades. Ground-level content never registers; each cascade's slice pullback already covers it.
Highest CAMERA-RELATIVE light-axis coordinate (dot(lightDir, center - camera) + radius) of any contributor. Consumed only when casterOverflow is set: the per-caster sphere list is the precise mechanism, this is the conservative global fallback that lifts every cascade. Kept camera-relative so ShadowLayer can re-base it to its ANCHORED light axis without planet-magnitude error.