Diagonal of the cube a synthetic LINES walk fills — the stand-in for
the position_bounds a compiled node carries, so the debug injector
can size its line primitive the way createLinesNode does (authored
width normalized by the node's own extent — see
types/line-primitive.ts). Without it a WIDE synthetic scene would
read as thin and build the capsule where production builds the quad.
The walk is re-anchored uniformly inside [-bounds, bounds]^3 every
64 steps and steps by at most bounds × stepScale, so the cube
diagonal is its extent to within the drift a run of 64 steps can
accumulate: negligible for the unbiased walk (~√64 steps ≈ 8% of
bounds at the default stepScale), but up to ~60% of bounds for
the near-ballistic turnAngle variant. That makes this a mild
UNDER-estimate of the true extent there — i.e. a mild over-estimate
of the width factor — which stays well inside the order-of-magnitude
tolerance the rule is stated at, and does not move any bench
scenario's resolution.
Diagonal of the cube a synthetic LINES walk fills — the stand-in for the
position_boundsa compiled node carries, so the debug injector can size its line primitive the waycreateLinesNodedoes (authored width normalized by the node's own extent — seetypes/line-primitive.ts). Without it a WIDE synthetic scene would read as thin and build the capsule where production builds the quad.The walk is re-anchored uniformly inside
[-bounds, bounds]^3every 64 steps and steps by at mostbounds × stepScale, so the cube diagonal is its extent to within the drift a run of 64 steps can accumulate: negligible for the unbiased walk (~√64 steps ≈ 8% ofboundsat the defaultstepScale), but up to ~60% ofboundsfor the near-ballisticturnAnglevariant. That makes this a mild UNDER-estimate of the true extent there — i.e. a mild over-estimate of the width factor — which stays well inside the order-of-magnitude tolerance the rule is stated at, and does not move any bench scenario's resolution.