PREFILTER_SAMPLE_COUNTS

const PREFILTER_SAMPLE_COUNTS:

import { PREFILTER_SAMPLE_COUNTS } from '@driftengine/core';

In depth

Powers of two, because the Hammersley sequence's radical inverse is a bit reversal. A count that is not a power of two leaves the sequence unevenly stratified, which does not error — it shows as a faint directional bias in the rough levels that reads as the environment being slightly wrong rather than as a sampling fault.

32 is where a rough level stops showing individual bright sources as separate blobs on the scenes in this repository; 128 is where doubling it again stopped being visible. What would make these wrong is a much brighter environment than any tested here — a small sun in an otherwise dark sky is the worst case for variance, and it wants the high tier or a source level chosen more conservatively than sourceLevelForSample does.