RadianceLab · PP33 · measured comparison

Two different curves.
The same path density.

The new cone + plane pair splits one total sampling budget. In this forward tracer, the balance weights reduce to allocation fractions: changing which coordinate is shared changes correlation, not the marginal PDF.

The pair is an orientation hedge, not a demonstrated caustic speedup. At equal cost, mixing independent equal-density sheet estimators cannot outperform the truly better single family's variance merely by applying balance MIS.

Theory and measured analysis →Native path measure and scope →Capture summary Native capture record

Variance stays near the single families; comparable render cost

320 × 200 linear HDR films, 524,288 launched rulings per image, six seed replicates per setting. Chart variance is the mean per-pixel luminance variance across independent seed images, not ground-truth MSE. Six seeds give a limited estimate; small differences should not be treated as a statistical ranking.

Scene 107 · direct surface paths