Carried across every bounce of the path walk, so anything derivable from a value the walk already holds is derived at the end instead of stored. What is left is only what the walk itself produces.

A product of k Householder reflections is orthogonal with determinant (-1)^k, so it is a rotation when k is even, and a rotation composed with any one fixed reflection when k is odd.

Primary surface replacement for perfect mirrors. The reflection chain has been followed to its first non-mirror hit; write that surface’s attributes, reflected into the mirror’s virtual space, to the DLSS RR guide buffers so the denoiser treats this pixel as directly seeing it. https://developer.nvidia.com/blog/rendering-perfect-reflections-and-refractions-in-path-traced-games/#primary_surface_replacement

Share of a surface’s total reflectance that leaves through the specular lobe, in 0..1.

Angle, in radians, that the shading normal changes over one pixel in screen space.

The rotation half of H(n) * F at even parity, or of F * H(n) at odd, where H(n) is the Householder reflection about plane_normal and F the fixed z-axis one.