On Dynamic Range

I have significantly "softened" on this stance.

The reason is that it is important to appreciate that when we are dealing with simulated light transport, we need to focus on the domain of the expression; the underlying metrics of amplitude of our proxy-of-light.

In the vast majority of cases, that metric is some relative RGB encoding. That means that we are manipulating values in a model that is rather distant from models that emulate electromagnetic radiation. Rather, the model itself is actually a tristimulus model. There is much nuance in this observation.

Perhaps most importantly, what is interesting is that this domain doesn't really have a "dynamic range" per se; it's all uniform with respect to a quasi-neurophysical projection anchored in the CIE Standard Observer. 600 units of the CIE Standard Observer X component is potentially 900 units in another RGB projection. It's somewhat arbitrary!

Given this rather strange and overlooked facet, our "ground truth" of proxy-energy bouncing around is always anchored in this relativistic projection, while at the same time, leaning heavily into the PBR math. PBR math is of course based on a working model of electromagnetic energy. It is my view that we cannot reconcile these two domains of "relative-absolute colourimetric quantities" and "electromagnetic radiation quantities"; they collide.

As such, when we consider the amplitudes / magnitudes of any tristimulus system, we should exert great care in drawing analogies to the electromagnetic amplitudes we have built notions of PBR rendering around. They are very different domains.

TL;DR: Suggesting we have a dynamic range by measuring a log2 increment in some arbitrary RGB colourspace is likely absolute folly. The best uniform metric in this domain is luminance for such an evaluation, given that it is constant between projections of the various RGB magnitudes.