Presumably a big part of this is because of a fundamental problem with the PQ brightness model.

PQ was defined by cinema people. Unlike everything else that I've ever worked with, its brightness levels are "display referred". That is, the standard actually says how bright (in nits) any specific gray value should be displayed. This makes perfect sense in dark theaters, but doesn't make any sense at all for most non-cinema uses. Imagine watching Youtube on your phone in bright sunlight or in bed at night, and having the screen be the same brightness in both cases, because that's what the standard (technically) requires.

Pretty much every other transfer function (Gamma, HLG, etc) is either explicitly or implicitly "scene referred". They tell you how bright the original scene was, and kinda leave it up to the display (and user's preferences) to determine how bright they should be displayed.

Practically speaking, though, pretty much everything that isn't in a theater ends up being displayed relative to a semi-standardized gray tone. Using SDR's "full white" (#ffffff in 8-bit color) as a reference point is usually easiest when doing HDR.

IIRC, PQ originally defined SDR white as 100 nits, and later moved to 203 nits. PQ was designed to be able to fit deep blacks up to a 10,000 nit white into a single 12-bit value without visible banding.

What I really want is a way to (reliably, compatibly) create image files that use Rec.2020 primary colors (because they're pretty close to the full range of human color vision) with the ability to show brighter highlights than SDR, but with a reasonable fallback for SDR displays. Practically speaking, there's no reason that Rec.2020 PQ *can't* do that, if you pretend that it's scene-referred just like everything else.

It just doesn't actually seem to work in practice today.