This is similar to what Scott did for the purple artifacts. (see his Nov 16 note in the color artifacts thread) You can create a matrix that ‘shrinks’ the color space available --oddly enough consider using a virtual set of primaries that are wider than AP1 and using the same white point. So just construct a standard transform matrix from AP1 to PX, then apply it in AP1 and treat the result as AP1. [works for modest displacements]. Note that you can play with the off-diagonal axis terms of a matrix to adjust RedxBlue for example, but the results have to be normalized and tweaked by eye. This is a bit risky.
The problem is a bit different than the artifacts referenced in Oscars note, partly caused by any set of primaries (ACES, Rec2020) that are perceptually out of balance
and then have a contrast building tone curve applied. The tone curve boosts the color contrasts as well as the luminance, and when red is perceptually twice as far out as before, you can get ‘electric’ reds. (or blues)
This was only partly addressed by some fixes in the RRT. (selective red desat and the ‘glow’ operation targeting a P3 look)
The group has thought about other ways to fix this, but don’t have a recommendation right now.