OK, so the last one: rotations. I have to admit, it’s too complicated for me to properly understand, there are just too many factors at play. Here is what I have understood so far.
Moving the red rotation slider changes either the green or the blue channel. The modified channel’s value always moves towards the primary we are controlling (so setting a positive red rotation influences the green in a way that brings the green value closer to red: increasing it, if it is currently below, decreasing it, if currently above the red; setting a negative red rotation does the same to blue). And of course each channel can be drawn towards two other primaries, which may both be larger (or smaller) than this channel, or one smaller, one higher. It’s hard to explain; open the tool, set some arbitrary RGB values like 0.1, 0.2, 0.4, and start rotating the primaries.
Again, we start with a red primary, and set red rotation to 10°. This seems to “generate energy from nothing”, as the red value does not decrease, but green is no longer 0:
Inset Applied: RGB: (1.000, 0.114, 0.000), Hue: 6.9°
This then goes through the usual steps, the log mapping decreasing the relative difference and increasing the hue angle, then the curve’s contrast and the linearisation increasing the difference and decreasing the hue angle back to 12.2°.
Since we now have energy in the green channel, the N6 kick in, skewing the result towards yellow. This explains the increased shift.
Setting red unrotation also to 10° just rotates the N6-skewed 12.2° to 6°.
Hue preservation and unrotation are both needed to restore the original hue.
If we turn off hue preservation, then of course the N6 increases with exposure value. Using the same 10° rotation and unrotation, the final hue angle is 38.4° when red output reaches 1 (at exposure +4.1 EV, given initial red = 0.18).
However, as our output is no longer pure red, the green and blue unrotation also play a role.
All I can say for sure is that without hue preservation, for high brightness, the hue shift is dominated by the N6, even if rotation and unrotation parameters are the same. With hue preservation at 100%, the result is determined by the rotation and unrotation, and does not depend on exposure, until at least one output channel starts clipping (relative white exposure being too low).
It’s probably best to explore this using the colour chart, instead of playing with RGB values, starting with no unrotation (as it is set in the blender-like and scene-referred default presets). smooth completely reverses rotations (but not the attenuation). Playing with the hue preservation slider is a further parameter in the mix.
Attenuation probably does not influence this, as increasing it reduces the value of the specified channel, symmetrically increasing the other two (but if you have rotations present, then those do mix take that decreased/increased value into account – and that’s more or less when my poor brain gives up).