Edit: haha, should have read to the end. It seems that the results look a lot closer to expectations after the latest fixes to DT – so there was a bug after all.
I don’t pretend I understand the intricacies, and I find some of the tool not at all intuitive to use. That said: one point you are missing is something I do understand, which is: Depending on the working colour space, what colour profile was assigned to the image at import, and what colour profile darktable uses to output pictures to your monitor, each of the primary colours has a slightly different “meaning”. That’s because the colour filters in front of your camera sensor are different from somebody else’s, the ones on different monitors are, too, and the spectra which activate the corresponding cells in your eyes are different still.
This means that if you started with an sRGB image with just 100% green channel, and everything else set to zero, and converted it from one colour space to another (e.g. to display it on a calibrated monitor using that screen’s colour space), it would light up not only the green subpixels, because they’re a different green than sRGB green. Darktable converts from sRGB to Rec2020, maybe performs some white balance, and after that there are no “pure” greens in that colour circle any more, and the green slider does not correspond to just the green subpixels on your screen.
That far, it’s very clear to me. Actually, if I try to minimize such conversions and tell Darktable to interpret the colour circle as REC2020, and work in REC2020, skip white balance and use linear Bradford for the CAT module, the channel swapping kinda works. The colours are still shifting quite a bit, though, and I’m not sure why they do that.
The brightness tab is maybe more straightforward, and it does let you darken the image based on the values of each of the channels, and works as I expect it to.
What I don’t understand is:
1: why are the colour shifts as big as they still are, because I would not have thought that those colour spaces are that different
2: if I go to the grey tab and use only the blue slider, I see the respective part of the wheel brighter, but even the opposite end (where the colour picker showed zero blue before), is rather bright. When I use only the red slider, the entire wheel has almost the same brightness – under what definition of “red” would that rainbow have a near-constant red channel?
Some of that may have to do with the bug(s) which @anon41087856 has already confirmed and fixed.