I would recommend to get your hands dirty and dive into the theory. There’s lots of handwaving in most sources available. Try to understand what the motivation behind the XYZ primaries is (to keep it short, they are three imaginary colors that, combined linearily, can create any perceivable colour), which followed from the fact that there are colours which cannot be “mixed” from RGB. I found the video series “Color Vision” by Craig Blackwell 1 2 3 4 5 to be very helpful and enlightening.

Now, once you’ve understood that, you have to understand that the “perfect” spectral colors (which means monochromatic light of different wavelengths) are the “boundaries” of human vision, so to say the convex hull of the XYZ-coordinates of all visible spectral colours (modulo lightness) is the realm of human vision. You can get those coordinates here. It turns out, these “coordinates” X(lambda), Y(lambda), Z(lambda) are called color matching functions (you can get them there as CSV’s).

The final piece in the puzzle is to convert these XYZ coordinates to common normalized formats like xyY or even CIELab or CIELuv, or even polar coordinate forms like HCL(uv). All these formulas can be found here, an excellent resource of transformation functions between all kinds of different formats, including XYZ <-> xyY and so on, by Bruce Lindbloom. He is a really awesome guy. Now the only thing left to do is plot the xyY data in ParaView or something and if you look at just the xy-plot you’ll see the familiar horseshoe. :)

If you look at the color transformations from RGB to XYZ, you’ll see that RGB data includes a lot of implicit information, including the reference white and which RGB primaries you actually choose. Most people, talking about RGB, are talking about sRGB, but you can “invent” an RGB color space easily by just defining three primaries. If you manage to convert an RGB coordinate to an XYZ, Luv or Lab coordinate, the latter are all “universal” in describing color perception. These spaces allow you to express any visible colour. With RGB, they over the years defined new colour spaces (Adobe RGB, ProPhoto RGB and so on), moving the primaries a bit to make the range of colours bigger, but in the end there are still quite a lot of colours left which just cannot be expressed within currently “popular” RGB colour spaces. There are drawbacks to working with anything other than sRGB coordinates (imaginary colours, no good software backing, lots of superfluous nomenclature, …), but you gain a lot of things. For instance, CIELuv is perceptually uniform, which means that if you select equidistant colors on a “line” within the uv-diagram, these colours will all be equidistant with regard to perception. If you work with HSL based on RGB, this does not work, which is very cumbersome if you want to select nice colours for diagrams and such. The reason for that is for instance that sRGB has a nonlinear gamma curve. You might want to work with Linear RGB then, but if you do any of that stuff on your data, you might as well work with the real thing.

I hope this introduction was sufficient to get you started. Send me a mail if you want to see some code. I might give a talk about it this year, but it takes some preparation.

not OP, but https://www.handprint.com/HP/WCL/wcolor.html is probably the most accessible treatment I’ve found.

Color theory is interesting, but (I feel) a bit of a rabbit hole. I learned a little about color spaces a few years back while writing a blog post about how the difference in color space and gamma value between older Mac and Windows computers might have influenced some of the popular color themes.

I found Danny Pascale’s “A Review of RGB Color Spaces” a useful (if somewhat terse) source of information about computer color spaces.

I assume you are referring to displays that can display colours beyond sRGB. Yes, these displays exist already. In fact, the current iPhone and most AMOLED displays fall within that group.