.... regardless of what they use to "color" the filters it is basically impossible to get a response curve that falls exactly in the spectrum they want it too. It may be "red" but I can assure you that it is reflecting to a greater or lesser extent in the yellow, green, and blue areas. In addition, each colorant (dye or otherwise) is unique in that once you decide on one you can't really change or the fingerprint will change..and all of your calcualtions are out the window.
Having said that, it is theoretically possilble to know what those values are that lie outside the target part of the spectrum and in what part of the spectrum they are, which would allow you to build the "perfect" data as a sum of the three layers.
I'm certainly no expert in sensor design (although I am one in color theory), but I suspect the difficulty is in the inherent noise contained in the system and the fact that in some cases the noise signal could very well be greater than the actual signal of the color itself in some areas. In those cases what is color/signal data and what is noise? You can make some base generalizations, and perhaps even establish a baseline of the noise signature of each layer and where it lies. I would think the theoretical issue is that you can't just assume that all of the data below "x" is noise and therefore irrelevant. It could very well have relevant data in it, which makes the mathematical calculations immensely more predictivespeculative vs. accurate.