Measuring a sensor with a monochromator is not difficult and would give direct information about the primary filters. It should be the preferred method.
But this is not what Doug said. Doug believes that accurate color reproduction is not what the manufacturers are aiming to. This is also my opinion. And if the manufacturers are not designing their back for accurate color reproduction, measuring the accuracy of color reproduction is futile.
The possibility to produce accurate color is bound to the sensor response. Final image color output is much bound to software, ie capture one and its profiles (or phocus/hasselblad and their natural color system). I don't know how much influence the back manufacturers have on bayer filter array design, I'd guess more for the newer exclusive sensors than what it was back for the P45+, for the KAF39000 I would guess the digital back manufacturers was not involved at all in sensor design (?). I'd guess Phase One has more influence on sensor design than Hasselblad has.
Anyway if you just can get the raw data you can tune the color any way the bayer array filters allows. As far as I know DxOmark measures raw data response in their color tests, ie they don't use any manufacturer provided raw converter and profiles, so they measure the sensor's ability to reproduce accurate color in some standard lighting condition, not the color you'll get if you use the manufacturer provided raw software and default profiles.
What's important is the sensor's ability to differ between nearby/similar colors, which DxOmark does measure. As long as it can do that you can tune the color in any direction you want and different colors will still be different colors. If similar colors are registered with the same RGB raw values no profile in the world can make them differ.
For me that aspect is a key performance indicator of a sensor. It's a bit difficult to evaluate though if you just get a single metric, as the sensor may be bad at separating greens but good at separating skin color, and thus may produce better results for you than a sensor good at greens but slightly less good at skin color, if your main subject is skin. A landscape photographer may prefer the sensor that is better at separating greens.
Color separation is reduced in the shadows, and I'm not sure about this but it seems to me that it's a bit separate from dynamic range, ie you can measure quite good dynamic range but still lose more color separation than another sensor which on paper has worse dynamic range. Some sensors get quite monochromatic in the shadows (usually towards green) before noise gets too bad.
To summarize, what we should be interested in and measure is "color separation" rather than "color accuracy". Different colors registered separately can be tuned, different colors registered as a single color cannot. In technical terms we want a "low metameric error". In general terms the Dalsa sensors have lower metameric error than the Kodaks, but I don't know how performance is in specific color ranges.