In my experience I've been able to tweak ACR/LR color temp sliders on a Raw shot of a flat spectra WhiBal card after clicking for R=G=B and not change the numbers but noticeably change the color cast of the scene so there goes your accuracy and repeatability.
Color temp data is the only manufacturer info as tags in the Raw's metadata that third party converters can use with regard to controlling color appearance but it's still interpreted.
Spectra data that describes exact luminance scaling of HSL appearance of backlit RGGB filtering on the sensor as more light is exposed to the sensor is still somewhat guess work as to how it should be appear reconstructed on a computer video system.
Think of the behavior of how stain glass window inside a church appears or changes hue/sat viewing at daylight, late afternoon and at night with car headlamps back lighting it. Does the green glass at night look like the same green when it's backlit by more light? Does it scale linearly? No one has proven what this looks like and how much it affects the hue/sat of color temp appearance. We have to rely on reconstruction on a limited gamut display.
Finally someone posted online a photo of an actual back lit RGGB bayer filter to demonstrate whether these RGB purities scale linearly exposed to more light as in the stain glass example...
http://www.siliconimaging.com/RGB%20Bayer.htm
