Hence, my language over the years (which you seem to find “offensive“ based on your kind words above) “a raw file is essentially a Grayscale file.” Not A Grayscale file but something like it. There is no question it has the potential to be, and will be a document with a defined and unambiguous color space.
Again, if you look at what I wrote above, I clearly said “the raw has no defined colorspace”. There’s nothing I know of in a raw file that tells any number of raw converters what the color space is (or is supposed to be). The converter creator has to make an assumption here which is what I got so clearly written from Doug’s fine piece.
Colorspace information is given by reference. The raw file does contain the identity of the camera, and the characteristics of the Bayer filters are the same within manufacturing tolerances for a given camera model and can be determined and applied separately. Apparently a method for determining color matrix is defined in ISO standard 17321 as reported by DXO for various cameras as per my previous post. The DNG spec does contain spaces for two matrices for two illuminants. They are not supplied by most cameras, but are filled in by the DNG converter. For example with my D3 they are:
33432 Copyright ASCII Public Domain
50708 UniqueCameraModel ASCII Nikon D3
50721 ColorMatrix1
SRATIONAL
9336/10000 -3405/10000 14/10000
-7320/10000 14779/10000 2763/10000
-914/10000 1171/10000 8248/10000
Since no sensor satisfies the Luther-Ives criteria, these coefficients are usually determined by a least squares fit and are not exact. Some raw converters might use a different set of coefficients to better optimize other colors. I agree that the camera space is not as well defined as in such spaces as ProPhotoRGB, but that does not meas that a camera color space does not exist. See Chapter 6 of the Adobe DNG Specification: "Mapping
Camera Color Space to CIE XYZ Space". Some colors would be rendered more accurately than others according to how these coefficients are determined. The situation is analogous to what might take place when I send some files for printing at a service bureau. I might not embed the profile, but could tell them the profile is for a specific printer and they could then assign the profile prior to printing.
I would submit that if we can’t see it (an imagery specification of a color), its not really a color. Again, it might be semantics but in terms of human vision, color is a perceptual properly. If we can’t see it, its not a color (it can be a particular wavelength of light, but a color coordinate outside the spectrum locus isn’t a color in the sense we perceive it).
As I mentioned previously, different people perceive color differently. Some have red-green color blindness and others are tetrachromats. What if I can see a color and you can't? Is it still a color?