I would like to _be able to_ recreate a scene as accurately as possible - i.e. if I had the possibility to inspect the print or my display "side-by-side" with the original scene at the time of capture, I want them to appear subjectively identical.
I might not choose to use that option for all images, but that should be my conscious choice, not some default camera/raw-developer behaviour.
*calibrated wide-gamut display (Dell 2711),
*Canon 9000 mk2 printer with paper-supplier profiles
*Adobe Lightroom/Windows 7
I was very dissatisfied with the default Lightroom profiles for my 7D after I got the wide gamut display: saturated reds appeared over-saturated and/or with the wrong hue. My own colorchecker profiles result in a lot more neutral look to me.
I have tried profiling in daylight, incandescent light, direct flash, flash pointed towards my white ceiling, and have concluded that a single profile seems to be sufficient for most of my needs.
Some random thoughts:
1. A given camera might not be able to differentiate between two colors that we humans are able to differentiate, due to differences in color filtering. Should a camera profile render as "color A" or "color B" then?
2a. The CIE observer response is based on measurements on a set of test people. There may be "noise" in that measurement, and any single user may have a response that is more or less different (I hear that some women have 4 primary colors - explains a lot to me)
2b. The CIE measurements are based on some idealized conditions (e.g. size of patches). In different conditions, the response may be different.
3. Color profiling is probably not only about "mapping each measurement to the closest available color". If you "sample" color by 16 patches, most codes will be some interpolated correction. So should you fit some parametric model? What happens if you allow for this model to have more free variables than warranted by the measurements? Perhaps banding? Perhaps visible clipping?