Finally back to this. I just beat a year.
My original post showed graph 2 (RelCol conversion) and graph 3 (AbsCol conversion.)
RelCol definitely does (and should) convert the printer profile’s whitepoint to ProPhoto’s L* 100, and in doing so also expands a* and b*, therefore increasing saturation.
The AbsCol conversion I was doing leaves the printer profile’s whitepoint where it is, and doesn’t expand a* and b* at all.
I wasn’t using a manually generated color list like you were, so although I knew the situation got better using AbsCol instead of RelCol, I was still getting the same results as in graph 3 after double checking.
Got frustrated and back-burnered the project.
Solved this bug while diagnosing a different symptom. My post today, “BUG: Wrong values extracted from images with dim > 50” is what caused this bug.
I was using the 956x752 pixel TIFF generated by i1 Profiler that is the actual test target, including the patches, the whitespace, the text labels X-Rite puts on it.
All that’s fine - all of that would still be converted to in-gamut values - of course adding black and white pixels, but still in gamut.
It’s the downsampling method of images over 500 pixels that is causing all those pixels in graph 3 to be displayed out of gamut.