Gerhard Fuernkranz wrote: > I guess you see an extraordinarily good agreement on 6500k, because the white > patch is heavily > overweighted when creating the ccmx (-> see ccmx.c, line 352). So the good > agreement on the 6500k > "white" is likely not at all representative for the average agreement > achieved by this ccmx, and > maybe the generalization even suffers from enforcing a better agreement for > white. I'm also not > sure whether the use of delta E CIE94 error metric (in CIELAB space) may have > impacts on the > generalization to 5000k as well. The recommendation of some papers I researched on the creation of 3x3 correction matrices was to weight the white somewhat more than outlying colors, and this makes some sense when the neutral accuracy (and white point) is seen as being more important than accuracy of saturated colors. If the source of the colorimeter inaccuracy was purely spectral filter errors, then a 3x3 matrix should make a perfect correction and (given perfectly accurate readings) the weighting of the test sample values shouldn't make any difference to the matrix (in fact you should only need 3 samples, R, G & B). But maybe eliminating the extra weight would make for a more balanced result, and/or perhaps I should change ccxxmake to use the current calibration rather than native device test values ? Graeme Gill.