my experience is that the verification result depends significantly on the distribution of the test set points used to assess the "resulting profile accuracy". My feeling is that I get in a majority of cases better average errors, whenever the same kind of distribution is used for both, the training set and the test set.
This sounds quite obvious, or am I reading it wrong? Basically you're saying "if I profile or refine with an ECI2002, and then I use an ECI2002 to verify, I get better verification results than if I use another patch set to verify". This is why I don't like those shootouts (IPA, Wuppertal, WMU) in which they always do that. You'll never know if an iterative-kind software (proofing or profiling) optimized the test chart patches in detriment of the profile's accuracy in other colors, or general smoothness. I remember someone posting a comparison (before and after) picture of PrintOpen 5.x CMYK separation of an RGB 'rainbow hexagon', in which after refining the CMYK profile by iteration (PrintOpen 5.x has this feature), it looked awfully un-smooth. My take: you should evaluate BOTH the same testchart which is used to refine, and another one composed of semi-random in-between points. And they should be evaluated separately, so that you can see whether the error decreases on both, or at least does not increases in any. -- Roberto Michelena Infinitek Lima, Peru