Hi Matt, sorry, I had promised Duncan to reply quickly, but forgot. The short story is that the results of the paper show that you have to change the parameters quite strongly to get any appreciable effect, and in the end the values will be quite different from the measured ones. You could just treat them as tuning parameters, and try to optimise your self-prediction, but the variation needed is probably small. Please also be aware that the paper mentioned by Duncan was calculated with older code, which had some calculation deficits for one side of the parameter space. We didn't do the calculations again with the new code. Hope this helps, Gregor
