[ccp4bb] different resolution cutoffs for integration and scaling

[Jorge Iulek]

Dear all,     I have been noticing, with many datasets, processed the duet xds/scale, that when one integrates to a resolution limit which is (a little) higher than the one used for scaling/merging, the statistics (and here I mean R-symm, R-meas, <I/sigI> and even CC1//2) get better (I might also advance that, in many cases, completeness gets a little better too).     Just to make clear, suppose I want to process to resolution x (according to any criteria/index I decide) ; suppose y is a little (how much is yet another good discussion) higher resolution, id est, numerically, x > y, I get better statistics when I integrate up to y and scale/merge to x, rather than using x in both cases, therefore, its seems to be advisable to integrate up to y and then to scale/merge up to x.       So the question is: why does this happen? Would this be related this the fact that the spot profiles gets more well defined? In this case, is it fair to do this and to obtain better data (and, I suppose, a better structural model)? In principle, I suppose this might be legitimate, as even CC1/2 gets better. Has anyone else ever observed such behavior, maybe with other processing duets?      Jorge