Hi Francis, the problem was in this line (and this is really our technical problem and not your fault): primary_map_type = "m*Fobs-D*Fmodel" which is in newer version looks like this: primary_map_type = "mFobs-DFmodel" It would not matter in refinement using non-twinned data (both will be recognized automatically), but in twin refinement phenix.refine uses some different internal routines that are less automated about this. The take-home lesson: it is always the best to use newly generated parameter sets to make sure you are up-to-date with all eventual changes in parameters that might come along with a newer version.
Command: phenix.refine *004.def --overwrite --unused_ok refinement.input.xray_data.r_free_flags.generate=True
Just a couple of comments on the above line: It is a bit dangerous to use "--unused_ok" because it will force phenix.refine to ignore the parameters that it did not recognize and so it will keep going, while you may still be thinking that those parameters are used. So, I wouldn't use this option unless you are very sure. This "refinement.input.xray_data.r_free_flags.generate=True" should be used only once at the very beginning BEFORE any refinement is done, otherwise it will re-create the test set reflections every time you start the refinement therefore making the Rfree statistics invalid. Pavel.