Dear Phenix people, is there a major improvement in the algorithm? I have refined a 2.2 Ang. room temperature dataset with very good data statistics (I think) Rmerge 3.7%(8.9%) completeness 96%(92%) I/sigI 16,6(8,8) in parentheses are the values for the highest resolution shell. After refinement I have a R/Rfree of 0.1294/0.1670. In some shells there are values close to 0.100 and in the lowest shell Rfree is lower than Rwork. The quality of the data are really quite good, what is probably one of the reasons, but I am close or below to the border of the Range of R values in the pdb. Great for me, but I wanna be sure that it is really true. It was a MolRep with a close homolog, but I run a torsion angle sim annealing to decouple R/Rfree and the MolRep was done with an poly-Ala model to prevent too much model bias. Best Regards Christian