Hi, I regularly encounter this problem with phenix.refine, happened to me with several structures, at wildly different resolution, so as described in the title, phenix.refine seems to break perfect geometry in its last cycle (put 3 5 or 10 it's the last), for no obvious improvement in term of other metrics, actually getting the temporary file just before the last cycle had given me very good result in the past. Is there a reason for this behaviour? A way to get back PDB at a certain point of the refinements cycles would be really great actually. example of what I refer to: [image: image.png] Cheers, Clément Dégut PDRA Dr M Plevin's group University Of York