Hi, I wonder in phenix.real_space_refine whether it may be worth adding refinement of a single overall anisotropic ADP to the model - or a single TLS group with modeled anisotropic ADPs - in order to model the effects of overall orientation errors (which affect resolution more substantially towards the periphery) and preferred orientation (which tends to give rise to anisotropic resolution across the volume). Maybe this would be helpful at least in generating a calculated map that is better for difference map calculation, for example? Cheers Oli