Hi Tim,

If you start from nearly identical copies (due to previous refinement), in the absence of a strong signal from the data, it is expected that the copies remain fairly similar even if you increase the sigma / lower the weight. We just added a new feature to phenix.refine that could be useful to introduce differences between the NCS copies, guided by the model density through a new real-space refinement protocol: strategy=individual_sites_real_space plus real_space_refinement.mode=lockit. At your resolution it is turned on by default. I.e. to try it out all you need to do is get dev-602 or higher and run phenix.refine again. Only trouble, those installers aren't generally visible since there were some minor problems, but watch out for a new one soon.

Ralf