I'm refining a model at 2.9 A, where four chains are in the ASU, and each active site has an iron with a few ligands; the configuration of this active site is the topic here. Currently the iron-ligand distances are not ideal (according to CheckMyMetal and the ideal distances generated by phenix.metal_coordination), but they are within a range observed previously for structures of truncated versions of this enzyme; furthermore, the more the metal-ligand distances change towards ideal distances, the more positive and negative density I observe in the Fo-Fc, so it appears that this deviation from the 'ideal' is supported by the data. However, when I use the 'Optimize X-ray/stereochemistry weights' option during refinement, the metal-ligand distances change (the metal appears to do most of the moving), getting closer to ideal, and this generates more positive and negative density in the Fo-Fc map). However, clearly it's nice that optimizing the X-ray/stereochemistry weights reduces R-free (and keeps the R-f/R-work ratio fine as well). So what is the most appropriate approach here, in order to keep the metal-ligand distances that best support the data, but still reap the benefits of the optimized stereochemistry weights? Some details of the structure and refinement are these: The iron in each of the four chains in the ASU is coordinated with atoms from two histidines, a glutamate, and a water (so four ligands per Fe) and there remains un-modeled density in the iterative build comp OMIT map used for modeling (because resolution is 2.9A and adding more water (which is my best guess, given what truncated structures at higher resolution show coordinating the Fe) does nothing to the Rf). I am working with an .edits file generated by phenix.metal_coordination. After some experimentation, I've left the sigma values (for Fe-O and Fe-N) at default values, and added a restraint for Fe-O (water), using an ideal distance of 2.3 and a sigma value of 0.1. Emily.