[phenixbb] Low resolution refinement - rotamers and Ramachandran

Dear developers I am refining a low resolution structure (3.8 A). After I fix most of the outliers (red bars) in Coot, phenix.refine is causing my structure to be worse than the start with 13 % Ramachandran outlier and 13 % rotamer outlier. Secondly, when I used phase_and_build to build my model, my R/Rfree is around 30/35 with atoms from some residues missing. When I fill them in with Coot, fix the outliers above and re-refine, R/Rfree jumps to 36/42 and the rmsd angles are too high (2.9). I have already used weight optimization and NCS. Am I doing something wrong? Thanks.