[phenixbb] Ensemble refinement

Using the FEM map, I managed to remove more bad waters and add instead ions from the crystallization mix. I tried phenix.ensemble_refinement but the Rfactor increased by 2 % (originally 19/22 % from phenix.refine). Before it crashes at the validation point all the time, it prints out the validation summary with 0.53-3.48 % and 1.87-7.22 % Ramachandran outlier (from 0.3 %, supported by density) and a Molprobity score of 1.57-1.92 and 1.82-2.14 depending on pTLS used (original single model score was 2.08). The clashscore in both settings (pTLS=0.6 and 1) are 0. Do I still go by the Rfactor and decide to not use ensemble refinement at all? Thanks.