Thanks Jesper for your response!  Yes, what I’ve found is that eddy_openmp does not give consistent results even with the —initrand flag, and I couldn’t identify an eddy version which gives consistent results and also has the —repol option.   We are doing some test-retest analyses and so we believe that deterministic results are preferable for our purposes, but perhaps the between run differences are too small to matter even when comparing data within subjects; we’ll have to further assess those differences as well as motion to determine what’s appropriate.