Attached is a time series of TMSX roll osems over the last year, which has been slowly drifting in the same direction the whole time.  There was a went in June, wich is the larger jump, you can see when the sat amp started to oscillate a few weeks ago, and me moving roll over the weekend. This drift is mostly seen in F1, I don't know if it is real or not.  

Rana has told us that at LLO the TMS QPD positions are sensitive to roll.  Before closing the soft loops we moved roll by 250 urad (accroding to the osems) and saw that this had no impact on the error signals of the soft loops.