I have looked a little bit more at the data from moving the ETM reaction masses on Monday (alog 54314), and I think that the attached plots show that the candidate positions that I was using are better than the nominal positions. 

On Monday, I did a chop / on-off test, going back and forth between the two positions, with 3 times in each position.  Today I separated the data by which position it was in, and concatenated those times.  In the attached plots I have a total of 26.6 minutes of data in the old nominal positions (0s on all 4 of the ETM R0 opticalign sliders) and 25.0 minutes of data in the candidate position (-200 on all 4 of the ETM R0 sliders).

In the first attachment, the scatter plots compare velocities of the ETMs versus the DARM BLRMS.  The DARM BLRMS are a pretty good proxy for when glitches happen.  The top row of all the plots is BLRMS #2, from 20 Hz - 34 Hz.  The bottom row is BLRMS #3, from 38 Hz - 60 Hz.  Each of the different figure windows is a different optic velocity (EX length, EX pitch, EX yaw, EY length, EY pitch, EY yaw). The blue markers are from times at the old nominal positions and the redish-orange markers are from times with the -200 urad slider values.  The first take-away I have is that for most velocities, the new position has lower peak BLRMS than the nominal positions.  The other take-away is that they seem pretty random for the most part, except for EX length which seems to have some interesting structure.  The y-axis data is the same for all of these plots, so the fact that there seems to be some structure to the EX L plot suggests that that optic could be the most egregious in terms of generating the scatter. 

Tomorrow I'll make some plots with also the wind and microseism incorporated in some way (I'm not yet sure how to best visualize those) to try to tease out whether this correlation is related to microseism levels at those times, or if it is more strongly related to optic velocity.

Something to note, I didn't think about the sign of the R0 offsets carefully enough on Monday, and I think that perhaps I did the pitch offsets in a good direction, but the yaw offsets in a not good direction.  If we're having bad glitches again I may ask to try again soon.  The R0 opticalign offsets should be of opposite sign from the M0 slider values to make the relative angle between the chains maximally large.  For both end stations, the pitch sliders for the main chain are positive, and the yaw sliders are negative.  Since I put negative slider values in for the R0s, I was increasing the pitch angle between the chains, but decreasing the relative yaw angle. So, overall I still was changing the angle between the optics and their reaction masses, but I could have gotten even more angular separation.