The last two transfer functions took on the lower stage of the beamsplitter under vacuum, were showing a mismatch with the model and Livingston's measurements (cf first plot attached). 

Before thinking of any mechanical issues, earthquake stops touching or misaligned flags, I wanted to check if the actuation and sensing side were acting as expected.

After the issue we had with tmsx few weeks ago (cables beeing swapped), I wrote a script to test medm-signs/cabling/magnet polarity all in one, and ran it today for the beamsplitter. It looks like there is something wrong with the actuation chain of M2 UR osem (cf the red curve of second attachement). The test consists in sending a positive ramp to the osems, one after the other, expecting to see a response that is following the drive. UR is the only osem seeing the contrary (not obvious on the plot because of the scaling)

Even though it doesn't match the convention any more, I swapped the sign of UR coilouput filters gain in the medm (+1 instead of -1) and took a new transfer function in M2 yaw to way with DTT (cf last attachement). The TF looks way better now, with a really good coherence.

TF will be running tonight for longitudinal and pitch dofs.