I've looked a bit more at the regions that Gabriele identified in alog 47656, due to INP1.  At least for several of our 35W locks, we have had consistent coherence with INP1Y and DC2Y. 

I made the 2Hz cutoff in the INP1 loops more aggressive, after determining that the loop gain at 0.2 Hz was already -15dB, so stuff around a few Hz is well above the UGF. This removed the coherence with INP1, but did not make any change in the coherence with DC2Y. 

The first two attachments show a lot of coherences with DARM both before, and then after, I change the cutoff.  The third attachment shows how the INP1 input and outputs change with this more aggressive cutoff.  In this third attachment, dashed lines are yaw, solid lines are pitch. Black is the before reference, pink is the after.  Both pitch and yaw outputs are reduced, but the yaw input motion is also reduced in a band between 4Hz - 25Hz. 

The fourth attachment compares the old and new cutoff filters.

Georgia is going to make a DC2Y noise budget injection, so that we can see how much it matters, but I suspect that we'll want to lower the bandwidth, or apply more cutoffs, to DC1 and DC2.  Recall that we recently lowered the bandwidths of the AS WFS centering loops DC3 and DC4, but didn't change DC1 and DC2.