-Punchline: I screwed up the measurement settings. These are critical in the calculation of the blend filters. I have loaded new blends based on old data, so ITMX is running, but not in an ideal configuration, as the controllers were designed around a different data set. I will work on getting proper data (last time, I swear!) this weekend. 

-More detail: If you look at the first attached pdf, the second page shows what the complementary filters should more or less look like, the gain for the sum (say, for the X loop) should be ~1 all the way across and the phase for the sum should be zero. The same page on the second plot shows what was causing the problems at ITMX. The gain for the sum does not stay at 1 and there is a large loss of phase in the 1-10hz region. This was causing loops to go unstable. 

-I didn't notice this at the time, because the commissioning script that calculates the filter produces a plot of the complementary form of the filters (shown in the first png file, the sum is the thicker line at 0 dB) that looks complementary. Something is wrong, somewhere.

-The form that gets loaded into Foton (second png, produced by the same script) should have alerted me, but it's a much less straightforward plot. The third and fourth pngs shows the same plots of the "good" blends, currently installed. 

-The only clue I would have had before trying to run things at this point is that the T240 and L4C blend filters should have been closer together on the Foton plots. Maybe there were other clues, but that's all I can see right now.

-How I figured this out: 

-I worked through the seismic commissioning scripts, designed my isolation loops and installed everything. Damping turned on (although, the first time round the St1 outputs were low) but turning on loops was difficult. 

-When I took open loop measurements in DTT my UGF was lower than expected. The magnitude shown in png #5 (the different traces are different loop gains, blue is a gain of 1, so look at blue) shows the UGF is much lower than design, I designed for 30 hz. The 6th picture shows that the loop has 0 phase at 30 hz and the shape wasn't right, it looked tilted relative to my design. My loop design for St1-X is in the last picture.

-Looking at blends, isolation loops and every other filter I could think of loaded in Foton didn't really yield any help. It wasn't until I ran JeffK's plot_current_blends script (which really should be a function, it would be nice) which produced the attached pdf's that I could see that the blends weren't complementary. After a lot of deleting, reloading and other shenanigans, as well as hours on the phone with Rich and Hugo, I decided to try loading old (Nov of last year!) data and seeing what blends I got. 

-Problems fixed. 

-Head-desk.