Again, I ran Gabriele's lsc_sensing_matrix.py from alog 44620, with some updates, this time at 25W input.

I also wrote a log radar plotter, so that we can actually see where the SRCL vectors are pointing. (on a linear scale that just look like zero)
You have to do some tricks to make the log vector always positive.
The log radar plot of our LSC sensing matrix is attached.

Also attached is our current sensing matrix.
Our current error signal logic seems to be 
1) PRCL uses POP A 9I (MICH also appears strongly in POP A 9I with opposite sign)
2) SRCL uses POP A 45I, with some PRCL feedforward to remove the PRCL motion in POP A 45I. (PRCL registers far more strongly than SRCL in POP A 45I)
3) MICH uses POP A 45Q, and is fairly independent of PRCL or SRCL motion.

PRCL could be slightly better tuned, but is fine.
SRCL is dead on.
MICH is at around 250 degs, which is 20 degs off from perfect but we have to choose either SRCL or MICH to be ideal for POP A 45, so we choose SRCL because it is weaker overall response and is contaminated with PRCL motion.

The unity gain hierarchy goes:
PRCL at 50 Hz
SRCL at 15 Hz
MICH at 8 Hz

Code now lives in https://git.ligo.org/aligo_commissioning/labutils/-/tree/master/lsc_sensing_matrix.