By driving a line in CARM at 400 Hz we saw that the 9 MHz demod phase was mistuned by 9° at 2 W of PSL power. We tuned the demod phase to within 0.5° and then powered up.
Once the interferometer reached its final power (44 W), we could see that the demod phase was again mistuned, this time by 5°. Over the course of 30 minutes it relaxed to ~0.5° of detuning.
Also, the REFL9I readback now agrees almost perfectly with the CARM control readback; not sure why (this was true even before changing the demod phase). The CARM error readback is still mostly junk.
I tried again to check for REFL sensing noise in DARM by reducing the CARM gain, it seems that we cannot reduce the gain enough to bring the ugf to a few hundred Hz without loosing lock. (I could reduce the gain by 8 dB, but not turn off the boost, the other day we turned off the boost first, were able to reduce the gain by about 4dB, then when we tried to reduce the gain by 3 more dB we lost lock.