Here I report the investigations done this evening while failing the lock acquisition several times.

As side notes, we could observe the B1p and B5 galvos opening often, and a small mistuning of the WI CO2 laser, but the amount looks fairly small.

At the beginning we unlocked during the transient to CARM_NULL_3F, so during the CARM handoff to B4_6MHz; the oscillation was around 300 Hz, which could be related to the CARM/RFC high gain margin; I (just slightly) reduced the gain in such phase, and all following unlocks happened later than here.

Then we unlocked between LOCKING_CARM_NULL_1F and in the first minute of CARM_NULL_1F, although no action happens in such minute; at the beginning this looked related to the MICH_PRCL coupling (so a mistuning of B4_56MHz demodulation phase), as it reached fairly high values (~10), but in the end this proved to be just related to the transient. In the lock that eventually reached LOW_NOISE_3 such phase is not well tuned, but hopefully this should be partially compensated by the diagonalization routine; for the next lock acquisition the usual phase will be reinstated.

Given the low frequency nature of the oscillation I started to look into the angular loops; at the beginning I focused on the input beam, as the ratio of the gains with respect to before the recent increase of the 8MHz sideband was quite asymmetrical (/6 for _X_ and /10 for _Y_); the test did not solve the issue at hand so I did not leave any change.

At this point, given that some unlocks happened after finishing the transients in LOCKING_CARM_NULL_1F, I tried to disable the change in the control scheme for SDB1_LC happening at the end of such state, as their ramptimes leak over the state itself. The interferometer started to oscillate again but without unlocking, possibly because SDB1 was more forgiving. Having the time to look at the thermalization I could see that DIFFp_TY was ramping up its gain, up to twice its initial value. After it reach a steady state, I changed online the SDB1 control scheme to what it was supposed to be, with no issue.

I modified the default DIFFp_TY gain from 8 to 12 in the automation, and the next lock acquisitions should be checked for consistency about this loop.