After Matt was gone, alignment degraded over 3 hours or so and ETMY output of slow servo for offloading VCO was getting large (lock filter output of the ETMY was more than 200000 counts).
Since IOP watchdog was alarming, I unlocked the cavity, reset the ETMY LOCK filter and relocked.
My guess is that the servo is running fine and it's just that the cavity length drift is big due to tidal or whatever. Regardless, I reduced the gain by a factor of 2 (H2:SUS-ETMY_M0_LOCK_L_GAIN=-0.1 instead of -0.2), as I want to provide more window to potential seismic/sus data mining for the night before the IOP watchdog potentially freaks out.
BTW the alignment got back to normal though I didn't realign. The poor alignment was apparently caused by the length drive with large length to angle drive coupling.
About the length to angle coupling: you can see at about -2.4h in the StripTool plot where I tuned the 4th digit of the L2Y coefficient. The L2P tuning stil needs to be tuned, though I have alrady confirmed that it is less than 5e-4. Unfortunately, it is not clear that 1 part in 1e4 is good enough (since the length drive is 1e5 and we are sensitive to alignment of ~10 counts on the sliders), nor is it clear that we can depend on linearity at this level. We should be offloading this to HEPI, but that is not yet possible. Some low-frequency WFS servos are probably the only hope for long OAT locks.