Following a suggestion from Michal, whose noise budget states that DARM is affected by the B4_56MHz sensing noise (either from MICH or SSFS), an additional increase of the modulation amplitude of the 56 MHz sideband was added inside the SET_MOD_AMPL state, bringing it back to 12 dBm as it was before the increase of the input power. This change was tested this both online (already when in LOW_NOISE_3) and during lock acquisition, and it seems to be working fine. Longitudinal and angular loop gains were adjusted accordingly.
The first stretch of good data is from 18:10:00 LT until the manual unlock.