I moved the soft degrees of freedom to maximize the power in the arm cavities, and reset the transmission QPD offsets to match the position.

For reference here are the recycling gain and power levels averaged over 30 seconds:

H1:LSC-PR_GAIN_OUT            37.43
H1:LSC-TR_X_NORM_INMON 1308.5
H1:LSC-TR_Y_NORM_INMON 1369.9

The new offsets are

H1:ASC-X_TR_A_PIT_OFFSET         0.074
H1:ASC-X_TR_A_YAW_OFFSET       0.144
H1:ASC-X_TR_B_PIT_OFFSET         0.057
H1:ASC-X_TR_B_YAW_OFFSET       0.196
H1:ASC-Y_TR_A_PIT_OFFSET         -0.018
H1:ASC-Y_TR_A_YAW_OFFSET       0.053
H1:ASC-Y_TR_B_PIT_OFFSET         0.083
H1:ASC-Y_TR_B_YAW_OFFSET       0.151

We could close the loops using the guardian and I measured the step response, by adding a +0.03 offset to each of the inputs. Time constants are approximately (see attached plot)

DSOFT_PIT        35s
CSOFT_PIT        45s
DSOFT_YAW     15s
CSOFT_YAW     20s

The error signals are now reconstructed with the new sensing matrix