I used the new script described in 44532 and notches at 30 Hz in the HARD loops as described in 44538 to measure and tune the angle to length decoupling.

A first set of yaw to length decoupling further improved the low frequency sensitivity and reduced the coherence of CHARD_Y with DARM.

Running the pitch to length decoupling improved the sensitivity below 15 Hz:

The plot below summarizes the improvements in the two steps of decoupling:

Attached a log of the script output.