I diagonalized the optical lever by setting the TM_SENSALIGN matrix to the inverse matrix of the coupling ratios.
In the case of L2P coupling, the coupling ratio was obtained by shifting the IP and measuring the shift in TM L and P. It was small, to begin with (~ 1%).
L2Y cannot be measured at the beginning because I don't trust the IP diagonalization, and the 0.65 Hz mode is pendulum mode so we can't tell whether it's P2Y or L2Y.
But after the first diagonalization, the peak almost disappeared in the Y spectra so it was probably P2Y coupling.

All cross-couplings ratios are now below/around the 1% level.
See figures for the spectrum before (references) and after the diagonalization
Fig.4 shows the TM_SENSALIGN matrix.

TM_SENSALIGN
[[ 9.99689693e-01, -1.41851361e-02, -1.40984481e-02],
[ 2.18535148e-02,  9.98995086e-01, -1.73635376e-02],
[ 8.90296895e-04,  4.06983605e-02,  9.99292622e-01]]

Measurements at /kagra/Dropbox/Subsystems/VIS/vis_commissioning/sr2/diagonalization/