The BS anti-reflective coating polarization P reflectivity amplifies the content in P-polarization on B5 by a factor 40. This means that when setting an upper limit on the polarization angle fluctuation in the locked interferometer based on the ratio of P and S polarization on B5, instead of 2.5e-8 mW/rtHz / 5.8mW = 4.3e-9 1/rtHz, on should use 2.5e-8 mW/rtHz / 40 / 5.8mW = 1e-10 1/rtHz.

The question is then how to convert upper limits on power fluctuation into upper limit on the angle. There is 0.34mW on B5P compared to 5.8mW on B2 S-polarization. Taking into account the factor 40 from the BS AR, the ratio of the two is 1.5e-3, which correspond to a polarization angle of 0.04rad. If one adds angle fluctuations of 1.2e-9 rad/rtHz to an angle of 0.04, then in the P-polarization there are power fluctuations relative to the power in the S-polarization of 1e-10 1/rtHz.

That is the same noise needed to create a noise at the level of the mystery noise when following the projections given in LIGO-P2300220  even when assuming very pessimistic parameters for coupling that would give a large coupling. To complete exclude out this scenario replacing B5P with a photodiode with 3 times less sensing noise could be a simple solution.