I have tried to run the noise budget for data from last night, I have adjusted the input power in the model to have 77kW in the arms, which should correspond to 340mW in transmission of the west arm. It doesn't work well because of high CMRF, as the frequency noise coupling is modeled by adding finesse asymetry in the plane wave model (while the correct model would involve higher order modes), and with high coupling the asymetry also changes significantly the DARM response used in the model. To work around the issue, I put a coupling in the noise budget a factor 10 than measured, and the by hand adding back the projection a factor 10 higher than it is computed.

Figure 1 shows the noise budget (with the SSFS noise a factor 10 lower than it should be), at 100Hz the dominating understood noise is quantum noise.

Figure 2 shows the total of that noise budget, compared to Hrec Raw, also the SSFS projection multiplied by 10 to match the SSFS coupling calibration lines at 227Hz and 1111Hz, and the mystery noise at the same level as before the input power reduction (1.7e-23*(100/f)^0.66).

So far the mystery noise level is compatible with having not changed with the input power reduction.

Figure 3 shows the same exercise for the data from the night before (Oct 25). To fill the gap in that case I need a higher level of mystery noise by 30%, 2.2e-23 at 100Hz. Looking at the calibration lines at ~105Hz and ~135Hz, the calibration lines are higher by 5% in this data, so the change in calibration is not the reason for the measured difference.

The mystery noise level seems to depend on the TCS tuning.