J. Kissel

While we continue investigate the parasitic cross-coupling with the low-frequency sensing function a.k.a. the DARM plant (see e.g. LHO aLOG 53378), we're stuck with just taking whatever the IFO gives us as far as a response below 20 Hz. Unfortunately this feature has resumed being time dependent (unlike the ~1 month between "after we installed the 100 ct offset on 2019-08-28" and "the end of O3A on 2019-10-01"). Obnoxiously, with only ~5 data sets so far, we can only say that that time dependence is bi-modal: in the first two weeks the response is different from the most recent two weeks.

Check out the attached collection of data, from 2019-10-31, 2019-11-04, 2019-11-11, 2019-11-20, and 2019-11-27. One can see two things:
    (1) The bi-modality of the data below 30 Hz (though, yes, the 2019-10-31 data set appears to have yet a third response) 
    (2) That the 2019-09-09 model used at the tail end of O3A is still quite good above 30 Hz, and the systematic error between these data sets and the model are still below the threshold in which we say "we really need to do something about this," which -- according to studies in G1901479 -- is "the systematic error of any one measurement show be no more than 8 (,16) % or 5 (,10) deg at 20 (,15) Hz (assuming the error is decreasing as a function of frequency as has been seen in the past)."

Further unfortunately, until we have a good model of what's happening between the mixing of detuning vs. L2A2L, and/or more than one metric such that we can separate the tracking, we have no good way to re-contruct the frequency dependent change in response with a few parameters derived from, calibration lines (like what we normally do with the optical gain, DARM coupled cavity pole frequency, the ESD actuator strength, etc).

Thus, we'll have to move forward with what has been done for the first parts of O3A, force the uncertainty in our unknown systematic error (i.e. the GPR fit to residual between model and collection of measurements and it's associated uncertainty) to be some value that we believe sufficiently covers the realm of possibility of this response (within a 68% confidence interval). Sad that we have to rely on such human metrics, but so be it!