I've verified the correlated DARM noise measured during the latest commissioning period.
I focused on a two hour period, starting at May 13, 2021 18:27:21 UTC.
I used "PSD-rejection" to statistically find and remove glitches from the analysis.

------------------------------------------------------------------------------------------------------------------------------------------------

Plot 1 is the correlated DARM comparison between the latest Pre-O4 correlated noise and the O3 commissioning paper correlated noises.
Each falls on a similar noise floor between 70 to 500 Hz.
The consistency of these noise floors may suggest we are measuring similar thermal noise in both interferometers. 
The thermal noise plotted comes from pygwinc, and the aLIGO thermal subbudget is here.

Plot 2 is the correlated noise budget for Pre-O4.
The high frequency region is likely dominated by intensity noise that's slightly higher than that measured for the noisebudget.
The middle frequencies are estimated to be dominated by thermal noise, and partially "mystery noise" which dominates at 40 Hz.

------------------------------------------------------------------------------------------------------------------------------------------------

The next plots are related to the correlated DARM verification process, to ensure we're hitting the true correlated noise floor and not the unresolved shot noise limit.

Plot 3 is the correlated DARM CSDs for different frequency resolutions, for the purpose of getting enough averages.
Plot 4 is the same as Plot 3, but with logbinning.
Comparing Plot 3 and Plot 4, we can see how logbinning the high frequency resolutions, with a low number of averages, gives the same result as a low frequency resolution with a high number of averages.
Logbinning accurately estimates the high frequency correlated DARM noise floor, providing artificially high averaging for the high frequencies.

Plot 5 is the coherence of the DCPDs with the DARM control loop removed.
Plot 6 is the same as Plot 5, but with logbinning.
This time, logbinning does not accurately estimate the high frequency coherence, but rather is overestimates it.
In any case, the coherence does not continue to go down below 10^-4 as the number of averages increases above around 10000, meaning we have hit the true correlated noise floor for the correlated noise DARM CSD.

------------------------------------------------------------------------------------------------------------------------------------------------

The code is in the aligo_commissioning/correlated_noise git repo:
https://git.ligo.org/aligo_commissioning/correlated_noise

------------------------------------------------------------------------------------------------------------------------------------------------

Other relevant alogs:

LHO O3 correlated DARM comparison: LHO alog 57284
Animation of August 2020 correlated noise plot: LHO alog 56942
Cross correlation comparison before and after discharging: LHO alog 56896
Initial LHO O3 correlated DARM: LHO alog 52387