Calibration wise it seems that a 0.4W step in ETMX RH power will reduce the 2nd order HOM frequencies by about 100Hz. This however isn't a simple linear change though, as changing the RH can also drops the circulating power in both arms which introduces more shifts in the HOM resonance.

You can see this effect in the orange and red traces. Here increasing the ETMX RH is reducing the arm circulating power from 370 to 350 kW, which in turn moves the Y arm modes lower in frequency. A large step was taken between orange and red of 0.8W the aim to move the HOM resonance through 10.4kHz quickly - this worked fine. There is a reasonable reduction in HF frequency noise couping in the red curve as can be seen. I didn't manage to get both arms to overlap their modes unfortunately.

Craig has some broadband measurements for the low frequency coupling I'm sure he'll upload.

To improve the build ups I tried some small annular CO2 changes but it did not really make any big improvement to the PRG and arm powers, as we'd need to be winning back 20kW. After this I tried reducing ETMY to bring the Y modes up in frequency to overlap better with the X arm. However this improved the buildups which pushed the X modes up slowly into 10.4 kHz which was enough to cause it to ring up the acoustic mode that eventually caused a lockloss. If I had spotted it sooner then maybe switching off the ETMX RH would have moved the HOMs up fast enough it wouldn't have been an issue.

During the broadband frequency noise injections, you could also see these X and Y HOM peaks increase significantly, so the reason we can see these peaks so clearly at the moment is because the laser noise is so high.

I have accepted H1:TCS-ETM{X,Y}_RH_INVERSE_FILTER_GAIN in sdf. 

Here are the frequency noise coupling measurements taken during this measurement.

https://lhocds.ligo-wa.caltech.edu/exports/craig.cahillane/Git/IFO/FrequencyNoise/figures/frequency_transfer_function_budget/frequency_transfer_function_budget.svg

I've also attached two PNGs with the most relevant traces highlighted.
The first PNG shows the frequency noise to DARM comparison traces with LSC feedforward OFF.
The second PNG shows the frequency noise to DARM comparison traces with LSC feedforward ON.

It's clear that the ring heater moves improved our frequency noise to DARM, at high frequency and low frequency, but not at 60 Hz where the high coupling is most important to our range.

With the feedforward off (PNG 1), we get a clearer idea of how true frequency noise to DARM decreases.
There is some robust hump at 80-90 Hz that makes the coupling worse, and is not improved too much with the RH moves.  
There was *some* improvement at 80-90 Hz at the very end, when the EY RH move was happening, but we lost lock before we could really test that (the EY RH never really settled at it's nominal value).

Dan also took a frequency to DARM measurement with the LSF FF on (PNG 2).
Comparing to before, there is improvement across the board, although maybe not as much as we might have hoped at 60 Hz or so.
Below 60 Hz, things are also much improved.  

Finally, our high frequency coupling sinks to basically nothing.  
I wonder what the contrast defect was at this time.
The coupling becomes entirely dominated by nonlinear effects (the injection is still visible in DARM, although at a lower level and there is little coherence).

Definitely worthwhile to check out what happens when the modes actually align, 
but this test seems to indicate the LF coupling at 60 Hz is not vastly influenced by the mode overlap.