After chatting with Sushant at LLO, I followed his advice to add a bit of weight to the minimization metric at 60 Hz, and was able to get cleaning parameters that do a pretty good job of subtracting jitter and laser frequency noise, but don't inject noise at the 60 Hz line.  The jitter subtraction is a little less good than alog 72276, but it's important to not inject noise with the cleaning, so this is a good balance. I similarly retrained the laser noise, and with both of them on (while other commissioning work was ongoing), I see good subtraction and no noise injection at 60 Hz.  This is shown in the first attachment.

After having collected a bit of time with the jitter and laser noise subtracted, I retrained the LSC modulated cleaning.  Unfortunately, when trying it in the online system (I don't have the offline mimic of this working yet), it needs a similar mysterious minus sign much like the jitter subtraction needed until I recompiled the code last week.  I'm starting to suspect that the LSC NonSENS c-code also needs to be regenerated and recompiled, and hopefully that will solve my minus sign issue.  Anyhow, when the LSC noise estimate has that minus sign in place, it gives a reasonable amount of subtraction, as shown in the second attachment.

The third attachment shows the improvement (when blue is above pink/orange) in range from the cleaning.  Around 19-20 hours on that plot I'm just subtracting jitter and laser noise.  Around 21 hours I add in LSC subtraction, and we get back to ~150 Mpc.  So, likely our LSC FF is becoming mistuned, and we'd benefit from either retuning the LSC FF or finally implementing the NonSENS subtraction. 

The cleaning review committee meets tomorrow.  All of this is turned off when the IFO is in Observing until given approval by the review committee.