We lost the AS36 phasing that I did yesterday in full lock with the computer crash last night.  I set the phase today during a DRMI-only time to zero the sum of the Q phase of AS_A_RF36, which gave me a phase 30 degrees different than my full lock phase yesterday. 

We were able to acquire lock with this new phasing.  As we thermalize, the Q sum is drifting away from zero.  Right now at about 90 minutes into the lock the ideal phase (from atan(Q/I)) is 10 degrees different than the DRMI-only phase, in a direction consistent with yesterday's measurement.  So, right now in full lock after 90 min, the ideal phasing is 20 degrees different than the value I found yesterday (and yesterday I might have overshot a few degrees).

So far, it seems as though Daniel is right that we could potentially servo the AS36 phase by calculating the arctan of Qsum/Isum.  Probably we could do this in guardian, since it only needs a ~1 minute cadence or less; it's not a fast thing that would require a model change (although we could do that too or instead).