I have been bothered by the problems we had  the other night  during the locking sequence, while trying to lock with the same parameters that worked very well the day before. Also, there have been several instances in the past in which some tuning parameters where good one day, and totally wrong the day after. I have been saying that it has all to do with being in different "alignment" states, but the changes we were seeing seemed too big to make sense even with this typical excuse. 

However, I think it does make sense, at least for some of the lock acquisition failures. Because we use the transmission DC signals as error signals for CARM, we somehow "force" ourselves to move to some fixed build-up value, and we tune our locking parameters accordingly. However, if for example we are not properly aligned, "forcing" ourselves to some pre-defined CARM offset means that we are essentially moving higher up in the sequence than we should be, similarly to the case in which we had high loss in the arms . Attached is one example of what I mean.  

The first plot shows a successful locking sequence, while the second one shows a failed one. In the first plot, for a power build up of X24 single arm, REFL_DC is still 20 and REFL_9I is ~1800. In the second plot, for a similar power buildup, REFL_DC had already started to decrease (17) and REFL 9I is much higher than before, ~ 2500. That would correspond to a much higher build-up (~100) in our "baseline" working sequence, and consequently a different locking tuning (gains, filters, etc). So, it is not surprise that we break lock.

So, the current plan of improving the initial alignment and closing WFS along the sequence is indeed a good one, and most likely all we need.