I've edited the guardian to use the sum of REFL A and REFL B 9I for CARM control, which should give us sqrt(2) lower frequency noise at low frequencies (below 1kHz) where the frequency noise is limited REFL shot noise.  

To summarize the story of these two sensors, in case Detchar is interested in looking at these:

We've had this REFL B channel available since it was installed in November and phased in early December.  Since then REFL B has been an out of loop sensor for frequency noise, since Refl A has been the only one used in the loop.  The Refl A readback was calibrated into volts on April 16th, so that REFL A +refl B have the same units. 

This tuesday (April 23rd) I switched the CARM loop from using REFL A to REFL B, as a temporary test because the better whitening on the REFL A readback means that it has better ADC noise clearance to use as a out of loop sensor.  This was only temporary, and we continued to use only REFL A in the CARM loop since then.

Craig and Georgia also increased the gain of the CARM loop by 3dB on Tuesday which helps to suppress the frequency noise (due to IMC shot noise) above 1 kHz.  Since then, it has been noted (Ed's shift summary) that there was extra high frequency noise at the beginning of the lock that went away a few minutes into the lock.  It may be that the optical gain of REFL 9 is higher in the first few minutes of the lock, and that the CARM loop has some gain peaking with the higher gain.  

In this lock, starting at about 19 UTC April 25th, we saw this happening again.  Jenne and I turned down the CARM gain by a few dB, and saw that this also had an impact on the power build ups. A few minutes later, the higher gain (12dB on IN1 gain) was fine.