Below is a measured gyro spectrum from today, plotted with the expected displacement noise sources as described in the document here. It is much quieter than we have had it before, though it has been reproduced enough times for me to believe it is in fact real. It occurs to me that we are now very close to the DAQ noise level, and this could be the broadband noise floor we are seeing. The few spikes above this floor are still a bit higher than calculated, but not by enough that calibration error in the noise estimates are inconceivable.
We may have accidentally done something to dramatically reduce some unexpected noise coupling by messing with the servo settings in both the locking loops and the PLL. We should preamp the gyro signal into the DAQ and get a real spectrum immediately.