Figure 1. I have looked at the RF quadrant time series for a lock from Sunday, and B1p QD2 might be saturating when in dark fringe. It does not reach 8192 count on the ADC, but this quadrant is connected to a demodulation board with the old AD8370 VGA that has 5 times less dynamic. The gain on the VGA is set to 2.3, so the ADC range is about the same as the analog range of the VGA.
Figure 2 Compares the RF spectra and B1p QD2 seems to have more RF harmonics around 100MHz than B1p QD1, so I suspect it stops being in the linear range of the RF dynamic.
B1p QD1 should be fine as it is connected to a demodulation board with the new VGA that has 5 times higher dynamic range. The gain is set to 2.3, so if the ADC starts to saturate (8192 count, while max is at 4000 in the lock I have looked at), so if needed the gain can be reduced by a factor up to 4 to increase the dynamic range. But it does not seem needed at the moment.
B1p QD1 should give a cleaner signal than B1p QD2 for alignment purposes (no risk of signal distorted by saturation).