It turns out that most of the power in the HWS path is just dumped. We installed a PDA55 to monitor this otherwise wasted beam so that we can measure the reflected beam power without being clipped.
According to this PD our contrast (1-min(locked)/unlocked) is 65%. I'm taking min(locked) because, when locked, the REFL power fluctuates much because of the alignment fluctuation (but not clipping).
We can refine alignment further, but doing it manually is no fun. We're working on WFS. Also, though there's no fast ADC monitoring the refl DC, we might be able to auto-dither connecting REFL DC to AUX channels.