I took the time to create a YouTube video on the SUNDAR. This gives an overview of the system:
And this is some more detail on the sensor head:
I also found that I could get much better accuracy if I narrowed the slit even more. It is now the width of a playing card. As such, the sun is only visible to the sensor for a very few motor microsteps. I collected data over an hour and found that the standard deviation in heading error was under 1/2 degree. I think I can live with that!