This machine performs a simple 1/3P algorithm.  We experimented with 1/6 and 1/18 as well, but in both cases the conditionals added up to a higher instruction total.

We also considered a longer arm.  The instruction count seemed theoretically superior, but we could not find a working geometry.