Document logic/behavior of mentorship pairs in practice

[kfanning]

The input data actually used for pairings for polaris is not as ideal as described in McVitie and Wilson 1970. I made a lot of personal decisions that worked out well for handling incomplete rankings within the same algorthim. I need to document what was done, why, and how it behaves (particularly recommendations for the minimum number of rankings needed to get reasonable pairs across the sets)

As a side note ~5 rankings/person for 40+ pairs this year seemed to be about the most extreme you can push this. Off the top of my head ideally you want 1/4 of people with ranks I think.