[Josh Adelson]

This is a fascinating thread; Great job, billw. I have one suggestion that I believe would make the model more realistic.

The initial deterministic ranking of teams from 1 though 96 implies a linear distribution of team 'merit' from top to bottom. In reality I suspect that relative team merit within a tournament more closely follows a normal distribution around a mean. What if instead as an input you assigned each team a baseline score between 1 and 96 with a determined mean and standard deviation? You would still be able to rank the teams 1 through 96 but the top team might have a basleine value of 92.4, the middle teams' baseline values would be more closely bunched, etc. Then continue the exercise as you have done.

I would be interested to see if this makes a difference in the simulation outcomes.