It would be useful to be able to test future matching and ranking algorithms to see how quickly and accurately they rank bots. Since matching and ranking affect each other, to evaluate them we need to simulate picking match, having a fight, and then adjusting the rankings. It would be painful to actually run all the battles needed to know an algorithm's performance.
I wrote a tiny python class that simulates battles based on the "Likelihood of Superiority" results from last contest. It handles server-day worth of battles in under a second.
https://gist.github.com/849248
Likelihood of Superiority Results:
viewtopic.php?f=3&t=1208