Predict winning probability per turn
With ~300,000 games, a machine learning algorithm might be able to estimate the winning probability after each turn with some reasonable predictive power. It opens up a lot of interesting possibilities if it was decent. It could be useful to implement an AI and to highlight bad buys.
AdminRob Renaud (Programmer/Parsing Monkey/Fake Statistician, Councilroom.com: Dominion Statistics) commented
Some advice on accomplishing this from metaoptimize.