Slitherine

It would be interesting to see stats on if the AI played itself 100 times, the win % for each side. If the game was perfectly balanced it would be close 50-50 but that would probably be hard to achieve. My feel is that it is pretty well balanced but it'd be interesting to see the numbers. Or even the human win/loss as each side in the different scenarios...

Also, I'm curious how the engine works. Does it search moves in a tree at all? That seems like it would be difficult to create a machine that could think more than a few ply (turns) deep because of the unpredictability of combat outcomes. I wonder if there are aspects of chess engines that could be carried over to games like this to make a stronger faster AI?

One other note is that it would be interesting to be able to see how the AI evaluates the position... (whether it thinks its winning, losing etc.)

We do have runs with all the AI's playing each other - they have different strategies, so it's not consistent which side wins.

Campaigns between humans (through GC) it's 50.5% Axis, 46.8% Soviet (balance: tie).

Typhoon is 50% Soviet, 40% Axis, 10% tie.

Note that we have no way to efficiently distinguish games played to completion from those resigned in the middle, so take those results with a grain of salt...

The AI generals with medals do use a tree-search, modified to address incredibly rapid branching, stochastic factors, and things like breakthrough. Even going a move or two ahead helps a lot -- we're not trying to win tournaments, just to give good players a good game!