Nice overview!

You gave a very nice explanation of play, and if you don't mind I'd like to link to your show from our site. I do have a couple of tiny nitpicks:

When you describe Paratroopers, you stated (correctly) that they can't just drop into the enemy base. However, your description gave the impression that they couldn't enter the any base at all, and that the only reason for their supply diagram was for them to form a supply line for other units. In fact, a unit of Paratroopers can deploy into the enemy base as though it were a normal unit, i.e. if the enemy base is empty and if it can trace a supply line from the enemy base back to its own base.

As was already noted, the AI doesn't card-count at all. What it does do is combine two things:

1. Lessons I've learned from over three years of playing the prototype. (I didn't do any of the programming but I provided a lot of input to Andrew about how to value things and where the computer's play could be improved.)

2. A computer's natural ability not to miss any possibilities.

Basically the AI will examine all of its legal plays and the resulting positions and apply a scoring system to them. That scoring includes some basic things about what you might be able to do back to it (for example, if there is a possible win for you, it will try very hard to avoid it) but it doesn't check to see what cards are left when making that evaluation. If you've played all of your Special Forces the computer will still apply the same rating to a board position in which a win is possible only if you have Special Forces in your hand.

Again, thanks for a very good and thorough overview/review of Hill 218.