I was waiting for sleep to come, and surfed over to the American Checker Federation website. As long-time readers of this blog might remember, I’ve been tinkering a checkers program together, which I tentatively named “Milhouse” to play checkers. This week’s problem challenge was a classic 2 on 3 battle where White is to move and force a draw from the down position:

It’s late, and been a full day, so I can’t really say that I understand this position. But Milhouse doesn’t get tired: when I load the endgame database, it immediately proclaim that the position is a draw. What’s more is that it identifies two different lines: their website suggests 23-26 as a drawing move, but 19-15 is also a draw.

Without the endgame database, White thinks it is down a man, but with a 25 ply search can’t find the draw. In addition to the two moves listed above, it thinks that 23-18 is also almost as viable. But 23-18 is a dead loss, presumably because it allows Red to king both of its checkers. Even with the database though, Milhouse can’t seem to find the win for Red in a 31 ply deep search. This seems like a good test case for future improvements to playing with the database.