While not as popular as GAYP or 3-move, 11-man openings are still used in some tournaments, and in the last couple of decades an 11-man world championship match has been played every few years. Since there are relatively few published games or analysis of these openings, I thought it would be interesting to see what kingsrow could reveal about this style of play. I have created an opening book, identified which ballots are losses, and developed a difficulty rating for each ballot.
The basic setup of an 11-man position is to remove one man from each side, and then each side makes one forced move. With some additional restrictions, there are a total of 2500 unique start positions. The NC Checkers website has a nice overview of the history and balloting details: Eleven Man Ballot Deck & System. I have adopted their numbering system for the ballots in the tables and pdn files here.
According to the NC Checkers website, Marion Tinsley is quoted as having said, "About 10% of them are probably losses". It's amazing how accurate this statement is. I have identified 247 lost ballots, and while I cannot guarantee that this is exact, it is likely within 1 or 2 of the correct number.
These files contain 11-man start positions. They can be used to load positions into CheckerBoard, run an engine match, etc.
The 11-man opening book file contains 1.1M positions and has play for all 2500 ballots. For technical reasons I have kept it separate from the normal kingsrow opening book. The opening book file is selected in CheckerBoard using the menu Engine, Options, More Options, Opening Book File. Both opening books are installed when kingsrow is installed. The normal opening book is for 3-move and GAYP play. There is very little overlap between the two books.
Table of 11-man difficulty ratings ordered by rating value.
Table of 11-man difficulty ratings ordered by ballot number.
At the ACF website there is this difficulty ranking of the 3-move ballots that was generated mostly by Richard Pask, with some additional input from a few other expert players for the 12 "mail play" ballots. I wanted to create a similar ranking for the 11-man ballots, but instead do it automatically by computer, as I am not an expert player, and even if I was, ranking 2500 ballots manually would be an enormous task. While I was trying various algorithms to rank the ballots, I decided that I could test them on 3-move ballots and see how closely their output agreed with the ACF's rankings. If I could find an algorithm that produces 3-move rankings that correlate well with the ACF rankings, then I could apply the same algorithm to 11-man ballots. After testing several schemes, and with helpful suggestions from Martin Fierz, I created an algorithm that gives reasonably good correlation between my 3-move rankings and the ACF rankings. My rankings are based on a linear combination of 2 values: 1) the sum of search scores along the opening book PV of each ballot, and 2) engine match results using 11-man start positions. The coefficients of the model were determined using linear regression.