A Database Dilema Resolved

Generating endgame databases is not an easy task! It requires substantial time, computing resources, programming skill and mathematical knowlege to do successfully, as I know from my own experience generating the endgame databases for Wyllie Draughts.

Both these positions are White to play. Before reading any further, checker players may like to ponder the question as to what difference the placing of the black king, which is on 3 in the left diagram but 4 on the right, makes to the outcome of the game with best play.

The position on the left was first posted as a draw at the WCC web site. Here is what was posted about the position .....

"With White to move from the position shown above, what does your program predict as the result? WCC Platinum knows this result instantly, without having to search. Amazingly, White to move can draw. This 7-piece position can eventually be swapped down into a permutation of a 5-piece Payne's Draw. Most programs would search for about a week and still get the wrong answer! Now do you see why the 7-piece database is so valuable? If the program was "in trouble" it would throw away a piece to get into positions such as this that hold the draw. There are hundreds of millions of draws in the 7-piece database!"

Now it is true that there are hundreds of millions of drawn 4v3 positions, and that these man-down draws make possible some neat escapes that programs without a 7-piece database would miss, but I was surprised to see this particular position posted as a draw, since the position is in fact a black win according to Wyllie Draughts. This meant that either the position posted was not the intended position, or, more seriously, that the either the Wyllie Draughts or the WCC Platinum databases were in error! The only other databases for 7-piece positions besides Wyllie Draughts and WCC Platinum are those of Chinook and those of Nemesis. I contacted Murray Cash, author of Nemesis, and asked him to check the position with his own independently generated databases, and he quickly confirmed the result given by Wyllie Draughts: there is no draw in this position! I concluded that either the position was given incorrectly on the WCC site or the WCC database for 3 kings + 1 man v 3 kings was flawed, which would of course be very serious if it were the case, for all the other 7 and 8 piece databases except for those with only kings on the board depend on the data in the 3 kings + 1 man v 3 kings database.

After I published my findings in an article on this site about the discrepancy Mr Ed Trice, co-author of WCC, told me that his databases agreed with the Chinook ones and challenged Wyllie to prove a win against WCC Platinum. The ante had now been upped in the sense that it now appeared that out of the only 4 existing 7-piece databases at least 2 of them appeared to be in error! However, it subsequently emerged that the position, originally presented by Dr Schaeffer, author of Chinook, had indeed been incorrectly displayed, and the king on 3 was meant to be on 4 instead, as in the diagram on the right. This minor alteration does indeed change the outcome to a draw, and resolves the apparent discrepancy between the Wyllie/Nemesis databases and the Chinook/WCC databases. Although it may seem to be wasteful of effort for each programmer to generate his own databases, it does help to remove any doubts about their validity when independently generated data can be compared and found to be in agreement.