Thirty-third KGS Computer Go Tournament

Sunday December 2nd, 2007

These results also appear on official KGS pages: Formal Division, Open Division which link to the game records.


 Formal divisionOpen division
format6-round Swiss6-round Swiss
board size19×1919×19
time18 minutes absolute18 minutes absolute


The first round started at 09:00 UCT for the Formal and 09:10 for the Open division.


As usual, the tournament was held in two divisions, Formal and Open, with more restrictive entry conditions for the Formal division.

Formal Division   19×19


Open Division   19×19


The "real" names of the bots listed above, and of their programmers, are listed here: programs which have registered for KGS Computer Go Tournaments.


MoGoBot3, playing in the Formal division, was running on four Xeon 4-core 2 GHz systems. MoGoBot2, in the Open division, was rinning on four 8-core 2.6 GHz systems.

ManyFaces2, in the Formal division, was running on a slow (1.6 GHz) laptop. It was an experimental version Many Faces of Go, doing a 3 to 5 ply alpha-beta search with many extensions, and deep search on patterns during quiescence.

CrazyStone was running on an 8-core system.

LeelaBot was running on a processor with two 2.2GHz Athlon64 cores.

AyaMC registered, and played in the first round, but had disappeared before the start of round 2. Its author Hiroshi Yamashita was busy in Tokyo, competing in the First Computer Go UEC Cup, and was not able to reset it. So I removed it from the competition before the start of round 3.

Formal division

In round 2, as mentioned above, AyaMC did not appear for its game with pagebot, and eventually timed out.

In the round 3 game between LeelaBot and pagebot, pagebot passed as black in the position shown. LeelaBot was happy to pass too. The players agreed that all the stones were alive, so LeelaBot won by the komi.

The round 3 game between MoGoBot3 and ManyFaces2 began as expected, with ManyFaces making conventional-looking moves while MoGo sprinkled stones on the sixth and seventh lines. MoGo killed a group with move 62, and eventually made the whole centre of the board into territory, while ManyFaces took all four corners and two whole sides. Mogo (using 13 times the processor power of ManyFaces) managed, in the familar UCT way, to win by exactly half a point.

In round 4 pagebot again made an early pass, on move 16; and again its opponent, break19 this time, responded by passing to end the game. This time it worked out better for pagebot, who as white won by komi.

Open division

In round 2 scottbot had some problem, and repeatedly disconnected from and reconnected to its game with SimpleBot. It never made a move, and SimpleBot eventually won on time.

In round 3 scottbot had the same problem in its game with IdiotBot. Again scottbot never made a move, and IdiotBot won on time.

I am grateful to Aloril for running his bots in these events. While the moves made by IdiotBot are not instructive, it does play fast and legally. A bot which fails to beat IdiotBot is in need of attention, whatever the reason for the loss.

Also in round 3, WeakBot50k reduced all of HBotSVN's groups to one eye or worse. Both players then passed, and disagreed about the status of the groups. WeakBot50k succeeded in removing all HBotSVN's stones from the board, while HBotSVN made meaningless moves and narrowly avoided losing on time.

Scottbot's connection problems continued in its round 4 game with HBotSVN. It eventually managed to connect and start playing, but by then less than 7 of its 18 minutes remained. Scottbot then achieved a won position on the board, but ran out of time.

In round 5 scottbot was again late in managing to join its game and play a stone, this time against MoGoBot2. But scottbot was polite enough to resign when it was clearly behind.

Top right corner of board.
white 4 captures 3
black 5 captures 2,
leading to:
Also in round 5, MonteGNU as white lost to Weakbot50k, when it failed to move and timed out after the sequence shown. My guess is that it wanted to start a "send-two-capture-one" sequence by playing midway between 2 and 3, and was prevented from doing so by the "positional superko" rule. It would also have been prevented from recapturing at 2 by the ordinary ko rule. It is unusual to see two of CGoban's "ko-forbidden" squares on the board at once.

Scottbot managed to join its round 6 game with StoneCrazy on time, but took more than half a minute each for its first five moves. Eventually scottbot resigned.


Towards the end of round five, the results from the First Computer Go UEC Cup in Tokyo were announced. First was Crazy Stone, second Katsunari, then MoGo, then AyaMC, then GGMC. So CrazyStone won three events in a single day (one in Tokyo, two on KGS).