In over a thousand games played in computer Go tournaments on KGS, ten have been ended by the Superko rule. In four of these ten cases, the effect of the rule effectively decided the result of the game.
There are four forms of superko rule:
PSK | Positional SuperKo | A player may not play a stone so as to create a previous board which existed previously in the game. | Creation refers to the position just after the play and consequent removals. |
---|---|---|---|
SSK | Situational SuperKo | A player may not play a stone so as to create a board position which existed previously in the game, and in which it was the opponent's turn to move next. | Creation refers to the position just after the play and consequent removals, and the rule also takes account of who moves next. |
NSK | Natural Situational SuperKo | A player may not play a stone so as to create a board position which existed previously in the game, if s/he played to create it previously. | Creation refers to the position just after the play and consequent removals. |
ChSK | Chinese Situational SuperKo | A play may not recreate a previous board position from the game by means of basic ko or sending-2-returning-1. If a cycle arises in another way and neither player varies from it, the referee may declare a draw or require a replay. | This, according to the sixth meeting of the International Go Rules Forum, is what the Chinese rules were intended to specify. |
All forms of superko rule refer only to playing of stones. It is always legal to pass, whatever form of the rule may apply.
Computer Go tournaments on KGS have so far always used "Chinese Rules", as implemented on KGS.
At the time the superko code on KGS was implemented, the intention of the Chinese Rules of Go about repeated positions was unclear. Section 6, as translated into English in "The Go Player's Almanac", Richard Bozulich, 1992, reads in full:
The application of the Positional Superko rule has several times affected a computer Go tournament game on KGS. These incidents have all be mentioned on the reports of the event, and are listed below.
date (link to report) | What happened | type of cycle |
---|---|---|
September 2005 | Viking5 was unable to play the move it wanted against gonzoBot, and so resigned, in what would otherwise have been a won position. | Sending two, returning one |
October 2005 | Dariush timed out in a won position against SlugGo, when prevented from playing the move it wanted. | |
November 2007 | Break timed out in a lost position against FirstGoBot, when prevented from playing the move it wanted. | |
December 2007 | MonteGNU timed out in a won position against WeakBot50k, when prevented from playing the move it wanted. | |
September 2008 | SimpleBot timed out in a won position against WeakBot50k, when prevented from playing the move it wanted. | |
October 2008 | HouseBot timed out in a lost position against ManyFaces, when prevented from playing the move it wanted. | |
October 2009 | Orego timed out in a lost position against WeakBot50k, when prevented from playing the move it wanted. | |
October 2009 | Wei2go timed out in a lost position against SimpleBot, when prevented from playing the move it wanted. | |
May 2010 | MCark timed out in a lost position against kiseki, when prevented from playing the move it wanted. | |
July 2010 | Orego timed out in a lost position against CzechBot, when prevented from playing the move it wanted. | triple ko |
May 2017 | Julie timed out in a lost position against Aya, when prevented from playing the move it wanted. | Sending two, returning one |