Disclaimer: This is an unfinished game but I have already obtained consent from the opponent before posting this.
I was playing the game as above and was excited that I finally saw an eternal life in a real game, but I was shocked when I tried to play at A4 and I couldn’t play.
Is this a bug or is this an expected behaviour on OGS?
I thought Chinese rules also have eternal life and it should be a draw or no result?
How should this be handled?
Please let me know if I should wait for moderator action instead of discussing here.
Thank you.
Not sure whether there was an eternal life on OGS in the past 10 years? The shape itself is so rare there are only 3 examples on Sensei’s library. All of them are Japanese or Korean rules and have been judged as a no result. I haven’t been able to find a case for Chinese rules, but according to the rulebooks, the judge may deem it to be a no result.
But I think the idea that “Chinese” rules on OGS have used superko since maybe 2017 (going by the github file) or earlier, kind of means that whether it was send two return one, triple or quadruple ko or eternal life, repetitions aren’t going to be allowed. If you see what I’m saying?
I guess whether it’s really a long-term bug, something long term on a to do list, or just a choice that was made probably for simplicity, is for @anoek maybe?
once we hear back from @anoek about whether or not it is a server bug, I guess the relevant decision is up to @gamesorry as tournament director to decide what the game result is and how to proceed.
Traditional Chinese rules judge some cyclical situations (including triple ko and eternal life) as “no result”, but applies a positional superko rule in most other situations. Moonshine life is dead.
OGS implements a simplified form of Chinese rules, where positional superko is always applied (and thus no cycles can occur).
I think the situation with OGS is a design choice, as it may be a bit tricky to properly understand and handle all of the potential special cases. One would have to distinguish some long cycles from other long cycles, and selectively apply the superko rule.
KataGo will accept any pseudo-legal move (i.e. legal ignoring history or multi-stone self-capture) if a GUI insists on saying it was the move made, including even multiple moves in a row by the same player. SGFs may easily be created to violate order of play or ko rules or other things for example, and it’s nice for the engine to not reject any variation that could show up in an SGF.
But the search should refrain from suggesting the move on its own. See here the absence of any analysis for A4 despite every other pseudo-legal move having at least one visit.
Just default settings, and changing the rules to chinese-ogs or any similar positional superko ruleset. You’re either not setting the rules correctly, or they’re getting overridden by your GUI or the SGF file, or something like that.
The interesting thing is I ran the position using rule = chinese-ogs on KataGo, and it still allows me to play at A4.
Maybe you have a typo? It should be “rules = chinese-ogs”, not “rule = chinese-ogs”.
KataGo also will output a warning to stderr if you have unused parameters in your config on startup as a guard against typos, but if you are using Lizzie, it won’t surface these outside of the engine console, plus I think it has a race condition on engine launch where it has a significant chance to just eat the stderr and never show it to you even in the engine console output. It’s possible more recent and maintained forks like LizzieYZY, have fixed the latter race.
Never thought I’d see an eternal life in a real game, let alone being a referee of it
In this case, I think we should end the game as Jigo and assign each player 0.5 points in the tournament (I think as a TD I can adjust the points by 0.5 points if the system doesn’t do that already). The reasons are that (1) this is typically how it’s resolved in real life games under Chinese rules, (2) I like the philosophy of ‘when in doubt, play it out’, which leads to an infinite loop in this case and thus a draw, (3) this situation is so rare that I don’t think punishing any of the players for it is necessary.
I think this is a longstanding misunderstanding in the West. Ttbomk normal long cycles are always draw or no result under Chinese rules - not really as an exception, more like the actual rule. (See ChineseSuperko on sensei’s and jaeup’s post on L19.)
I also think OGS is “incorrect by design” here, but it doesn’t seem that hard to fix this.
Only forbidding repetition if the player lost more stones since the last occurrence would already be reasonably good. And if you switch to the current behavior AFTER the first stop / resumption you also cover moonshine (which, according to the incident mentioned on sensei’s, should still be draw if played in the main game).
