This would only be the case if the players agreed that white and black are both alive, see §8
Empty points surrounded by the live stones of just one player are called “eye points.” Other empty points are called “dame.” Stones which are alive but possess dame are said to be in “seki.” Eye points surrounded by stones that are alive but not in seki are called “territory,” each eye point counting as one point of territory.
According to the Japanese rules the players need to confirm L/D before the scoring can commence (See §9.2 as well as the commentary). Fox assumes that the players would agree that the white group is dead because it is dead. OGS, on the other hand, seems to assume that the players think the group is alive and therefore depicts it as seki, which is unexpected (to me).
I generally agree with this but it becomes a difficult topic when we are talking about ranked matches.
I understand that. I still think we need to be very careful to not give AI hints to players during the OGS scoring phase. I can see your point that beginners need all the help they can get but we do also have to acknowledge that beginners play ranked for MMR. Any kind of hints for the players during a ranked game scenario could be seen as a taint for OGS competitive integrity.
If this is the Fox-style autoscoring, then it appears to be wrong, because it believes that white’s 4 stones are dead while black’s stone is alive. In this position as it stands, white’s 4 stones are alive (alive in seki). The new OGS marking is correct and Fox is wrong here.
Black can collapse the seki by playing out the capture before the end of the game, in which case Black will score 2 more points than leaving the seki on the board. So black usually should do so.
The Fox marking would be fine if White mistakenly agreed that their 4 stones are dead, but white should not agree to this and the AI should not be supplying this as the marking by default.
(edit: @Regenwasser If you’re confused why white is alive in seki here, you might be missing that there is an under-the-stones here. See the japanese rules at Article 1. The game of go, 1. Positions Related to Article 7, Clause 1, Example 1. Traditionally this position was ruled for black as “3 points without capturing” but under modern Japanese rules it instead is a seki if uncaptured, so black should capture before the end of the game and win 2 points instead.)
You are correct. I am wrong and Fox also appears to be wrong here. In my defense the rules do appear to be a little self contradictory there but I will need to reread that section when I’ve got more time.
Do you have a citation for this? I was under the impression that the Japanese rules didn’t change since 1989?
Do you have a citation for this? I was under the impression that the Japanese rules didn’t change since 1989?
The citation is the exact same rules document you just read in the link. It states that the position if scored directly is a seki. It also shows that Black (note: colors flipped compared to the document) playing first would result in net gaining 2 points (or actually 1, depending on a ko). Therefore Black should play and do so.
Black would prefer of course for White to play first, then it’s at least 3 points for Black, even better. But White would just not play, leaving it as a seki, so Black might as well play and accept the 1-2 points.
Torazu sanmoku is Japanese for “Three points without capturing” and refers to the following situation:
The traditional Japanese ruling, attributed to Honinbo Shuwa, was that this corner position was worth 3 points for Black at the end, without his having to capture the marked white stones. However…
… the Japanese ’89 rules make this sekiif it is left on the board at the end of the game. Therefore Black should capture before the game ends (See diagram; may be delayed). The name is misleading under current Japanese rules and the same is true for another similar shape, Torazu Gomoku.
Black has gained 2 points in gote (White may play at a instead of and fight the ko to save 1 point).
So this is a case where scoring (and thus optimal play) under current rules deviates a bit from traditional rules.
Current Japanese rules says seki (both chains in the corner alive), so 0 points for both. This means that it’s better for black to collapse it and play it out before the game stops to get ~2 points. White shouldn’t collapse it and play it out before the game stops, because then black would get ~3 points.
Under traditional rules black gets 3 points as is. This means black should leave the situation on the board, because collapsing it and playing it out before the game stops would lose ~1 point (as black would be getting only ~2 points instead of 3). If OTOH white collapses it and plays it out before the game stops, black would still get at least 3 points, so white should also leave it on the board.
Fox autoscore seems to deviate from both current Japanese rules as well as traditional Japanese rules. That diagram suggests black gets 9(!) points if the situation is left on the board. I suppose that means that white should collapse it and play it out before the game stops to reduce black’s score to ~3 points (instead of 9).
Here’s another set of interesting positions that I think the new OGS seems to handle okay right now but might be useful to consider as test positions if anything is ever changed later. I think these should all work with komi 6.5, Japanese rules.
These are the markings OGS gives, and if scored as is, Black wins by 0.5 (3 white stones captured at G6/G5/G4).
This is an approach ko rather than a direct ko, so Black is NOT obligated to capture B9 before the end of the game, which would result in Black losing one point of territory.
White also has no way to win this game by playing it out because White has no ko threats. If wA8, bPass, wC9, bA9, wPass, bD9, White did force black to play it out, but White had to give away an extra capture to do so, so Black still wins by 0.5.
These are the markings OGS gives, and if scored as is, Black wins by 0.5 (2 black stones captured at C6/D6).
This is an approach ko rather than a direct ko, so Black is NOT obligated to capture B9 before the end of the game, which would result in Black losing one point of territory.
In this position, white DOES have a way to win this game by playing it out. White can initiate the two-step ko, and then use the multiple ko threats in the upper right to force Black to give up something. So White should do so rather than agreeing to end the game here.
These are the markings OGS gives, and if scored as is, Black wins by 0.5.
This is a direct ko, so Black is not supposed to be able to get away with omitting a defense here, even though White has no ko threats and Black has multiple big enough ko threats (starting with H4).
In this position, white DOES have a way to win this game by playing it out. White captures at A4, and then whenever Black spends a ko threat and retakes, White passes. If Black passes, White disputes the game ending, play resumes, and white takes A4 again. Eventually Black runs out of ko threats and has to defend, or else White lives.
Yep, that’s why chose the word “okay” here rather than saying it was perfect. But OGS isn’t obligated to follow J89 logic in positions where play is unfinished. Indeed J89 rules would fight against many of the scoring improvements of the beta OGS algorithm approved of by many people here (e.g. marking a killable territory border as alive where a beginner didn’t see another move was needed) and in many cases favor results that those here would consider worse.
It’s in the finished positions that it’s most important for the scoring to match the stricter rules, and particularly where the side that would be disadvantaged would have no power to force the correct result via play. In all these positions, the correct result is still forceable by play.
We also have a funny thing where it doesn’t require two new passes to go back to scoring after a resumption.
Probably the code looks for two consecutive passes, and so if you get Pass Pass (resume) Pass you’re back in scoring even if the other player would’ve played.
It’s an interesting variation that I think deserves some more thought. I’d be open to exploring that in the future, but it’s out of the scope of this update.
I see. I didn’t really have time to read it in detail until now and got confused cause you said “modern” rules but apparently you were referring to the 1989 rules.
Thanks for linking this but the actual rule I missed was this one:
§7.1 Stones are said to be “alive” if they cannot be captured by the opponent, or if capturing them would enable a new stone to be played that the opponent could not capture. Stones which are not alive are said to be “dead.”
I did not realize that stones are also considered alive if “capturing them would enable a new stone to be played that the opponent could not capture”. Somehow I overlooked that part.
That is the reason for why this board position is a seki. I understand that now.
I realized that too now thanks. Calling the mess of a ruleset from 35 years ago “modern” did have me confused but I guess the game has a long history so it was meant as “relatively modern”.
![[Diagram]](https://senseis.xmp.net/diagrams/2/e36a6263b07f9f70000c447008411940.png)
![[Diagram]](https://senseis.xmp.net/diagrams/41/3ed735a805b7429a2231cc64f24c6841.png)
may be delayed). The name is misleading under current Japanese rules and the same is true for another similar shape, ![[Diagram]](https://senseis.xmp.net/diagrams/19/5625cbbc2c0c35c9075870b49ecf5f1e.png)
and fight the 
