Yeah, I agree it's not an exact match. I'm a little sad though to implement fourth/fifth/etc ko rule to select between. It seems to me that in practice, Leela Zero already works usually fine for Chinese rules despite having several differences, one gets even closer by forbidding suicide and using Japanese simple ko. At that point, it's not really in a player's interest to try to waste time in a sending-two-returning-one in the first place - the main reason you'd ever try it was only if it somehow kept the game from terminating.

Maybe it's worth adding, still? Also, is sending-two-returning-one actually the *only* 3-move cycle? My understanding was that Chinese rules prohibit sending-two-returning-one cycles, not 3-move cycles, in which case it matters if they are the only one.



So we have "current" rule: only unblock opponent non-ko-moves after your move.
And "changed" rule: unblock all player's non-ko-moves after your move.

Under the changed rule, if you notice in your example, unblocks black's ability to capture because for a brief moment, everything of white is in atari. So black is actually the one winning because black can continually recapture everything without spending turns to unblock, whereas white cannot.

Under the current rule, your example seems to make sense, but I think black makes an error. Black should not capture at , instead black should unblock taking . Then after , black takes and becomes safe. Black can then just cycle forever using the 3 ko mouths, always one step ahead of white, always refusing to take the second-stage ko.

So there are four ways this can play out - black first current rule, white first current rule, black first changed rule, white first changed rule. I think this gives:

Current rule: black first is safe but white can force no-result if white is losing and black cannot give up the group and still win. White first kills black.
Changed rule: black first kills white, white first kills black.

Do you agree with this? If this is correct, then there is some difference, and is not clear how to handle it, but if anything, the changed rule is a bit more symmetric here.

Usually, I think we will not get to cleanup phase with this situation in the first place though. In normal play, either player can force a no result by insistently capturing the ko mouths over and over if their opponent is unwilling to sacrifice their whole group, so whichever player would lose, if they are losing by less than the value of the whole fight, can do so. I think this is generally true of situations where two groups share tons of ko mouths and have no other liberties, so I'm less worried if this kind of situation differs between this formalization and "true" Japanese rules. Actually, I don't really know how "true" Japanese rules would handle this situation either if somehow the game ended during main play without either of the players having attempted to force the no result.

What do you think?

A player cannot force no result in regular play.

Click Here To Show Diagram Code

[go]$$cB
$$ ----------------------
$$ | X X X X X X X O . . .
$$ | X W X . X . X O O . .
$$ | O . O X O X O X O . .
$$ | . O O O O O O X O . .
$$ | O O X X X X X X . . .
$$ | X X X . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

Assume whte has played the circled move. Black has no way to continue the cycle without making a a move elsewhere. Then he can make another cycle, but he needs aother move elsewhere to continue. If he passes white could also pass and finish the regular play. I think it would be seki under Japanese rules.

I will return to your other points later.

Ah, you're right. So then this position becomes extremely complex. Because, under "changed" rules for example, white does not *want* to finish play, because then black going first during cleanup would kill white. So we get some passfight-like dynamics, and it's not clear to me what happens. Hmmm...



Would it be really? The Japanese rules as written seem too ambiguous to work this out. They too specify a "pass for ko" rule, but it's unclear how that applies in staged kos, when one of the kos dissolves and then reforms And none of their commentary covers a case like this one. I wouldn't be surprised if it was declared to be a seki by "fiat" though, regardless of what the rules as they've attempted to write down would actually imply.

I wonder if there's a way to make this consistently evaluate out to a seki during cleanup without affecting other cases too badly. It's far from clear to me.

I'm definitely looking for help and guidance here. One way or another, there actually need to be some final rules that I do end up implementing in KataGo. There are going to be some inevitable differences. Three points without capturing is already a known difference where I cannot match the official Japanese rules that I'm just going to tolerate and accept, for example, but minimizing them where possible would be great.

My interpretation of the 1989 Japanese rules is, that a player must make a pass to recapture a stone in a ko shape for each stage of the ko separately. With this interpretation the position would be seki.

Under that interpretation, would that mean that https://senseis.xmp.net/?MolassesKo would be a win for whichever player to play first in a hypothetical sequence at the end of the game, and therefore not a seki? Maybe something like "both players are dead" since either hypothetically playing first kills the other, if this position is left on the board at game end? Because if you have to pass for ko in both of the stages, you can't keep up with the opponent.

If so, then while the Japanese rules could still just say "both sides are dead, so we don't score this", this gives some troublesome cases for computer playout, since in computer cleanup if actually trying to play out to determine cleanup, the position is actually unstable depending on who plays next.

The earlier case you presented is really quite troublesome too.

I'm curious - do you have thoughts as to if there is a simple modification to the rules that would be easy to apply/check that would collectively handle these cases "nicely"?

I think that this is seki under Japanese rules, but I agree that the Japanese rules are ambiguous. Since they are, you have the leeway to interpret them as you see fit.

BTW, I have shown that this is a seki under the komaster approach. moha's questions have helped me clarify and improve the approach. There is no pass for ko rule. No ko ban can be broken in hypothetical play to determine the fate of a single ko. Then we may resolve that ko and start over with a new one.



You are getting the traditional result, aren't you? I regard that as a virtue. But it should have hardly any effect upon training, and in play humans can easily adapt, just as they did until 1989.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

I think you're right, but I'm not an expert. Can someone confirm this? In any case, I think it would be nice to have proper Chinese rules available, especially since they are so easy to implement.


A 3-play cycle must include 2 captures of one color and 1 of the other. There is no way to achieve this other than sending-two-returning-one.

Playing around with it - how about this? It's certainly not a normal sending-two-returning-one (the third move isn't even a capture!), but it is a sequence of moves that does the 3-cycle repeat. I should or shouldn't be prohibiting this one too?

Click Here To Show Diagram Code

[go]$$Bc
$$ ----------------------
$$ | . . X O . X O . . . .
$$ | . . X X O O O . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc
$$ ----------------------
$$ | . . X O 1 X O . . . .
$$ | . . X X O O O . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc
$$ ----------------------
$$ | . . X 2 X X O . . . .
$$ | . . X X O O O . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc
$$ ----------------------
$$ | . . X O . 3 O . . . .
$$ | . . X X O O O . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

I'm not sure if there are other cases to also be found. Categorizing/proving all the cases seems a little fiddly.

Thankfully, nobody in real life probably wants to play this sequence, so maybe it doesn't matter if this one also gets prohibited too.
(Now watch as someone now posts a rules beast where doing this sort of sequence is correct...).

This is why I say you can drive yourself nuts worrying about how to score non-scorable positions. Non-scorable in the sense of having a temperature greater than -1. This is a gote that gains ½ pt. If Black makes a silly mistake and plays , should simply capture. Or play somewhere else and hope that Black repeats the procedure.

By the J89 rules this is pretty plainly both sides lose. Don't worry, be happy.

----

Note: Sometimes you can end play by Japanese rules with a position that has a temperature greater than -1, because of a ko ban. Both the J49 rules and the J89 rules force the komaster to win the ko at temperature 0, in effect, by different means. Neither does so without criticism.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Bill, that one wasn't about scoring a position or Japanese rules. That was about maybe adding "Chinese superko", which would apply at *all* times during normal play, scorable or not, to implement *Chinese* rules.

The way I see it, if the ko rule forbids repeating the position that existed one play ago, the Chinese superko rule forbids repeating the position that existed two plays ago, and I assume the player to move is not included in the idea of position (since the rule used to be misinterpreted in the West as describing positional superko).

In other words, in your diagram would be allowed, as it doesn't repeat any position. If White then passes (a pass is not a play), to the left of would be forbidden, as it repeats the position after .

I'm not sure where this idea comes from, but is almost certainly wrong. Chinese ko rules forbid a player to prevent game end with any cycle where the stones lost is not symmetrical, regardless of the length of that cycle. Consider a sending-two-returning-one with a double ko seki, for example.

One common approach in computer go is that cycles (that are not forbidden by the normal ko rule) end the game with a draw, loss or win depending on the stones lost in them. Which is not perfect but may be considered sufficient.

OIC. As Emily Litella says, Nevermind.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

That's not my understanding, see Chinese superko at Sensei's Library.

That paragraph is even rougher than the official texts, which show the simplest case and states a clear intent to generalize. It also works the other way: superko is mentioned, forbidding all repetition, then triple ko and a few other valid cycles are cited as possible exceptions with draw result. No referee would declare draw for an unbalanced cycle.

Yes. The official text certainly seems to be a positional superko rule. In particular there is no mention of sending two returning one, or the like.

　　第六条 禁止全局同形再现

Aside from eyeballing this I ran it through various online translation sites. Google gave this in clear, but technical English.

Prohibition of global isomorphic reproduction.

Several years ago I saw a version that mentioned that a referee might declare no result, despite the text. It may have been back in the 90s, though. I would not trust anyone who said that the intention of the rule was different from the plain text, when it is so short and clear. It may be that top Chinese players do not like the positional superko rule, and so referees defer to them.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Maybe we could call this Taking one, losing three.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

The widely circulating old English version does show it as an example for this particular issue, and its phrasing gives useful hints about the real intent (just like where triple ko and similar cycles are mentioned).

I think the text simply has a hard time explaining actual practice. These positions are rare but still frequent enough to have a decent number of precedents in pro games, maybe even dating back before written rules.

Look again. It *does* repeat the position at the start.

Right, I think what's going on is that like the Japanese players, the "official text" is not actually the rules that are used in reality. They write the text, and then ignore it, and one must look to what the actual practice is to try to guess the "real" rules.