Odd Cases 🤔 in the Japanese Rules

[elsantodel90]

Explanation of the previous problem solution:

Black E12 is the best move. Black loses if they do anything else (well, apart from playing a ko threat now for no reason, and then E12).

White’s best response is N8. Then, since the D7 and H5 kos would normally “cancel out” each other, the only remaining move would normally be black E11 to finish the game. But black loses by half point playing this way.

No matter how many ko threats the players have, the two kos D7 and H5 cancel out, as long as there is a finite number of ko threats. Here, due to the double-ko in the lower-right corner (white is alive and black is dead due to double-ko), black has an infinite number of ko threats.

So, after playing Black E12 - White N8, black now takes the ko at D7, instead of playing the normal E11. Black now has taken the three kos. You can imagine black exclaiming out loud: “I have an infinite number of ko threats, thus, I win THE THREE kos. So, I win by half a point”.

White cannot finish the game and avoid black winning the three kos, because black has infinite ko threats. To avoid losing now, the best white can do is to keep taking the kos, after which black keeps playing a ko-threat in the lower-right and then recapturing. So, the game enters a long cycle and is thus judged as no result.

Note that black’s infinite ko threats are NOT unremovable. White can at any time capture at N2 in the initial position, and black’s ko threats dissapear. However, that costs white one point (because they filled their own territory, as the black stones were already dead) and a whole turn. Since the game is so close, if white tries at any time to eliminate these ko threats, the loss is enough for black to win the resulting game.

You can check that other lines by either player do not work: Black cannot win without using this “let me take the three kos or void the game” trick. And similarly, no matter how White tries to prevent Black from using his infinite ko threats to capture all the kos, White ends up losing enough points that Black wins the game.

Note that this game would be winning for white if Black had any finite number of ko threats, no matter how large. If black had a thousand ko threats, white wins by just playing a lot until black runs out of ko threats.

[elsantodel90]

After further investigation, this determination I did (based on the Jasiek 2003 model) does not agree with actual Japanese rules as the pro’s apply. The new “enabled stones” do not count, and the group is not deemed alive. So the final Japanese result is also the final result of Katago rules in those three positions!!

[yebellz]

For quick reference, let me copy these diagrams here:

Source: How do Japanese rules handle this? • Life In 19x19

[yebellz]

My understanding is that the Jasiek 2003 rules are meant to accurately and formally capture the life and death determination of the official pro 1989 Japanese rules. Jasiek even seems to assert this in the thread linked above:

Apply the Japanese 2003 Rules to determine the correct statuses for the Japanese 1989 Rules.

However, it seems that @elsantodel90 is pointing out that these particular cases illustrate a divergence between Japanese1989 and Jasiek2003 rules:

@elsantodel90

this determination I did (based on the Jasiek 2003 model) does not agree with actual Japanese rules as the pro’s apply.

Is that an accurate summary of the situation?

[jannn]

J2003 has known differences to Japanese rules (see here for example).

For J89, variants of the above examples exist where an extra reinforcement move is actually necessary to avoid problems with double kos in confirmation. But in these particular cases B first needs to capture E18 (before going against the corner), and W can replace that stone later as an “enabled” stone to invalidate B claims.

[elsantodel90]

Yes indeed, that is the claim.

[elsantodel90]

I will add a nice example that lightvector showed me today:

Assume black has infinite ko threats so has the luxury of leaving that ko open. Can he claim the extra point of territory of not playing C9 to defend and kill everything?

This is almost the same as life and death example 9: Article 1. The game of go

It seems very likely, the way the Japanese pros justify their reasoning and just like the above three positions are explained by pros, that this extra liberties do not change the life and dead example 9, and it is still ruled as black needing to defend.

If you apply J2003 however, white stones are dead (not uncapturable, not capturable-1, not capturable-2) but all the black stones are capturable-2 (white wins the capturing race easily if they play first, thus controlling the whole corner, thus black stones are neither uncapturable nor capturable-1. But the local-2 covers the whole corner up to the outer black group, and thus it is easy for black to add extra stones here on the outside, and similar to the F19 stone in the above three examples, J2003 model counts that as “enabled” by the dying black group and thus the group is capturable-2 and thus alive. There is the capturable-3 exception that is meant to cover exactly some other examples very similar to this one, but in this one the ko stone is not capturable-3 and thus the model deems black simply alive). Thus under J2003, black does not need to defend.

I have thought of an alternative formal definition for which stones count as “enabling”, but I have been unable to get a definition to match all of these:

• The seki in life and death example 2
• The 3 points without capture position
• The above three examples I posted before
• This lightvector example

Normal snapbacks and such are easy because you can specify that the player manages to control an intersection right under the group. The essential hard cases why one can’t simply “drop” the “enable new stones” rule of 1989 (or replace it with the simple rule that it enables control of the intersection right under the original stones, that is, local-1) seem to be positions similar to the three points without capture and the seki in life and death example 2.

[elsantodel90]

Sorry to go back to the three unfinished-direct-ko positions again, I just see that I hadn’t paid enough attention before to the details of your explanation, and I am not sure that I follow. You seem to explain why the single white E18 stone is alive, and that is clear in J2003 and Japanese 1989: white passing for the ko can easily retake and play a new permanent stone right at the same place that the original stone was, so E18 is clearly alive as is.

The crux of the matter in these three positions is whether the A19 string of 5 white stones is alive or not, which determines whether white needs to defend or not. The Japanese rules definitely say that position A needs an extra move by white, and that position C does not need such an extra move. And the rules technically also say that position position B requires defense too, although some pros had doubts when consulted informally, probably because under “normal” ko rule with continued play, as in the new Korean rules, position B would not need defense, but it needs defense in Japanese rules that use the hypothetical ko rule.

Now, one could reasonably argue that white’s corner stones should be alive in all positions, due to the “enable” clause: when black goes for the capture of the corner stones, white manages in the process to play a permanent stone at F19, in all three examples. White’s ability to force that stone even if black starts certainly has “at least some relation” to the presence of the five white stones in the corner that will be captured by black: if we replace the white C18 stone with a black stone, for example, then it is completely impossible for white to force control of F19 if black starts, so in this sense that stone is “enabled by the capture”. However, the Japanese rule is that white needs to defend in position A, so that new local stone does not count as enabled.

There seems to be a firm unwritten “principle” that trumps over any technicality of the “enables” rule: the principle that a direct ko that is unresolved by itself in normal play, can never be left without defense at the end of the game. So it seems that "must always resolve direct ko’s " trumps the “enables new stones” rule, as far as Japanese practice is concerned.

[jannn]

I don’t think F19 is particularly promising for enable. What I wrote is that E18 itself is the enabled point. B cannot do anything to the corner without taking E18 first. And when W replaces it later, it will be an enabled stone for obvious reason: B could have prevented the new uncapturable stone at E18 simply by not capturing the old stone from there. It’s a kind of snapback for the whole corner action which depends on taking E18 first.

So B cannot capture the corner without enabling W to play a new uncapturable stone at E18.

[yebellz]

@jannn and @elsantodel90, could you please clarify which specific figure/position you two are talking about?

I believe it is in reference to these three diagrams: Odd Cases 🤔 in the Japanese Rules - #64 by yebellz

Does it matter which one in particular?

[jannn]

@yebellz

Does it matter which one in particular?

I think the above logic (which was not my idea originally) may work for all three.

[elsantodel90]

@yebellz

I believe it is in reference to these three diagrams: Odd Cases in the Japanese Rules - #64 by yebellz

Yes, those three.

@jannn

So B cannot capture the corner without enabling W to play a new uncapturable stone at E18.

So you mean that all three are alive, and white thus does not need to add an extra defensive stone, because of the “enable rule” and enabling E18? I am sorry, but that is not the way Japanese rules these positions, that is what J2003 would say instead, which is the whole reason why I brought the positions up again, to showcase this difference.

In Japan, position A and B actually require defense, and only position C does not. That is my whole point: some at least reasonable interpretations of the enable rule (like the J2003 one or whichever one you are using in your post) would deem white alive without need to defend here (because of the “enables” rule), but Japanese enable rule “does not count” here (for whatever reason). I asked some Japanese friends and also Antti Tormanen, who had previously consulted another pro who knows the technicalities of rules better, and they all agreed that the direct ko just cannot be left unresolved because white would be captured, “enabled” stones do not matter here (the discussion was informal though and by no means an official statement).

What and when exactly something counts as “enabled by the capture” is very tricky, as I say. You can’t just take it to mean “control of any new intersection with a stone, that you did not control before” for example, as then any one sided dame on the board would make all your groups alive. It somehow has to be directly related to the stones that are being captured.

But also, it seems that it is a pretty established thing that “you cannot use the enabling new stones clause to justify leaving a direct ko unresolved”, there are many precedents of that, like these three positions for example, or the modified “life and death example 9”, or some other simpler ones where the ko stone ends up being capturable-2 in J2003, which is the reason why the extra special case of capturable-3 is added there, but as I say that case fails to cover these three positions or the modified “life and death example 9”.

[jannn]

I doubt you will find a logical interpretation for enable besides the obvious one: something played in course of the capture, which wasn’t possible (could have been prevented) without it. A new W stone at E18 seem to qualify (B could have prevented it originally, but cannot prevent it if he goes for the corner capture - this is not like onesided dame).

I also thought for some time that W defense is necessary in some of the 3 cases, and it looked like the most serious manifestation of the known old double ko reinforcement problem. But now I think this was an oversight - maybe your friends made this same oversight as posters on L19 when lightvector asked originally (I doubt these are direct kos btw).

[elsantodel90]

@jannn

B could have prevented it originally

What do you mean by that? It is easy for white to unconditionally force a permanent stone in E18: If black starts by taking the ko, white passes, and now black cannot fill the ko: filling the ko leads to all of the black stones being captured.

@jannn

I doubt these are direct kos btw

Position A is definitely a large direct ko for the life of the corner white 5 stones group, during the main game under normal play. If white adds a move winning the direct ko, the position is completely resolved. If black wins the direct ko, black plays D19 and the temperature drops sharply as the stones are now captured. Yose moves remains: white retaking E18 is sente-ko forcing black to actually capture, and then only much smaller yose moves remain. So the main ko is a direct ko by play, with much smaller yose continuations then in case that black wins the main direct ko.

@jannn

But now I think this was an oversight - maybe your friends made this same oversight as posters on L19 when lightvector asked originally (I doubt these are direct kos btw).

Well, until we can point to an actual precedent of a referee ruling a position like this in an actual professional game, it is really hard to be sure. I think though that the informal opinion of two Japanese profesional players, even though they might have made a mistake and we cannot be 100% sure, is the best I have right now.

[hexahedron]

That interpretation of enable is a little weird, morally speaking. The normal kinds of cases that ths rule handles are for example like:

• Snapbacks - where for example nominally some white stone may be captured, but actually if you read further, the white side regains everything in return, including everything black was attempting to gain in the first place.
• This seki, where nominally black can capture two stones, but actually if you read further, white gains something in return. White gains territory that genuinely beyond what they would get if white was merely ruled alive, if black tries to kill white.
  

In both cases, informally, when white claims they don’t need to add a move, they can genuinely defend 100% of the territory and 100% of the survival of stones that they claim… OR in the second case they genuinely kill something and/or gain new territory in return that they would not otherwise be awarded.

In the position being discussed:

under @jannn’s interpretation of “enable” which is indeed a reasonable possible English-language reading (no idea if that matches the Japanese language reading, or if it matches what pros would say when asked directly), a white stone at E18 qualifies because if black captures the stone and white later establishes a new stone there, that stone is not the same as the original, and it was not possible without the original stone being removed. White cannot play a new stone there without the help of black to remove the original stone, so it is “enabled”.

But it’s a bit weird, because here we’re talking about a case where white cannot defend 100% of the territory that they would be awarded if black was ruled all dead and white was ruled all alive, nor can white gain anything in return relative to what they would be awarded under that ruling. It is a pure gain for black - so it doesn’t seem to fit the same spirit as the “normal” cases for this rule.

We can make the example even simpler like this:

If white has sufficiently many ko threats, can white avoid finishing this ko to gain an additional point? Under jannn’s interpretation, I believe the answer is yes, because if black tries to proceed during the hypothetical play, it involves capturing A2, and then white can re-establish a new stone on A2, which white cannot do if black doesn’t remove the old stone on A2 first.

But again it does not quite match the “normal” cases that the rule is intended to handle, because if adjudicate that black is all dead and white is all alive, then white is awarded this entire corner… but without adding a move white cannot defend that award in hypothetical play, nor can white in return gain anything at all that black would be awarded.

How would Japanese pros rule on this position?

[hexahedron]

Hmm let me edit that position a little to eliminate a bit of under-the-stones that results if black captures the snapback right away.
Edit: done

[elsantodel90]

Very well put hexahedron. Indeed, I think Japanese players more or less reason along those lines, which are of course informal and intuitive.

@hexahedron

White gains territory that genuinely beyond what they would get if white was merely ruled alive, if black tries to kill white.

This kind of reasoning helps to understand why matching Japanese intuition for what exactly is “enable” is so hard to define precisely: the “if white was merely ruled alive” part considers what intersections white would control, but that ALSO depends on which of the other groups we consider alive, not just this one, sort of “breaking” the kind of completely independent alive / dead reasoning group per group one at a time that is typical of the Japanese rules. I can’t speak of professionals, but I have seen 5dan players reason in exactly those terms before for similar positions: “well, IF we ruled white alive, then so and so would happen” [which implicitly assumes the status of the OTHER groups, which luckily in the particular position are not under doubt too! ]

[jannn]

@hexahedron

Snapbacks - where for example nominally some white stone may be captured, but actually if you read further, the white side regains everything in return

This is what matters here imo, the 3 (or at least the last two) examples seem a partial snapback. B cannot capture any stones in a way that at least some of them won’t get reestablished. And torazu3 shows a partial snapback IS considered a snapback. Captures need to be 100% clean, and without any compensation, to count as valid.

@hexahedron

We can make the example even simpler like this:
…
If white has sufficiently many ko threats, can white avoid finishing this ko to gain an additional point? Under jannn’s interpretation, I believe the answer is yes

Nice examle. But imo W can never leave this unfinished, regardless of threats, since B has infinite threats (passes and resumptions). Hypothetical play doesn’t matter here, B will resume as many times as necessary to force W reinforcement. This is also generally why unresolved kos can be prevented in modern rules.

[elsantodel90]

@jannn

This is what matters here imo, the 3 examples seem a partial snapback. B cannot capture any stones in a way that at least some of them won’t get reestablished.

Your phrase “any stones” (I assume you mean any GROUP?) makes me think that you are analyzing the life of many white groups at the same time. The Japanese rules do not seem to work that way: all the precedents explanations show that the analysis is carried out in a completely independent way, group by group. The analysis explains why each group is alive/dead in turn.

In a snapback, the new permanent stones appear right BELOW the original group. Those cases I think are 100% clear: nobody disputes those, if you can force a permanent stone in one of the intersections of the original group (what J2003 calls local-1), your group is definitely alive. That is indeed the case for the E18 stone in position A: white can easily re-establish a new permanent stone where the old one was, like normal snapback.

But that only proves that the E18 single-stone group is alive. The A19 A18 B18 C18 C19 five white stones group is analysed in a separate step.

When analyzing the A19 A18 B18 B19 C18 C19 five white stones group, white cannot force control of even a single one of THOSE intersections, so it is not even a partial snapback for that group. The only thing that could save the independent analysis of life for those five stones is the enabling of control of some new intersection somewhere else, as a result of the capture. Here is where one must interpret whether any of E18 / F19 stones “count” or not as “enabled by the capture” for this rule. So far, no matter whether that is logical or not, and even though J2003 claims that it counts and the group is alive as is, when consulting actual players I have seen much more consensus that it does not count and cannot be claimed, including the informal case of two Japanese professionals. I think we can’t get a more “definite” resolution than this unless we find a precedent of an actual ruling by an actual referee in a real match, unfortunately. But my understanding (from the need of the capturable-3 distinction in J2003, which is an ugly extra case that is there just to “fix” some similar shapes pretty similar to this one, where the ko stone would be considered alive by the enabling rule but the Japanese tradition says that you CANNOT leave a direct ko open) is that there are at least “similar” positions for which precedents exist.

@jannn

Nice examle. But imo W can never leave this unfinished, regardless of threats, since B has infinite threats (passes and resumptions). Hypothetical play doesn’t matter here, B will resume as many times as necessary to force W reinforcement. This is also generally why unresolved kos can be prevented in modern rules.

Your argument of “just use resumptions to force eventual ko defense” can be almost always used to force the defense of any direct ko in practice (at leasts if the players are well-versed in rules technicalities and resumptions), and also applies to the original “position A” to show that even if your interpretation is the J2003 or any other that says that you do not need to defend, you will almost always need to defend anyway because the opponent can keep taking the direct ko after each resumption making you consume more and more ko threats until you have no more ko threats and you are forced to lose it, which is worse than just losing one territory by defending, so you should instead defend.

But still, it does not completely solve the situation for us rule enthusiasts: there is the posibility of the player defending the ko to actually have INFINITE ko threats (double ko death or double ko seki both work for this, one to unconditionally get the point and the other to create a no-result if the opponent does not yield the point). Then, even after any number of resumptions, the opponent can never force the player to defend the ko. So it matters whether the rule says that the player must defend or not in order to be considered alive in the final position, as if the rules say that defense is necessary, then the extra point won’t be claimed even in the presence of infinite ko threats.

For all the positions that we mentioned, there seems to be many (including some pros) who agree that, no matter what the enabling rule says, you can’t leave any of these direct kos without defense, no matter ko threats anywhere else on the board.

[jannn]

@elsantodel90

Your phrase “any stones” (I assume you mean any GROUP?) makes me think that you are analyzing the life of many white groups at the same time.

No, I mean string by string. But under J89 (as written) B needs to show a sequence where he does not capture and then allow reestablishment of any W stone, for the claim that he enabled no new stones with that sequence (cf J89 example 5).

For infinite threats, a double ko seki could make the last example a triple ko repetition. But adding a double ko death providing threats for W only may be interesting indeed.