Territory Scoring / Japanese Rules (Counting)

General Go Discussion
1

Wait a minute!! I’ve been playing for over 3 months and I got the point system wrong?? I was trying to count points in simple situations, like 1 eye or simple life/death scenarios and getting some inconsistencies where I thought I should get the same number of points. There’s an order to counting the dead stones and territory: dead stones first, then territory later, is that correct? This means that indirectly, spaces with dead stones count as 2 points. I was always thinking the opposite, counting the territory first and then removing the dead stones from the board.

Here are some things I found on reddit:

you are incorrect to count spaces occupied by prisoners as 1 point. By japanese rules, such space will give you 2 points: 1 for territory and 1 for the prisoner."

and the most surprising one:

In area scoring the score is territory plus living stones, so there is no penalty to playing inside your own territory. So if there are any problems with deciding what is alive or dead you can just keep playing on until things are absolutely dead.

In territory scoring (like Japanese rules) you cannot do this because as you have observed the score changes. To fix this issue the Japanese rule book has a very complicated system for deciding how to judge whether groups are alive or dead. A basic summary is you say the game is over and then you ask your opponent do you agree this is dead. If they don’t agree you hypothetically play it out asking would it live or die if both sides play perfectly, and then once you have an answer you reset the board to before you hypothetically played it out."

I had no idea about this pretend make-believe phase. I thought that after a pass-pass sequence, the game would continue (if there is disagreement) and points would change accordingly (OGS doesn’t work like this?). And also in the middle of the game, anyone can play anywhere and the points would change too in life/death situations.

Is there an easy way to think about this counting situation? Some video with simple examples or a book? I think this counting/scoring phase is very neglected and should be one of the very first things to the taught. When I taught a friend in a real board, we agreed about the dead stones in the end of the game and then filled the territory with the dead stones. (Ok, I think this “double-counting” of the dead stones can be irrelevant because in the scoring phase we subtract both points). But if we had disagreed about the dead stones, I would say “let’s continue” and ended up teaching him incorrectly?

1 Like
2

Under Japanese rules there is indeed a “pretend make-believe phase”, as you put it. Unfortunately this phase is not properly implemented on OGS (nor on any other online server that I know of).

Frankly I feel this kind of convoluted rules is a key weakness of Japanese rules; thanksfully, such disagreements are rare once you’re past a beginner stage.

5 Likes
3

Note that there is only no penalty after all other worthwhile moves have been played (i.e., when the game is strategically over, and players can just pass). Otherwise, during the middle of the game, needlessly playing inside of your own territory usually loses points under any of the common Go rules.

In nearly all situations, the ideal move is the same for either Chinese or Japanese rules. There are convoluted and rare situations where they differ dramatically, but I wouldn’t worry too much about those for now, if you are still working on learning the basics of the rules and strategy.

Under both Chinese and Japanese rules, these parts of the end game procedure are (broadly speaking) the same:

  1. Both players pass when they believe they are no more strategically worthwhile moves to be made.
  2. They discuss which stones are dead (and should be removed from the board).
  3. If they are in agreement, they proceed to tallying up the score.

In most situations, the life/death status is the same under Chinese and Japanese rules. However, there are some rare cases where they are different. In such cases, the Japanese rules do have a rather complex set of rules to determine life and death status that considers by evaluating what would happen under perfect hypothetical play, while also applying special ko rules and other traditional conventions. Disputes for these rare situations under Japanese rules have to be handled by applying a rather complicated set of rules. The game should not be simply resumed to settle life and death disputes under Japanese rules, despite many people doing so. It’s simply not possible to algorithmically support all of the complexity of resolving life and death under the Japanese rules, so on OGS, you just need to work things out through discussion with your opponent. Often, when dealing with the rare edge cases, players might misunderstand how to properly apply some rules and may be unaware of such things. Even as a somewhat experienced player with an above average interest in the technical details of the rules, I’m not 100% confident that I could even properly resolve every complicated situation / potential dispute in the Japanese rules.

Under the Chinese (and other area scoring) rules, if you have dispute about life and death during scoring, you resume the game, and each player must prove that each disputed group is dead by playing on to actually capture it.

If the life and death status is the same, the scoring under Japanese and Chinese rules is almost the same:

  1. Under Japanese rules, each player’s score is their territory plus stones captured from the opponent.
  2. Under Chinese rules, each player’s score is their territory plus their own living stones that remain on the board.

Note: the scoring for both rules are done after removing the dead stones on the board.

In most games (if one player does not pass more than the other), the players will either play an equal number of stones on the board (if Black passes first), or Black will play one more stone than White (if White passes first). Thus, counting the captures versus counting living stones are often complementary (or just off by one).

Note that when continuing a game to prove that something is dead, one player might start passing while the other continues to play stones, so this almost equivalence no longer holds.

5 Likes
4

Note: the scoring for both rules are done after removing the dead stones on the board.

I believe this has been a gross mistake of mine (which is what I said in my first paragraph). So, I should say that so far it can be seen that white has 17 points here? (8 dead stones + 9 territory points).

Shape 1
Shape 1215×203 43.9 KB

Or equivalently if it’s the end of the game, it could be counted as their territory minus their own captured stones, since what matter is the subtraction of points. But always dealing with the dead stones first.

Note that when continuing a game to prove that something is dead, one player might start passing while the other continues to play stones, so this almost equivalence no longer holds.

Yes, I was thinking about this and also situations when someone can exploit a difference of 1 point. I’ll try to think about some simple example.

1 Like
5

There are unreasonable cases and there are legit differences between rules and cases in between. One unreasonable case is your above example, black insists the 8 stone is alive and just passes, and asks white to prove they are dead, and when white captures black playing one extra stone to capture them, the territory reduces to 8. This is one of the “exploits” some people with bad faith playing online would do (or in reverse mark your alive groups as dead, because they claim they don’t want to play inside to reduce their score). In these unreasonable cases, normally you can call for moderators to resolve it (or simply blacklist these opponents and never play games with them in the future)

More subtle cases like the bend 4, I’ve run into opponents who are used to the Chinese rules, and claim the more complicated version of bend 5 that can be reduced to bend 4, and had filled all the dames, and refused to play the ko, insisted it is a seki, and asked me to fill all the liberties to prove it dead. And to prove it I had to count all the ko threats and at least fill the outside liberties. the act alone would reduce points. Under Japanese rules, there are special cases like this where the hypothetical plays would come in, declare them as dead, and return to the final state to count the score.

And finally, there are legit differences not related to the “live and death” or involve hypothetical play but change the scores where one side can play more stones at the end result in area scoring has more scores for one side, is the winning the last ko as the komaster and as a result get to play one more stone on the board where the other had no threat left but pass. There are also situations, like in actual seki, the Japanese rules don’t count false eyes or true eye space in seki as points, but the Chinese rules do. If the seki is very huge, this issue would get out of control with way more than 1 point difference.

3 Likes
6

You maybe better start from the area (Chinese like) rules which may be less elegant but which is very clear: play as you want until we agree that we fully decide what is yours and what is mine (stones and emptyness together).

Now you can start thinking on the japanese one: if we manage to have the same number of stones for black vs white on the board, we can afford to not count them and just count the emptyness (territory scoring). So prisoners will go back on board, reducing the points.

Maybe this could make the rules sets a bit more clear for you?

4 Likes
7

Practically speaking, area scoring is what people should be using online because Japanese/Korean scoring is never properly implemented. Even Go AI have to approximate it because of how complex the full rules are. But if you want to delve into the nuances of the rules differences, Robert Jasiek’s website has extensive commentaries and examples. (They are well written enough that I was able to understand them before I was actually playing the game and only had an interest in the AI and combinatorial game theory aspects of the problem.)

His “New Amateur-Japanese Rules” are probably the best way to think about territory rules conceptually because they capture the essence of what makes territory rules different: using hypothetical play to assign a meaningful score to the position in which the players passed. (Whereas under area rules, if players disagree, play just resumes and the score will be based on the new final position.) Those model rules only give different results from the professional rules in positions that happen once or twice in a player’s lifetime and are probably what should be used for online territory scoring rules instead of the mess we have now.

To summarize, you do two play-outs from the final game position, one with black going first and one with white going first. In the play-out where black goes first, black tries to limit the area that white controls at the end of the play-out. The final white-controlled space is then overlaid onto the actual final position to determine which intersections score for white. Vice-versa to determine what intersections score for black when white goes first in the play-out.

IRL, those rules are less practical because they require making a copy of the game onto two extra boards in order to resolve a scoring dispute. Making an IRL resolution of hypothetical play more practical is where all the complexity of the real professional rules comes in.

Pragmatically, area vs territory almost never changes the outcome of a game or even the correct move that should be played, and there are only two differences that professional players seem to actually worry about: 1) that territory scoring does not care about whether the game ends with an even or odd number of dame and 2) that certain very late end-game ko positions are worth playing out under area but not territory making late end-game calculation is slightly “harder” under area.

As close as pro games tend to be, you’d think that there would be a lot of examples of games where these differences matter, but I haven’t ever seen statistics to back up the worries. It seems to be a corner-case even at that level of play. (Though, there is at least one international professional game where a Japanese or Korean player messed this part up and lost a game by passing when he shouldn’t have.)

For non-pros, we should all aspire to be good enough where our games are consistently decided by the half-point margin where any of these concerns would even come into play.

5 Likes
8

This is a common misunderstanding of their relation. The scores usually agree - up to a point of difference. But that one point (and the komi difference that stems from it) can and do reverse a lot of close games. In particular, B+0.5 territory (with komi 6.5) normally becomes W+0.5 area (with komi 7.5).

This is less important for beginners, but its effects are known for stronger players, and around 2020 computer go tournaments stopped using normal area scoring and switched to a button go variant (dynamic komi), citing statistics that otherwise W wins 2/3 games.

3 Likes
9

I was unaware that the dame parity issue was such a big problem at a very high level. Is this born out in professional games? Or is it only an issue in computer competition? Either way, fixing it is fairly trivial, as those rules attest.

Tangentially, I seem to remember that there was some issue with some rare end-game positions under a similar provision in the 2008 world mind sports rules that made them add a rule about not passing when certain ko are on the board (or something to that effect).

Is that not an issue with this variant? Or is it just that it’s rare enough that people don’t care?

To be clear, I did specifically mention the dame parity issue in my post as one of the main differences people cared about. But, as I said there, I haven’t seen any studies or evidence that this matters for human played games, even at the pro level. By, e.g., comparing professional games played under the two types of rules and showing a disparity that would be fixed by a correction. Or even showing the percent of games where the correction would even come into play.

If you know of a source that does explore this data, I would be very interested in seeing it.

There is a proof that the two are equivalent for non-handicap games if there are no seki with eyes (and no player played after a pass). Specifically, to quote sensei’s:

“If White made the last move, territory scoring and area scoring give the same result. Otherwise there is a difference of one point (to Black’s advantage under area scoring and to White’s under territory scoring).”

The parity correction you get with the kind of adjustment used in the computer tournament you cited gives white the extra point of komi only if black made the final move. (Though the rules are stated in terms of the first pass.)

10

Between the Chinese rules and Japanese rules, under a normal situation (no seki, or pass and get one extra as komaster), where there is a difference before komi, it is always black that gets one more point under area scoring.

If the area score difference (before komi) is 7, it can only be a territory score of 6 (different where black plays the last move) or 7 (the same where white plays the last move). In this situation with 6.5 komi under territory rules, the first case white wins, but the second case black wins, however, under the Chinese rules with 7.5 komi, both cases white wins (hence when white plays the last move, the winner is different). If we add seki, the situation gets complicated since it is now possible to have an odd number of shared seki points, as well as seki eyes counted in one rule, but not the other. In those cases, the difference can add up to more than one point difference.

Then we need to count how many games using the Chinese rules ended up with W+0.5 (which can be W+0.5 in the Japanese rules or B+0.5 in the Japanese rules), and how many games ended up with B+0.5 in Japanese rules (which almost always be W+0.5 in the Chinese rules, only on rare occasions we see even difference with seki or other situations). I think the rate of winning by half a point as black under the Japanese rules is about 2.9% or something, And under the Chinese rules, the game ended in W+0.5 maybe slightly higher say 3%, but maybe about half of them give a different result say 1.5% (white played the last move). Assume games played in Japanese rules (and Korean rules, they basically give the same result) and the Chinese rules are about the same amount, the average rate of seeing a game that would give a different winner would be (2.9+1.5)/2 = 2.2%, which is about 1 in 50 or so. There are some 5000+ pro games played each year (those got recorded, the total is probably close to 10000 if we include all preliminaries), so the number of games that ended up with different winners at pro levels using different rules are at the order of 100 to 200 games a year.

I’ve manually scanned through the records on https://kifudepot.net/ for 500 games, and I found exactly 10 games that ended with different winners using different rules, so this estimation is pretty accurate I think (I didn’t even look for more rare cases like with seki or get last move being a komaster that need to dig deeper with larger winning margin, which definitely brings the ratio higher). Regardless, hundreds of games a year with different winners is definitely not something we can simply just say it doesn’t matter.

2 Likes
11

I think this scales with playing strength. Area komi 7.5 is likely fine for amateurs. For pros, I have heard anecdotal stories about top players like Ke Jie for streaks of dozens of (area) games where W won in all but 1-2 games, but this is likely exaggeration or partly a psychological effect, considering the 2/3 winrate reported for the (even stronger) computers.

But since area komi 7.5 usually gives the same winner as territory 7.5 would (this is why it was set, between an odd value and the next even value, to avoid granularity issues and error masking), it is not surprising that it favors W - as territory 6.5 is balanced.

I think Chinese pros simply learned to live with W advantage - just like chess pros do.

I consider button go somewhat inferior to the Taiwan Rule, but this problem is much rarer than the parity bias itself. I also heard some vague mentions that some people were unsatisfied by how some late kos play out under button/taiwan (probably in different examples than the one on sensei’s), but would also be interested if you could remember more or recall some links on this.

1 Like
12

The overall winrate for white using the Chinese rules at 7.5 komi is not that high of a difference, in the 1980s when Ing’s rule first used it, it was like 52%, and many decades later another survey I remember from 2000 to 2016 was still like 52.4%. It doesn’t change that much for pros (and using 5.5, it was like the reverse, black gets 52+% something, and only like 3% of the game ended with W+0.5).

I actually didn’t remember many details about the old rules we used to have. I know that there used to be black 184 wins (and removing last black move would make it match the result of the territory score of 6 points before komi) and 184.5 wins (also deal with seki and seki eyes). It can get a bit complicated to define the last meaningful play to remove it. It has more to do with matching the territory scoring results, even if it is area scoring.

1 Like
13

I think individually there are players who play better with white than black? Though that may be more of a psychological thing and I don’t have the exact numbers

2 Likes
14

KataGo developer (@hexahedron on this forum, but lightvector on L19) posted this on the LifeIn19x19 forum about studying komi with KataGo: AI opinions on "fairest" komi by board size • Life In 19x19

It seems that Chinese rules with 7.5 komi gives roughly a 60% win-rate for White, but 7.0 komi is roughly balanced.

1 Like
15

I think FineArt gives almost a 70% win-rate for white with 7.5 komi? Would be interesting to see the same analysis by FineArt though that may be difficult

16

Yes, if you formulate the rule that way. But to me it seems better to simply say that B pays an extra point of komi if he played more stones than W. This is a logical extension of the compensation points that B pays in handicap games for his handicap stones under most area rulesets.

1 Like
17

I think the winrate also gradually increases as computers get stronger each year.

Area komi 7 is balanced - but at what cost? It treats both B+6 and B+7 (territory) as the same performance. Just compare these two: territory opening vs area opening - no wonder computer tournaments switched to button instead of simple komi 7. This dulling is also why the parity issues are not just about matching territory results.

2 Likes
18

The phrases we used are 黑先黑後 and 黑先白後. The issue is that if we just assume black and white would get the same amount of stones, It is technically possible for one side to keep passing under the territory scoring, while the other side keeps filling the dame, and they wouldn’t change the scores with territory rules. However, with the same kifu put to the Taiwanese rules which are area scoring in nature, it would give the side to keep playing more scores, it doesn’t always just favor black. The issue is to keep the number of stones balanced, while at the same time identifying if these extra stones from either side are “meaningful” under the territory scoring. And at the same time, we still need to compare them with the threshold of 184. We didn’t compare them with each other like the Japanese rules, we compared half of the board plus komi with 184. (I think that is the main issue, that we also want to have a compatible score with area scoring half counting even though it will not be the same komi used in the Chinese rules)

19

To me, this is desirable, as it allows draws. Whether draws are (in moderation) a good thing, or if any possibility of drawing is necessarily bad, is a matter of some controversy, of course, but I just want to highlight that not all people see this as a “cost”, and indeed, the people (such as me) who prefer 7.0 komi area scoring are I would think quite likely to see a small number of draws as a positive thing, not a cost

Of course, computer tournaments are a different matter and there is makes sense to vary komi from game to game to get closer to 50% winrate for both sides, as it seems they are

20

I think Ke Jie at his height before AI had a very high white win rate compared to his black win rate (although at the time he was stronger than probably everyone, so his win rates were both above 50%. Just white winrate is so high to nearly undefeatable as white).

However, for players like Shin Jinseo, his winrate is already super height combined nearly 90%, so regardless he played black or white the winrate are both nearly undefetable.

1 Like