:: Igo Hatsuyôron 120 (2019)

KataGo's Bad Shape Move Again –

Flying Below the Radar of Area Scoring (2019)

KataGo – like all Go playing programs, which are based on Artificial Intelligence – uses area scoring (typical for Chinese rules) for the determination of a game's result, because this kind of scoring is much easier to implement than territory scoring (typical for Japanese rules).
However, Igo Hatsuyōron originates from Japan; so strictly speaking, we have to use the Japanese method of counting.

With area scoring, occupied board points matter. This implies that the parity of played stones (even or odd) plays an important role for scoring.

  • If the total amount of played stones is even (both players used the same number of stones, which means that the player who moved second occupied the last neutral point), the area scoring result is identical with the territory scoring one.
  • If the total amount of played stones is odd (this means that the player who moved first also occupied the last neutral point), the area scoring result is one point better for the player who opened the game than the territory scoring result.

The problem's starting position contains 71 White, but only 70 Black stones, which will be interpreted by area scoring that White (and not Black) started the game. So, for example, a result of White + 2 (area scoring) is equivalent to either

  • White + 2 (territory scoring), when Black occupied the last neutral point, or
  • White + 1 (territory scoring), when White occupied the last neutral point, providing her with a surplus of one stone on the board (= one point) in area scoring.

This equivalence is valid in all games that do not contain any surprising (and rare) endgame positions like special kinds of seki or a final ko that can be deferred until all neutral points have been filled. We can safely assume that these "exotic" shapes will not emerge in Igo Hatsuyōron 120.
Therefore, we must keep in mind that KataGo is NOT able to distinguish between e.g. "White + 2" and "White + 1" under territory scoring. This implies that it is easily possible that KataGo is having Black play a line that is slack by ONE point by territory scoring, because such a line would still be exactly identical in score and so just as "optimal" under area scoring.

Whilst further analysing variations after KataGo's new bad shape move at the left, we accidentically stumbled across a corresponding application case, which we explain below.

Joachim's Tesuji in the Bamboo Joint After Black's Tenuki 1093f
Too Much Noise for KataGo? 1093f

We ask for your kind understanding that we have not been able to adjust the structure of our website according to the new correct order of moves, but only added a corresponding note in the affected variations.

Joachim's Tesuji in the Bamboo Joint

After Black's Tenuki

White's bad-shape move at is gote locally, so Black – after inserting some forcing moves – might play tenuki with , instead of choosing one of his local options at and at .

The exchange of and makes sure that White will choose the Capture Variation.

If Black plays away, White's wedge at is best.

White did not only destroy all of Black's territory in the left centre.
She also keeps her liberty at , and created an additional approach-move liberty at .

These two won liberties are compensated for by White's loss of two moves in the Main Semeai, due to Black's forcing exchange for that was inserted before.

White and Black are miai.

Finally, White returns to the top right, wedging in at .

Black might be the reply that comes to mind first (it is also KataGo's preferred move in this situation), but the subsequent sequence will not change the final score.


However, Joachim suggested answering White's wedge () by giving atari from above, with the move at here.

In principle, this move is a kind of Michael Redmond's kikashi, just played on the other side of the bamboo joint. Black gains an additional liberty at , being worth one destroyed point of White territory.
(Funnily enough I had to convince Joachim from the validity of the above mentioned effect in this case. Usually, it is Joachim, who has to remind me of findings that we had a decade ago.)

White's cutting point at now is an approach-move liberty for Black, because his group has an eye.

Black will capture at , only after White connected at .

Black's tenuki at lost a move in the Main Semeai (it did not occupy one of White's liberties at the left, which is the reason that this move must not be played in Our Solution; otherwise, Black will fall prey to the Punishment Semeai), but White also lost a move in the Main Semeai (it did not occupy one of Black's liberties at the right.

Black's additional liberty at and the additional approach-move liberty at (due to Black's eye) allow Black to connect at , losing these two liberties again.

In total, the liberty relationship of the Main Semeai is not affected.

We add some liberty-occupying moves of the Main Semeai in order to simplify the determination of the effect of the combination of Black's tenuki and Joachim's tesuji.

Compared with KataGo's preferred line ...

– White gains two points at the top (), because Black is no longer able to reduce her territory there.

– Black loses one point of territory at the left (), due to the loss of his eye there.

– Black gains one point at the left (), where he captured a White stone.

– White loses one point in the centre (), which she otherwise was able to surround as territory.

– Black gains two points in the top right (), where he got one point of territory after capturing a White stone.

In total, Black is one point better after Joachim's atari in the bamboo-joint. KataGo's bad-shape move at the left is no longer effective.


White will win the Capture Variation by one point.

Too Much Noise for KataGo?

Joachim's tesuji in the bamboo-joint seems to be a quite unusual move, because KataGo did not even consider it.

The diagram at left shows a visualisation of KataGo's evaluation of the position.

is KataGo's preferred move; the points marked are alternative moves that are also mainly considered.


Because the other considered moves were predominatly part of the liberty-occupying part of the Main Semeai, ...

... I wondered what would happen, if the other liberties were occupied first, and the bamboo-joint area was left untouched until shortly before the moment when the tail is captured.

The result of KataGo's analysis at this moment is visualised here.

Now, Joachim's tesuji is also KataGo's preferred move!!!


I would like to assume that – besides the possibility of a very rare occurrence of the shape that includes Joachim's tesuji – the sheer number of permutions of the liberty-occupying sequences in the Main Semeai causes too much noise for KataGo's analysis of the position.

KataGo is a very powerful tool for analysis. But it seems to me that you have to be very well aware of what you are doing while using this tool under territory scoring, especially in extremly close positions wherein one point difference matters.

Copyright © 2020 Thomas Redecker.

Design by Jan van Rongen, modified by Thomas Redecker.

L e g a l   n o t i c e