KataGo's primary 2019 version played my Guzumi just before the start of the Crosscut Sequence in the left centre.

We examine the chronological order of my discovery of the Guzumi as well as the additional support through Harry's later added Hasami-Tsuke Sequence.

Finally, we try to answer the question why KataGo identified my Guzumi so late during its initial training.

Referenced by ( 2022MainLine)

Let's enter the world of KataGo ...

: ( 2022MainLine)
This is the position when KataGo's original 2019 version played my Guzumi.

Please note that Harry Fearnley's Hasami-Tsuke Sequence (, ) has been played already.

Let's go back in time, and enter the world of amateurs ...

:
After a three-month search for a move that would bring Black additional points, I found my Guzumi two years before Harry's Hasami-Tsuke (after the previous exchange of for ) saw the light of day.

Please note that at that time, the kikashi in the hanezeki were not yet played.

:
It is best for White to complete the Guzumi Exchange.

However, we were very unsure about White options for a tenuki, because Black was ahead by only a very small margin at that time.

in particular worried us, which is also the strongest move for White in the top left corner, according to Yoon Young-Sun 8p, who kindly supported us on this issue.

For further details, please refer to CHAPTER "Tenuki after the Guzumi (2022)" ( 1439)

... and / or the history of our investigations ...
432

:
The Crosscut Sequence follows, as in the original solution of Fujisawa Hideyuki 9p.

In the Guzumi Area there are now two false eyes for White, which is equivalent to one approach-move liberty for Black's large group in the upper right.

This means that Black has not lost any Main Semeai Liberty because of my Guzumi.

Liberties:

Fujisawa Hideyuki (at left):

• Four direct liberties in the Guzumi Area.

My Guzumi (at right):

• Three direct liberties + one approach-move liberty in the Guzumi Area = four liberties in total.

No difference.

Territory (in the Capture Variation):

Fujisawa Hideyuki (at left):

• Three points for White in the Guzumi Area.

My Guzumi (at right):

• No territory in the Guzumi Area.

Black earned three points by my Guzumi.

About two years later ...

:
Harry Fearnley's discovery of his Hasami-Tsuke Sequence (, ) made our work much easier.

Please note that at that time, the kikashi in the hanezeki were not yet played.

We thought that the Guzumi must be played latest now, just before the start of the Crosscut Sequence, but KataGo has proven us wrong.

For further details, please refer to CHAPTER "Timing of the Guzumi (2022)" ( 1813).

However, even KataGo's initial version of 2019 was wrong in its assessment of the latest moment until which the Guzumi could be delayed ...
222an

:
As before, completing the Guzumi Exchange is best for White.

To our great relief, Harry's Hasami-Tsuke provided Black with an additional territorial advantage, which offered a sufficient safety buffer for our analysis of White's tenuki options after Black's Guzumi.

Harry's Hasami-Tsuke Sequence created another White false eye at the top.

This "own" approach-move liberty at the top allows Black to connect both hasami-tsuke stones (, , , ).

, :
Harry also found that Black's kikashi in the hanezeki should be played latest now, immediately after White's local reply to the Guzumi.

:
As usual, the game continues with the Crosscut Sequence in the upper left centre.

Liberties:

Fujisawa Hideyuki (at left):

• Six direct liberties in the marked area.

Guzumi & Hasami-Tsuke (at right):

• Four direct liberties + two approach-move liberties in the marked area = six liberties in total.

No difference.

Territory (in the Capture Variation):

Fujisawa Hideyuki (at left):

• Ten points of White territory in the marked area.

Guzumi & Hasami-Tsuke (at right):

• Three points of White territory in the marked area.

In total, Black won seven points by combining Harry's Hasami-Tsuke and my Guzumi.

Let's return to the era of KataGo ...

:
My Guzumi was found very late during KataGo's initial training in 2019, just in time before its end.

Possible explanations:

- Over a very long period of this training, KataGo came to the conclusion that Joachim's Ko-Semeai would end successfully for Black. This semeai is only possible in the absence of the Guzumi.

For further details, please refer to section "Joachim's Hane at the Top (Joachim's Ko-Semeai #2)" ( 2030)

- It can therefore be assumed that KataGo had already identified the new moves on the left side of the board that were advantageous for White. In this case, the Guzumi Exchange only resulted in a one-point advantage for Black, which was not enough to turn the game around.

The latter will be explained below.

///
With a manually suppressed Guzumi Exchange, KataGo reaches this position before the final mutual occupation of opponent's liberties in the Main Semeai.

The diagram is taken from CHAPTER "Timing of the Guzumi (2022)" ( 1813)

///
This is the comparable position that KataGo reaches with the Guzumi Exchange played.

The diagram can be found in "The Solution" => "2021" ...
9

Capture Variation (right half of the board):

Without Guzumi Exchange (at left):

• Four points of territory for White.

With Guzumi Exchange (at right):

• No territory for White.

White lost four points of territory in the right half of the board through the Guzumi.

Capture Variation (left half of the board):

Without Guzumi Exchange (at left):

• Four points of territory for White.
• 15 points of territory for Black.

In total, Black has a surplus of eleven points of territory.

With Guzumi Exchange (at right):

• Four points of territory for White.
• 12 points of territory for Black.

In total, Black has a surplus of eight points of territory.

White gained three points of territory in the left half of the board through the Guzumi.

Taken together, Black gained (only!) one point by the Guzumi on the whole board.

2022MainLine