My guess is that - to a pro of Ishida's generation -being squeezed as black is in that joseki sequence seems so unbearable and 'obviously wrong' that it affected the evaluation.

...Maybe it is that joseki appeared fishy to pros?

This is Relentless chapter 1, variation 10, a sequence that didn't actually come up in the game:

Click Here To Show Diagram Code

[go]$$Wc19 Unfavourable capturing race
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . X O 7 . |
$$ | . . . O . . . . . . . . . . . X 4 5 . |
$$ | . . . . . . . . . . . . . . . X O O 6 |
$$ | . . . O . . . . . , . . . . 2 1 X X . |
$$ | . . . . . X . . . . . X . 0 9 3 8 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

So how are we supposed to interpret this?

It looks like to cut first at a, get captured and give away a bunch of solid territory in the corner is not merely OK, it's positively better than not giving away that territory!

I don't see it at all as White giving away a lot of solid territory. I see it as ajitsuki - adding aji. Furthermore, White retains the option of adding even more aji here. The game in Chapter 34 illustrates the sort of aji that exists. I would even say that it is dangerous to call White Q4 a cut. Of course it is one, but using that word alone brings in too readily connotations of cut-and-capture or cut-and-kill. I would suggest the right way to express Q4 here would be to call it "a probe with the cut at Q4". I also infer that katago might think making this probe early is good because it is more likely that Black will play the hane on the outside instead of the inside (which seems like a plausible line, after all). So, is the result of this "probe" a loss for White? I hardly think so.

know when to take leave of absence from an ongoing joseki and when to return to it later, something AI bots do all the time [...] I've often wondered if it's part of a randomising function in the bots

Black's hold on the territory in the corner is conditional on black playing another move there. That move is not urgent but it still needs to be played eventually. We can interpret that to mean black has at least 50% chance of defending the territory later.

Now if KataGo plays first in the corner, which my KataGo thinks it should after unspecified number of playouts, then white gets a chance to start a ko. We can interpret that to mean there is a 25% chance that there will be a ko and 25% chance that black will defend here. Now we are at 75% black territory and 25% ko.

I do not understand how black can kill white with only one move. I think if black plays first then the position becomes yose ko and white has still a chance to live.
OC if white plays first the resulting position is a simple ko.

Click Here To Show Diagram Code

[go]$$B
$$ . . . . . . O . .|
$$ . . . . . X O O .|
$$ . . . . . X X O .|
$$ . . . . O X O O X|
$$ . . . X X O X X 2|
$$ . X X X O O X 3 O|
$$ . . . O O . O X 1|
$$ . . . . . O X X .|
$$ -----------------+[/go]

Suppose Black ko captures. Better. Afterwards, White cannot remove the big black string but Black can remove the big white string. White is dead in perpetual ko. So again: what ko?!

At first impression there appears to be this ko. White has local threats. I learned long ago that with ko it is not about finding the ko as much as it is about finding the right ko.

Oh no, Gerard beat me to posting this.

Click Here To Show Diagram Code

[go]$$B
$$ . . . . . 4 O . .|
$$ . . . . 5 X O O .|
$$ . . . . . X X O 2|
$$ . . . . O X O O X|
$$ . . . X X O X X .|
$$ . X X X O O X . O|
$$ . . . O O 3 O X 1|
$$ . . . . . O X X .|
$$ -----------------+[/go]

Click Here To Show Diagram Code

[go]$$Wm6
$$ . . . . . O O . .|
$$ . . . . X X O O .|
$$ . . . . . X X O O|
$$ . . . . O X O O X|
$$ . . . X X O X X 3|
$$ . X X X O O X 2 O|
$$ . . . O O X 1 X X|
$$ . . . . . O X X .|
$$ -----------------+[/go]

Click Here To Show Diagram Code

[go]$$Bm9
$$ . . . . 5 O O . .|
$$ . . . 4 X X O O .|
$$ . . . . . X X O O|
$$ . . . . O X O O 1|
$$ . . . X X O X X 2|
$$ . X X X O O X X .|
$$ . . . O O 3 . X X|
$$ . . . . . O X X .|
$$ -----------------+[/go]

Click Here To Show Diagram Code

[go]$$Wm14
$$ . . . . X O O . .|
$$ . . . O X X O O .|
$$ . . . . . X X O O|
$$ . . . . O X O O 3|
$$ . . . X X O X X O|
$$ . X X X O O X X .|
$$ . . 2 O O X 1 X X|
$$ . . . . . O X X .|
$$ -----------------+[/go]

Maybe I am wrong but what is your own evaluation of the local temperature?

It is one of those cases when it is moderately hard to calculate but knowing if you got the right answer is much harder. So I can calculate for you but it would take too much effort to verify beyond my intuition about what is happening. It is good practice so I'll humor you

I only noted down some numbers but otherwise did it without playing out variations. Since I did this in my head we have to go backwards and I'm not going to explain how I reached the conclusion which I'll try to summarize.

The way I see it, if black takes and wins the ko he is at +17 but if white starts the ko there are two choices. White can choose a big ko to drive up the value per move or a small ko. The thing is that if he chooses the big ko he will forego an endgame move here but if he chooses the small ko he gets that endgame as part of starting the ko.

Most ko are trades, I assume this one is also, therefore I don't want to give up that endgame to drive up the value per move. The small ko then it is.

I have black at +16 if he ends the small ko, I have white at -17 if he ends the same ko. I'll ignore fractions and say the value of the ko will be at +4 when white plays to start it.

How about if black stars a ko immediately? In that case he can win the ko to be at +16 (not sure why I got one less here, I'll just go with it). White, however, can only go back to needing a move to start the small ko.

We, therefore, have a situation of black being able to move to +17 in two moves and white being able to move to +4 in one move. The value of a move in this situation is just 4 or 8 as you like to say

Black won't make a move here for a long time.

OK, OK, I'm ignoring that there is an approach ko but I doubt it is a good idea for black when the situation otherwise isn't urgent. I'm ignoring the approach ko for the reason that white doesn't need to spend an extra move, white in fact gains a move but needs to come up with lot of threats.

There is something important that I have saved for last. Which is the territory in the corner. Since the move was worth ~4 and white were to move to +4 then the territory would seem to be +8 but that is for an arbitrary region. I think we can count some naive +25 points in the corner before white locked-in the aji. Now it is +8, some naive 17 points are already erased. Maybe it is no wonder that the score evaluation is -3 to -4?

It is a little difficult to understand your figures without the corresponding varaitions.
Let me try to show you my result after some more analysis.

I can see if I can add diagrams later but I'm not sure it is so interesting for me

Your analysis is similar to mine but I'm suspicious of choosing the larger ko and not assuming that white is playing it to win. Possible the difference between these two ko isn't enough to make a difference. I mean that it is possible white also shouldn't play the smaller ko unless he means to guarantee that he wins the ko and in that case it is likely better to win the larger ko. My point was about that black should or could wait for white to start the ko, for now. Black appears to gain a move if white plays first in the corner. White would after all be playing two moves in a row in the corner.

Click Here To Show Diagram Code

[go]$$B Black plays away, white has choice between a larger ko with A and a smaller one with B, and of course tenuki.
$$ +--------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . O . . . . . . . . . . . . . . .|
$$ | . . . , . . . . . , . . . . . X . . .|
$$ | . . X . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . X . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . , . . . . . , . . . . . , O . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . O . .|
$$ | . . . . . . . . . . . . . . . X O O .|
$$ | . . . O . . . . . . . . . . . X X O .|
$$ | . . . . . . . . . . . . . . O X O O X|
$$ | . . . O . . . . . , . . . X X O X X b|
$$ | . . . . . X . . . . . X X X O O X a O|
$$ | . . . . . . . . . . . . . O O . O X .|
$$ | . . . . . . . . . . . . . . . O X X .|
$$ +--------------------------------------+[/go]

We both used the method of finding a value for the ko and value per move but the calculus would change if we assume one or the other player could win the ko using good threats. I think white is close to being there, maybe he could win the ko. I did assume that white can't guarantee this outcome and is interested in a fair trade. Possibly this assumption is not good but it was what I thought was a reasonable assumption but these ko are large

It still seems likely to me that black shouldn't play first in the corner. They say every dog has its day, but usually you can't make up for playing more moves when less would have done. Without checking the principle variation that KataGo provides I can only guess at what happens if white tries to win the large ko.

Now that the questions have become so complicated I have began to wonder what KataGo says. I looked at it briefly before and recall that these questions weren't critical. Now that I'm checking again it appears they are even less critical than what I recalled.

Let's expect that white will be able to play first in the corner and will be able to win the ko.
The question is now the following: when white will start the ko?
The question is difficult indeed. If the ambiant temperature is T then, in compensation for the corner killed by white, black will play tenuki two times and will gain 2*T points. That means that white wish to wait for a ambiant temperature as low as possible. In the other hand, if white wait too long then her good ko threats could disappear and black could also build her own ko threats.
In addition, if white wait too long then a difficult fight may take place elsewhere on the board and the temperature could stay quite high.
For those reasons I think white should start the ko quite early in order to win this ko and let black play tenuki two times.

It is not only a difficult question, it's a question that starts from lot of assumptions.

I'm very suspicious about arguments that start by declaring something like "white will win every ko". If you assume this, then ko becomes a one sided move that black can't play. If you state instead "white will win every ko while <some-condition> is true", then there is still a very strong assumption there.

Why not instead assume the player that starts the ko wins it? Possibly by ignoring a ko threat. When we do this we would eliminate capturing in the ko on the two conditions:

• If winning the ko in this way is not better than other moves in the game. We'd say it's dominated by some other move.
• If winning the ko ends up being worse than letting the other player play. We'd say it is reversed.

It is not clear which of these condition applies when we say that white will win the ko. Maybe that is why it appears as if white's fortunes could suddenly change if few moves later we say that black will win the ko? I can't relate this argument directly to this complex position but I think it is likely that if starting the ko immediately isn't white's strategic choice, then the change in who we say is winning the ko is somewhat fictious.

Take for example a gote endgame move. What happens if we declare white will play this move and then declare a few moves later that black will play there? We can often make good arguments about who is likely to play an endgame move. In this case nothing really happened unless either player made a mistake, our argument about who will play first might help us identify some mistakes but it isn't an assumption that we strictly require. It is also an argument that can lead us astray if we turn it into an assumption.

At the risk of relativize the argument one can also ask what happened in this case if it was never part of white's strategy to be certain about who wins this ko? In this position it doesn't seem reasonable to claim to know what the optimal strategy is or to claim that we could discover it. I think we need to be very careful when optimal strategy is so subtle.

I ran KataGo for 1 million playout on the four most interesting candidates. For some reason the playout statistics sums to a higher number in the following screenshots but it was the same 1 million in each case. Take a look at the attached image.