Kirby's Study Journal

[Kirby]

A few caustic comments on the last game.

By now I am reasonably certain that the last corner tedomari is a thing. Besides that, even the ancients who played with no komi thought that this approach was objectively suboptimal, but did not want to give Black an easy game.

It's probably correct that playing the last corner is best. But I don't like playing this way as white. It might be an artifact of being Inseong's student - he has a lecture on white's opening, which is about breaking up the board and preventing black from playing classical openings like mini-chinese, kobayashi, etc. That all said, before AI, pros were fine with playing white in the mini-Chinese, Kobayashi, etc., so they obviously don't have a problem with it. But I like to prevent those formations, because I'd prefer to get into a game position without giving my opponent of some sort of pre-studied setup.

Maybe I lose a couple of percentage points when you evaluate by a bot, but if I get into a board position that I like to play, I think that's worth it.

Long before the AI era, Shuei showed that leaving this two stone shape to fend for itself was fine. Seize the last open corner. (And the bots have shown that the three space extension rule is not correct. Jowa and other top players did not follow it, either.)

What's wrong with a 3-space extension here?

What special opening did you prevent Black from playing? He got in the Shusaku 1-3-5, didn't he?

I like it better than mini-Chinese shape. I like playing mini-Chinese (I like it more than the probably more popular Kobayashi, or modern AI openings), and I don't like playing against it - even if bots disagree.

I guess he likes that pattern, too. Probably because his opponents are content to extend three from two stones.

/shrug
I still like the 3-space extension, and I'd be comfortable playing that way again.

[Kirby]

Generally speaking, opening preferences like this probably have merit in the spirit of playing optimally. But I don't think they have much impact on the chance of win/loss for my games.

I'd rather focus on things like the fact that I put my right side group at risk - that could have really changed the outcome of the game. And it makes my play unstable. If I played move 4 as an approach instead of an empty corner... I'm happy to go with the route that makes me feel more comfortable, unless I can see a strong reason showing that it's worse.

[Bill Spight]

A few caustic comments on the last game.

By now I am reasonably certain that the last corner tedomari is a thing. Besides that, even the ancients who played with no komi thought that this approach was objectively suboptimal, but did not want to give Black an easy game.

It's probably correct that playing the last corner is best. But I don't like playing this way as white. It might be an artifact of being Inseong's student - he has a lecture on white's opening, which is about breaking up the board and preventing black from playing classical openings like mini-chinese, kobayashi, etc.

Go Seigen had a similar attitude. White should play the opening to keep Black from getting an easy game. He said as much in his 21 Century Go series. In fact. he disliked defending against the mini-Chinese so much that he recommended playing the two space high approach as when was a 3-4. It's hard to argue against Go Seigen, but I think that the bots have shown us, at least with 7½ pt. komi, that Black's premature development on the sides is questionable, if not an outright mistake. Anyway, it's nothing to fear, as Shusai understood.

That all said, before AI, pros were fine with playing white in the mini-Chinese, Kobayashi, etc., so they obviously don't have a problem with it. But I like to prevent those formations, because I'd prefer to get into a game position without giving my opponent of some sort of pre-studied setup.

What makes you think that this opponent does not have a lot of study and experience with the opening you played?

Maybe I lose a couple of percentage points when you evaluate by a bot, but if I get into a board position that I like to play, I think that's worth it. . . .

What's wrong with a 3-space extension here {:w8:}?

First, it leaves the last open corner to Black. As a corollary, .White is mildly overconcentrated on the top side. As a corollary to that, Black does not gain much from going into the top side (Edit: if you play in the last open corner). Also, if you do play on the top side, closer to the lone White stone in the top left is probably better, on the side star point, for instance. We are not used to seeing top players making those long extensions, because of the dogma of 3 from 2, etc., but top players before the 20th century played them (on the 3d line, OC), and played tenuki. The bots agree with those players about that.

I still like the 3-space extension, and I'd be comfortable playing that way again.

It was your level of comfort after that bothered me. You felt OK, but maybe you should have felt sick to your stomach. It's not like you made a few suboptimal plays that took Black out of his comfort zone.

We are all familiar with the high approach, low attachment joseki that ends with a 3 space extension on the side by the second player. Even Segoe, in his day, knew that joseki was not good for the second player. The argument, attributed to one of the Fujisawas (I used to say, Hideyuki, but maybe I misremembered and it was Hosai), is to play that joseki in all four corners, in which cases it is obvious that the first player is ahead. Now. the kick joseki chosen by your opponent may not make as much territory for Black, but that does not make the three space extension a good play (so early in the game). In the other joseki, Segoe recommended playing in an open corner, not instead of the three space extension, but on the previous move. Still, I think he would have recommended the open corner for . Now, you did not play the three space extension four times, but you did play it three times. That's too many times. You should have felt sick to your stomach.

[Bill Spight]

I'd rather focus on things like the fact that I put my right side group at risk - that could have really changed the outcome of the game.

Yeah, you picked a fight where you were outnumbered. After the Black jump, , the White peep looks questionable, and then the push and cut looks like an overplay. and would probably been better on the left side, or, if you are worried about your right side group, just a one space jump yourself.

[Kirby]

You should have felt sick to your stomach.

I still prefer white, and I think black lost out. Maybe a bot disagrees, but I feel good playing from this opening.

I have a review with Blackie this Sunday. If he shares your sentiment, I’ll reflect more on it.

And again, I don’t think the opening is game deciding there.

[Bill Spight]

And again, I don’t think the opening is game deciding there.

It seldom is.

[Kirby]

I could be wrong. Part of my basis for thinking black lost out can be seen from Ryan’s example at 9 minutes into this video:
https://m.youtube.com/watch?v=cWlL5WpJMmw

There, he compares to a more common joseki where the corner is bigger. He notes that not everyone will agree, but he thinks that the kicker incurred a loss.

That being said, he noted that it may be a good idea to do if, eg., the side was not developable. So the 3-space extension may be less effective. I’d imagine that taking a corner may be more valuable there.

So maybe the opening wasn’t as much of a loss for black as I thought.

That being said, I did achieve an opening I was comfortable with. Tenuki after a kick is something I do sometimes, but being pincered isn’t always comfortable. So from that perspective, I could relax a little bit...

So I guess I’d say that I’m leaning more toward thinking it’s possible that I lost out some in the opening. Though, I’d still maintain that getting into a position I was comfortable with was valuable, even if suboptimal.

[xela]

(Edit: a couple of other posts have appeared as I was writing this. I don't think it changes my point though. I'll be really interested to see Blackie's review!)

Some balanced analysis of Bill's caustic comments (-:

And the bots have shown that the three space extension rule is not correct.

Initially I thought: I'm with Kirby on this one. The bots have also shown that amateurs frequently make -10% moves in the middlegame. Splitting hairs over +/- 2% in the opening feels like the wrong priority.

At move 7, KataGo says black is leading very slightly, 53.6% or +0.2 points. In human terms, a completely even position. At move 35, after mistakes by both sides, we're back to 53.4%, +0.2 points. No need for either player to feel sick to the stomach. (Or perhaps both should feel sick, after all the lost opportunities? I think this is business as usual at our level.)

However, the swings turn out to be bigger than I expected. So if the pattern is (sometimes) this common, maybe it is worth learning more about it.

What's interesting is that KataGo evaluates the three sequences differently. At top right, the three-space extension is a small mistake, losing 4.9% winrate or 0.8 of a point. Of course KataGo wants to play in the empty corner instead. Looking at follow-up sequences, if black were to attack at the top next, white would ignore it and approach the bottom right. Best play for both sides, accoding to KataGo, looks like this:

$$W
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . a b . . O X . . . |
$$ | . . . O . . . . . , 3 . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 2 . . |
$$ | . . . . . . . . . . . . . . . . c . . |
$$ | . . . 1 . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$W
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . a b . . O X . . . |
$$ | . . . O . . . . . , 3 . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 2 . . |
$$ | . . . . . . . . . . . . . . . . c . . |
$$ | . . . 1 . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

So the three-space extension isn't terrible shape, it's just not urgent, and high is slightly better than low. This position is 53.3% for black, or +0.1 point. If white plays a instead of , it's 54.5% or +0.3, not much difference. If black plays b instead of , white will reply at c, ending up with B+53.3% or +0.2, again pretty much the same.

Well, I do not yet have my new computer to back me up, but I bet that just about any top bot with at least 4k rollouts for each of these options will say that the keima approach to the bottom right corner has a better winrate for White than this three space extension.

Bill would win that bet. But it's not a big difference. The keima approach is B+57.2 or +0.7 points. The game move is B+59.2% or +1.0 points. By the way, I'd normally round these figures to the nearest whole number. The decimals are not that meaningful: I'm just including them to show the extent to which we're splitting hairs here. (If you restarted KataGo with a different random number seed, on 10,000 playouts you'd see differences of more than 0.1 in the evaluations.) All figures here are with 10,000 playouts or slightly more.

Finally, at bottom right, move 26 really is a clanger. But not because it's bad shape locally. The reason is that KataGo sees the top left as still being hot: white should play B17 next. The game move loses a whopping 7.7%, or an entire point, and puts black up to +69.1%, +2.2 points. But then black's reply is about equally terrible (black also should have gone to the top left, at G17), and black's one-space jump later (move 35) is also a small mistake, so we end up back on an even keel.

For what it's worth, each of black's kicks (Q17, E3, R4) is also a small mistake (-3%, -1%, -3% respectively).

Overall, I'm not seeing any evidence from KataGo that the three-space extension is locally bad shape, just that it's much less urgent than we may have thought (if we weren't paying close attention to Jowa and Shuei).

[Bill Spight]

I could be wrong. Part of my basis for thinking black lost out can be seen from Ryan’s example at 9 minutes into this video:
https://m.youtube.com/watch?v=cWlL5WpJMmw

Thanks.

There, he compares to a more common joseki where the corner is bigger. He notes that not everyone will agree, but he thinks that the kicker incurred a loss.

Yes, that other joseki is the one that I was referring to that has been known to be generally inferior for White since some time in the early or mid 20th century. I know that Ryan was covering a lot of material, but he could have said that the extension that he showed is now usually omitted. Instead he said that pros now avoid that joseki, which I guess they do. I agree that the kick joseki that he showed is better than other one for White, but, having seen bots prefer the kick in many cases, I can't blame the kick. This is a case where I am having to reconsider things. Still, even if White is better that does not make the extension right. The bots generally avoid it in both "joseki". Out of curiosity I looked up the kick on Waltheri. It turns out that White tenuki instead of any particular extension is the most frequent choice.

Early on Ryan spoke of the importance of rapid development in the opening. I agree, in general. But I think he should have applied that principle to White in both of theses cases. Instead of extending, White should usually tenuki in the early opening. (Side extensions are still good plays in the opening, but they are usually second tier plays.)

[Bill Spight]

Thanks, xela, for analyzing this opening with KataGo.

What's interesting is that KataGo evaluates the three sequences differently.

Yes. With the type of evaluation the bots do, the question is always how well a play stacks up to par, and par depends upon other plays besides the one which is being evaluated.

At top right, the three-space extension is a small mistake, losing 4.9% winrate or 0.8 of a point. Of course KataGo wants to play in the empty corner instead.

0.8 pt. is a small difference in the game, but a relatively large difference for a play in the opening. IMHO it is like the difference between shodan play and a 4 dan play. Small differences can add up.

Looking at follow-up sequences, if black were to attack at the top next, white would ignore it and approach the bottom right. Best play for both sides, accoding to KataGo, looks like this:

$$Wcm8
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . a b . . O X . . . |
$$ | . . . O . . . . . d 3 . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 2 . . |
$$ | . . . . . . . . . . . . . . . . c . . |
$$ | . . . 1 . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Wcm8
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . a b . . O X . . . |
$$ | . . . O . . . . . d 3 . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 2 . . |
$$ | . . . . . . . . . . . . . . . . c . . |
$$ | . . . 1 . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

{Diagram corrected. WLS}

So the three-space extension isn't terrible shape, it's just not urgent, and high is slightly better than low.

Right. The question is not whether the extension is good or bad locally, but how it stacks up with other plays on the board at the time. Edit: Note that after the two corner plays, the time is ripe for the extension.

Well, I do not yet have my new computer to back me up, but I bet that just about any top bot with at least 4k rollouts for each of these options will say that the keima approach to the bottom right corner has a better winrate for White than this three space extension.

Bill would win that bet. But it's not a big difference. The keima approach is B+57.2 or +0.7 points. The game move is B+59.2% or +1.0 points.

Sure. Large swings are rare in the opening. However, it was relatively easy for me to make that bet, or, to put it another way, to choose what is probably a slightly superior play. (Although I think that there are many more playable plays than we sometimes think. ) While a 10% swing for KataGo on a single play in the opening may be unusual, it is not unusual to gain that much in a few plays on the basis of intuition, experience, and judgement. What is difficult is not the thinking or reading in a single game, but the study and training over time. In addition, these qualities that produce slightly better plays, play after play, are generally significant well into the endgame.

[Kirby]

0.8 pt. is a small difference in the game, but a relatively large difference for a play in the opening. IMHO it is like the difference between shodan play and a 4 dan play. Small differences can add up.

Small differences can add up. Big differences can add up more. I spent some time letting the opening churn on leelaz this evening. For the first several moves, I get a graph like this:

The places were the graph jumps up are places where I am giving black an advantage - i.e., my move was bad. The first *big* jump is where the cursor is in the screenshot, from move 17 to 18. Namely, this:

$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X X . . . . . . O . . . O X . . . |
$$ | . . O O W . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X X . . . . . . O . . . O X . . . |
$$ | . . O O W . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

With ~40k playouts, Leelaz thinks I even maintain an advantage if I play this sequence:

$$Wc
$$ ---------------------------------------
$$ | . . . . . 8 . . . . . . . . . . . . . |
$$ | . . 9 7 2 3 6 . . . . . . . . . . . . |
$$ | . . X X 1 4 . . . . O . . . O X . . . |
$$ | . . O O 5 . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . 0 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Wc
$$ ---------------------------------------
$$ | . . . . . 8 . . . . . . . . . . . . . |
$$ | . . 9 7 2 3 6 . . . . . . . . . . . . |
$$ | . . X X 1 4 . . . . O . . . O X . . . |
$$ | . . O O 5 . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . 0 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

lz says that white is slightly ahead here at 53.6%. I suppose the reason for playing the double hane is because the top is already low for white, so it's good to just take the corner territory.

Second biggest jump was a little bit later, when I extended instead of following up in the top left:

$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . . X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . W . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . . X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . W . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

lz thinks top left is urgent, and wanted me to continue there. After black also answered on the right, and I returned to the top left, game came back closer to even:

$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . 2 X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . 2 X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

---

After all this, some takeaways I have:
1. Three space extension is more tenuki-able than I initially thought.
2. I still don't think it's a game changer, but it'll depend on what's urgent in other parts of the board. In this example, the bottom right extension, in particular, was not urgent - the top left corner was urgent, so it would have been better to continue there.
2.5) I want to also point out that *black* also lost points by choosing the kick variations.
3. The bigger issue here was extending the third stone after the 3-3 invasion in the top left.

---

I agree with the premise that small differences add up. But I also think there's something to be said for prioritization. Like many others, I'm work with software for a living. One of the common things folks work on is performance optimization - you don't just want applications to work. You want them to work fast.

And one of the techniques for identifying performance optimizations is to look for bottlenecks. You look for the biggest problem area, and try to see if you can fix that. You can get a lot of bang for your buck that way. Then after fixing the biggest bottleneck, you can work your way down.

I think it's similar for go. There are several moves in a 200 or 300 move game that can require performance optimization. And the little ones can add up a lot. But when there's a big spike in your graph where you totally change the expected winner of the game... I think it's good to prioritize that.

Here, I don't think the three space extensions constituted any such spikes. But it's good to know that tenuki is a very viable option here. I'll try to keep that in mind.

While I haven't drifted too far from the performance optimization analogy, sometimes in software, after you optimize all of the big hitters, your code is still too slow. And maybe you need to rearchitect stuff. I'll probably have to do that with my go someday. But I don't think I'm there, yet. There are still a lot of spikes to flatten, which are probably easier to grab.

[Bill Spight]

0.8 pt. is a small difference in the game, but a relatively large difference for a play in the opening. IMHO it is like the difference between shodan play and a 4 dan play. Small differences can add up.

Small differences can add up. Big differences can add up more.

We agree on that. OC, in the nature of things, there are more small differences to add up.

I spent some time letting the opening churn on leelaz this evening.

Thanks. This analysis is very interesting.

For the first several moves, I get a graph like this:

The places were the graph jumps up are places where I am giving black an advantage - i.e., my move was bad. The first *big* jump is where the cursor is in the screenshot, from move 17 to 18. Namely, this:

$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X X . . . . . . O . . . O X . . . |
$$ | . . O O W . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X X . . . . . . O . . . O X . . . |
$$ | . . O O W . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

With ~40k playouts, Leelaz thinks I even maintain an advantage if I play this sequence:

$$Wc
$$ ---------------------------------------
$$ | . . . . . 8 . . . . . . . . . . . . . |
$$ | . . 9 7 2 3 6 . . . . . . . . . . . . |
$$ | . . X X 1 4 . . . . O . . . O X . . . |
$$ | . . O O 5 . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . 0 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Wc
$$ ---------------------------------------
$$ | . . . . . 8 . . . . . . . . . . . . . |
$$ | . . 9 7 2 3 6 . . . . . . . . . . . . |
$$ | . . X X 1 4 . . . . O . . . O X . . . |
$$ | . . O O 5 . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . 0 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , O . . . . , . . . . . , . . . |
$$ | . . . X O . . . O . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

lz says that white is slightly ahead here at 53.6%. I suppose the reason for playing the double hane is because the top is already low for white, so it's good to just take the corner territory.

Another possible reason is that Black already has taken two corners (thanks in part to getting the last open corner) and has played first in the third. I admit that I have not figured out when to play the two step hane joseki.

Second biggest jump was a little bit later, when I extended instead of following up in the top left:

$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . . X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . W . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . . X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . W . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

lz thinks top left is urgent, and wanted me to continue there. After black also answered on the right, and I returned to the top left, game came back closer to even:

$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . 2 X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . 2 X X X O . . . . O . . . O X . . . |
$$ | . . O O O . . . . , . . . . O , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O O . . |
$$ | . . X , O . . . . , . . . . . , X . . |
$$ | . . . X O . . . O . . . . X . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

This last example was one where, after xela's post, I was going to point out that learning about what is usually a small difference can lead you to plays that make a large difference. I have noticed this phenomenon when studying the Elf GoGoD commentaries. This is a good example. If you have already learned, mostly from positions where the differences are small, that the default should be tenuki instead of the three space extension, then your eye is automatically drawn to the top left. Not that you make any play in the opening automatically, but it helps to know what to think about. The extension on the right side approaches Black's strength, so that's a strike against it. Maybe your stomach should churn at the thought. In any event, if you are already primed to tenuki, you have a good chance of switching to the top left, even if you do not understand its importance very well. Sometimes what usually produces small advantages produces a big one.

After all this, some takeaways I have:
1. Three space extension is more tenuki-able than I initially thought.
2. I still don't think it's a game changer, but it'll depend on what's urgent in other parts of the board. In this example, the bottom right extension, in particular, was not urgent - the top left corner was urgent, so it would have been better to continue there.
2.5) I want to also point out that *black* also lost points by choosing the kick variations.
3. The bigger issue here was extending the third stone after the 3-3 invasion in the top left.

---

There is another takeaway, I think. It is good to study the opening. In fact, my argument about the accumulation of small advantages is made in favor of that proposition, because those who don't like to study the opening point out that the gains for each opening move are small, as a rule. Yes, I reply, but they add up. The fact that sometimes the gains are large makes opening study even better.

I have no argument with the following. I am quoting it because I think it is good.

I agree with the premise that small differences add up. But I also think there's something to be said for prioritization. Like many others, I'm work with software for a living. One of the common things folks work on is performance optimization - you don't just want applications to work. You want them to work fast.

And one of the techniques for identifying performance optimizations is to look for bottlenecks. You look for the biggest problem area, and try to see if you can fix that. You can get a lot of bang for your buck that way. Then after fixing the biggest bottleneck, you can work your way down.

I think it's similar for go. There are several moves in a 200 or 300 move game that can require performance optimization. And the little ones can add up a lot. But when there's a big spike in your graph where you totally change the expected winner of the game... I think it's good to prioritize that.

Here, I don't think the three space extensions constituted any such spikes. But it's good to know that tenuki is a very viable option here. I'll try to keep that in mind.

While I haven't drifted too far from the performance optimization analogy, sometimes in software, after you optimize all of the big hitters, your code is still too slow. And maybe you need to rearchitect stuff. I'll probably have to do that with my go someday. But I don't think I'm there, yet. There are still a lot of spikes to flatten, which are probably easier to grab.

Well, in my case I was fortunate enough to learn from Segoe that the high approach joseki with the three space extension was not in general good for the second player. That gave me a small but concrete advantage over those who thought that it's joseki, so the results must be even. But then AlphaGo came along and said, in effect, that the problem with that joseki is not, as Segoe and others had thought, the solid connection but the three space extension. Wow! That was an eye opener. (OC, sometimes you do make an extension, but not necessarily the three space extension and tenuki is the default.) Furthermore, that is a lesson that generalizes. It's one of the low hanging fruits of the AI era. It was something that even those who were unaware of the problem with that joseki could take away from AlphaGo's play.

Studying the opening is not easy these days, because, while the bots have overturned a lot of the old dogmas, they don't explain anything. What Takagawa taught in the 20th century was the result of human study over centuries. What Go Seigen taught was largely the result of his own genius. (But, as Rui Naiwei, who attended his study group, once told me, he often changed his mind from week to week. Creative till the end. ) Now, the bots have overturned old dogmas, but we humans have yet to rebuild opening theory. It is going to take a while.

I agree that LZ has indicated that the top left corner deserves study. But making the default tenuki instead of the three space extension has been low hanging fruit for a few years.

[xela]

learning about what is usually a small difference can lead you to plays that make a large difference. I have noticed this phenomenon when studying the Elf GoGoD commentaries. This is a good example. If you have already learned, mostly from positions where the differences are small, that the default should be tenuki instead of the three space extension, then your eye is automatically drawn to the top left. Not that you make any play in the opening automatically, but it helps to know what to think about.

QFT :-)

[Kirby]

This last example was one where, after xela's post, I was going to point out that learning about what is usually a small difference can lead you to plays that make a large difference. I have noticed this phenomenon when studying the Elf GoGoD commentaries. This is a good example. If you have already learned, mostly from positions where the differences are small, that the default should be tenuki instead of the three space extension, then your eye is automatically drawn to the top left. Not that you make any play in the opening automatically, but it helps to know what to think about. The extension on the right side approaches Black's strength, so that's a strike against it. Maybe your stomach should churn at the thought. In any event, if you are already primed to tenuki, you have a good chance of switching to the top left, even if you do not understand its importance very well. Sometimes what usually produces small advantages produces a big one.

I agree with the principle. But saying that the default is to tenuki is not clear to me from analyzing this single game - maybe it would be for others. I played the same sequence 3 times, and it stuck out as problematic just once in the analysis. I don't know why that should dictate stomach churning on my part.

Well, in my case I was fortunate enough to learn from Segoe that the high approach joseki with the three space extension was not in general good for the second player.

This is a different joseki. Is it possible that there are different nuances here?

[Bill Spight]

But saying that the default is to tenuki is not clear to me from analyzing this single game - maybe it would be for others. I played the same sequence 3 times, and it stuck out as problematic just once in the analysis. I don't know why that should dictate stomach churning on my part.

Sorry, I didn't mean that you should have learned that from this game. That's why I pointed it out to you.

But if you had noted that AlphaGo omitted the three space extension in the similar joseki and had also noticed small preferences by bots against side extensions early in the opening, then you would have been primed to tenuki. In particular, I think you might well have occupied the open corner with .

As for the stomach churning, I intended to introduce a note of levity. It seems that I failed at that. I apologize.

Well, in my case I was fortunate enough to learn from Segoe that the high approach joseki with the three space extension was not in general good for the second player.

This is a different joseki. Is it possible that there are different nuances here?

Of course. But to tease out nuances you have to look at small differences in a number of games.