I guess the reason is that after A, black will cut with atari and what should white do then? Connect and make an ugly shape? Play a ko?
Many generations of students know that their teachers would scold them for making an empty triangle, so they would avoid it at all costs. But AI disagree. They say that connecting is the only move and it’s fine for white.
This is similar to many other cases where AI recommend a move that humans considered bad for centuries (such as early 3-3 invasions).
There are two type of triangles: an empty triangle and a full triangle. Both look the same, but the first one is useless (wasting a move) and the second one is useful (adding an essential reinforcement).
Yes, but Waltheri’s and OGS joseki library are based on pro games. The policy of the OGS joseki explorer is to ignore AI joseki until pros play it. So that explains how it can deviate from AI game reviews.
I did a bit more research on Waltheri and I think I’ve got to the bottom of it.
In reference to the diagram below, we don’t see the farmer’s hat variation because hypermodern professionals have no interest in (1) except when △ is on the board. Then, quite naturally, after (2)–(3) they play the lighter ko shape B, not the farmer’s hat A.
So it seems that the answer to the original question is that White plays the bad hane in an attempt to avoid entering the farmer’s hat variation (due to its empty triangle), having become convinced by AI not to play the original atari-cover line. And that the root cause of White’s issue is the mistake (1).
Yes, that makes sense.
But when the marked stone is already in place, I think white can consider the hane in the corner instead of 2, because the marked black stone will probably end up overconcentrated (too close to black’s outside thickness) when white lives in the corner.
In @bugcat’s farmer’s hat variation, white makes a full trangle and an empty triangle. That empty triangle would be a reason for generations of strong players to avoid it. They would feel the shape is painful for white. But AI say white should do it nevertheless, because the alternatives are worse.
It is #3. The original attachment fell out of fashion and then AI killed it (and the low one space pincer) even more. That diagram would probably be one of the main variations if the 3-3 attachment were considered by professionals.
I probably have some internet pro games with that variation or I may have put it there when looking at a database of kyu games and seeing how common the 3-3 attachment is and wanted to put a comment for the AI move there.
I noticed how often fox 4kyus play C17 and either your move or the 3-2 point next to it. I guess they really want to settle and avoid a drawn out fight that a lot of other joseki choices lead to.
Cursory AI review show that your diagram move is the worst local move(by a lot). Basically every other reasonable move is better in that corner. 3-2, 4-2, crosscut, connect as a stick, or the marked A move.
1.) 7.5% loss means probably less than a point. In amateur games (especially for at least weaker than high dan), imho, everything up to 75-25 (or even 80-20) can easily be seen as an even game.
2.) 1kyu/1dan is not as strong as a player as you might think.
3.) Ask the players what they thought and discuss with them. I dislike that tendency. People stop talking to each other about the games, but just ask their bots.
4.) I think the hane comes from the “new” 4-4 joseki at which white hanes into the corner. People tend to just copy moves in other situations, like this one, ignoring 1 line lower makes a significant difference.