Attachment as a response to the two-space approach to the komoku

In my last game, I encountered twice, and I responded to it with an attachment: and

I examined four main variations. We’ll call the corner-owner Black, even though I was White in the game.

  1. Transposing to a high-approach joseki, so even


  1. Seems better for Black


  1. Seems a lot better for Black


  1. This one is tricky, though: I don’t know to evaluate it. Perhaps it’s this variation that’s lurking against the attachment.


According to Waltheri, the attachment’s only been played a handful of times. The pros seem to like playing this variation, which looks even:


Well, I’m not going to suggest adding this to OGS Joseki because the pros aren’t actually playing it, but I wondered if anyone could explain to me why the attachment might be a suboptimal move.


It’s variation 1. The attachment can be suboptimal precisely because it transposes to a joseki. The long knight’s approach is more of a special-purpose move for when the opponent has other stones nearby, to prevent being pincered too severely or to serve some other special need.

The dictionary should probably mark the long knight’s approach as “good” rather than “ideal.” I just haven’t found a source that says so yet. But you can see it clearly with most modern AI (which recommends tenuki in many cases) and also in its virtual disappearance from professional play.



This really illustrates the the importance of something I read yesterday on Sensei’s Library, about “double-loop thinking”.

It went like this:

When you reason about something using the assumed truth of certain parameters [in this case, that the two-space approach is ideal] to reach a conclusion [in this case, that the attachment is also ideal], you’re using single-loop thinking.

When you question the parameters that you’re using to underpin your reasoning [in this case, considering the idea that the two-space approach is not ideal], you activate double-loop thinking and can reach a different conclusion that was previously invisible to you [in this case, that the attachment is also not ideal].

Another example is 3-3 invasions. In this case, the parameter we were thinking under and not questioning is that was the best variation of 3-3 invasion that White could achieve. When AlphaGo showed us, we were able to question this thought parameter and access a wider plane of reasoning.


Also worth noting:

  • Bot analysis in a simple sample opening position suggests that for diagram 3 and 4, black move 5 is a terrible mistake. White should play as in diagram 3 and then for white move 8, push through and cut black to the right of black’s move 7. Black can sacrifice two stones and still come out with good shape, but the trade is overall good for white.

  • My personal first instinct on seeing your diagram 3 was that the result as presented is better for white, not black. White can block on the second line in pseudo-sente, which means that relative to the joseki, white will be a lot stronger. A lot of the bad aji for white in the joseki stems from the fact that if/when white plays a 3-space extension, black can later invade. But if white has a second line stone, it takes a lot of the fire out of the invasion because white has an easy connection under. See for example, which I’ve seen Guo Juan 5p give a lecture about before as a possible technique for white, where white even voluntarily sacrifices more points and outer influence in order to get the second line stone in sente. Plus, while black’s corner is a bit bigger, black actually hasn’t hopped out as far in the joseki, black didn’t get to play at the 3-6 point.
    Checking with AI: seems to confirm this too. If white misses the sharper way to punish black, but plays this way, and black obediently follows along, the result is good for white still - black’s capturing of the white stone this way is worse for black than joseki, not better.

  • For move 4, white can consider the hane under black’s stone as well, as you noted in how pros seem to play. White is low, but black’s stones also feel a little inefficient here, so again it’s quite possible in many positions I’d like this more than the joseki for white.

So white can often do even better than simply transposing to joseki. While the surrounding stones and the overall board position will obviously affect these things, it seems like white will often have ways to do better.