Today we’re going to try something new for this blog: a game review! I’ve written up a little introduction for this new category here.

This review will reference evaluations via KataGo. Some less important positions for the discussion might have evaluations with as few as 10k playouts, but generally most positions that are discussed in detail will have evaluations based on >20k playouts.

This will be a review of the opening only. I could continue onto the middlegame, and there are plenty of interesting things to say about it, but then the post would be extremely long given the level of detail I tend to go into, and the opening alone was already plenty taxing.

Full game and background

Below is the record for the full game. The review covers up through move 46. This is a ranked Tygem match between two 1-dan players. The game is played under Japanese/Korean rules with no handicap. I am playing White; the result is W+R.

I’ve chosen this game because I adopted an opening strategy that I don’t often take, so I found it instructive to work out how to execute the plan in practice. It was a relatively balanced game (for games at this rank) without either side taking a decisive advantage for most of the game. Such games are also nice review material because in close games you need to play closer to optimal consistently throughout the game (though, again, we’ll only cover the opening here). There are things to learn from more imbalanced games too (e.g. how to safely close out games you’re winning, or complicate things in games you’re losing). But for practising general, consistent quality of play, a close game offers many teachable moments.

Just a heads up: in this review I intersperse what I was thinking during the actual game together with the conclusions I drew during the postgame review. Hopefully it will be clear which is which from the context. I am hoping it will be helpful to hear why I played as I did and then to see where the AI differs in its assessment.

The Review

The first thing that is unusual (for me) in this game is that I opted to tenuki when Black approached the bottom left corner. Of course, this is not uncommon overall, especially in the modern era of AI analysis; in another person’s game it wouldn’t be a point I would dwell on. It’s just uncommon for me to play this way. Normally I prefer to stick to moves I understand pretty well, and to be perfectly frank my understanding of the 4-4 low approach after tenuki is far from complete. But in this game I just had the sudden urge to play the tenuki - after all, I’ve got the top left corner and an opposing 3-4 point, and my head knows that tenuki from a 4-4 low approach is viable (even if I don’t understand it in my heart), so how badly can this possibly go?

I wasn’t quite expecting the 1-space jump along the side from Black, but I figured that this wouldn’t significantly change the outcome of the 3-3 attachment, and I didn’t think there was any particularly crazy jōseki Black can attempt from such a low position this early, so I opted for the typical jōseki here. I figured this is probably a minor loss for Black but nothing to write home about; in the end Black settles very solidly, while White gets what I consider a slightly more active position on the side with good (but not complete) progress toward settling. Move 12 is an example of a principle I discussed in my post on extensions and I was gratified to see that the principle held up in the post-game AI review, though it’s an extremely minor difference at +0.08 points for the three-space extension over the two-space extension.

I was now expecting Black to double approach the bottom left. I knew in my head that this would be an even outcome according to AI (neither player has deviated all that significantly from AI play yet) and was primarily considering which attachment I would play in response to the double low approach.

However, Black instead opted to slide, making me reconsider. (My opponent appeared to have a more classical, less AI-influenced playstyle than mine.) This is a move that, while not particularly dangerous or unusual, is something I’m less well-versed in compared to the double approaches, simply because I know in my head that the double approaches are AI recommended and therefore I’ve devoted specific effort to study them in the past. I’ve never bothered to study the slide specifically, so I had to play by instinct here.

Here I made my second uncharacteristic play this game, which sets the tone for the rest of the opening (and the game overall): I ignored Black once again and invaded the lower right corner. My reasoning at the time went like this:

I’m pretty sure that Black’s last move is a small local loss, but I don’t specifically know what the best local continuation is. I’m considering attaching on top, knight’s move, two-space high extension, and the pincer, most of which are probably reasonable moves, but I have no idea which one is best and if any are optimal.

I also don’t feel great about any of the local responses I’m considering. I feel like if I respond locally, I should expect to either lose sente or give away influence. For instance, if I play the knight’s move, then Black could take the 3-3 point and then I owe Black either an extension toward the side or attaching on top. If I attach on top right away, Black strengthens this direction and I start to erase some opportunities for continuation on the bottom side, which is something I’d prefer to avoid this early on. Losing sente isn’t unexpected given that I’m now a stone behind locally, but it’s early enough that I want better compensation for it.

I happen to know that Black’s slide doesn’t really pressure my 4-4 stone that well, as part of my general understanding of the 4-4 low approach slide. Knowing how AI likes to play, and given that Black’s last move isn’t putting that much pressure on me (it’s not a double approach or attachment), invading the 3-3 point is likely to be close to optimal here. I have no concrete rationale for this other than being used to AI analysis, but I’m confident that this is at minimum a viable move. As a bonus, I’m quite well versed in 3-3 invasion jōseki, so I’m comfortable that I’ll get out of the invasion without taking a loss.

If I invade the 3-3, the most likely outcome is that Black gives me the corner in exchange for sente, so Black will play first in the lower left. Drawing on the same principles as in my article on tenuki in 3-4 pincer jōseki, I don’t think Black can cleanly subjugate my 4-4 stone even with an extra move. The only reasonable options here are to pincer, which leaves me the option to escape to the center, or press from above, which gives me a clean escape to the left side, or take the 3-3 point, which cleans up a lot of the associated aji but is submissive and still leaves both escape routes. So I don’t think I’m risking any severe punishment, and I should be able to construct a reasonable (i.e. not significantly losing, which is honestly all I care about) game plan on the fly since I have some theoretical understanding of a similar type of position.

Fortunately, my instincts proved reasonable according to KataGo, and the 3-3 invasion is indeed a decent response to the slide. What’s interesting to me is that KataGo’s preferred option is B, the large knight’s move, by about +0.35 points. I don’t fully understand the rationale here, but my best guess based on the sample playout shown is that the large knight’s move is the best local continuation at progressing toward settling without needing to invest another stone or strengthening Black locally. The small knight’s move can’t settle without investing an extra stone anyway, so there is no point going with a narrow extension; this is in accordance with the principle that weak groups should extend far enough to claim enough space to settle. The large knight’s move also doesn’t force a local response out of Black (unlike, say, attaching on top), and following the principle “play away from strength”, Black should not bother to spend an extra move here either as now it has access to the 3-3 point for guaranteed life. So this manages to progress toward settling without damaging White’s prospects on the bottom side. It’s an interesting possibility that I would not have naturally thought of. Still, I am overall satisfied with the 3-3 invasion as an alternative answer.

I was not expecting the hane at R2, but I did happen to know that jōseki pretty well, and that connecting at R3 was a mistake, though not the precise numerical value. After White concludes the jōseki with the tiger’s mouth, Black’s outcome is inferior to the 3-3 invasion, knight’s move variation. White’s corner is more conclusively alive, the tiger’s mouth radiates strength on the right side that neutralizes the effect of Black’s strong group in the top right, and the hane-connect exchange doesn’t strengthen Black’s group enough to compensate for the fact that White’s corner is also stronger. The added White strength makes Black’s group weaker, as it no longer gets access to certain sente moves against the corner, and it is well-known that this Black group is a potential target for attack in other lines of the 3-3 invasion. So I came out of the jōseki pretty satisfied overall, and starting to consider possible continuations on the bottom side where I could look to exploit the potential weakness of Black’s wall. KataGo agrees that the outcome favors White, though I did not expect a number like W+7.3 - I was thinking that the advantage was more modest.

But of course, Black does emerge with sente, and comes around to make the third uncontested move in the bottom left. Here my first complicated positional evaluation task begins, ultimately (spoilers!) leading to my first significant mistake.

I think it’s interesting to view KataGo’s view of the position. I was surprised when I got to this position in the review, as I did not expect the position to be quite so sharp this early in the game. In most games there are many viable moves at move 26, but here KataGo says that there’s a pretty clear-cut answer. As such, this seems a good point to pause for a careful direction-of-play analysis.

First, let’s take a brief glance at the overall position. White’s stones overall are mostly fairly safe. The bottom right corner is alive. The top right group probably needs one more move to secure life, but it’s not under much pressure at the moment so it can easily be left alone. The top left corner is undeveloped, but it’s a 4-4 point so there’s nothing catastrophic that can happen to it if Black approaches (even twice), so while we can’t call it a strong group there’s not much weakness there either. The stone in the bottom left is obviously weak, but that’s by design so it’s ok.

As for Black, its group in the top right is alive. The wall in the bottom right is potentially a weak group as discussed previously, and White wants to exploit this fact somehow, at some point. The bottom left corner is fairly strong.

As an additional point of consideration, there remains some aji associated to the D4 stone, as it is not completely subjugated. Whether it’s a good idea to move it right this moment is a different story. What’s important here is that Black really wants to avoid capturing this stone in a small way (like E5) if at all possible. To do so would be overconcentrated, and there remains enough aji in the corner that Black is still subject to reduction. Ideally Black would like to capture this stone indirectly, basically for free as a consequence of stones placed elsewhere as part of another engagement. White, on the other hand, still needs to extract some value from this stone for its two tenukis to really constitute a gain over the local loss sustained here.

Therefore White has an important question on its hands that will probably drive the central narrative of the approaching middlegame: how can White profit from pressuring Black’s weakness on the bottom right while making sure that it can still extract value out of its stone in the lower right?

With this in mind, let’s examine some of KataGo’s continuations and get an understanding of its rationale.

First, the recommended move (A) is to pressure Black’s lower right from the third line, forcing it to secure an exit route to the center. White establishes a presence in the bottom side, neutralizing Black’s potential while taking sente; it then gets the opportunity to come back to H4 and threaten follow-ups in the lower left. The key observation here appears to be that Black ultimately needs two moves to fully subjugate the White stone in the bottom left, so there is no actual worry that White will lose the opportunity to exploit its aji (in this variation, by swooping under from the left side). Therefore White should pressure Black’s wall more directly and extract as much profit as it can before the opportunity disappears. Essentially, the lower right is urgent, because Black can settle the situation if given a single move, but the lower left is not because Black needs two moves.

The next best move, B, threatens the aji in the lower left more directly, and basically asks Black to finish subjugating the White stone in a way that is small. If Black agrees, then White can extend back toward the bottom right and return to pressuring Black’s weak group, and once again, the aji in the corner is not completely dealt with. It’s worth noting that this sequence and the previous, “optimal” one both get to play H4 and M3, just in opposite orders; the difference appears to be that in the optimal sequence, White gets an opportunity to further pressure Black. Black also has other options after B, such as attempting to settle the bottom right immediately, in which case White has several options now to exploit the bottom left stone in a big way. In the sequence shown, White settles in the center and takes sente while Black connects under in the corner. For having had a 3-to-1 advantage in the area, Black’s corner is surprisingly small, so this is acceptable.

The remaining moves which are a bit shy of B in terms of score appear to be pointing out that H4 is the best move in terms of pressuring Black to address White’s aji in the bottom left, while M3 is the best move to pressure the bottom right. The threat of F4 makes it hard for Black to address the aji after H4 without playing directly against the D4 stone, and M3 is the appropriate distance to pincer the bottom right even from a classical perspective. The other moves in the area live in a limbo where it falls short in one of these two aspects, generally both.

As it turns out, in the actual game I picked one of the worst among similar choices in the area.

Move 26 has the same problems as described in the suboptimal choices above. It has the further problem that it is on the third line, and White should fully expect Black to extend from the lower right and pincer, but in that case it is preferable that White be one line higher. White cannot expect to settle on the side and therefore is better off being a step closer to the center. This is an important principle of invading into strength and something I should have considered.

My rationale for J3 was that I wanted it to achieve a similar effect to H4: my guess was that Black would address the aji in the bottom left directly and leave it overconcentrated. But again, the urgent direction here was the bottom right as Black can take care of the weakness here with a single move, and Black identified this well.

That said, now I was committed and I decided to activate the D4 stone right away. Simply jumping out would not be sufficient as White would then form a floating group, allowing Black to attack and surround the D4 stone in sente, and it is not harder for White to lean against the bottom right group as it was recently strengthened. Again, it is important for this game that White prevents Black from taking the D4 stone for free. Fortunately this turns out to be the correct decision. Overall I was pretty happy to see Black connect under at C4, as I judged that this is not sufficient compensation locally in a corner where Black got 2 free moves.

My next mistake came a few moves later, when I attached on top at L4. Here I was expecting the hane-draw back-connect exchange, but neglected to consider if Black extends upward. This outcome isn’t terrible, but I was also not in a position to see it due to a major misconception I had. I was playing under the assumption that the D4 stone is not large and not worth connecting to. After all, it is only a single stone, and my group is still a ways from life while being beset on two sides by strong groups, so I don’t know that I can spare the time to connect to what doesn’t seem like a valuable stone. For this reason I had also not bothered to read out the aji against the corner; my mind was too biased toward treating it lightly.

KataGo’s analysis here is different. The D4 stone is still important for two reasons. First, the aji in the corner is not finished: C5 and C3 are locally sente and thus D4 still has a role to play in reducing the value of the corner. White can in fact activate this aji right away to overconcentrate Black’s corner while providing additional sente moves to aid White’s center when it needs to settle. Second, White is weak enough locally that pushing through at E4 is sente against White’s group. This means that if White doesn’t connect the D4 stone soon, then Black will have the opportunity to take it and erase the corner aji in sente. This means that White doesn’t quite have the time right now to be performing the L4-hane exchange. It doesn’t fundamentally solve White’s issue with settling - in fact, by strengthening Black on the right, it weakens White.

I seem to rely on the L4 exchange more than I ought to, mainly because I know what to expect. But that is problematic, because if I think through the outcomes carefully then I come to the conclusion that this move usually strengthens the opponent’s stones more than mine, especially if the opponent can afford to extend instead of connecting. Moreover it is often (as is the case here) a loss in terms of territory, so I should reserve it for more specific situations. This was not an appropriate situation for the L4-hane exchange because I have to settle the entire group, not merely the J3 stone. In the end, however, Black actually opted to extend instead of hane, and I felt ok with that exchange overall during the game as I don’t mind Black playing submissively when I am the one who’s locally weak.

H5 was a mistake with similar origins. Here I was still misjudging the value of D4, and thinking that it doesn’t really matter if Black captures it, so I am better off protecting the G4 cutting point right away. But this is bad not only because Black will happily take D4 and clean up the corner aji if it can do it in sente. Moreover, now that H5 is in place this also manages to overconcentrate White. Better options here include pressing Black further from above at M4, strengthening White’s center and also taking away the turn from Black, and connecting at E4, allowing the cut and starting a fight. These are viable because the exchange of L4 for M3 is a good one for White when a fight breaks out, allowing White to viably settle two groups simultaneously.

However, I don’t particularly regret my decision to stay connected here overall. At this point in the game, I was confident that so far I had managed to come out of the opening not necessarily ahead (I certainly didn’t think I was a full +4.9 points ahead as KataGo says), but at least with a reasonable outcome and within striking distance. Staying connected would avoid a fight on two fronts within the enemy’s sphere of influence, and I had already gained in the bottom right while also limiting Black’s profit in the bottom left to a reasonable degree. So while I wasn’t sure if staying connected would be a loss (I realized it was a few moves later after Black pushed through), I was fairly confident that it would not be such a loss as to overcome White’s gains and put White firmly behind. My feeling was that there was a good chance that this game would go all the way to counting, since no substantially weak groups were forming and significant one-sided local advantages seemed unlikely to form, so staying within striking distance of komi seemed an acceptable outcome and staying connected felt sufficient for that.

In any case, Black took sente again and approached the final corner, beginning the final phase of the opening.

Here I made another direction of play mistake: the pincer is not appropriate at this juncture.

My rationale behind the pincer was that I expected (as KataGo suggests) Black to take the corner, and I would take the outside and neutralize Black’s influence from the bottom left corner. But this is failing to adhere to basic direction of play. The bottom left corner is extremely strong and White has no business putting stones in that direction. There is little development potential for White here as White gets absolutely no sente moves in this area.

My main worry was that the other options, such as drawing back or the kick, were most likely to result in Black strengthening the group here, and then I would need to consider an invasion of the left side. But KataGo suggests that this is not something to be afraid of, specifically for the kick. Black cannot secure the left side in a single move after the kick, and in addition White has forcing moves to build influence on the bottom side since Black responded submissively to L4. Taking all these into account, White doesn’t need to fear strengthening Black via the kick, as it can freely invade the side anyway.

There are two important points to note here, which I failed to in the actual game. First, I failed to notice that White’s bottom side group, now being quite strong, was a major source of influence, so I could took toward playing for a moyo in the center (as KataGo recommends in the kick variation). I was still in the mindset of thinking about this group as potentially weak, and moreover I was still in an opening-oriented mindset and focusing on sides and corners. I think there are two takeaways here. One, when the strength/weakness designation of my group changes, I need to re-evaluate this designation more quickly, and re-evaluate the consequent impact on the global position more quickly as well. Two, I need to be faster at the transition from an opening-oriented mindset to a middlegame-oriented one, which ultimately means I need to be thinking a few moves farther ahead in time. Here, by being too focused on opening goals like side and corner territory, I overly neglected the possibility of making profit in the center, leading to overlooking the kick and its immediate followups.

Second, I failed to recognize that all of White’s groups are fairly secure, and thus I should have been not be afraid to strengthen Black to begin with. Since I would have had only one weak group to take care of post-invasion, I should have been confident in my ability to emerge from an invasion without taking serious damage. Instead, I erred too strongly on my habitual side of not jumping into opposing strength, which is another reason why I dismissed the kick prematurely.

This misconception would back back to haunt me again. I did not expect the Black jump, though that wasn’t a problem; I knew that this move is a local loss. The problem comes after the slide: I should have thought through why the jump is a local loss. If I had, then I would again have realized that as White is generally strong around the board, it is not difficult for me to deal with an attack on the pincer stone. In addition, because Black played the one-space jump, it has to be a little selective on how it chooses to attack the pincer. Therefore the left side is not that urgent and I can afford to take the big corner point. I also failed to take into account that taking the 3-3 point as Black weakens the entire top side overall, and begins to expose the weaknesses of the top right (such as N17), so it’s very close to (if not actually) sente for Black. This also fully settles Black’s group on the top left, making a subsequent attack on White’s left group more powerful.

For move 46 I considered blocking Black off, which is the locally optimal move, but it seemed obvious that Black would then play the top side, which KataGo confirms is a valid direction of play. In particular I was concerned with what happens if Black splits my two groups, as I would have to settle separately (an option KataGo confirms is viable, though it’s not shown here). But again, I think I failed to account for the strength of my stones here. After the two-space extension, the White group on the left is fairly safe; it is in the midst of Black strength, but it has room to slide to B7. So after Black splits the top and I settle one of the two top groups, I am again working with essentially only one weak group. Taking into account that I should have sente moves against Black’s splitting attack, and that I have pretty good support in the center, I should have been willing to at least consider the fight rather than immediately opt for a guaranteed local loss. Moreover, there are enough vulnerabilities in White’s formation that Black can profit without starting a fight, as we see in the sample AI continuation above.

I was mostly focusing on the potential fighting variation, however, and for me it was fairly difficult to evaluate the fight at the time. In such an event, the board would likely finalize into one where lots of little groups settle separately and nobody has big chunks of territory anywhere; I find these types of games relatively hard to count. By claiming the top side I could instead give myself a fairly sizeable chunk of territory that I can evaluate easily, and I could instead try to approach the middlegame with a clearer understanding of the win conditions. I think this decision was strategically reasonable at the time, but to improve I think need to develop enough strength so that I can play a bit more tight in these situations rather than play while absorbing a loss that I’m fully aware of.

We conclude our opening review at move 46, right before Black provokes a fight to signal the start of the middlegame. There is quite a lot about the middlegame to say as well, with the lead changing multiple times and the game never really getting out of hand for either side until Black misplays toward the end, but this review is quite long enough already.

Conclusion

To conclude, let’s summarize the actionable ideas I’ve extracted from this review.

• Tenuki is pretty viable in the opening, and honestly pretty fun! I should consider it more often in situations where I don’t know the local response but I’m not immediately under pressure to respond locally.

• The principle of extending far enough to secure eyespace applies in more places than I was aware of.

• If one enemy group can fix its weakness with a single move and another group needs two, this pushes attacking the first one up the priority list. I’ve got some time to spare before attacking the second group.

• I don’t pressure the wall after the 3-3 invasion as often as I should in most games.

• I should think carefully about the surroundings before entering my opponent’s sphere of influence on the 3rd line. I should consider if I can realistically expect to settle on the side or if I’m more likely to need to run to the center.

• I should check if the opponent can make me overconcentrated via their sente exchanges.

• Against a one-space approach from the side on the third line, reconsider before attaching on top. The hane-draw back exchange often favors the opponent locally, so I need a good reason to play it.

• Aji has concrete value, I need to read it out before I decide on whether I allow it to disappear.

• Group strength and weakness changes dynamically. I should be quick to re-evaluate, and be quick to factor this into my positional evaluation as well.

• It is worth it to start thinking about the transition from opening to middlegame early. Look ahead!

• It’s not that dangerous to play with a single weak group. I need to make a conscious effort to count how many I’m really working with.

• Backing off to reduce the game to an understandable position isn’t inherently a bad idea, but I should try harder to understand the tighter but riskier plays.

Return to the Reviews page.

I am starting a reviews category on this blog, in which I review my games with the help of AI analysis as well as (hopefully) input from stronger players.

Like anybody else, I have my own philosophy on what elements go into an effective review. I also have my own opinion on how best to incorporate AI analysis into reviewing. In this post I’m going to explain my overall thoughts on both of these topics, as well as how they will influence reviews that I write for this blog. Naturally, this post will be more subjective than most on this blog, so I ask that you not take these words as gospel. My opinions are also liable to shift over time as I improve at the game, and I may return to this post at a later date to adjust my thoughts and recommendations.

How I review in the age of AI

I focus on strategic mistakes in the opening and middlegame over tactical mistakes in complex situations.

This is as much a reflection of my personal playstyle as it is a philosophy on what I consider important in reviews. Although I do not dispute the importance of correcting tactical errors, I believe that for amateurs around my level and below, there is more benefit to studying the mistakes that occur before the game devolves into complicated, tactically sharp positions. This means that I generally focus on issues such as direction of play, jōseki choice, and strength/weakness analysis, and less on what the correct sequence of play was during an important middlegame fight.

My rationale here is that I think these types of strategic concepts in the opening are areas where we can draw the most actionable lessons from during review. Scenarios in the opening often re-occur with relatively minor changes across multiple games. Correcting for strategic mistakes we make in one opening scenario often leads us to better play across other games going forward. But a tactical error during sharp fighting isn’t always as applicable across many games. Sometimes, it’s hard to draw a lesson from such an error other than “I misread, I should have read more deeply,” since outside of shape issues it’s rare that a fight develops in the exact same way across two amateur games, and small differences in board position have an outsized impact on local fighting. And sometimes, the real error in such a scenario comes from earlier, in allowing the situation to develop in the first place.

Moreover, opening mistakes are often relatively straightforward to recognize and to correct compared to tactical errors in complicated fights. For example, it takes a strong player mere seconds to identify a direction of play mistake in most opening positions. The remedy to a direction of play mistake is primarily to identify what the correct direction of play is, which also often doesn’t take long, as it is based on a fairly simple principle (“play away from strength”). Once this is finished, we’ve done about 90% of the work, since in many cases we can already get significant improvement by merely playing anything reasonable in the correct direction, even if we don’t find the absolutely optimal answer. A mistake during a complicated battle, on the other hand, often requires exhaustive consideration of deeply read variations. Not only is this far more effort spent on a single move compared to a direction of play error, it has relatively low returns. Even if I find the correct variation in the review, in many cases the mistake is highly specific to that game, and in any case I cannot fully address problems in my game arising from shallow reading via review alone. In actual play I cannot expect to read as deeply as I can during reviews, or with the leisure that unlimited time allows me, or with the advantage of being able to play out the variations in front of me.

Tactical errors are not to be dismissed off-hand, but I generally focus my tactical attentions on issues where general principles can apply; the primary example is good shape vs. bad shape, and recognition of life/death, all before the complicated fighting begins.

Small mistakes according to AI are often the most important mistakes.

This is in line with the earlier statement, just with AI analysis now adding additional context. We often tell players looking to implement AI analysis into their reviews to focus on the big mistakes (commonly we point at -5.0 score as a boundary) and less on the smaller mistakes (like -1.5). Most AI review interfaces (such as provided in KaTrain, or implemented in places like AI Sensei) do something similar, setting a point cutoff to define a mistake. There is some amount of sense to this suggestion, as it is relatively rare for amateur games to be decided by narrow margins like -1.5, so why bother focusing on these small mistakes over bigger ones?

However, I think this is shortsighted for all the reasons mentioned above. Generally, the strategic mistakes of the types mentioned earlier occur in the opening and early middlegame, and generally they constitute relatively small losses. Even larger direction of play mistakes in the opening are often around the -2.0 range. However, not only do I consider these the most actionable mistakes, I also think that these numbers are misleading to some degree.

When an AI tells us that a mistake is worth -2.0 points, it means that it is -2.0 points assuming a superhuman AI takes over for both players from that point on. But when a human commits a strategic error, it is rare that the human plays perfectly from that point forward. What is much more likely is that one strategic error cascades into another error or into a concrete disadvantage that generates further losses. Part of this is that we do not always recognize a strategic error after we play one. Another part is that such mistakes are often born of a misconception about the game state, and unless that fundamental problem is corrected we are liable to continue to act on said misconception and commit more mistakes.

Your mileage may vary here, but my personal preference is that I play a relatively error-free opening and early middlegame, even if I am weaker when coming to individual situations in the middlegame. Ultimately, nothing I can do with AI analysis replaces the need for reading ability in the middlegame. And I prefer to make those tactical errors in the middlegame rather than come out of the opening with a crippling disadvantage and not have the opportunity to commit those middlegame errors at all.

AI isn’t the solution to everything.

I love AI analysis, but it’s important to realize that it cannot be a substitute for all other aspects of improving at baduk. My opinion is that AI analysis is particularly adept at pointing out misconceptions in opening strategy and shape, and these are consequently the areas where amateurs can derive the greatest benefits from AI. But it cannot, for example, substantially improve your raw reading ability. The best AI can do is show you the sequences it has read out. But nothing improves reading ability more than reading out moves yourself; watching someone else do the reading for you is quite useless. Similarly, AI can point out life and death issues that arise during the game, but nothing will substitute simply knowing what the basic live groups are or having the raw reading ability to determine group status yourself in-game. Thus AI is not a replacement for tsumego, or for sequence study, or for memorizing common life/death scenarios. Consequently, when I review I tend to de-emphasize aspects of the review that are better addressed by these other activities.

It is not always helpful to focus on playing optimally.

When I break with the AI’s recommendation it is often because I distinguish between the optimal move and the move most likely to help me win the game. AI recommends moves under the assumption that a superhuman player is playing on both sides, and thus either side can realistically expect to generate significant advantages over the other. However, I am a human playing other human players, and thus I am fully willing to take on even significant losses in score in order to create better opportunities for victory. Sometimes that means intentionally playing suboptimally when I have the lead, in order to produce a more stable position that I understand more completely. Sometimes that means intentionally overplaying because I am behind and I must take a risk instead of playing the theoretically optimal sequence toward a guaranteed loss. I believe it is ok to disagree on such grounds with the AI during a review. The AI is an assistant; it is not a god dictating the law.

There are prerequisites to studying effectively with AI.

This is likely going to be a controversial opinion. I believe there is a minimum level of game knowledge before one can obtain concrete benefits from reviewing with AI assistance. In particular, I think there is no point to using an AI for reviews until you understand the following concepts:

• strong and weak groups;

• attack and defense;

• direction of play.

If you do not understand these concepts, you do not have the appropriate background to accurately interpret the AI’s suggestions, and you are better off continuing to learn via actual play, human teachers, book study, and tsumego for the time being. And though you might argue that one does not need the ability to interpret to derive some benefit from seeing the AI’s suggestions, I would disagree strongly on this point. I think one derives very little lasting benefit from merely viewing sequences. The only way you can hope to apply this type of training to your games is to memorize all AI responses from a given situation and hope that you are never caught in a position that has deviated from a position you know.

This is not only unrealistic from a sheer human memory capacity point of view, it is inferior to looking to properly understand the AI recommendations from a memorization perspective as well. Human minds are much better at committing large amounts of information to memory if they can provide a rhyme and reason to how that information is structured. The player that understands why the AI suggests its moves is much more likely to remember what the AI suggested several games down the line. This is like how one derives very little benefit from memorizing jōseki sequences. It is well-established that encyclopedic knowledge of jōseki is inferior to correct judgment of direction of play and applying it to jōseki choice with a much smaller pool of jōseki.

And although one might hypothetically be able to reason out justifications for an AI’s moves without understanding the 3 fundamental concepts given above, the much more likely scenario is that such a player will only be able to provide an incomplete justification at best, at worse a flat-out inaccurate one. This is also not helpful, and if anything it could be actively harmful. Misconceptions about the game inevitably lead to errors down the line, even if those misconceptions have their roots from a superhuman AI’s suggestions.

Does that mean I am fully prepared to be studying with AI? Not necessarily, no. At my level I am still overall better off studying with a human teacher, and I have a lot to gain still from non-AI tools. However, I like to believe (and hopefully this will be corroborated in the reviews) that I am at least capable of conducting a useful review with AI assistance.

I like to believe that I have a sound understanding of the basic concepts listed above, and that my problems with them largely come down to consistent application, but it is important to be wary of overconfidence. Thus I do not always entirely grasp the rationale behind an AI’s suggestions (and if I did, I’d be at a significantly higher rank). But I am fairly confident that I can produce a reasonably accurate and partially complete explanation for these suggestions in sufficiently simple board states, and that is a fair place to get started. I can always turn to other, stronger players in order to complete these explanations.

How reviews will work on this blog

Reviewing on a blog presents its own set of challenges that I’d normally not have to deal with when reviewing purely for myself. I’d like to provide a level of detail in reviews that is helpful for players below my level, but this is fairly exhausting work to do for a full game. I expect most reviews to cover the opening and perhaps the early middlegame, but going beyond that may bring posts to lengths that are frankly unsustainable for me. Still, the opening and early middlegame are where I prefer to focus my attention anyway, so this isn’t necessarily a bad thing. There’s always plenty to say about this stage of the game. The focus will be on extracting concrete and actionable insights and I will try to summarize them at the end of each review post.

As mentioned previously, I am not the strongest player and I am not always going to come up with the most correct or complete interpretation when I employ AI analysis in my reviews. As such, I welcome comments from stronger players on anything I say, and with their permission I will incorporate their comments into an updated review whenever possible. This is as much a learning experience for me as it is content for an audience. I may consider performing reviews for other people as content but naturally I will limit these to players below my rank.

Return to the Reviews page.

The directional bias of 3-4 approaches

Barring an urgent local situation, approaching an unenclosed 3-4 point is almost always one of the biggest moves on the board, an idea which AI analysis has only reinforced. As such it is essential for every baduk player to learn at least some portion of the myriad 3-4 approach jōseki, with the one-space low and high approaches being standard elements of every player’s repertoire. Some standard continuations from these approaches are given in the above diagram.

However, the standard approaches do suffer from two major flaws which you need to be aware of in high-level play. The first is that typical approaches to the 3-4 point are inherently directionally biased. Unlike the 4-4 point, there is no symmetry axis with a 3-4 point. As such, the approacher in most relatively peaceful 3-4 point jōseki has a tendency to gravitate away from the side with the original 3-4 stone, as is the case for White in the two examples in the diagram. If we assume that the direction of play in the diagram is the bottom side, then it can be difficult for White to get stones in that direction starting from a high approach without giving up too much locally, especially if Black is also contesting this direction of play. As for the low approach, getting stones on the bottom side in a natural way is nearly impossible using standard methods.

The second problem is that most approaches to the 3-4 point are not necessarily sente. Although in most early opening scenarios one should respond to a 3-4 approach, it is not uncommon for there to be a sufficiently important play on the board that overturns this heuristic. Part of this is that ignoring a standard 3-4 approach is not all that dangerous locally. With the low or high approaches for instance, ignoring the approach merely transposes to a 3-5/4-5 point, 3-4 invasion jōseki: the approacher cannot force any extremely significant local advantage if the defender plays elsewhere once. In one line, responding to a standard 3-4 corner approach is big but not necessarily urgent. This ultimately ties back to direction of play: if the direction of play is elsewhere and sufficiently important, it is acceptable to tenuki once from a distant 3-4 corner approach.

To summarize:

Standard approaches to a 3-4 corner can be bad at controlling the direction of play.

This means that it is important for you to have options in your library that give you agency over the direction of play as the approacher to a 3-4 point; we saw how crucial this can be for the 4-4 point in our article on the 4-4, 3-4 invasion. For all the strengths we discussed in our recent article on the attach-draw back jōseki, you cannot play high-level baduk if this is your only option on the approach. Therefore in this article, we will begin developing some options for the approaching side to dictate the direction of play against a 3-4 point, with a particular focus on the 4-4 attachment.

Three methods of choosing a direction

In the above diagram, we begin with a Black 3-4 stone and assume that White wants its stones to be directed toward the bottom side. White has three typical ways to play locally in a way that may allow for this to happen.

The first option is approaching from the bottom side directly, e.g. A. This is a situational move that was most popular against Chinese-style fuseki, but such fuseki largely disappeared from top professional play in the post-AlphaGo era, taking this approach with them. From a purely local perspective it is a loss for White: allowing the corner enclosure is very large and the approaching stone is unlikely to get equal compensation on the side even with another move. It should be viewed as part of a special strategy rather than as a standard approach: White prioritizes the bottom side so heavily as to allow Black a pretty big corner enclosure in exchange.

The second option is B, the high approach into the avalanche. AI’s most likely local response to a 3-4 one-space high approach is to attach below, as this is the most that best secures a decently sized corner against the high approach. Instead of going into the attach-draw back jōseki, White instead plays the 4-4 point and enters the (pre-)avalanche shape. This is one of the most common methods used by AI engines to alter the direction of play against a 3-4 point; however, it has two disadvantages. The first is that Black need not go along with White’s wishes and make this shape: Black could move out along the bottom side first, or even tenuki. The second is that this sequence can lead to the small avalanche, a notoriously complicated jōseki. Complicated jōseki are not only more challenging to learn, they tend to be very sensitive to the global board-state. Even if the avalanche gives you the correct direction of play, it is possible for you to get an unfavorable outcome based on several other factors on the board. Therefore it is a bit difficult to rely on the avalanche as a go-to method to change the direction of play.

The 4-4 attachment is a good middle ground. Unlike an approach from the side, the 4-4 attachment does contest the corner and prevents the defender from getting a major local advantage. Compared to the high approach, the 4-4 attachment discourages tenuki and has a better shot at giving you a position in the direction of your choice: moreover, the lines are generally fairly simple. We will see that not only is the 4-4 attachment a situationally viable way to approach a 3-4 point, there are times when it far outshines any other standard approach.

The 4-4 attachment

We can summarize the idea of the 4-4 attachment as follows:

1. The 4-4 attachment offers the defender a choice: take a locally favorable position at the cost of the direction of play, or give up the locally optimal outcome to contest the direction of play.
2. Compared to the low or high approaches, the 4-4 attachment demands an immediate response.
3. As the first corner approach of the game, KataGo rates it as about a 0.4 point loss - it is not good to play it willy-nilly, and the loss must be made up by direction of play.

There are three main lines to the 4-4 attachment: the hane (A), inside extension (B), and outside extension (C). Before we describe each line in detail, let’s get a basic understanding of their implications.

The basic idea of the 4-4 attachment is to offer the defender a choice between the hane and the other two options. The decision breaks down along the following lines of thought:

1. From a purely local perspective, A is the best option. It is the response that keeps the largest corner and in fact the outcome is locally favorable for the defender. However, it also introduces a symmetry into the position and thus gives control of the direction of play to the approacher.

2. B and C let Black continue to contest the direction of play. However, in return Black does not get as large of a corner - from a purely local perspective A is preferable.

Main lines

Hane

The hane is the main line to understand. It creates a local symmetry in the position which hands over the choice of direction to the approacher (White in this case). White will hane in whichever direction corresponds to the direction of play.

A few different continuations are possible by Black, but there isn’t any pressing need to learn all of the continuations in the diagram. If you want to keep things simple, then this is all you need to remember:

1. Atari after the hane (move 3, A). White connects and Black connects the corner. White can tenuki from here or push a few times.

2. Do not try to cut White. Black has two weaknesses to cover after the cut but can only cover one.

The result after A is slightly favorable to Black. We can see this via tewari analysis, comparing to the 3-4 low approach kick jōseki. White has made an unnecessary exchange that removes some of the aji in Black’s corner and filled in one of White’s liberties. This is certainly a significant advantage, but not game-breakingly large. In compensation, White gets to settle in the direction of its choice. Before entering the hane line, Black needs to judge carefully whether this exchange is worth giving White the correct direction of play; of course, the decision will depend on the surroundings heavily.

Inside & outside extension

If Black does not want to totally cede control of the direction of play, then its main options are to extend either toward the corner or the outside. Extending toward the corner transposes to a 3-3, 4-4 approach jōseki, which can then proceed in many ways; we show only a few lines to illustrate the possibilities. Extending toward the outside generally gives White a group facing the right side, away from its desired direction of play. However, in return Black cannot get as large a corner as in the attach-draw back jōseki or the hane variation discussed above. White does not necessarily need to finish the sequence in either case: simply forcing one of these two responses already limits the size of Black’s corner and gives White a useful outside stone.

Simulated example

Using KataGo, we simulate a scenario in which the 4-4 attachment is a viable approach option to the 3-4 point, and show how play continues.

Here the point of concern is the incomplete double low approach jōseki that has been played in the top left corner. White and Black each have a weak group here, White on the basis of having at best one eye and a weakness at E16, and Black by virtue of having a single stone close to stronger White stones. Therefore the direction of play here is clearly the top left. As such, it is possible to play the top left directly, and indeed KataGo rates moves such as B15 as among the best options.

However, there is also an open 3-4 corner in the bottom left, and in the context of this article we ask: what happens if we approach first? KataGo’s response is quite clear-cut: both the low and high approaches are significant mistakes, losing anywhere from 0.8 to 1.0 score. However, the 4-4 attachment is considered right up there with B15 as another essentially optimal continuation.

In this particular case, the top left is so urgent that KataGo is willing to tenuki in response to the low and high approaches: after all, even if Black plays twice in the bottom left, at worst this only transposes to a 5-3/5-4 point, 3-4 invasion jōseki. In fact, against the low approach KataGo is even willing to tenuki three times from the bottom left corner: spending four moves to capture the corner is not worth it for Black at this early stage, especially given that White still has some aji remaining there.

The 4-4 attachment invites White to take the locally favorable trade in the bottom left, in exchange for Black gaining some extra strength on the left side. Normally in the 4-4 low double approach, the outcome is acceptable to good for Black if Black can settle both sides cleanly. Meanwhile, since White starts on the back foot (having one fewer stone locally), White wants to settle as cleanly as possible and usually does so by attacking whichever of Black’s two groups is weaker. By gaining some strength in the bottom left, Black discourages any severe attack (such as a pincer) by White against its stones in the top left. Instead White is forced to stabilize its group by pressing from above, which lets Black settle cleanly while taking a little territory to boot. As a result, Black essentially uses its locally unfavorable outcome in the bottom left as payment for a locally favorable result in the top left.

Not all examples are so clear-cut as this one, as in this example the top left was particularly urgent. In more typical scenarios, the difference between the 4-4 attachment and either the low or high approach is often very slight, on the order of 0.1 score. As a rule of thumb, the 4-4 attachment as the first corner approach is a loss of about 0.4 score, but if there is a weak group in the corresponding direction then it is often enough to bring it to at least even with the low or high approach.

Conclusion

As we have seen from the example, the standard approaches to the 3-4 point are not always viable options due to their lack of immediate pressure and lack of ability to transition the direction of play. As always on this blog, we emphasize the importance of finding the correct direction of play, and having options to shift the direction of play when necessary. For that the 4-4 attachment is an essential tool, and is relatively easy to learn as well. The key to making good use of the 4-4 attachment is to get a feel for when direction of play considerations outweigh the local sacrifice intrinsic to this move.

Return to the 3-4 Jōseki page.

This is the fourth article in the series “The 3-3 Invasion.”

Previous article, Next article

Outside & sente vs. outside & sente

In the previous articles of this series, we gave a fairly thorough account of the double hane line of the 3-3 invasion, which (to summarize) results in the defender taking the corner in gote, while the invader gets a ponnuki on the outside in sente. We now turn to the other lines of the 3-3 invasion, which generally involve the defender taking the outside and sente. For the reasons we discussed in the first article, we restrict ourselves to the lines where the defender extends from the push, as shown in the diagram below.

In later articles we will discuss each line individually in detail. However, in this article we will take a moment to answer a question that you’ve probably asked in your own games: what’s the difference between the knight’s move and the push? After all, both results appear similar on paper: the defender (Black in this case) takes the outside in sente and gets to choose the direction, while the invader (White) takes the corner in gote. Does it matter if we choose one over the other, and if so, what ideas go into the choice? By the end of this article, we will hopefully have a good understanding of this common early-game decision.

The short version: It really doesn’t matter

Let me just put the answer up front:

The distinction between the knight’s move and the push is extremely minor and realistically only matters in professional play. You can play whichever one you prefer and be perfectly fine deep into dan level.

If you’ve got things to do, you can feel free to stop reading here: with complete seriousness, that’s all you need to know. Our initial assessment of the difference between the two jōseki is correct: the most important aspect of these jōseki is that the defender chooses the direction of play and takes the outside in sente, while the invader takes the corner in gote, and what differences exist are fairly minor relative to this fact. So for example, if you only play the knight’s move as the invader, nobody will yell at you and tell you you’re totally wrong, including our lord and savior KataGo. Most of the time, if one is indeed better than the other, it is only by a difference of 0.1 score, and almost never more than 0.2. The most important reason to learn both lines is simply that half of the time, you will be the defender rather than the invader, so you at least need to know how the defender responds to both lines.

Consequently, this article isn’t going to be the most important article in this series. We are having this discussion mainly as a fun little side excursion to warm up for the detailed analysis of the individual lines. However, it will still be instructive for you to follow along, as we are going to introduce some basic concepts that we will refer to when we do start these individual analyses.

However, does that mean that there’s literally no difference between the two lines? Not at all! There are fairly specific, albeit subtle, differences between the knight’s move and the push, and we will be spending the rest of this post describing them: there are some interesting strategic concepts here that are good to keep in mind as we continue learning about the 3-3 invasion. It’s just that if all we’re concerned about is winrate and score, then these differences are still sufficiently minor that amateurs can get away with either jōseki and come out perfectly fine.

The medium version: The knight’s move is never bad

For those of you who don’t have anywhere to be right now, but also aren’t interested in watching me explain something that literally only makes a 0.1 score difference, here’s the working summary of how the choice between the knight’s move and the push works.

1. In the push variation, it is often a priority to be the player that returns to the corner first. In the knight’s move, it is rarely urgent for either side to continue locally.
2. The knight’s move is an all-purpose option. It is essentially never a mistake: neither side can expect to get a significantly favorable or disadvantageous follow-up.
3. The push is slightly more swingy. Given the right board-state, some lines can be judged to favor one side over the other.

Ultimately, what this means is that if you don’t want to think too hard about your jōseki, you might want to adopt the knight’s move as your main option. If you know the lines of the knight’s move well, it is basically impossible for you to make a serious mistake and come out with an uneven result. You don’t need to worry all that much about timing a local continuation, since it is relatively rare that a local continuation from a knight’s move is overwhelmingly better than any other option. However, it also means that there is very little opportunity for you to punish your opponent for a mistake. It is a jōseki with a very even keel.

With the push the probability of an unfavorable outcome increases slightly, and you need to match it to the global board-state a little more carefully. However, if you play it well, you also have the possibility of coming out slightly ahead. It’s a jōseki where you want to know the ins and outs pretty well to get the most out of it. Realistically, if you’re at the level where you’re capable of getting the most out of the push variation, then you’re way too strong to be reading this blog.

The long version

Now, let’s finally get into the details. The nature of the difference between the two jōseki can be stated quite briefly:

In the push variation, the invader owes an extra move locally, and the jōseki has a tendency to favor whoever returns to the position first. In the knight’s move variation, neither side gets any significant local advantage by playing the first follow-up.

Explaining this dichotomy, however, requires us to understand some basic ideas specific to these jōseki. The first of these is to understand the strength of the groups involved.

Why is tenuki possible for the defender?

To explain the strength of the groups involved in each jōseki, it is instructive to explore the question: why is Black able to tenuki at this stage? Let’s first consider this question for the knight’s move.

In the knight’s move variation, Black’s group is able to withstand quite a lot of pressure from White before having to respond. This is because in the knight’s move, Black has access to sente moves against the corner which revolve around the critical shape point at the 3-2 point and the jumping tesuji at the 2-5 point. Since Black’s group is able to survive multiple approaches from White with ease, it is considered to be a strong group.

The push variation tells a similar story. Even if White attempts to pressure Black, the weakness at the 3-2 point means that Black has sente moves against the corner to help it settle. Therefore the Black group in the push variation is also considered fairly strong. Of course, in both cases it is also fine for Black to extend from its wall before it comes under pressure: the point is that Black is at least not obligated to do so.

If the invader plays first

So we see that in terms of the outside standing up to pressure, both lines are essentially equivalent. The first difference between the two will arise when we consider how each line evolves after each player returns to the corner down the line, and can be summarized as follows:

The invader gets a lot out of playing first in the push variation, and not that much in the knight’s move variation.

First we will consider how the knight’s move variation develops if the invader gets the first move locally. The typical local continuation for White is the diagonal move, which aims to pressure Black by repairing some of White’s aji and taking away Black’s aforementioned escape routes. However, here we run into a weakness inherent to the knight’s move: because of the low position of White’s stones, and because White still needs to mind Black’s jump tesuji in the corner, ultimately White can only exert so much pressure on Black with this move. We see here that White is unable to mount a fierce resistance if Black decides to press from above, and ultimately Black’s stones remain fairly secure. The takeaway here is that in the knight’s move variation, the invader tends not to get a whole lot locally even if it gets to play first. Consequently, in high-level games this diagonal move tends to go unplayed until the mid-late stages of the opening.

However, if we run the same exercise for the push variation, things change rather drastically. When White plays first in this jōseki, White gets a nice atari in sente and a very important descent to the 3-2 point (funny how that point keeps coming up, eh?) which is deceptively large by virtue of accomplishing important goals simultaneously. The takeaway here is that in the push variation, the invader gets quite a lot out of coming back to the corner first - the first significant departure from the knight’s move. This is a sequence that is well worth playing early if the invader gets the opportunity.

If the defender plays first

Now let’s try this exercise again, but this time assuming that the defender gets the first move locally. This discussion is a bit more complicated, because there are many ways for the defender to play locally, especially in the knight’s move variation. But the executive summary is as follows:

The defender can essentially force a sequence in the push variation which is favorable with the ladder. In the knight’s move variation the defender cannot force such a favorable sequence, and the invader can generally come out with an even result.

The idea here is to note the key shape differences from the defender’s perspective in both variations. In the push variation, the invader’s group is pressed up against the defender’s outside group, and consequently has a noticeable shortage of liberties if the defender takes the 3-2 point, which we noted comes to great effect when the defender needs to settle against pressure. The invader effectively owes a move locally (the 3-2 point), and with first move the defender can collect on this debt. As a consequence of this shortage of liberties, the defender is better able to dictate the invader’s response. The knight’s move variation essentially removes two stones from the push variation, which adds an extra liberty to all of the defender’s stones. This extra liberty makes it hard for the defender, even with first-move advantage, to force the invader into any particular sequence, so the invader is generally able to dodge sequences it deems unfavorable in light of the global board-state.

The above diagram shows the forcing sequence in the push variation when the defender plays first. Because of White’s shortage of liberties, it is quite difficult for White to deviate from the eventual double hane, which allows Black to take advantage of its favorable ladder. Moreover, note that in playing this forcing sequence, Black is also denying White the first move, which we previously noted to be big for the invader in the push variation.

For the knight’s move variation, the defender has a multitude of ways to play the first move locally, but options B and C in the above diagram are fairly peaceful. Black generally lets White leave easily in the corner in exchange for expanding its moyo, and none of the outcomes significantly favor either side locally; as usual, the status of such territory-influence exchanges comes down to the global board-state. Option A is the most combative option, and there is a single line here in which Black benefits from a favorable ladder. However, White has the freedom to dodge this line, at two separate points no less. Although A takes the same critical 3-2 point that we keep coming back to, in the knight’s move variation taking A does automatically give the defender all the agency in the choice of jōseki.

The game plan

So, to summarize:

1. In the push variation, returning to the corner first tends to be fairly valuable. With first-move advantage, the invader gets to play a very large descent, while the defender denies this descent and can potentially force a locally favorable jōseki with a favorable ladder.

2. In the knight’s move variation, returning to the corner first tends not to be all that valuable. The invader cannot do much to extract a huge advantage even with the first move, and the defender doesn’t have many ways to force any specific type of favorable response.

With this we are finally ready to describe how one chooses between the two variations, and how one develops their game plan around them.

The invader plays the push variation with the intent of spending sente fairly early on to return to the corner. This will usually happen via the defender spending sente to play a large move elsewhere, and the invader will look to arrange for this to happen rather than the defender playing the corner first.

The invader plays the knight’s move with the intention of leaving the corner alone for quite a while. Neither player will find it urgent to follow up in the corner first, and both players will wait until bigger moves have been played or some global need to decide the local situation quickly arises.

At amateur level, these differences are unlikely to really matter. Kyu players can’t be expected to consistently play the push variation optimally: you need a pretty sophisticated understanding of opening theory to successfully arrange for you to get sente just as returning to the corner becomes the biggest move on the board. But all you really need to know to effectively play the push variation is that returning to the corner tends to become the biggest move on the board pretty quickly, so you should make it a relative priority. And all you really need to know to effectively play the knight’s move is that you can use it essentially at any time, but unless you have a specific reason, you should stay away from the corner once it has been played as it is not urgent for either player.

Examples from simulated play

As usual, we end this article by putting our ideas to the test with the aid of AI.

In this KataGo experiment, we set up a typical opening scenario and simulate a number of games from the same position based on White taking the knight’s move or push variations in the top right corner. Our goal is to identify general trends in how KataGo changes its play based on which variation we select.

Our simulations demonstrate virtually all aspects of the ideas we have described above. When using the push variation, KataGo makes it a priority to return to the corner first, barring some specific cases as the defender in which the first move is discouraged by an unfavorable ladder and direction of play. In the top left corner, we even see an example in which Black manipulates the direction of play and a favorable ladder to discourage White from playing first. With the knight’s move variation, we see that KataGo is perfectly happy to let the rest of the board play out before returning to the corner, if it does so at all: the first time this occurs is top variation, move 30. We also see an instance of KataGo using the defender’s emergency escape option in the knight’s move variation when the defender comes under pressure from the outside.

Conclusion

With this we have covered the main strategic differences between the push and the knight’s move variations of the 3-3 invasion. Ultimately, at amateur level it is perfectly acceptable to use either one as you see fit. The important aspect of these variations is not how to choose which one to play, as they are nearly equivalent in value in most situations. Rather, the key point is to understand that though these jōseki are usually similar in terms of value, they do play fairly differently in the continuations, and you must adapt your play accordingly.

In the next articles in this series, we will start looking at the individual variations (knight’s move, push, one-space jump) in greater detail.

Return to the 4-4 Jōseki page.

This is the first article in the series “The Attach-Draw Back Jōseki.”

Next article (to be written)

The attach-draw back jōseki (main line shown above) is beloved by amateurs for its simplicity, yet it stands up very well to AI analysis and is a common sight in professional play as well. It is a fantastic all-purpose opening jōseki that every baduk player should have in their arsenal. Despite its ubiquity, however, it is common that players do not have a complete understanding of the significance of the position even all the way to SDK level. In this series we will aim to develop a fairly thorough understanding of the main lines of the attach-draw back jōseki and how one plays with them in the middlegame. The first article will serve as an introduction to the main lines and the overarching principles governing their middlegame followups.

Main variations and decision points

First, let us establish some basic knowledge of the main lines of this jōseki. We will only aim for surface-level knowledge at this point. Our aim here is just to learn enough to recognize which outcomes are jōseki, because as we progress through this series we will discuss why one might choose one variation over another. We leave a detailed discussion of individual lines to later articles. For brevity we will refer to the approaching side as “White” and the defending side as “Black” in this article, and the diagrams will be consistent with this usage; hopefully it is clear that what we say applies when the colors are reversed too.

The above is a substantially abridged tree of the attach-draw back jōseki, focusing on the main lines and its minor variations. Once we enter the jōseki (move 5) there are, roughly speaking, 3 main decision points - two for White (moves 5 and 7) and one for Black (move 6). The decisions that need to be made at these points are generally fairly similar between the variations: for instance, in all cases at move 7, White is mainly choosing between extending or tenuki, along with some more minor options which are more situational. We will address what ideas go into the decision-making process in later articles.

The primary objective

Ultimately, this means that most outcomes in the attach-draw back jōseki basically boil down to the following:

Black gets a decently-sized, fairly secure corner. White settles on the outside (possibly in sente, if White is willing to tenuki).

The key to playing the attach-draw back jōseki well is to constantly keep this outcome in mind and treat it as the goal, especially as White (and more generally, the approaching side). We will summarize these objectives in the form of two principles:

1. White just needs to settle in some form with decent shape to get an acceptable result.
2. White effectively has no territory on the side. It is not worth defending nor is it worth invading during the opening.

The most common failure with this jōseki in kyu-level play is to continue naïvely without consideration for how the follow-ups interact with these objectives. DDK and high-SDK players in particular are prone to interpreting White’s extension as territory and focusing too much of their effort into reducing/invading it or guarding against such reductions/invasions. Playing the attach-draw back jōseki requires you to make a very important step in your development as a baduk player: the appreciation of the hidden and incalculable value of things like tenuki, influence, and settling, and becoming comfortable with giving up concrete value (secure territory) for these things. In the rest of this article we will give a high-level explanation of how these concepts apply to the attach-draw back jōseki. We will revisit many of these ideas in greater detail in subsequent articles in the series dealing with the minutiae of individual variations.

Aside: These ideas are not unique to the attach-draw back jōseki. They have to do mostly with the fact that the extensions in this jōseki are three-space extensions, which inherently are more about settling than about securing territory. We would fight much harder for the side territory if we were working with two-space extensions. We develop these and related ideas in the article Choosing the Right Extension.

The extension is resilient

As a simple example, let us consider one of the lines of the attach-draw back jōseki in which Black invades the extension.

In the secure connection, high extension variation of the attach-draw back jōseki, the above diagram describes the standard invasion. (The details do change if Black has another stone nearby, but let’s keep the discussion simple for now.) Attaching to the invasion stone on the side is a fighting-oriented option, best used if White has support nearby. Here, however, Black is the one who has support from the corner, so White’s best option is to attach on top, letting Black’s invasion connect to the corner in exchange for thickness, decent shape, and in this case sente.

Depending on playstyle, kyu players have a tendency to dislike this result for White, going so far as to spend sente to add an extra stone here as White or invading right away as Black. In reality, this result is generally better for White, especially in the early game where influence and sente have the greatest value. The reason for this dissonance is that weaker players are more likely to think of the area under White’s extension as rightfully belonging to White, and thus it feels like a great loss when Black can slide under just like that. The correct way to think about this position is to remember our two principles: let go of the idea of making territory with White’s stones, and focus on settling with good shape. If we adopt this paradigm, then the invasion jōseki starts to make a lot more sense.

Let’s see this play out in practice with the aid of KataGo. Here we simulate a game in which Black attempts the same invasion at the earliest possible opportunity. Now, the invasion itself is not the best move on the board, but at only -0.2 score it is also not a major mistake by any means. However, if we take the jōseki to its conclusion then White has amassed a significant advantage of +8.9% winrate and +1.4 score. The big problem occurs at move 5 (all other moves are close to optimal). KataGo agrees with our assessment that finishing the sequence is far better for White given its extremely clean shape and sente, and its solution is to leave the invasion unfinished as a sente exchange. Since White is unlikely to spend an extra move locally to prevent Black from connecting back just yet, it is not as if the invasion has failed just yet for Black. Black can look for an opportune moment to resume the invasion sequence or exploit its aji in some other way, omitting all this funny business of helping White make good shape. In the meantime Black can spend its sente accomplishing bigger things on this early board, and perhaps even prepare a better boardstate to continue the invasion sequence. White is not too badly off either if Black plays elsewhere at move 5: the H2-H3 exchange is better for White in a vacuum, because it connects up White’s three-space extension and discourages a Black approach from the bottom right, which is much less likely to be sente now.

So resilient, you can tenuki

The great thing about abandoning any attachment to the side territory in the attach-draw back jōseki is that it opens up a world of options to us, the best of them being tenuki. If you’re not afraid of losing the side territory, that makes it a lot easier to ignore someone threatening to take it! This is an essential feature of the attach-draw back jōseki for high-level play: it does depend on the surroundings and the specific type of approach, but usually it is resilient enough for White to ignore most approaches once, as long as we stick to the game plan of just aiming to settle with good shape.

In the above diagram, White is playing with the secure attachment, high extension variation, and ignores Black’s low approach from the right. Black’s main followups consist of invading the middle of the 3-space extension, but there is nothing Black can really do to force White into a weak position. Notice how White has no interest at all in taking the side territory, nor in denying it to Black. Not having to waste moves on the side allows White to make good shape and settle with ease. The point of this diagram isn’t about memorizing the sequence of moves: instead, try to focus on what general outcomes White is aiming for.

Let’s see if these variations bear out in practice. Here White ignores Black’s approach in favor of enclosing its corner. This happens to be essentially optimal, by the way - KataGo likes the tenuki just as much as we do! That said, Black’s approach is also considered essentially optimal; just because White can ignore it does not mean that the move is bad for Black.

Right off the bat, KataGo disapproves of Black’s invasion, especially on the third line (which tends to result in a much cleaner shape for White and a low position for Black). This basically boils down to what we were saying with the previous AI simulation. Ultimately, all outcomes here lead to White settling with good shape on the outside, while Black at best manages to carve out a small chuck of side territory (which, remember, we weren’t thinking of as belonging to White in the first place) and slightly enlarge its corner. Keeping in mind that White has played elsewhere once, these outcomes are eminently playable for White. In fact, KataGo considers all of the above outcomes as fairly significant improvements from the starting position.

Also, just to drive the point home regarding the side territory: after Black 1, KataGo considers all moves defending the side territory to be serious mistakes for White. K2, for instance, consists of a change of -1.0 score for White (which, this early in the game, makes it one of the worst among semi-reasonable moves for White). It would mean that Black’s initial approach at M3 is a sente exchange, in which case Black gets the better end of the deal since White is overconcentrated; and even after all that, there is still a ton of bad aji at H2, so White will probably be forced to spend an extra move locally at some point. So not only is defending the side territory as White unnecessary: there is a good chance that it is an outright mistake.

Conclusion

To summarize the above discussion:

1. The main objective for the approaching side in the attach-draw back jōseki is to settle on the outside with decent shape. Most outcomes of this type are acceptable.

2. The side territory is relatively worthless early on. Giving it up is acceptable (or even downright optimal), and doing so makes it easier to achieve the main objective.

3. The outside group tends to be fairly resilient, and can possibly withstand one tenuki while achieving the main objective.

4. There is nothing to fear about an early invasion of the 3-space extension. Such invasions are generally premature and help the approaching side more than the invader.

Keeping these four points in mind is already enough to serve as a decent basis from which you can play the attach-draw back jōseki by ear. Of course, in this article we have only scratched the surface of this jōseki. In the remainder of this series we intend to discuss several of the lines in detail, and refine some of the concepts we have developed above. However, the philosophical framework of all of these subsequent discussions will be based on the principles and objectives for the jōseki we have described here.

Return to the 3-4 Jōseki page.

I will be the first to admit that 3-4 pincer jōseki terrify me. They are rife with complications and many lines are quite delicate: a single misstep and the position is irredeemable. Not only that, it’s quite difficult to find a unifying theme for the different variations or even the different types of 3-4 pincer jōseki, in contrast to 4-4 pincers which all work more or less the same way. And even if you somehow manage to devote enough time to memorize a majority of the lines, ultimately this isn’t all that useful: many 3-4 pincer jōseki are extremely sensitive to the global board state and there is no reason to expect a good outcome even if you follow them to the letter.

Fortunately, there is a way around dealing with 3-4 pincers that is more fruitful than religiously memorizing their myriad lines. It turns out that in the vast majority of scenarios, there is no significant need to dive into the rabbit hole of 3-4 pincers: you can almost always dodge playing out a 3-4 pincer if you so choose. And lest the responsible among you fear this is a sign of laziness or weakness: the conclusions we draw here bear out perfectly well with AI analysis and professional play. Even pros will admit that they do not necessarily have an encyclopedic knowledge of 3-4 pincers, and I can point you to times when they have gone on the record saying as much. The objective of this post is to dispense with the idea that one needs to understand all the complications of 3-4 jōseki, and liberate you from the pressure of the pincer when you consider playing the 3-4. As a side objective, we will demonstrate the danger in over-reliance on advice from jōseki dictionaries.

Nobody’s making you pincer

Principle #1: There is rarely ever a need to initiate a pincer yourself.

First let’s discuss the pincer from the perspective of the player who does the pincering - the pincermaster, if you will. Obviously the pincermaster has all the agency in the world - if the pincermaster doesn’t want to pincer, then that’s the end of the story! So if you don’t like or don’t have a good understanding of 3-4 pincers, then at least half of your problems are already solved: you can confidently dodge the pincer in those cases where you are the pincermaster.

But that all sounds a little too obvious and trite. If you want to improve your game you should naturally ask the follow-up question: “Is there ever a situation in which I am the pincermaster and I should pincer?” That is, can an early-game situation arise where the 3-4 pincer is so good that it blows its competition out of the water? In that case, in order to play optimally you would have no choice but to pincer even when you are pincermaster.

The answer to that question is: probably not. Such a situation would be rather highly contrived. The thing to understand is that pincers are usually not the absolute best move on the board unless they are supported by a certain global board-state. That doesn’t make them bad moves outright in most early game situations: they remain in the class of reasonable opening moves which differ from the optimal moves by only a few decimal points. But it does mean that usually you don’t need to feel like you have to pincer. Here are a few examples to illustrate the point.

Here we’ve put down some fairly typical first five opening moves, and had KataGo analyze some of White’s follow-ups in the top right. As we can see, KataGo slightly prefers the corner and side-oriented moves A, B, and C over any of the pincers. This is a consistent pattern with AI analysis, which tends to dislike 3-4 pincers like these in the very early opening, and which explains why they have largely disappeared from top professional play. This is also consistent to a degree with classical theory, which would state that the pincer on an empty board like this isn’t bad, but pincers are really best if they also serve as extensions from a position of strength in the opposite corner. Here White does have a stone in the opposing 4-4, which is better than nothing, but not nearly as powerful as an enclosure for example. Of course, these winrate and score differences are not significant enough to make a big difference at the amateur level, but it does mean that at least in this situation, you should not feel like you need to play the pincer as White.

How about if we change the position to one where classical theory would recommend a pincer? Here Black has spent sente to enclose the bottom right corner and White approaches the bottom left 3-4. This time KataGo rates the pincer at A more highly as we’d expect, but it still struggles to choose between this and the knight’s move at B. C, D, and E are the next best options according to KataGo, and all other options are rated less positively: in particular, A and D are the only pincers KataGo seems to like.

What we gather from here is that even if the situation is pretty favorable for the pincermaster to pincer, that doesn’t necessarily make it better to pincer as opposed to another option. Here the choice between A and B is largely up to playstyle and jōseki knowledge. If I am playing White in this position, then I would elect to play B, since if neither one is theoretically superior to the other then I may as well play the jōseki I find easier to play (and most players should find the lines of B far easier to understand than A).

A secondary piece of information we gather is that not all pincers are equivalent. There are something like 8 different viable pincers in this position for White, but KataGo only thinks A and D are an even result for White. (I can’t say for sure, but this probably has to do with the opposing enclosure being a two-space high enclosure, and White wanting to keep a high and flexible position early on.) This is an extra burden that the pincermaster needs to deal with: even if you are comfortable with playing the pincer, it’s a separate thing entirely to understand just which pincer is the appropriate one at any given time.

Let’s go through one last example to illustrate the idea from the pincermaster’s perspective. Here the notable feature is the 3-3 invasion jōseki in the bottom right. This time, when White approaches, KataGo does consider a pincer to be the best option, preferring the distant pincers A, C, and D above the other options. (This has to do with the Black wall, which can be prone to turning into a weak group later in the game; the pincer serves as both an extension and reinforcement.) Nevertheless, KataGo still considers the knight’s move at B to be essentially as good as these pincers.

The takeaway from these examples is that at least as far as AI is concerned, if you do not want to play a 3-4 pincer then you’re in luck. Roughly half the time you will be the pincermaster from a 3-4 corner, and these examples illustrate that the pincermaster is almost never required to pincer from a 3-4 stone. As long as the game has not progressed extremely far elsewhere on the board, you will almost certainly have a non-pincer option that is at least as viable as the pincers, if not outright better than them. So then the only question that remains is: how do I avoid the pincer if I am not the pincermaster?

Dodging the pincer I: When in doubt, tenuki

Principle #2: Tenuki is usually a viable response to a 3-4 pincer.

If you are around or not yet mid-SDK level, you may be surprised to learn that tenuki is not only a viable option against a 3-4 pincer, it is often one of the best responses! This is where we break with typical jōseki dictionaries such as Kogo’s, which traditionally look very unfavorably upon tenuki in response to a pincer. As an example, let’s see what Kogo’s has to say about tenuki in the low approach, one-space low pincer:

Kogo’s characterizes the one-space low pincer as the “severest attacking move” and claims that it “does not permit White to switch lightly elsewhere.” Upon White passing, Kogo’s claims that this “invites a loss” without the ladder. This is an example of the worst kind of advice that is commonly offered in a typical jōseki dictionary, and a good reason to avoid relying on them to decide what to play. There are some elements of truth to what is said here, but there’s so much missing context that it is easily subject to misinterpretation.

We will use this jōseki as a springboard to illustrate some general principles behind how the tenuki from a 3-4 pincer (of any variety) works. In particular, we will disavow the notion that White cannot easily switch lightly elsewhere here, and that the tenuki without the ladder “invites a loss.” Instead, switching lightly elsewhere and doing so without the ladder is precisely what we are going to do!

Let’s examine how things go if we tenuki here. Black’s most common local continuations are the kick and the attachment on top. Let’s consider the kick first.

After the kick, White main options are twofold. The most obvious local continuation is to form a pillar. This outcome is fairly straightforward: Black grabs a little side territory and chases White out toward the center, perhaps looking to play both sides. White is admittedly weak but not in a huge amount of danger, and will look to balance the tax it pays for this weak group with the gains it made earlier by playing elsewhere.

However, the most incredible thing about this position is that White can tenuki one more time! In that case, Black’s best local continuation is to attach on top, completely smothering this White stone (move 3, top variation). You may well ask: isn’t this a disaster for White? What was the point of approaching at all if White’s just going to let Black take such a commanding position in this corner?

Actually, this is a clever sacrifice play from White, which we can unravel using some tewari analysis. The idea is that after Black attaches on top, this position transposes to if Black started with the squared one-space high enclosure. In that case, the triangled stone is overconcentrated: it is too close to Black’s enclosure, and should really be one space further away. Granted, the 1-2 exchange is a bad one for White in the presence of the high enclosure, but it is bad mainly in terms of deleting aji (aji-keshi); Black has preemptively limited how large a corner it can build here with the narrow extension. So ultimately, what happened here is that White and Black each added a suboptimal move to a one-space high extension, but White also played elsewhere twice and has sente. If you put it that way, things don’t sound so bad for White anymore, do they?

Not only that, there is still a tiny bit of value left in the White stone. For instance, White could continue locally at A to put a few stones on the outside in sente, or perhaps threaten a connection underneath to get an approach from the side in sente. So despite Black having invested a total of four stones in this area to White’s one, there’s still some aji left in this position for White to exploit!

So to summarize, to capture White’s stone by attaching on top is overconcentrated and smaller in value than at first glance. Of course, Black knows this as well. Therefore Black can also tenuki from this position (bottom variation), and the local position becomes very interesting. Black’s goal will be to capture the White stone indirectly, namely by omitting the attachment on top. White’s goal will be to force Black to capture directly, forcing Black into the aforementioned overconcentrated position. The final evaluation of who comes out on top may not happen for quite a while yet.

If Black opts for the attachment on top instead of the kick, then these are some of the outcomes that can arise. We first note that by the same logic as before, White does not need to respond immediately to the attachment. After all, what’s Black going to do - spend another move locally to kill? We’ve already established that it would be overconcentrated for Black to do so. This time, there is a difference in that Black can force this corner into a ko if White plays elsewhere again. But remember that White is already treating this stone lightly by playing elsewhere twice - forcing Black to play a ko to kill is perfectly fine by White’s standards.

If White does respond locally then the outcomes are generally some type of Black-thickness-for-White-territory exchange. Generally these exchanges are locally good for Black, but this is to be expected - Black has at least one extra stone locally, and globally we still need to account for the value of the White tenuki and sente (which White tends to get in these variations). And as with any thickness-for-territory exchange, ultimately the evaluation of this position isn’t really about who came out on top locally. All such exchanges can only be evaluated in conjunction with the global position and direction of play. If Black cannot get its thickness pointing in the correct direction of play, then these outcomes are generally good for White.

Once we understand these ideas behind the tenuki in the one-space low pincer, they adapt easily to other pincers as well. The point here is not to memorize these specific variations, but just the general idea behind why tenuki is a viable response to a 3-4 pincer. It should of course be said that the variations we’ve shown are far from the only ones that are possible in the one-space low pincer, tenuki variations. The gist of it is that there is a good chance that the pincermaster cannot capture the pincered stone cleanly with one move. For narrow pincers it might even be overconcentrated to kill, while for wide pincers your stones have a bit more breathing room to survive a tenuki in the first place. Let’s now see a few examples of how this works in a simulated game.

Here we have created a hypothetical early-game position in which White has just pincered in the top left, and we will aim to showcase just a few of the possibilities that can happen if Black ignores the pincer. We preface by saying that this is not intended to mean that tenuki is the only option. As a matter of fact, in this starting position it is also viable to respond locally at C; A, B, and C all have similar winrates and are in fact the highest rated choices by KataGo. The point of this exercise is merely to demonstrate that tenuki is a viable option, to the point that AI considers it among its top choices in this particular board state.

The first line we explore is the kick (top variation), and here to prove our point we in fact take the line where Black plays elsewhere twice. We see that by completely foregoing any attempt to save the pincered stone, Black can use its two free moves to quickly establish a presence everywhere on the board and even begin going on a counter-offensive in White’s opposing corner. White comes out of this with an extremely strong group in the top left, but otherwise it is so far very limited in what parts of the board it lays claim to. As a result, Black actually comes out slightly ahead in this line. There are many other ways to play the kick from this starting position, including responding locally, all of which KataGo considers at least playable.

The second line we explore is the attachment (bottom variation). Here we note that Black, by virtue of having the bottom right corner, has the ladder and therefore can play the wedge. By the end of the sequence, White has some decent influence and sente while Black has a respectable corner. If we do some tewari analysis, this isn’t that different from if Black first plays the top left corner in a way to exchange the corner for influence in sente, and then uses sente to enclose the bottom right corner, as opposed to vice versa. Such a sequence of events is, of course, nothing new if you’ve played enough games at the SDK level, and it makes sense that this variation also turns out to be ok for Black according to KataGo. Again, this is not the only line KataGo considers viable after the attachment, and things can play out in several ways (including all of the ways we have discussed above).

Dodging the pincer II: Make a base on the side

Principle #3: Making a base on the opposing side is usually at least playable.

If you are DDK or high SDK, you might find the previous discussion somewhat intimidating, and that is fair. To benefit properly from the tenuki requires a certain amount of global intuition that you may not have developed just yet. However, that does not mean that all hope is lost if you are confronted with a 3-4 pincer There is another, far simpler way than tenuki around a 3-4 pincer. This is to simply give up the pincered stone and make a base on the opposite side.

A major reason for why complications arise in 3-4 pincer jōseki is that many lines involve both players working with cut groups, and usually one cannot afford to sacrifice one of the groups right away. Abandoning this goal entirely drastically simplifies the jōseki. The basic ideas for why this works are the same as with tenuki; in a sense, we are doing a semi-tenuki, in that we are playing somewhat locally but not locally enough to connect with the pincered stone. The preceding discussion for the tenuki explains why this is ok: it is rare that the pincermaster can cleanly kill the pincered stone in one or even two more moves, so you can make up for the loss of the stone using the establishment of an outside group and the aji left in the pincered stone. The tradeoff for this simplicity is that sacrificing the pincered stone immediately is typically not locally optimal. However, it is usually not so unfavorable to the point that it is unviable. Often the difference is as small as 0.5 points, which is not something that tends to decide a game at the SDK level, whereas confusion with a jōseki can easily send a game out of control.

We illustrate this with the two-space high pincer to the one-space high approach. Here, instead of responding directly to the pincer, White elects to establish a base along the bottom side, inviting Black to cut White apart and take a large corner. The initial jump feels very similar to other variations of the two-space high pincer in which White jumps out. The difference is that the jump is intentionally too wide to prevent the cut. The reason that complications arise after jumping out of a 3-4 pincer jōseki is that normally White wants to continue doing work with its inside stone after the cut, and therefore must commit to a potentially complicated fight to keep it alive. Abandoning this goal entirely drastically simplifies the jōseki.

The differences between the two variations essentially come down to who takes sente; however, even in the bottom variation where Black takes sente, Black in fact owes White an extra move locally. White’s job will be to use the aji in this position and sente/half-sente to make up for local loss it incurs by allowing Black to take this large corner. This is, of course, not specific to the two-space high pincer, nor to the high approach: the principle of sacrificing the pincered stone for an outside base works in general.

Let’s see this idea in action. In response to White’s two-space high pincer Black appears to sacrifice the pincered stone. Now, as we have mentioned previously, such a move is usually not locally optimal, but overall the loss is just 0.5 points and 2.5% winrate - not the kind of difference that decides amateur games. All moves following the initial jump are essentially optimal according to KataGo, and we see no subsequent loss in winrate or score.

It should be noted that this position is helped a lot by Black having the opposing corner to support Black’s wall. If the position is changed so that the wall faces a White corner, then the score drops by another 0.5 points against Black. This support allows Black to play a corner enclosure as an extension and establish an imposing Black moyo.

Here we see some of the aji present in the pincered stone come to life early, and this is the kind of aji you need to be aware of to use this strategy, as White’s corner is indeed too large to give away completely for free. There are some lingering questions left on the board: is White’s corner large, or is it overconcentrated? Are Black’s stones in the center in prime position to limit the scope of White’s expansion, or will Black bleed points as it attempts to stabilize the group against White harassment? Can White spare a few extra moves on the left, or is Black’s moyo too threatening to completely ignore for much longer? All in all, Black has managed to create a fairly stable position without falling into a labyrinth of complications, and can look forward to a more strategically defined rather than tactically oriented opening.

Summary and conclusion

In this article we explored ways one can avoid dealing with complications when faced with a 3-4 pincer jōseki. As pincermaster, there is rarely ever a need to initiate a 3-4 pincer if you are not comfortable with it. If you are not pincermaster, then tenuki is almost always viable and usually among the top options, while those of you who are less comfortable with the idea can still sacrifice the pincered stone for a position on the side in order to avoid all complications. In addition, I hope this article reinforces the need for active learning when it comes to jōseki, and the danger of taking the word of jōseki dictionaries as gospel.

Return to the 3-4 Jōseki page.

A proverb that does more harm than good

When we first start learning baduk, we are commonly taught the following principle governing the length of side extensions:

Extend two spaces from one stone. Extend three spaces from a two-stone wall. Extend four…

We then learn to judge extensions based on this N+1 spaces from an N stone wall rule, as well as (to some degree) extensions from enclosures. e.g. “Two spaces from two stones is bad! It should be three spaces, so two spaces is overconcentrated!” And so on.

However, this proverb is one of the most damaging proverbs in existence. The truth is that the proper spacing of side extensions is tied to a far more nuanced rule than a formula like N goes to N+1. Sometimes you do extend two spaces from two stones, and sometimes you extend ten spaces from four! In this article, we are going to undo the damage the proverb has wrought, and explain the real principles behind how one decides on the appropriate distance to extend. The goal is that by the end of this article, you should be ready to throw the proverb in the trash, and play your extensions with confidence and finesse.

The real meaning behind the proverb

First let’s get an idea of what the real value in the proverb is.

The proverb does exist for a reason, but it’s not really about ideal side extensions in the context of a whole-board strategy. It is a proverb that is more tactical in nature rather than strategic, meaning its recommendation is more concerned with local considerations than global ones. This is its true meaning:

The N+1 spaces from an N stone wall rule is mostly about maintaining the connectedness of your stones. It is a rule of thumb that tells you roughly how many spaces you can extend on the third line while keeping your extension connected to the wall.

Above we see that the two-space extension, in the absence of extra surrounding stones (beyond those already on this board), cannot be cut. However, that’s the only thing that the proverb says regarding two-space extensions on the third line. It says nothing about whether these stones are safe or whether this is an efficient formation. In fact, in this particular scenario Black is actually under a fair amount of pressure: flanked by White stones one space away from each side, this Black group can only expect maybe one eye on the side if White has sente, and possibly not even that considering the aji at B or C. Too many beginners would leave this Black group alone, thinking it is safe because “the proverb says 2 spaces from 1 stone is ok!”

The same goes for the three-space extension from a two-stone wall. Without additional White support nearby, this extension cannot be cut. But that has nothing to do with whether these stones are weak or strong or claim any territory or even whether or not it’s good for Black to stay connected. Some of these connecting sequences shown here might actually be detrimental if we account for the board at large. Moreover, everything is extremely sensitive to the presence of additional White stones. A White stone at O3, for instance, will significantly alter the outcome of the White invasion at (A).

With three-stone walls and higher, the value of the proverb becomes even more suspect, because the wider the extension, the more sensitive it is to surrounding stones. Once we are at four-stone walls, it is difficult to claim that the wall and the extension are actually connected - again, ignoring for the sake of argument whether you would even want to connect.

So how do we actually figure out the proper length of an extension?

Decide on the purpose of the extension

Broadly speaking, there are two types of side extensions in baduk, delineated by purpose.

The first type is an extension whose primary purpose is to help a group settle. These extensions generally reinforce weak groups and help them establish eyespace on the side; they also claim some side territory as a bonus.

The second type is an extension from thickness. These are typically extensions from strong groups or walls. The purpose of such extensions is to make good use of the influence generated by your strong group.

The length of an ideal extension is tied to what kind of group you are extending from. So, before you do anything else, you need to first take a look at the group you are extending from, and ask:

Is this a weak group that needs to settle, or a strong group projecting lots of influence?

Only once this question is answered can we start to study the problem in greater detail.

Extending to settle

Now let’s assume that you are working with a weak group, and the purpose of your extension is to help it settle. In this case, the length of an ideal extension is determined by the following principle:

When extending to settle, extend as far as you need to secure sufficient eyespace for life, but no further.

Here is the logic behind this principle.

• If you extend further than absolutely necessary to secure eyespace, then your thinness can introduce additional weaknesses into your group. This will give your opponent sente exchanges against your group which are more valuable than the value of extending one or two spaces further.
• If you don’t extend far enough to secure eyespace, then your group will be weak by virtue of having limited eyespace. This will again give your opponent sente exchanges against your group. In this case you might as well extend further, since either way your opponent will get sente exchanges but extending further may secure extra eyespace or territory.

Let’s see what this means by studying some examples from jōseki.

In this sample game, Black approached high to White’s 3-4 point in the top left, and White has just played (1). To finish the jōseki, Black needs to make an extension to settle its three stones. Two options seem most reasonable (and are both jōseki): two spaces at (A), or three spaces at (B). Which is the better option?

If we are following the old rule of N+1 spaces from N stones, it is tempting to treat the three Black stones as a two-stone wall: after all, Black may as well be connected at E17. This would suggest that (B) is correct, as this would be three spaces from a two-stone wall.

However, if we follow our new principle for determining the length of an extension, then the actual best answer here is (A)! The key is White’s last move at B15: in particular, it is not at B18. Because White does not have B18, this means that extending to (A) is already enough for these Black stones to secure enough eyespace to settle: there is no need to go any further.

Why does this matter? Wouldn’t it be more efficient territory-wise to go one space further to (B)? Actually, it’s the opposite. Extending three spaces creates a Black weakness at G17. White can exploit this weakness in order to get favorable sente exchanges against the group. For instance, after (B), White could approach from the right to get an extension from its stones top right in sente. Or White could instead press from above at F15, also in sente, which builds White influence toward the center. Both of these exchanges are better for White than for Black - whatever extra territory is claimed by extending one space further is unlikely to match White’s approaches in value.

What happens if Black instead extends conservatively to (A)? In that case, Black’s group is already settled, and does not need to respond to any White approaches. If White does approach, then Black can simply ignore it and take the next big point on the board, pulling ahead. So in actuality, playing solidly here is playing more efficiently - once we take the weaknesses of our group into account.

KataGo’s evaluation confirms our analysis. Admittedly, it is not a large difference by any means - KataGo prefers (A) by only a tiny margin, as in either case Black’s group is secured. Any White approaches after (A) are ignored.

Let’s put our principle to the test by switching up the local situation. This time White has descended to B18. Once again we ask: should Black extend two spaces to (A), or three spaces to (B)?

Again, the key lies in White’s descent to B18. This stone destroys much of the eyespace that Black previously had when this stone was at B15 instead. In particular, this means that neither (A) nor (B) secure enough eyespace on the side for unconditional life in a single move. Therefore the answer here is (B) over (A).

The reasoning is as follows. Suppose Black opts to extend conservatively to (A), and White approaches from the right at K17. This White approach is not necessarily sente in the usual sense of the word, but it is a form of delayed sente: Black owes White an extra move here at some point. This is because thanks to White’s descent at B18, Black has not yet settled its group, and needs at least one more move to do so. Which specific move Black chooses in order to settle is not so relevant: the point is that Black needs to respond in some way to K17 eventually, effectively turning K17 into a sente exchange.

If Black cannot settle locally in a single move anyway, then Black might as well extend one space further, in order to compensate for the eyespace destroyed by B18. It is still the case that White can get sente exchanges from above or from the right. However, these exchanges have the additional effect of helping a weak Black group settle, so Black has nothing to complain about. When White did not have the descent at B18, these exchanges helped White more mainly because the Black group wasn’t that weak to begin with, so Black gained much less out of the exchange compared to when White did have the descent.

Once again, KataGo agrees with this analysis, this time favoring (B) over (A) by a more noticeable margin than in the previous scenario. KataGo’s response to (B) is indeed to approach right away at L17, but this time KataGo sees this as perfectly acceptable for Black rather than a one-sided sente exchange.

Extending from thickness

In the last section we were working with extensions from weak groups that needed to settle. Now let us consider the opposite scenario, in which we are extending from a strong group projecting lots of influence. It turns out that for this scenario, the principle for deciding the optimal extension is quite different:

When extending from thickness, extend further than normal.

Here is the logic behind this reasoning:

• Directly building territory out of thickness is inefficient in terms of territory-per-stone. Making a normal-length extension risks your opponent simply letting you have your territory and playing a big point elsewhere.
• Extending further than normal invites your opponent to enter your moyo by threatening to build more efficiently than usual. This bring your thickness to life, as you can use it to attack the invader and derive benefits in sente.

Let’s test this principle in a real-game context.

Here Black has set up a three-stone wall from the lower right corner and has sente. This wall is a fairly strong group, a topic we discuss in the early 3-3 invasion series. As this is still an early stage of the game, lots of different moves are viable; KataGo likes playing the top side in particular. However, if Black were to specifically choose to extend from this wall, what would be the best distance to extend?

To answer this question, we have listed all of the reasonable-seeming extensions from this wall consisting of at least four spaces. For each extension we have listed the resulting change in winrate and score (from Black’s perspective). If the N+1 spaces from N stones principle were correct, then the four-space extensions (A) and (B) should be the best extensions.

But in reality, the four-space extensions are the worst extensions among these choices from KataGo’s perspective! Even (C), a five-space extension, is third worst among all extensions. We can see that KataGo’s evaluation tends to improve the further the extension is from the bottom right. ((N) is a noticeable exception, but this has mostly to do with the fact that it is an unorthodox and non-optimal approach to a 3-3 corner.) The best “extension” is in fact a shoulder hit against White’s top right corner, and the next two best extensions are (I) and (K) - a whopping 8 and 9 spaces respectively from Black’s three-stone wall! Now, part of this favorable evaluation does have to do with the fact that the longer extensions double as corner approaches. But even the extensions that are a fair distance away from the top right are seen as better than the narrow extensions (A), (B), and (C).

How does play continue after these approaches?

• KataGo ignores (A), (B), and (C), choosing instead to turn at G16, which is a very big move. White does not bother invading this extension; instead, White simply treats the area between the wall and the extension as a no-go zone for a large portion of the game.
• The approaches (I) and (K) are met by White protecting its corner, which is fairly standard as 3-3 jōseki go. Then will Black extend back to (D) or (F). Now the right side is actually worth a substantial amount of territory, since Black has claimed a much wider chunk of it.
• (O) is met by a standard 3-3 4-4 approach jōseki. This could go in a few different ways, but KataGo seems to prefer Black taking the top side and White taking the corner and parts of the top right side. Neither side gets to claim much of a moyo on the top or right in this scenario.

The major problems with the narrow extensions (A), (B), and (C) are the following:

• They don’t claim enough territory to be efficient on a per-stone basis. Black has already invested four stones after the extension, but it is still not secure: there are far more secure combinations of four stones to secure this amount of territory. Black probably needs another three or four stones before it can really claim that all the territory between its wall and its extension belongs to Black - at which point it becomes obvious that Black is not getting much value out of each individual move.
• Because it doesn’t claim enough territory, White has no incentive to go anywhere near Black’s extension - it does not require much reduction or invasion, especially not right away. This makes the early investment into Black’s wall in the lower right for naught, as influence shines brightest when it can be used to attack weak groups, but Black’s wall will have nothing to attack if White doesn’t enter the extension.
• Because White doesn’t need to do anything about Black’s extension, this immediately gives up sente.
• These narrow extensions are better than the wider extensions at one thing: they reinforce the Black wall, giving a potential way to secure eyespace or support in the case of a fight breaking out nearby. But as we mentioned earlier, the Black wall is already a fairly strong group by itself, and hence does not really require additional reinforcement. Thus the narrow extensions are somewhat redundant.

These considerations are fairly universal for extensions from strong groups.

Summary and conclusion

To summarize what we have learned:

• The proverb “extend N+1 spaces from N stones” is mostly about connectedness. In terms of whole-board strategy it actually tends to produce bad recommendations. It is best not to rely on it.
• When looking to extend, first you need to identify the purpose of the extension.
  • When extending from a weak group with the purpose of settling, you should extend as far as you need to best secure life, but no further.
    • Extending further than necessary tends to expose weaknesses in your group, leading to favorable sente exchanges for your opponent that could have been avoided.
  • When extending from a strong group, you should usually make a significantly wider extension than normal.
    • Narrow extensions from a strong group tend to be redundant and inefficient.
    • Wide extensions from a strong group allow you to make good use of your thickness rather than letting it go to waste.

With this, you are now hopefully ready to throw away the N+1 spaces from N stones proverb for good, and start playing extensions according to fundamental strategic principles rather than via a formula.

Return to the Concepts hub.

This is the third article in the series “Strong and Weak Groups.”

Previous article, Next article.

In the first article in this series, we identified how to recognize when groups are weak or strong. In the second article, we discussed how one exploits imbalances in the strength and weakness of groups to play profitable moves. In this article, we will explain how to synthesize these concepts into the strategic principle and decision known as direction of play.

What is the direction of play?

When we first start learning to play baduk, we are commonly taught the following basic principles of baduk strategy:

• First play the corners, then the sides, then the center.
• Play urgent moves before playing big moves.

Direction of play is a concept that consolidates these and all similar concepts into one. It asks a question with a simple premise: in what general area of the board should I play my next move? Of course, the simplicity of this question is deceiving: anyone who has played more than a few games of baduk knows the feeling of being lost in the vastness of the 19x19 board. However, it may surprise you to learn that determining the direction of play is actually fairly straightforward a lot of the time. The key to it all is the concept of strong and weak groups.

The most important proverb in baduk

The fundamental principle in the determination of direction of play is summarized by the following proverb:

Play away from strength.

Simple and understated, this proverb is in fact the deepest and most important one in all of baduk. Nearly all opening and middlegame strategic concepts can be boiled down to this one. Its meaning is precisely as it states: the correct direction of play is away from the strong groups on the board, and close to the weak groups. This means to play away from both your opponent’s strong groups as well as your own, and to play near your opponent’s weak groups as well as your own.

The logic behind this idea is the following:

• You should not play near your opponent’s strong groups, because by definition a strong group will not be pressured by such a move.
• You should not play near your own strong groups because your strong groups do not need to be reinforced.
• You should play close to your opponent’s weak groups, because this will put pressure on your opponent’s groups.
• You should play close to your own weak groups, because this reinforces your weak groups into strong groups.

We’ll do a full accounting of these four ideas later. For now, there’s no better way to learn this concept than to see some examples.

Example 1-1

Let’s examine the above position from a sample game. White has just played (1) as part of the 4-4, double low approach jōseki. At the moment the game is very even. In which general direction should Black play next, and why?

Determining the weak and strong groups

Before we do anything here, we first need to take a look at the state of the board, and account for the strength of each of the groups. Remember our criteria from the first article in this series for determining which groups are weak and which are strong:

• Groups are stronger the closer they are to unconditional life, and weaker the farther away they are from it.
• Groups are strong if they can ignore many moves from the opponent before feeling threatened, and weak if they have many moves against them that threaten them.

With these criteria firmly in mind, let’s go through this board carefully, corner by corner.

White’s corner in the top right is strong. Generally enclosures are strong shapes in the early stages of the game. There is nothing Black can do to this corner at the moment that will make White fear for the status of its stones. Later on when Black has more stones nearby Black might aim at some of the aji associated to the large knight’s move enclosure, but to do so at this moment is premature.

Black’s corner in the bottom right is strong. The reasons are the same as for White. With no White stones in the vicinity, White does not have the necessary support to really threaten the safety of these stones. Again, that’s not to say that there’s no aji here - there is a lot! But if we’re purely talking about life and death, then Black can afford to ignore multiple White moves in this area before it really needs to start taking care of these stones.

White’s group in the lower left is strong. These stones have a fairly nice shape for making two eyes, as well as emergency exits to the side and to the center if things get hairy. They should not struggle to make life unless White ignores several moves against them. Even then, there is the possibly that White treats part of this group lightly: killing the entire group should be a near impossibility.

Black’s group in the lower left is strong. Like the nearby White stones, the Black stones have a nice little base for making eyeshape, as well as ways out to the left side and center. If Black gets C2 then Black might as well just be considered unconditionally alive.

Lastly we turn to the top left corner, and here things get a bit interesting.

White’s group, marked with triangles, is weak. The first reason is the cutting point at (A), which is an active weakness thanks to the presence of the circled Black stone. Although Black does not necessarily want to cut at this very moment, it does mean that White owes Black a move here if White wants to stay connected. The second reason is that even if White connects up, White’s entire group is a bit far from making two eyes. White does have some eyespace in the corner, but if you work this out carefully then this eyespace only amounts to one eye in gote. (We will cover the details of this statement in our 4-4 double low approach series.) Therefore White has some work to do before it secures its second eye.

Black’s group on the left, marked with squares, is weak. This largely has to do with the cutting point at (B). Again, White doesn’t want to cut right away, but the aji behind this cut makes even just a casual White approach a serious threat to this group’s survival.

Lastly, Black’s lone stone on the top, marked with a circle, is weak. This stone is clearly a long ways from life: in fact, bunched up so close to so many White stones with a high position, it is under quite a bit of pressure. A White pincer at C or nearby could rob this stone of its eyespace and force it to run out to the center to survive, but with White having a towering position over it, this stone will likely need several moves before it can run sufficiently far out to collect its breath.

Using weak and strong groups to determine the direction of strength

So let’s summarize what we’ve learned:

• All groups in the top left corner are weak to some degree.
• All other groups on the board are strong. In particular, the Black bottom left corner is a strong group.

Now let’s apply this to determine the direction of play. The goal here is not necessarily to guess the best next move: merely the general area where the best next move is located.

Black’s typical continuation from in this jōseki is either to extend from the group on the left, or to reinforce the stone on the top. Here this is consistent with direction of play, as both Black groups in this corner as well as the White group are weak, and direction of play suggests that we play near both our own and our opponent’s weak groups. However, since we cannot play two moves at once, we only have the opportunity to take care of one of the two sides. Which should it be?

The key to answer this question lies in the bottom left corner. Here there is a strong Black group, and since we want to play away from our own strong groups, this suggests that the correct direction of play is to reinforce the stone on the top, such as at (A) or (B), rather than to extend on the left, like at (C), (D), or (E).

This is indeed the case. (A) and (B) result in no significant loss of winrate, and even a gain of winrate in the case of (A). (C), however, results in a small loss in winrate and score, while (D) and (E) are significantly worse. As we can see, the closer our stones get to the bottom left corner, the worse KataGo evaluates it, but that’s not the only reason: (D) and (E) are also farther away from White’s weak group on the top right, hurting Black’s ability to fight on the top side.

The sequences following each choice are just some of the many possible continuations, which serve to illustrate how these differences in winrate might manifest in play.

Just to further verify that out theory on the direction of play is correct, let’s check out how KataGo likes it when we really mistake the direction of play.

• (A) is reinforcing a Black enclosure that doesn’t need it, so is value is purely in terms of the territory it claims and the influence generated by the enclosure. It is certainly valuable, but with the situation in the top left still unfolding, now is not the time.
• (B) is slightly better in that it puts a little pressure on White’s group, which does have some bad aji at (A) and (B) with this Black stone nearby. But ultimately, this is still a strong White group that Black is approaching, which is not nearly as important as the three weak groups in the top left.
• (C) is also too close to Black’s strong corner. It does limit the potential of the top right corner, but it is just another move that purely claims and denies territory without doing anything for the stones it is supporting.
• (D) is in the right direction, but it is one step too far from Black’s weak group. Since both White and Black are weak in the top left, it is better for Black to play more solidly here. Black becoming stronger here correspondingly makes White’s stones even weaker. White will invade at A (which is supported by its strong top right corner), hoping to settle its top left by exploiting the thinness of Black’s stones.
• (E) is interesting - unlike the others, it leads to no drop in winrate. Its meaning is just as a sente exchange - Black wants White A before returning to the top right. This is a minor loss locally as it eliminates all the aji in White’s group, but Black’s reasoning is that having this outside stone may help the fight on top or just help Black get out to the center later.

Example 1-2

Let’s now try this again, with a small modification to the previous scenario. This time White has played a 3-3 invasion jōseki in the lower left corner. Once again, in order to determine the direction of play we need to evaluate the strength and weakness of each of the groups. Since the only difference is in the lower left, we simply update our evaluation there.

The White group in the lower left corner is strong. It is obviously unconditionally alive and can even escape out to the side or center if it needs to.

The Black group in the lower left corner is strong, although this is not nearly as obvious as for White. This is based on prior knowledge of the aji in this jōseki, which will be covered in our early 3-3 invasion series. The basic idea is that Black’s wall has a few emergency escape options to help it settle even if White denies its eyespace on the bottom side: (A) and (B) are sente against the corner, and the aji there give Black a way to secure either eyespace on the side or a way out to the left side.

Based on this analysis, we arrive at the same conclusion as before, which is that both groups in the bottom left are strong. Therefore our direction of play should be the top side, same as before.

Indeed, KataGo confirms our analysis when we examine the same five moves from before. Either play reinforcing the top Black stone is near optimal, while each of the extensions on the left are received poorly. Once again, the further the extensions on the left, the worse the reception by KataGo - because they are closer to the strong stones in the lower left, and farther from White’s weak group in the top left, hurting the ability for Black to fight on the top.

Example 1-3

To demonstrate that this isn’t just a case of the top side always being better, let’s change the situation one more time. This time in the bottom left, White played out a 4-4 low approach jōseki from the left side. Is there a difference in the direction of play, and does KataGo think differently about the top left corner?

First, let us evaluate the strength and weakness of the groups once more. Again, only the bottom left needs re-evaluating.

The Black group on the bottom left is strong. There is some aji against it, but it has a decent base for eyespace and access to the center and the side.

The White group on the bottom left is slightly weak. The issue, as always in this shape, is the peep at (A). If White is forced to connect in response to the peep, then the whole group does not have two eyes and is liable to be chased around.

With this the status of the bottom left corner has changed: instead of both groups being strong, now there is a slightly weak White group. This should now introduce some changes into the direction of play. The extensions from the Black group on the left should better than before, because they are now also approaches to the weak White group on the bottom left.

This time, KataGo evaluates the side extensions on the left more positively, (E) particularly so. This is because the extensions further weaken White’s bottom left, with the threat of the peep looming ever closer. (C) is likely evaluated better than (D) because neither (C) nor (D) are close enough to White’s bottom left to really pressure it, in which case they are more evaluated for what sort of effect they have on the fight on the top side. In any case, with Black having settled one weak group, now White has two weak groups to Black’s one, forcing White to find a way to somehow settle both groups while Black only needs to focus on one.

These examples should hopefully convince you that the intuition from the principle “play away from strength” bears out in actual play.

Summary

Once again, there is only one rule you need to follow: “play away from strength.” Let’s take a moment to summarize how things can go wrong when you, and how things can go well when you follow it.

Playing close to your opponent’s strength

What happens when you play close to your opponent’s strong groups?

• Since by definition strong groups don’t really need taking care of, your opponent could just ignore your move and play something more valuable - in other words, you would give up sente.
• Since your opponent is strong in the area, your move is unlikely to make much territory, as your opponent could use their strong group as support for a reduction or invasion.
• If your approaching stone is weaker than your opponent’s strong group, then in the worst case scenario your opponent could go on the attack.

Playing close to your own strength

What happens when you play close to your own strong groups?

• Since your own strong groups don’t need reinforcement, an extra stone close to them would be a redundant move that could have fulfilled a more meaningful purpose elsewhere.
• Since strong groups tend to consist of a multitude of stones in close proximity, adding another stone nearby could lead to overconcentration.

Playing close to your opponent’s weakness

What happens when you play close to your opponent’s weak groups?

• Since weak groups are still a ways away from settling, your move is likely to be sente against the group, meaning at the very least you can keep the initiative.
• Using the basic ideas outlined in the Attack and Defense article, you can exploit the weakness of your opponent’s group to derive benefits in sente.

Playing close to your own weakness

What happens when you play close to your own weak groups?

• Adding an extra stone near a weak group will help it get closer to settling; if it settles entirely it may even turn into a strong group, which can support your other stones.
• Reinforcing a weak group will make it more difficult for your opponent to derive benefits from attacking it, and may even forestall such attacks permanently.

Caveat

This sounds very simple and all: as long as you can accurately determine which groups are strong and which are weak, it is easy to determine the correct direction of play. There is, however, one thing that makes this a nontrivial problem. Determining the strength of groups is already pretty hard!

In fact, this is the greatest obstacle to being able to correctly determine the direction of play. The strength of groups is not always as readily apparent as in the examples we have chosen to explain the concept. Most mistakes in direction of play arise from miscalculating the true strength and weakness of groups.

Since strength and weakness of groups is ultimately tied to their life and death status, this makes life and death practice fundamental to all baduk players. In fact, this is the single most important reason for you to practice life and death problems: it allows you to properly evaluate the strength of each group, which allows you to determine the correct direction of play.

Conclusion and takeaways

In this article we have introduced the concept of direction of play, and how it can be determined using the concept of strong and weak groups. We have also seen simple examples of how this calculation works in practice.

Direction of play is a deep concept, and this is far from the last time we will touch upon the subject. Later articles in the series will further develop this concept with more varied examples of using direction of play to inform opening and middlegame decision-making.

Return to the Concepts hub.

This is the first article in the series “The 3-3 Invasion.”

Next article

The early 3-3 invasion is probably the most front-and-center change that has taken place in baduk strategy since AlphaGo took the world by storm. Nowadays it is studied and played by professional players and amateurs alike. In this series of articles, our goal will be to build a working knowledge of the 3-3 invasion which will be sufficient for you to play with and against it well into dan level. This first article will serve as an introduction to the basic lines of the 3-3 invasion, which will be studied individually in subsequent articles in the series.

Old theory

For the sake of perspective, let us first start with an explanation of the old understanding of the early 3-3 invasion.

This is what was classically considered to be jōseki in the event of an early 3-3 invasion. It is a straightforward tradeoff of territory and outside influence. The reason it was frowned upon as an early-game move is because Black is quite thick. White may have gotten the corner in sente, but is also totally sealed in and the Black tiger’s mouth makes it very hard for White to successfully pull off any operations on the bottom side. The tiger’s mouth also alleviates a lot of the bad aji at the squared cutting point. Since Black’s lower stones have quite a good shape, Black may decide based on the situation to ignore a peep at the cutting point, aiming to settle its two sides separately if need be.

The paradigm shift regarding this position brought about by AI is that the hane-connect is a bad exchange for White. Although it is sente, it is not valuable enough to compare to the thickness White gives away in Black’s tiger’s mouth. Instead, the AI suggestion is that White omit it altogether and tenuki.

These are the possible lines for Black followups if White plays elsewhere. In all cases, as long as White plays correctly, White can make life. Black is still thick, but Black would always have been thick, so there is no opportunity cost. Therefore it is slow for White to spend an extra move here, since the reinforcement is unnecessary, and also slow for Black since ultimately the White corner is not under threat. White can instead tenuki in a way that, in addition to the inherent value of the move, helps to mitigate the influence of Black’s thickness. Leaving the position unfinished also gives White more options to reduce Black’s influence with plays on the bottom side and the aji inherent in Black’s weakness at Q5, whereas previously the strength of the tiger’s mouth discourages such maneuvers.

The ultimate result of this analysis is that the extension after White’s hane at S5 has largely disappeared from high-level opening play. After all, Black wants to do better than not-all-that-much influence in gote. This sets the stage for the new and improved early 3-3 invasion.

The three main lines

The AI-era 3-3 invasion is a fairly complex beast when considered in its entirety. However, for the vast majority of circumstances, you can actually get away with knowing only the basic variations from three main lines:

• the double hane;
• the knight’s move;
• the push.

There is one other major line seen in professional play, but it offers miniscule advantages at the cost of immense complications, and can be totally avoided by both players. We mention it at the end, but we will not devote serious study to it.

The double hane

If Black wants to keep the corner, then Black can use the double hane. This comes at the cost of White getting an outside ponnuki and sente.

The knight’s move

If Black values sente, then Black can extend instead of the hane. Then it is White’s turn to influence which jōseki is played. The knight’s move is a fairly straightforward trade of outside influence and sente for the corner.

The push

White could also choose to extend and invite Black’s hane. After White’s hane Black can tenuki, but also has several options for playing locally. This can lead to a number of different outcomes, some of which are sensitive to ladder considerations.

Keeping it simple

There is a final line worth noting, in which Black plays the knight’s move instead of extending after White pushes. This line is home to many complicated variations that are difficult to play correctly.

Fortunately, neither player needs to know anything about this line if they do not want to deal with these complications. Black can avoid it entirely by not playing the knight’s move, and White can avoid it entirely by transposing to the push line instead of attaching below the knight’s move.

This line sees some use in professional play, but for amateurs the variations can be too complex to use properly as part of a whole-board strategy. There are very few, if any, circumstances in which this line is significantly better than the aforementioned three lines to a degree that matters for amateur play. Certainly at kyu level, it is perfectly acceptable to study just the three simple lines, and avoid this one if it is offered. We in turn will not include it as part of this series.

Conclusion

In this article we laid down the thought process behind the shift in opinion regarding the early 3-3 invasion brought about by AI. We then introduced the three main lines of the 3-3 invasion: the double hane, the knight’s move, and the push. One can play the 3-3 invasion all the way through dan level knowing just the basics of these three lines.

Future articles in this series will begin to examine each line and its variations in detail, explaining the whole-board contexts in which they are appropriate.

Return to the 4-4 Jōseki page.