Wacky 9x9 book refinements

[hexahedron]

I’ve been working on growing out the 9x9 opening books a bit more for a future update to https://katagobooks.org/ and a few days ago I noticed that the newest version on the machine I have computing it suddenly had a different eval for one of the subvariations in a major common opening, “Bean throwing” ( Weird and Wonderful 9x9 Openings ), under Japanese rules.

Under Chinese-like rules the scoring is more lenient allowing other options for draws by one or both players so this discovery there doesn’t result in so big a change.

Anyways, I looked a bit down the variation and it’s really sharp and interesting. As we get a bit deeper into the variation, I’d guess that some dan players might have fun trying to find the tesujis or solve the endgame tactics themselves before peeking at the answers, so I’ve layered them as nested hidden blocks.

Okay! So, in Japanese rules, with the fair komi of 6, the current book thinks this move (white 8) is bad, and quite possibly losing. But the new book suggests it’s probably fine for white! What tactic did the new book find?

Well, stepping forward a few moves…

Both books agree that pushing through A is bad for white (black would respond tightly at D6). The current book thinks every other move is bad and/or likely losing too. The new book disagrees and thinks white has a unique move that equalizes. What is it?

Answer

…

…

…

Clamp!

Okay, what next?

Both books agree that F7 is a good reply for black 13. Now white pushes through at 14, and black 15 still replies tightly at D6.

By tewari, compared to white pushing directly, we see that by playing the clamp first, white has managed to exchange E6 for the passive F7, which is clearly a nice gain for white (black would normally just cut or atari white 3). Assuming of course that white doesn’t have some better way to use the aji of G6, or shouldn’t play for life with G6 directly, or that black can’t get a better result by resisting more, which is non-obvious.

Anyways, the old book thinks this position is still totally winning for black while the new book disagrees and again thinks white has a unique move to achieve a draw. What is it?

Answer

…

…

…

Clamp. Again!

This second-line sort of clamp isn’t usually one of the moves in my repertoire, but I figure it’s the kind of move that is definitely under consideration for players that are better than me at the game. Still, I’d guess this kind of move is rarely ever an easy or automatic move, and usually requires a bit of reading.

Anyways, at this point we drop out of the old book as it doesn’t have this clamp. But, even for moves that don’t end up in the book, it spends thousands of playouts searching them for whether they are worth adding, and it turns out this clamp had a whole 19% of the policy probability of the raw policy of the neural net, so it was definitely considered by the old book, the old book just thought it was bad and didn’t include it. What blind spot did the old book have in rejecting this clamp, that the new book has busted past?

Well, let’s suppose black 17 now just ignores the clamp and cuts and eats the white stone.

What should white do next? Grab a stone back at A? Atari at B? Something else?

Answer

…

…

…

Ignore the clamped stone and hop deeper!

This apparently was unintuitive to the policy net, but the new book claims it’s again the unique move for white to draw. It’s also unintuitive to me, but I can’t speak for stronger players. Is this move instinctive?

It’s not the end of fun stuff though! Going forward a few more moves:

Now, time for a puzzle the other way around. Supposedly now it’s black that has to find a unique move or be losing. What should black play?

Answer

…

…

…

Peep! Magic at the 1-2 point! Supposedly the clamp at A or the extend at B instead for black both lose the game.

So I think one of the meanings of this peep is that if white ignores it, it has a much sharper followup than the more normal-looking black clamp at A would, winning black the stone at A6 like so:

But the other meaning is that if white responds, now black extends at B…

and apparently the exchange of 23 and 24 is a good exchange for black!
Maybe to offer a final puzzle in which that exchange now plays a big role, White’s next move to maintain a draw is apparently again unique. What should it be?

Answer

…

…

…

First line diagonal!

If KataGo is to be trusted this is again a unique move for white, all other moves lose. (Of course, given how picky this endgame has been so far, maybe KataGo’s still missing other possibilities!)

So, what do you think?

There must be tons more blind spots in deep variations elsewhere in the 9x9 book, but I think it’s wild that we’re able make this much progress to maybe resolve endgames with unique moves and tesuji as sharp as this, to where we’re forming opinions about major plausible moves as early as move 8 being losing or not. (Or even move 1, in the case of opening on 3-4!). Even if mathematically-rigorous solving is astronomically out of reach, the 9x9 board doesn’t actually doesn’t seem that far out of reach from solving all the major opening lines to a decent practical degree of confidence, if enough people wanted to assemble a big project to throw a few more orders of magnitude of compute at it.

[hexahedron]

Hm, the hiding blocks aren’t working for me - they’re starting expanded (?)
Test

Summary

This text will be hidden

Edit: not sure what’s going on. Did the amount of nesting or images break the forum software or something? Or is it just my browser that the collapsible blocks for hiding stuff are starting out as pre-expanded?

Summary

Test

Anyways, I can’t seem to get the blocks to hide properly, so I guess I’ll just leave the post as-is for people to enjoy and maybe someone who knows about the forum can help if there’s a way to fix it.

[corner.square]

@hexahedron

not sure what’s going on

in old way to create posts it looks like that:

[details="Summary"]
test
[/details]

---

in new way

if you create hidden text that is opened during publishing, it would stay opened by default

Summary

test

you need to close it before posting

Summary

test

___

I guess if you edit post, close them and click save, it will be fixed

[hexahedron]

@corner.square
Thanks! I think that works. But that is both unintuitive AND I tried that and due to a forum interface bug at least on my browser, I was convinced it didn’t work and so I presumably undid it in the process of trying other things.

So here:

I click to edit, and close the box:

Then I click to save the edit, and here’s what it still looks like after editing:

It looks like it didn’t work!

However, if I reload the page after that as well, then it worked:

So presumably when I tried it I didn’t click to reload the page so I assumed it didn’t work, and made more edits that undid it.

@anoek - Is there any possibility to make the text “This text will be hidden if this block is not open when submitting the post”? (Instead of "simply “This text will be hidden”)

[Sadaharu]

I think this is more of a discourse functionality?

[hexahedron]

Aaaanyways, now that that’s resolved, (whew, thanks everyone!) if anyone has any thoughts or comments about the actual book variation or tactics, feel free to post.

[mark5000]

Fascinating line! Just some background on it to help others who might be interested:

@hexahedron

The Bean Throwing variation of the Black Boomerang opening is well-studied and appears in books like Amazon.co.jp: The Final Edition! Go Complete Guide for 9 Paths (Gojin Books) : 安斎 伸彰: Japanese Books. That book recommends f2 and h5 for White 8 rather than g6, reasoning that since Black has many stones in the area, it’s a bit difficult for White to make sabaki after g6.

Above is one way White can make sabaki, but the author considered Black 8 in the variation as too big, with Black having good attacks against both white groups. The KataGo books sort of confirm this. They give White a bad win percentage in Japanese rules and a probable draw in Tromp-Taylor rules but with a slightly disadvantageous win percentage and an unfavorable score evaluation.

In this light, preemptively occupying Black 8 in the variation makes sense for White, but it’s a novel idea for OGS. I’ve never seen it played here. The closest game I’ve logged was https://online-go.com/game/23132536/8, with KataGo playing Black, which featured the moves in a different order, KataGo answering White f2 at g3.

Black tenuki to c5 has been known to human players for a long time. (See example OGS games one and two.) But somehow, the tactics were never put together, possibly because Black 9 in the original position signals Black’s interest in the lower right.

Nice to know that these ideas are compatible without sacrificing accuracy!

[hexahedron]

I still find it magic that the sequence from move 23 is a beneficial exchange. I’ve never seen that shape before, and my first instinct would be to think that it’s a losing exchange, not even a neutral exchange. But no, it magically restricts white even after white responds.

[garlicsoup]

Yeah this is hard to grasp…

Edit: Is it that white has one less liberty when connecting B7 back to B4? I guess later on black would play C8 followed by white B8 and black B9 …

[hexahedron]

Yeah there’s a few things going on that the stone is just magically positioned for somehow.

Summary

For example, here if white now plays C9, black has a throw-in ko at B9 because of A8.

Monkey jump leads to ko too.

If black omits the exchange white’s diagonal is the obvious followup, and now if black were to peep A, white would play B rather than C and make an eye in the corner.

With the exchange played though, white’s diagonal now is an empty triangle, and A8 is in the perfect place to throw in and destroy the corner eye and white is very short of liberties.

What’s hilarious is that even in some variations where white does make an eye in the corner, A8 is still providing value by making white owe an extra defensive move! Here, eventually once the dame are filled, white will then owe a move, otherwise black makes a throw-in ko at A (!)

If we take away the exchange, then white will no longer owe a move inside. If black plays A now, white will respond at B, not B8. And then white will be able to connect at C and not owe any moves inside. (Black can’t play C or else D becomes white’s sente instead of black’s because black is also short of liberties).

[garlicsoup]

Astounding, thank you for elaborating on that.

Just for the sake of the hypothetical I wonder if there is a reason Black should play A8 before G3, rather than G3, A8, H3.

I am hoping to internalize something from this scenario and later to delve into your other work.

[mark5000]

White would answer a8 at h3, leaving Black a nasty liberty shortage in the lower right and granting White further advancement along the right edge.

[garlicsoup]

The position after B G3 and W G2 is more vulnerable than it was before with just B F3? White can’t ignore A8 in the other scenario and play G3 to take the corner?