Discussion about Q16,R16,R15,R17

At position Q16,R16,R15,R17, @aesalon started the conversation:

Given that R16 tries to dictate the direction of play. B and C are swapped here. Black B would be giving white a ponnuki on the right side (or undercutting it). A and C are consistent with the joseki

( FYI @aesalon )

When going against our lowest level of sourcing (evidence of professional/strong play), I think that is when we most want sources.

1 Like

Do you know of a good source for this position? I only know of:

Fan Hui says that B equals a 3-3 invasion and that Blackā€™s direction can become weird. I interpret that not to mean B is bad but rather the R16 attachment can be good, giving that benefit. Does anyone speak more directly about this?

I havenā€™t, Iā€™m surprised that itā€™s not in an AlphaGo game that Redmond has done. It shows up in the teaching tool as A->C->B but even with pro games alone, the descent should be Ideal. You can use the description for C on move B (or 50 other joseki) that may be called Ideal.

I looked at the games where this situation appeared and am less on the fence about Q17. I think putting it as Ideal if there is any question is the recommendation. It would be great to get a good source on this position. R16 is an interesting move.

I was surprised too. At present, it seems like a credible and current source for this position doesnā€™t exist. Since this is the case, I ran test positions with LeelaZero and KataGo, and they didnā€™t like C as much. Thatā€™s the reason for the GOOD label. I confirmed this by looking at AlphaGo Teach, where AlphaGo wonā€™t play C until you make a 10x10 square or smaller. That suggests to me that C is context sensitive and thus situational.

I also tested every pro game that I have downloaded that chose to block. It was a mistake more often than the pro games who chose the descentā€¦ but I was under the impression that AI evaluations arenā€™t a source we wanted and AlphaGo Teach is little better.

What we have are professional records and:

Figuring out when to use a Good move type is one of the trickiest parts of Josekipedia. We would have loved to not force this distinction, but the cost would have been too great. Some moves simply cannot be called correct or joseki, and yet cannot be called bad.

The basic rule is not to use the Good move type unless absolutely sure. There are few of these moves compared to the Ideal choice.

Sorry, Iā€™m getting confused. Are you using the term ā€œblockā€ to mean the Q15 solid connection or the Q17 push? More clarity on what youā€™re proposing to change would be great, too.

EDIT: I removed path B. Iā€™d like to hear Eugeneā€™s opinion before removing C.

If I understand correctly, the argument for including C as ā€œGoodā€ is that it once was joseki (is in Kogos) so it has a reason to be in, but we believe that it is no longer ā€œIdealā€ - partly because we canā€™t find a modern source to play it, and partly because the AI doesnā€™t like it

That makes sense to me.

I want to note that this doesnā€™t mean ā€œCall it Ideal unless you are sure it is not Idealā€. It actually means ā€œdonā€™t put in moves that are not Joseki unless you are sure that they are really worth includingā€.

IE ā€œuseā€ in this context means ā€œinclude in the dictionaryā€. In my opinion.

ā€œIf it is not Ideal, then it is not Joseki, and should not be inā€
ā€œBut this move deserves to be there, it is not bad, it is correct, itā€™s just not joseki anymoreā€
ā€œOh, OK, we will include it, but call it Good. There should be much less of these than Ideal josekiā€

1 Like

I think there is an interesting and important question to explore here actually.

I was thinking about the analysis: ā€œC is no longer ā€˜Idealā€™ because the AI does not like that moveā€

This can be paraphrased as ā€œwe no longer think that C is joseki because the AI does not like that moveā€.

BUT I am thinkingā€¦

Moves are not ā€œjosekiā€. Sequences are. << Right?

This lead me to paraphrase the definition of joseki like this:

ā€œA sequence of moves is joseki if it results in an even position when both players chose joseki moves, and one player would be worse off if they did notā€.

With this definition, it doesā€™t matter if one of the choices of joseki moves is ā€œless favouredā€. What matters is whether, after playing that move, the opponent can do no better than ā€œevenā€ at the end of that sequence.

Based on the current data in our Joseki Explorer, it appears that C meets this criteria. It leads to https://online-go.com/joseki/18487, which is described as ā€œIdealā€. That means that if Black plays C, white has to play Joseki moves to stay even, and it does lead to an even position.

By this argument, we would conclude that either:

Curiously, one would think that the latter ought to be true: if the AI thinks that C is worse than A, and A leads to an even position after playing the best moves we know, then how can it be that C does also?

This seems to mean that there is an as-yet unfound refutation of the sequence to https://online-go.com/joseki/18487, or that this position itself is not as even as we thought.

1 Like

I may have misunderstood you, but Iā€™ve been using ā€œJoseki: Position is settledā€ as synonymous with ā€œPosition is settled, and this sequence is a corner sequence, not fuseki.ā€ I didnā€™t mean to represent the end position as joseki there.

Good discussion to have then! :slight_smile:

At some point I added

" Please put the ā€œJoseki: Position Is Settledā€ tag on the final position of a joseki sequence"

to the doco, in an attempt to clarify what I had intended with that tag.

This may be my own misunderstanding of what the Josekipedia tag of the same name is: I have always thought that this is what it means.

I think itā€™s useful for it to mean that, because then you can tell the filter ā€œonly show me sequences that are josekiā€.

Hereā€™s a curious thing. You said you take it to mean

ā€œPosition is settled, and this sequence is a corner sequenceā€.

Combined with ā€œIdealā€ (which the position under discussion is marked as) isnā€™t this the same thing?

ā€œHere is an Ideal position that is settledā€ appears to me to mean ā€œThis is Josekiā€.

So for this discussion, the argument is the same: C is part of a Joseki according to our Dictionary, so how can it be anything less than Ideal?

It might not be the best choice usually, which is why it would be ā€˜Cā€™ not ā€˜Aā€™ (and ā€˜3ā€™ not ā€˜1ā€™).

Good discussion to have indeed! I noticed ā€œFuseki: Doneā€ and thought it was the opening equivalent of ā€œJoseki: Position is settled.ā€ That is, any position is either in the nature of joseki or the nature of fuseki. I donā€™t mind interpreting it your intended way, if clarity comes in the documentation.

The ā€œIdealā€ tag is another thing. I had marked the last move ā€œIdealā€ because itā€™s a perfect move and thereā€™s nothing better for that move. Iā€™ve never used it to mean everything that came before is also ideal. The presence of the ā€œGoodā€ tag on an earlier move should have put that to rest. Or so I thought. Otherwise how can we mark mistakes from both sides?

OK - letā€™s establish that the ā€œJosekiā€ tag means "this is the end (*) of a ā€œJoseki sequenceā€ and all that this brings with it.

I think actually the ā€œposition is settledā€ part of that tag needs to be taken off and put into a sibling tag of the proposed ā€œfightingā€ tag, because some Joseki do not end in a settled position.

Then the discussion about the position in this thread seem to boil down to the fact that the final move in the sequence to https://online-go.com/joseki/18487 is indeed ā€œIdealā€ - a better move cannot be found in the previous position BUT it turns out that end position is not ā€œJosekiā€ in the full sense of it.

Therefore, if that is agreed about this position that the end position of C - if the experts agree it is not Joseki - then ā€œJosekiā€ tag comes off that position and move C is indeed only ā€œGoodā€ not ā€œIdealā€.

I would put it out there that no sequence with a ā€œGoodā€ in it can be Joseki, right?

(Josekipedia says ā€œA series of Ideal moves forms a joseki.ā€)

*: note - ā€œthe endā€ also needs to be defined: it means ā€œposition at which you have achieved the goal of playing a Joseki, which is that it is the best you can play and now you can play away, or commence an even fightā€. Interestingly, some Joseki appear to have continuations that are also Joseki, so it doesnā€™t mean that there are no continuations (!)

1 Like

That all sounds good to me. Separate ā€œJosekiā€ and ā€œPosition is settledā€ tags would be useful.

1 Like