It is currently Thu Mar 28, 2024 6:25 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 1 post ] 
Author Message
Offline
 Post subject: game forensics
Post #1 Posted: Wed Jul 27, 2022 1:11 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
This was mentioned by Akira in the Chinese version of Hikaru no Go. Having seen the result, can you deduce the order of moves that created it? In this way, many why questions can be answered.

Not something I have any skill/experience in, but just some thoughts.

A sort of narrative could sound like this, based on seeing just the end position of https://katagotraining.org/sgfplayer/ra ... es/745024/ and then scrolling back a little:

bear in mind that in AI scoring, dead stones must be taken off the board.
observations and guesses
Why are these four white stones captured by black? Perhaps it could be that they were threatening the black group between it and the corner, but that black group escaped on the 3rd line, after which it wasn't so big to hold onto the white harassing stones, and capturing them became the easiest way for black's centre group to escape, while white probably attacked black's centre to make eyespace for the isolated weak white group below it.

Dead stones tend not to be deep inside the opponent's area as normally an effort is made to rescue them or reduce their loss.

In black's lower right corner group, why has black filled in an extra 5 stones in black's own territory? Probably white cut previously as ko threats, so black had to pull back to capture them, after which the cutting point remained, and black had to play inside again in response to that ko threat, after which the corner is finally pass-alive.

It looks like B has all the corners, but the solidity of black's lower right is suggestive of a B shimari, and the crampedness of W in the upper right suggests it was a B corner that got attacked. The lower left looks like a W 4-4, but it doesn't seem like a 3-3 invasion as W has both 3-4 points. Instead, B must have entered more carefully with an approach on the left, not depending on the 3-4 points (attacking the 4-4) for eyespace, but instead on the side access. Meanwhile, W had enough eyespace locally and was alive enough and low enough (B is alive) not to try to develop a moyo with this group but just expand locally. The upper left looks like a W 3-4 B low approach, W keima which developed into the upper side.

Why does white have 2 groups while B has 3? The low numbers suggests a moyo game, parallel fuseki. B will likely have more if B tries to invade W's side at the start.

Bulky shapes such as in the upper left on centre of B tend to be either dame or a ko. A ko in the upper left is fairly typical. One reason the centre has so many stones seems to be because B captured many small W chains.


The key tools to use are that each move must be the biggest (or one of them) when it was played, finiteness of space, influence value spreads in 4 directions, all groups need eyespace, but one way is just having so much access that the opponent is unlikely to be able to block all of them from becoming eyes (as the opponent needs eyespace themselves to block).

For example, the least useful stones on the board are likely to have been played during dame filling.

just endgame points. points given are gains rather than the swings I am more used to for exercise.
So much drama happens when temperatures drops to 1, since there are tricky kos. (move 223-4)
251: B has 7 ko threats, W has 2. This with follow up gains 5pts + (6pts follow up + 10pts follow up)/2= 13pts.
249: necessary. W first then B can make 1 eye with a liberty helped by W's having only 2 liberties, but B's other group only has 2 liberties now and 1 is only temporary as W fills from the outside. Hence B's other group dies giving W an extra liberty to two liberties on the outside, enough to kill B's 1 eye group too.
245: bad shape for ko threats though better for immediate and clear life. Playing at 248 would leave 1 less ko threat (namely 248) but instead of seeing 2 eyes immediately, B would need to read if W can make seki with a threat. It seems that 245 may be better if it saves computing power in the sense that a smaller space needs to be read out. If B has two eyes, W needs two eyes to live inside, which is clearly impossible, but if B only has 1 eye, W only needs 1 eye for seki aji. In my kg analysis, taking the ko is considered best, but locally 248 is the top move, but if B is behind, 245 has a higher wp though lower score.
244: normally we end the ko if there are only dame left and we can't win the ko. There are 9 dame left and a 0.5 pt move (which must as usual add one dame where the point is). W taking the 0.5pt will also get the last dame in exchange for losing the ko for 0.5+5-4-1=0.5. B can't take all the dame since too many ko threats depend on them. If W ends the ko, each get 5 dame and W loses the ko for -0.5+1=0.5. Probably taking dame is the "stronger move" in case B doesn't find good ko threats and W is confident this isn't somewhere W can go wrong.
243: strange. 246 is a clear point that also reduces the ko size from 1/3 in sente to 1/3 in gote. (nb W's threat of 2 moves in a row gains 5pt) Then 10 dame would be left with 2 symmetric kos. In the game, B instead gives up 1pt here for winning an extra ko (2pt) but loses 1 pt (last dame) to balance. Unusual enough to surprise me, but probably good instinct. You can't win 2 kos, but if you have many ko threats, filling one certainly threatens to win both. A ko concerns 2 squares, so it has a slightly larger swing though the gain is normally smaller.
242: 1 5/6. curious that even after this a cutting point remains. This often happens in the corner as it is valuable and a fortification may be secure even if there is a cutting point so leaving a cutting point is common for efficiency to escape the cramped space. Before this the cut didn't work but there is a tesuji combination. If W ignores the tesuji, B gains a further 4 1/3 points, unusually big as B can still get an eye in the corner.
241: 2pt sente. No better move as though W has a weak point, B is too far from it. Also, W's weak point below didn't help before.
240: W is lucky to get the last move of this size. It prevents B's double threat which gains 2/3 above and 1 2/3 below (if W doesn't respond). As this was the first move of its size, W had the chance to take it before B played it all in sente.
239: approx (ko) 2 2/3 pts reverse sente
238: 1pt sente
237: 3pt + 2/3 swing from the ko (1/2 from the pt, 1/6 from the ko) + more aji/cutting point on the right so follow ups are larger even though we expect that W will answer anyway. Hence 1 5/6 gain. (bigger than the upper right corner as that would take 2 moves rather than only 1)
236: 3 1/2 gain
235: 1pt sente + 1 ko threat of size around 2 2/3 (and increase size of previous threats). Good practice to play before a gote even as large as 1 5/6 as it is actually the largest other move. (next largest is 2/3 sente in the upper right). Though bad practice otherwise to not maximise locally. Taking the stone would gain 5/6 in gote instead.
234: 15 1/2pts + (14pt follow up )/2=22 1/2 swing. 11 1/4 gain.
233: As 235.
232: 3 + 3/4 (5/3)+ 1/4(1/3) - 2/3 + 6 locally = 9 2/3 swing or 4 5/6 gain. But it affects 237 too, increasing B's profit from it by at least 1/3 as W gets a cutting point.
231: As 235.
230: around 1/3 in sente, but as B has many more ko threats big and small, it could gain more like 1pt (assuming B is komaster for the smaller two-step ko but not for this larger 5pt+ direct ko). At the moment, W doesn't dare to take the ko as B will make ko.
229: swing: 3 + follow ups (2 - 1)/2 + (4/3) /4 (follow up to follow up) = 3 5/6. Total 1 11/12 gain.
228: 0 pts sente.
227: 1pt gain. very standard, but there is aji inside for ko threats that this helps solve too. The 1x2= 2 pts swing makes 2pt + 2pt (2 moves compared)=4pt area scoring. And 4/2=2 squares changed. Of course one is N9 itself. The other is M9. N10, N8, O9 are unaffected being alive, but influence extends a little further via M9 to the L10 cutting point aji.
226: 2/3 reverse sente (with sente up to a 2pt gain follow up which is much larger than 2/3)
225: if B can't fight the ko, this is very large. Like a double sente. If B responds to W's follow up, it is reverse sente 7pts. If B ignores, it is like 5 gain ish locally + the 2/3 in the upper left corner. AI says that connecting the 3 stones in lower right is also fine as that also gains average of 5.5 if ignored (11 with follow up), and B has more ko threats too.
224: standard 1pt gain, but also 2 ko threats size around 4pts + which become W's only threats.
223: 1.5pt gain if I read the corner correctly.
222: 0pt sente, but is a way to save the centre 12pts too, so can see this as 6pts gain.
221: this early on and as B has many ko threats, this is gain of 1 1/6. If W were ko master, it would gain 1 1/2. It would often affect the W group ko by taking the eye, but in this case, W's weak point is too far from B's position to matter, while also close enough to owe a move regardless. This is a case of the defender having an advantage since one move defends both sides.
220: B has a big follow up with the kosumi too. Assuming game is best, this gains 2 1/2. Actually kg, says the 2nd line bump dominates here though the same as it works too, with a bigger threat.
219: If W wins the ko, I can't see any life and death issues in the corner, though seki is close. More it is W gets K2,K3 in sente. So locally the swing is 3pts, and K2/3 is 2pts, and there is a follow up of 2 1/2 (220). A move then gains around (7 1/2 )/3 = 2 1/2. If B had enough ko threats, B would risk K2 first as the next largest move gains much less. However, K14 is too small, W has A7, and critically even J7 isn't a threat yet since if W gets P2 for O1, W gets K2/K3 in sente even if B wins the ko. B risks around 33 pts, so S7 is similarly not a threat.
218: doesn't look that big compared to the ko, but local swing is 4pt + (2 1/6 follow up) /2 = 5 1/6 or 2 7/12 gain. Especially as all these 4pts are gained since W must connect to maintain the threat of ko. kg says taking the ko loses 1pt. I think I slightly overestimated the ko size as K2/K3 isn't sufficient to secure K1. Note that W fighting the ko here is pointless not because you can't fight 2 kos (this ko is bigger than the one below), but because B has too many threats. S7 and its T5 follow up threatens 8pts + aji on both sides (nonzero for points).
217: as 221
216: at least 2 1/2 gain , but there is ko aji inside where B cuts with 4 liberties. If W kills, W gains 3pts from extra dead stones but if they have to fill all the liberties, that costs 4pts as well as B's gain from 3 extra forcing moves (one forcing move at T5 B already had). Likely W loses 4 pts if W loses the ko. W backing down from the ko also loses 4 pts. This is the last direct access to W's territory, but the combination with W's weak point at T5 can be severe too. This helps P9 by cutting and killing B's Q9
215: sente 2pts.
214: a reverse sente, around 5pts + 2/4 (if B adds 2 moves, B also captures P9) - 1/8 (if W responds, that threatens to make O10 a point) = 5 3/8. This is smaller than I thought. In fact it is even smaller as W can respond to Q8 at S8 which gains around 1pt + 1pt in the corner. So 3 3/8 overall.
213: swing: around 3pt capture + saving 11 + 4/3 /2 = 14 2/3 at least. Actually it is bigger by perhaps 2 1/6 from M6 and I assumed B gets K16. If W gets K16 that swings around 3pts. So perhaps 18 1/3 in total or 9 1/6 gain.
212: 1pt sente
211: saves 8 stones (16), 3 dame, follow ups (3 3/8 + 2 + 2 1/3)/2 = 22 41/48 or 11 41/96 gain. But this is with 2 moves of this size both for W, when a response would ensure none. This is why ko is confusing. It isn't resolved with just one move, and if it is large, then it is possible for the winner to get all the big moves. So komaster gets the swing of the ko (minus temperature) rather than just a 3rd of it.
210: 214 is also an option. swing is 1 1/6 + 2 + 2/2 - 1/4 = 4 1/12. 2 1/24 in sente.
209: 8pts swing, 4pt gain.
208: 1pt sente.
207: curiously two sides have different best moves if they play first. This is uncommon between gote moves in the endgame, but can happen if there is a ko. W should take the ko and just exchange P2 for O1 apparently. Locally the swing is 2 + 3 3/8 + ko for 7 pts compared to 1 1/16 game (ko in B's hand). Normally estimate as 2 + 3 3/8 + 7/3 - 1 1/16 = 8 37/48. So gain is 4 37/96. Probably forcing the ko once already gains 2 pts in the corner + 1/3 from ko, + outside forcing moves + removing need to play 220, gaining perhaps 2 1/2. It's a bit difficult to understand, but it seems that B playing first expects to be able to hold onto the ko by getting more ko threats with 207. The main line seems to be that W gets B to retreat in the ko but B gets Q8 and T4 which still has a follow up at T6.
206: reverse sente 3 + 1/2 ko value, perhaps around 5pts. The size of follow up if ignored is 7 pts. I say 1/2 as W is likely to get most of the profit after just taking once.
205: Compared to the game, where W needs a move to force B back, probably locally sente 1 2/3 + 1/2 of ko value
204: slight tesuji, using the weakness of P8. We have reached the point where the opponent's best move is not your own as W's best move captures B's best move. This happens if two moves (P9,P8) can capture B's best move but one move can't (or else B wouldn't play there). In particular, a move closer to open space also captures Q9. The move gain is the same as calculated for 214 at 3 3/8 in reverse sente. We see just a few moves is the difference between temperature around 3 (2nd line) and 1 (1st line). P8 instead would be flawed as B's atari may have a cut but it is efficiently defended by N9. W only gets 1pt + ko for 5 more, but B also gets 1pt and recall B has many ko threats. Then W gets around 5/3 = 1 2/3 in sente which isn't enough gain.
203: saves around 24 pts in reverse sente. Easily the biggest move considered so far. The follow up W has is around 40 pts if ignored.
202: surprising but W intends to take gote locally as there aren't bigger moves elsewhere. This ensures that B must defend at P2 since this chain can't make an eye at N1 anymore. So N1 is valuable, but not for territory (which we normally focus on) but for eyespace, so just getting B to fill it in gains something. If the B chain is only left with one liberty other than the P2 cut, then B eventually must fill in P2.
201: painful but actually very big. Instinct is to take G16 in 2pt reverse sente. But W's push and atari make it reverse sente more like 1 1/3 pt. And if W takes N9, B must defend again. Locally defending the push is already 2 pts reverse sente, so it is simply more than 1pt better than G16. But a move must be made or else W captures 10pts next move. (5pt gain)
200: 2pt sente
199: Around 9pts reverse sente. Note that N2 is around 7 1/3 pts swing. But if B isn't careful, W will get that in sente. Block isn't just big locally with the monkey jump (already around 7pt swing), but the important thing is not to let it happen in sente.
198: 2pts. double sente at this point. W's follow up is very large. B's is 6pts + good aji (though perhaps reduced as B also wants to play Q8). But even if W responds S8 to Q8, it is less of a tesuji as B has T6 to gain 2pt in sente again which wasn't possible if W had gotten 1 more liberty by crawling. Normally we might count a1 + a2/2 + ... for the size of Q8 and b1 + b2/2 + .. for the size of S5, but here we mustn't overcount as S8 reduces both simultaneously, though perhaps only by 1pt. It is as if a2 conflicts with b2 but a2 was so big already that Q8 was already counted as 2a1. So the main effect is to reduce b2. Note that it wasn't necessary before as if B played Q8, such a stone could be killed or otherwise W lives so that B at S8 isn't a threat.
197: Around 15 swing or 7.5 gain.
196: O6 is 3pts worse as this way W still has positive (but very small) points locally rather than 3. But losing gote loses 2+3 already for S5, and around 2 more from not being able to capture Q9.
195: like 197 but around 3pts smaller in swing, perhaps 6pts gain.
194: Starting to leave endgame that I can't read at a glance. But still very standard, good shape with R9. If we assume B responds, then locally, it is the same as if B played here in sente except B has lost 1pt if B gets both Q12 and T7, which is less than 1pt gain (though close as both are valuable B moves by themselves). Otherwise, W gets 198 and prevents B there, so overall nearly 6pts gain. But B could get this in sente. It seems that the sente is bigger than what seems like a larger gote. This sometimes happens if the next gote is also pretty large compared to the rest of the endgame. Namely 204.
190-193: forcing moves that help life too just in case. This is the only profitable direction to use the aji. Good to prevent B stubbornly holding the ko. 190 first is good practice to connect so that H5 is secure territory. kg suggests B8 as a first forcing move, probably to maximise the threat at A7. A16 becomes less threatening but it never was.
189: 6pts + L1 aji reverse sente. Or can take B1 tesuji for a swing of 11pts. Biggest for now. B loses 2 pts by playing P7 or S6 probably due to the L1 aji.
188: valuable (though almost always forcing first) to secure E3 eye so there is no chance of seki.
187: If w responds, B gets 3pts in sente since there are 2 cutting points adjacent as well as W's main area.
186: W is alive anyway, but B can take 8 pts - L1 aji + N2 aji (worth perhaps (4 2/3)/2 = 2 1/3)
185: 3 pt+ L1 aji sente
184: like 5pts reverse sente, ish.
183: 3pt sente ish.
182: N7 might lose a point. This looks clearly biggest. Assuming W can't fight the ko, this is 8 + (7 1/3)/2 = 11 2/3 swing or 5 5/6 gain. Apparently W can fight the ko but taking gote on the right will still lose at least 1 pt and often 2pts. But see 163. Apparently this may have been W's last chance to win with a tesuji.
180: The ko is worth around 4pts but W gets it in sente, threatening 36 + large follow ups. It seems W's several dead stones create aji for many ko threats worth 20+ pts, whereas B has barely any large ko threats. Later on this reverses as W uses up large ko threats for this ko and runs out of ko threats larger than 4 pts while B still has several that are more like 8 pts.
170: good timing as otherwise B plays E1 and E2 and makes the ko heavy on W. For now, W has many local ko threats that keep W's local group healthy even upon losing the ko.
169: subtle but the only move kg considers. This helps the weakest group when play continues, expecting that the throw in A4 doesn't gain W anything (though B loses a liberty), expecting that W will use the ko to squeeze the outside so C7 will be quite dead and B can fix all the problems by capturing C7. Otherwise, the cut B3 will remain as a very large ko threat and bad aji.
168: now it is a real flower ko.
165: probably bad instinct to block rather than empty triangle back though kg only rates retreating as slightly better. In the game both sides get one more point and B can still fight the ko, but W is more clearly alive, rather than potentially needing to spend another move inside. kg thinks this could have been a losing move if W found the centre tesuji.
164: 4-5 pt sente. kg also suggest J12 as a tesuji to seal the centre with M8 in sente. It no longer works as well in the game as W loses a liberty with the centre ko threat, but it still works and kg thinks it gains a point over 182.
163: bad shape but trying to fight the ko loses 3 points since there are so many ko threats in the lower left. It is effectively double sente 1 1/3 pts, with big (6pt+) follow ups for both sides.
148: questionable as unnecessary. direct kosumi is devastating (W+2), as W threatens L10. Kg seems to still have blind spots in life and death.
159: around 6-7 pts sente + L1 aji.
158: around 7 pts reverse sente locally.


having seen how the corners started now
I suppose the solidity and largeness of B's lower right suggests B got it in an exchange by attacking rather than a shimari.

The upper right was a shimari instead, which is perhaps visible by the fact that W's access on the top is very cramped all on the 2nd line and not really threatening B's connection on the outside.

The upper left was actually a solid W kick, one space jump but I was closer there. This made the top smaller so it is more understandable why B never invaded the top side.

In the lower left, I had some correct points but perhaps I should have recognised the kick. And the shoulder hit to the small knight's move on the lower side.


hard to avoid thinking about influence functions

116: not strictly necessary but such a quiet move can be so powerful when B is thin nearby. Already Q9 is difficult to rescue when it could have caused trouble before and M9 is a double threat too.
182: W has a weak stone on the boundary but it is only a small corner so only medium size endgame. Likewise for the C18 follow up. Just one white connection will help both weak stones and secure the area.
183: can be thought of as a probe. B doesn't want to play bad shape at 185 immediately as it walks into a tiger mouth when N9 was also possible. But the combination exposes 3 cuts so that B is securely connected without needing N9.
194: a bit subtle to get more 1st line security when W is already secure on the 2nd.

M2/O2 combination. The meaning is that W doesn't get much eyespace by playing O2, though big for territory/attack. So B is comfortably able to connect while W is still thin on the outside, though in gote.

It seems to be endgame training is good for the whole game because when expanding territory/development, an area must be close to territory to be valuable, so you get some intuition as to how to count it and how much further moves change the situation. It feels like a soap bubble that can stretch as you pump moves into it or pop when it solidifies into territory.

Q7 eyes the R5 expecting to get a move in sente from it sooner or later which will greatly help life and death and fighting (prevent B P6 directly). However it takes W another move to prepare it.


why is the middle game so important and violent? When a territory is one move away from secure, the value of moves gets higher. Especially if you don't have the skill to completely surround it without a long term attack or exchange, and you are uncertain as to what the result should be or will be, then you might overly retreat or over attack.


20220906

tips: look for recognisable joseki. In which order did they play it and why?

an empty triangle should only appear on the board when forced at the boundary of groups.
Otherwise it is too inefficient to play alone.
If one side has more stones at a fighting area than you expect, it is probably because one side is trying to make that territory, or is strong enough to tenuki and still live.
This may appear from a battle to live/kill in middlegame.
If one side has an extra move when they are already alive on the other side, they likely played it when the other side wasn't secure yet.

Check where the eyespace of groups is. If in valuable space such as sides and corners, they probably had a base there first. If they have more stones locally they probably always had more eyespace locally. If they only have eyes in the centre, they probably had to run there. If they are unusually clumpy near the boundary of groups, they probably already made thick shape nearby from attacking both sides and in the endgame had more control of that boundary through being thick though it may look overconcentrated.


This post by dhu163 was liked by: Elom0
Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 1 post ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: Majestic-12 [Bot] and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group