At What Point Is A Match Largely Decided?

Usually, if the game lasts long, by move 100 you can tell if it’s one sided.

If it is a peaceful game, then usually you hit endgame by move 80.

If it’s a violent game then you tend to see someone resign by 80.

3 Likes

The question seems pretty meaningless to me.

Consider one of the most high level human games possible, Iyama Yuta vs Ichiriki Ryo in this year’s Kisei title match. Currently Go Ratings considers them both to be in the World Top 20, and the title they were fighting for came with the richest domestic prize in Go.

Yet if I remember correctly, AI thought that Ichiriki threw a winning game, in which he was as much as seven points up, through an unfavourable trade deep in the midgame, I think past the hundredth move.

2 Likes

I think that one could make this question a bit more objective, while I think keeping the same spirit of the question.

  1. Gather a bunch of N-by-N game records played by professionals and/or top amateurs (high dan levels).
  2. Show only the first X moves of each game to some strong (like a professional or top amateur) observers (assuming that they are not already familiar with those specific games).
  3. Ask the observer to guess which color won each game, and record their success rate.

With data from enough games and observers, while varying the parameters X and N, one could put together a table of statistics that would contain information like “very stronger observers on average have a Y% success rate of guessing the winner after observing X moves of an N-by-N game by similarly strong players”.

7 Likes

Yes you are correct. So let me come at this slightly differently. My response to @Bugcat is intended for you too @Vsotvep. I’ve also added an addendum to the original post.

I agree, this sort of thing isn’t remotely a science. But I do think there is something tangible in the idea being explored. If you have two players who are very skilled, trying their hardest, and, this being very important for this hypothetical question, both players play the rest of the game with no serious blunders, can a player “feel out” how a game will conclude. And if so, by what move on the different board sizes. Even if we simply approach this as a game, I’m curious what people would guess.

That would be wonderful. Unfortunately, I doubt any such research study is coming. All I have is everyone’s best guess. So, what do you think the 13x13 or 19x19 “determining move number” would be?

I was considering some basic math, which I don’t think really applies to this question very well. But if there are 81 moves on the 9x9 and move 8 is the point of game development where such a definitive estimation becomes possible, how do I calculate that? I really have no idea, so looked at the difference in board sizes.

9x9 - 81 Coordinates
13x13 - 169 Coordinates
19x19 - 361 Coordinates

Blank Is What Percent Of Blank?
Equation: (1stBlank*100)/2ndBlank
8 is 9.88% of 81
17 is 10.06% of 169
36 is 9.97% of 361

Mathematically, these numbers may paint one sort of picture. But I don’t think the 13x13 and 19x19 results can be estimated in this manner. The 9x9 is small and there simply is only so much growth and wiggle room each player can bring to bear. However, the larger sizes change what is possible in a very profound way. For that reason, I don’t think Move 17 on a 13x13 feels anything like Move 8 does on the 9x9. Much more possibility exists on the 13x13 at that point. Still, I find the notion of this sort of estimation interesting nonetheless :thinking:

6 Likes

Apparently Lee Changho used to say on occasion that he could tell he had a half-point advantage as early as a hundred moves in.

Of course, we only have his word for it.

2 Likes

It sounds like…
*lowers :dark_sunglasses:*
We need some polls!

When is a 9x9 game decided?
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20

0 voters

When is a 13x13 game decided?
  • 1
  • 4
  • 7
  • 10
  • 13
  • 16
  • 19
  • 22
  • 25
  • 28
  • 31
  • 34
  • 37
  • 40
  • 43
  • 46
  • 49
  • 52
  • 55
  • 58

0 voters

When is a 19x19 game decided?
  • 1
  • 7
  • 13
  • 19
  • 25
  • 31
  • 37
  • 43
  • 49
  • 55
  • 61
  • 67
  • 73
  • 79
  • 85
  • 91
  • 97
  • 103
  • 109
  • 115

0 voters

4 Likes

Not my experience. Based on the thousands of games I played, I feel the game has just started after move 8. I only play online so congress games might be different.

3 Likes

Or if you really want to phrase it as a question and get an answer similar to @Kaworu_Nagisa’s then maybe something like:

You’re observing go games at a competition, and there is a prize for observers who can guess the most outcomes of games correctly where tiebreaks are decided by the earliest guesses. You must pick a move number and then predict the outcome of the games at that point in each game. If the game is over before the move number you’ve chosen you don’t get a guess. What move number do you pick?

You could pick a really high move number and then have lots of information about the game to decide, but you might miss out on some guesses, or someone else might do an equally good job with less information.

Similarly you could just guess everything from move 0, but it’s plausible one could do a better job by watching for blunders or just unusual play in the opening etc. Or at the very least, in theory one wouldn’t lose out by gaining information before making the same guess, except maybe the first handful of moves by the above rules (I guess if someone plays the black hole opening in the first 4-8 moves, it might affect your decision one way or another :slight_smile: )

2 Likes

There was a funny game between Ma Xiaochun, the top Chinese player of the time, and Lee where Ma tried to beat Lee at his own game, played really boring and started the endgame early. Ma lost, by half a point iirc.

5 Likes

In both cases about 10 % of the intersections are occupied, but yes, on larger boards those 10 % seem insufficient to determine the outcome to me too.

Maybe we can try to change our perspective. After move 8 on 9x9 there are 73 intersections left (ignoring captures). On a 13x13 board, it would need 169 - 73 = 96 moves to arrive at that point.

Different approach, still not satisfying I think, because 96 moves seem like too many now.

1 Like

I agree that 96 feels like too many, but I like your reasoning. I think one of the aspects of the 9x9 that makes “feeling it out” easier and more intuitive is that the middle of the board usually possess a light framework by Move 8. I don’t play the 19x19 yet, but on the 13x13 I notice a great deal of early game effort is spent developing corners and sides before the middle starts to receive any real attention.

Logically I know that you can win a match with corners and sides. That is why we usually develop them first, right? Yet, concerning my own “feelings” of how a game will turn out, I really do rely on how the middle is developing to develop such feelings. With a big hole in the center I don’t know that anyone will feel strongly about how the game will end. I could be very wrong, but that is true for myself.

Maybe another way of coming at this is by asking how many moves it takes on the 13x13 and 19x19 before the middle starts to see development that matters. “Development that matters” obviously isn’t a scientific measurement and being so generic all answers are going to be subjective. Establishing a baseline for development that matters is pretty much impossible. Regardless, it may be worth considering when trying to quantify how a person intuitively determines how a game is progressing and what the odds of each players success is.

1 Like

Pro 19x19 games are resigned roughly around move 170 (±40 maybe), so I would say that for a human, 19x19 games are generally not decided before move 170.

4 Likes

When Park Junghwan and Ke Jie play, before the first move, but after the nigiri. White nearly always wins.

1 Like

My gut feeling is that its 10/40/120 moves on 9x/13x/19x boards, or close to that ballpark. By around that point you often can feel if you’re getting crushed or if its gonna be an easy victory… or if the game is still close and the outcome depends on end-game.

2 Likes

Well, I’m a new player, playing 9x9, and on many of my games by move 8, I’ve lost (-; . A very deep game of which I’m not even scratching the surface.

2 Likes


Move 8, who will win? Area counting, komi is 6.5.

Answer

KataGo - Game ID 30367129 (run kata1)
This is a katago training game. I just went to the katago website and picked the first 9x9 game I saw. At this point black winrate was 80%. Around move 35 the percentage went back to about 50%. Finally white won by killing everything.

5 Likes

When I look at that I think White has won. White owns two corners, at least. White has the advantage of Komi, so unless White makes some mistakes costing them territory, I assume at this point that White will win. The fact that KataGo thinks Black has an 80% chance of winning at this point is ridiculous to my human mind. Simultaneously, it also makes perfect sense as I’ve seen KataGo give huge percentages to one side or the other when the board is in the early stages of development.

1 Like

For context, with 6.5 komi KataGo gives black ~85% winrate from the empty 9x9 board :slightly_smiling_face:

2 Likes

We could apply the procedure that I suggested above (see At What Point Is A Match Largely Decided? - #8 by yebellz) using KataGo training games and KataGo as the judge.

For each value of X, how many training games does KataGo correctly guess the winner from observing the first X moves?

1 Like

While reading this thread I was also thinking in KataGo, but using it with professional games and check at which move his winning probability crosses 90% on average.

3 Likes