That’s because I wasn’t accounting for prisoners correctly … should be fixed going forward.
That looks like the autoscore bug, which has been around since August 2020.
@Conrad_Melville I believe the bug you are talking about is where autoscore just doesn’t run, which might be due to temporary network issues. This looks the same because autoscore is intentionally not running - the game never made it to scoring at all. In this case, however, the black and white squares you see are a display artifact that has nothing to do with the final score assigned to the game. The reported numbers like 45.1 or 10.4 are, as far as I know, just KataGo score estimates, and they don’t have any relation to any particular determination of life and death.
One suggestion for the anti-stalling feature, I think it should only work if the player who passes 3 times is also the player who is almost certain to win. In a recent game, I was passing because I thought 2 of my opponent’s groups were dead. But it turned out the engine didn’t think so, and ended the game just calling them live and declaring yhe territory neutral rather than mine. The engine was right that I wasn’t going to win the game, but I would have preferred to play out the groups in question for practice’ sake (and again, my opponent wasn’t even passing, so he wasn’t trying to end the game either).
I totally disagree with the logic behind the anti-stalling, not every person places stone to stall, if a player leaves a weakness that can be exploited to win, or lessen the loss, the player should have the option to take said option. The anti-stalling gives carte blanche, to sdk’s (single digit kyu) a fast track to preserve thier rank
If you think there is a weakness left to exploit, you should be able to do it with one move. If you cannot do it with 3 moves, you are just stalling.
It didn’t appear to work for me in this 9x9 game:
On two different occasions I passed three times in a row, these all occurred after turn 27 and the endgame analysis gave me a 100% chance of winning each time I passed.
Edit: I’ve just had this happen in another game. Has the system been switched off?
Could you give an example of the kind of thing you want to do which is prevented by this new procedure?
I think the idea is some people might not close the tab but just navigate to some other part of OGS and e.g. play a new game, unaware you are supposed to resign a game not just leave it (seems an odd thing to think and I don’t recall every thinking that was a thing when I was a noob, but apparently some do). So it’s an education point for them, but yes useless for those who close the tab.
No, the autoscore bug is not when the autoscore doesn’t run. The autoscore does run and randomly fails to score some completely enclosed territory, which is left unmarked. Originally it affected only territory, then it began not marking dead stones in unmarked territory. This was confusing because the new cases involving unmarked stones looked like possible score cheating (i.e., as if someone had clicked the stones alive—there was no tool then to tell the difference).
The scoring update in 2021 greatly reduced the autoscore bug (based on my observations), but then it began returning after some months, and it appears to have become increasingly common in recent months, based on the number of examples I see. I have been sounding the alarm and posting examples of this in the scoring threads for the last 3 years. To give due credit, Eugene was the first moderator to notice the bug, in August of 2020.
In some cases it takes more than one move to exploit a weakness. There are some situations where, after you fill all the dames around a group, it is possible to kill some stones, or make a seki, or start a ko. When you play the dames, the opponent often becomes suspicious, thinks about it, and adds a move to defend, but not always. Sometimes they don’t see the threat, they lose some points, and they learn a lesson.
With the anti-stalling feature, the opponent wins without learning the lesson, and later when they play in Go Congress they lose a game because of it.
Nice catch, I really failed to consider teire moves.
So the anti-stalling opens up the possibility to skimp on teire by passing prematurely.
Let’s remember that you would need to be ahead already in the eyes of AI and also willing to exploit the loophole and your opponent would need to not report you. But it is a curious edge case.
Maybe one could take that into account by checking if the pass move lost a point with chinese rules? Like: only count pass moves that don’t change the result regardless of rules set.
I’m not sure I understand your second paragraph, Animiral. Take this game for example. After move 289, White passed, but White still owes me a move around G15. I was filling liberties to force White to defend, but White passed 3 times instead and the anti-stalling feature triggered. Without the feature, maybe there was 20% chance of a reversal but now it’s 0%. In some sense, OGS gets it backwards in this game: the “staller” is the player who repeatedly refuses to defend, not the one filling dame!
In my view, filling dames should not be considered stalling. As richyfourtytwo said, this can be detected by calling the AI with the Chinese rules. The load on the AI would be greater as it would need to be called repeatedly in a “dispute”.
Yes, I also don’t understand the part about reporting in the message from @Animiral. What is there to report? ‘My opponent filled dame, in the end I missed to make a required defensive move and my opponent used that to capture some of my stones and win the game. Please annul the game and ban that player’?
I fully agree with the earlier posts, i think its very weird to grant a victory for the person who trollishly passes multiple times before the game is actually ready to be scored. If passing was a mistake, then it should not count as a pass.
Theres usually a situation where passing is the only reasonable thing to do at the end of the game, because all other moves would negatively affect players score tally by reducing ones territory or by giving captures to opponent. In these cases, if somebody refuses to pass (or resign) and just keeps playing -1 point moves, then it makes sense to launch that anti-stalling system and have the game declared as someones win.
But there are also situations where the game is not finished and there are plenty of moves that should be played before the game can be properly scored, but one player prematurely passes. Those passes should be treated as blunders, and the other player should keep playing normal moves until the territories are sealed and the game could be scored.
We should only have that anti-stalling system kicking in when the game is actually over and one player is stalling, but not with the unfinished games which still have moves that need to be played.
Good example. FYI you can link to a particular move, just add the move number after game url:
https://online-go.com/game/57364935/289
Hmm it’s not stalling to fill dame, but with a differential of 20 pts, 1 pt moves can feel kind of pointless. If one came across a similar situation without this feature, they’d probably just play overly safe moves (and still not “learn the lesson”)
frolag’s example is excellent, because whilst white would still win if white did defend once all the dame are filled which is what KataGo is assuming, in practice humans are proud and don’t like to defend inside if they don’t need to, and can make mistakes in working out if a defensive move inside is needed, so it is quite plausible for a human to not defend inside, lose those stones, and lose the game. So OGS is preventing the proper ending of this game. Some “polite” people will play the defensive move before all dame are filled to save time and out of mutual respect of each other’s ability to read, others wait until forced to do so and might well choose to pass rather than play other dame in the interim.
Also note that it can be pretty hard sometimes to find out if a protective move is needed, certainly for weaker players (whereever that range begins ).