Is the scoring clock reliable?

Feedback Bug report
1

No, not unless it has been fixed recently. Some imprecise information relating to this question appeared this morning in a thread that is now closed (How to undo the results of game if AI mistaken? - #9 by Kosh), so I have started this thread to try to clarify the issue. The following was said:

“After either party makes adjustments there is a pause that prevents users from immediately accepting so in theory you can’t be tricked into accepting the wrong result with a last minute change by your opponent.”

This quote describes how accidental acceptance of a wrong score is supposed to be prevented, but acceptance isn’t usually the real issue—the scoring clock is. When I rewrote the scoring section of the OGS Guide (subsequently altered by wiki gremlins), during its large-scale revision a couple years ago, I was told that the scoring clock was 5 minutes and that it reset after every alteration to the scoring of the board. However, I found that this is often untrue. On many, many occasions I saw the clock time out long before 5 minutes, sometimes only a few seconds after a score-cheating alteration. These incidents often occurred when both sides were rapidly altering the board (because victims who don’t know how to deal with score cheating sometimes react by “cheating back”). I saw these incidents because I would arrive on the scene as a moderator, take half a minute or a minute to figure out who the winner was (not easy when the board is flashing a new state every second, like a neon sign), and then have my decision preempted by the early time out of the clock. (This didn’t always happen, of course, I also had a lot of success at stopping in-progress score cheating.)

We shouldn’t be surprised at this malfunction, because the scoring clock has also been buggy in the opposite direction, by timing out but not ending the game as it is supposed to. The mod team used to get loads of “stuck clock” reports, where the game is left in limbo and has to be manually decided by a mod. I don’t know whether or not this has been fixed.

In my opinion, wrongly scored games, where the wrong person won, should be annulled. Most are the result of score cheating, and not to annul them just abets cheating. (So far as I know, governments don’t give a free pass to swindlers by saying victims should be more careful.) Moreover, most beginners, many DDKs, and even some SDKs, don’t know what to do to counteract score cheating. Technically, one could argue that it is their fault for not reading the documentation, but I am not a fan of blaming the victim, or rewarding cheats.

8 Likes
2

I thought someone mentioned the possibility of having the correct winner assigned after the end of the game. This would be a better solution!

@shinuito said in the other thread:

As a quick question to @Kosh , I have a vague memory before the recent rating update, because of the sliding window, the best one could do with an incorrect result was annul the game as @groin mentions. Since we were supposed to have gotten rid of that, I thought I remembered somebody suggesting one could change the result to wins now, although I can’t seem to find that (if it exists). That said I do believe I’ve seen some games marked as “Black wins by moderator decision” etc.

Is this a thing?

1 Like
3

But if you see Kosh’s reply I was mistaken

Things just have to be annulled. Although it would be great to be able to fix the rating and award the win in the case of score cheating, I guess it’s not possible for some reason (maybe if more games are played in between the incorrect result and when a mod sees it? - not counting if the mod tool just isn’t available).

3 Likes
4

Ah but games labelled “blah blah by moderator decision” has always been there as Kosh said but I think that is different to (now?) being able to retrospectively change game outcomes. You and I cllearly both recall this being discussed. If I get time I’ll do some digging!

1 Like
5

We’re able to decide ongoing games in favor of one player which will show as “X by Moderator Decision”, but we still are unable to change the outcome of games that have ended.

1 Like
6

I already know that. However, people keep talking about the responsibility of the victim.

7

So how should we deal with it?

Now that we have a superhuman score estimator, it would be nice if auto-score would be the default when the players disagree, instead of deferring to the cheater.

1 Like
8

Click autoscore, which resets the original score. Then call a moderator.

9

And then just let the clock tick down so he gets the win if nobody shows up?

10

If your opponent keeps cheating by altering the score, then you keep hitting autoscore. Score cheating in progress is a high priority, so hopefully a mod will come if anyone is available. If not, it is also possible that your opponent will give up and escape. However, if that happens, don’t leave the tab. Wait for the clock to run out, confirming you as the winner. Otherwise, your opponent might return after you are gone.

2 Likes
11

Thank you @Conrad_Melville for raising this important issue :slightly_smiling_face:

The root cause of this issue is score-cheating. There is a simple solution that will prevent score-cheating completely…

When both players pass and we enter the scoring phase, then the scoring should be conducted in a turn-based fashion (instead of the free-for-all that we currently have).

This would work as follows:

  • One player has some time, say 2 minutes, to mark what they think are the dead stones and click “submit”.

  • This proposal is presented to the other player. They then have some time (probably the same, say 2 mins) to decide if they agree. If they agree, click “accept” and the game is scored. If not, they can submit a counter-proposal (hopefully with a chat message to explain, though probably best if this is not required).

  • Original player then has the option to accept the counter offer (e.g. if they realise they were indeed mistaken first time round) or request to resume play.

  • If they players cannot reach agreement by this point, then they settle the dispute by resuming play.

  • Traditional score cheating of changing the score at the last moment or forcing a scoring phase time out becomes a thing of the past.

  • All that remains is stuff like never accepting, always resuming then passing straight away so nothing changes and getting stuck in a loop that way. There might be a clever way round this or this might still have to revert to “call moderator” as with other trollish behaviour.

Thoughts?

3 Likes
12

I like your idea of a turn-based stone removal phase a lot. It would eliminate the back-and-forth cheating battles, making it much easier to resolve the issue. I would refine the idea by not forcing them to return to play, however. If they discover they need to close boundaries, for example, they can already reject the score and return to play. However, commonly disputed situations such as seki or nakade can’t be resolved by returning to play. I would suggest that when the players can’t agree on the score, the system would freeze the game and send a report to the mod team. In the short term, this would create more work for the mods, but in the long term, I think it would decrease the work load, due to the elimination of score cheating reports. Moreover, such frozen games would be easy to resolve, probably within a minute, because the mods would not need to send any feedback PM, since the system was the reporter rather than a player. Best of all, the virtual elimination of score cheating would be a great boon to OGS.

4 Likes
13

I would add one chance more to get the right scoring before the freeze. Everyone can make a mistake sometimes.

2 Likes
14

Why not just auto-score it if the players disagree? That’s zero workload, much faster, and probably even more accurate than mods.

2 Likes
15

The problem with autoscore, is that it is wrong too often. When done by a strong AI, it sees aji that both players missed and assumes continued play to exploit that aji. And the score estimator is often too weak to correctly determine obvious life and death.

3 Likes
16

But that doesn’t mean that it is impossible. FlyOrDie does autoscore very well (it’s their only scoring option and their is no appeal possible), but I assume they devote much less programmer resources to go than OGS does.

17

I am only proposing to defer to auto-score when there is an unresolved disagreement. Missed aji isn’t a reason to disagree; one player can just resume the game and do the invasion or secure his border.

But once the game is actually finished, with all the borders closed and secured, auto-score is almost perfect, isn’t it? Probably the main thing it would get wrong is territory in seki, and that’s only going to affect the result once in thousands of games.

And even in the extremely rare case where there is an honest disagreement AND auto-score doesn’t give the right answer, players could still report the game.

1 Like
18

But there has previously been plenty of discussion about how if the auto-score reveals information about the game state that the players didn’t have (i.e. because neither is strong enough to know) then it’s not right for the game to resume once the players have this new information.
Maybe this thread is a example, I think there were others: A compendium of OGS's terrible scoring system confusing beginners

3 Likes
19

Right, but is that a problem for deferring to the score estimator in case of disagreement? Most of the problems in those other threads seem to come after the players have agreed on the wrong score.

And if it gets better someday, the improved estimator will probably still be a great thing to default to.