A good autoscore would not do this, and I don’t think you’ll easily find such a case in v3.
It occurs to me that if we have instant forced autoscore that follows the rules of Go, the losing player has a cool new strategy of passing early to try to trick their opponent into passing and giving up territory due to unclosed borders.
We already have a cool new strategy of passing 3 times in the middle of the game to get the server to declare the game your win with no cooperation needed from your opponent.
I am not up-to-date with a lot of things — is this still the case currently?
Yes, depending on your definition of “middle”. I don’t think that feature has changed since the first week or so after release.
Apparently @Uberdude likes it so much he’s advocating for more forced early win strategies? 
I’m not sure that sarcasm is of much help here 
I’m not sure why you chose to point this out to @Feijoa when he was responding in kind to the ongoing sarcasm campaign by @Uberdude
I tuned out of this conversation some time back when it veered into this sort of territory, and ironically only popped back in just now because I am keen to help with the actual problem if constructive suggestions are available and getting some concensus.
Are they?
Not at all, but I think it was interesting and surprising to hear @gennan’s perspective on it. It hadn’t even occurred to me that people would be in favor of instantly ending a game after two passes, possibly with a huge open border, without giving an option for the players to resume.
There was a lot of discussion but here’s my take on a few key points:
Fixing mistakes after passes: I agree players should be able to fix mistakes after passes. Being able to resume a game is part of Go, this should not really be controversial. Anti-stalling mechanism is an exception, based on one of the player being perceived as acting in bad faith, but that does not change the general rule.
Score estimator and scoring algorythm: I agree with @Uberdude that those are two different things. Players should be able to resume the game and fix their mistakes after passing, but if they do not, then the scoring algorythm should not do it for them: they’ll get a huge dame and so be it.
Help provided by the scoring algorythm: This one is more difficult, as it is a balance between the convenience of automated scoring vs. the risk of providing new information to the players.
On opened borders, I think the point is moot: this is never an issue except for pure beginners, and I’m completely fine with the scoring tool revealing that they forgot to close a border so that they may fix it - they will quickly learn.
On life and death, I’m less comfortable on this one and understand the purist’s perspective that the scoring tool should say nothing and let the players mark each group as dead or alive. That said, (i) this is not the way OGS operates, and (ii) there is no clear majority at all in favor of such change. Pending the day where such material change is agreed, I do believe @Feijoa’s algorythm is at least an improvement on the current tool as it provides less info and will cause less confusion among beginners.
Thanks!
I mean, that is kind of the reason why we are having this thread and this discussion.
How did you determine that? Reading through this thread it seems that most players are not satisfied with the current feature and have agreed that pure manual scoring or pure auto scoring should be better. Currently users get AI input through the scoring algorithm during the game which also seems to violate OGS’ own ToS
You can NEVER use Go programs (Leela, Zen, etc.) or neural networks to analyze current ongoing games unless specifically permitted (e.g., a computer tournament).
Then how about this?:
After 2 consecutive passes, the players should mark dead stones. The players should not need to or be able to mark empty intersections. The players can only toggle the status of strings of stones.
During this marking phase, the scoring system does not provide any pre-markings of dead stones, but it does display a 100% accurate score (including a clear marking of all territory intersections on the board), given the current markings of dead stones by the players. So there should be no surprises on the scoring result when both players agree about all dead stones during the marking phase.
If there are any open borders, the score presentation during the marking phase reflects that, so attentive players can resume to fix those. One could say that the system is assisting the players by showing the score during the marking phase, but I think this is quite low level assistence, because counting during the marking phase is simply using floodfilling, which does not require any tactical go skills. It does require the system to understand seki to score the game correctly when the game is played under Japanese rules. This is complicated, but OGS can use the KGS algorithm, which is very robust and its source code is available.
The system may provide some additional time-saving assistence by marking all other stones dead that can “see” friendly dead stones and marking stones alive that can “see” dead opponent stones. Such features also don’t require tactical go skill. It’s just basic logic that can be programmed into a “dumb” algorithm, perhaps some sort of Wave Function Collapse algorithm.
When a player clicks the “agree” button to finalise the marking phase, the system should be completely clear on what they agree to. To avoid mishaps in the finalisation of the marking phase, there should be a short agreement cooldown period after being shown the last marking change (during which time any previous agreement is cancelled).
If the players cannot reach an agreement about dead stones, but they also both don’t play any moves after resumption (so they enter a resumption-pass cycle), the game ends then and there and the game is scored by an autoscore algorithm that uses an AI, possibly @Feijoa’s algorithm. I think this feature would make the scoring system better than KGS’s.
I think such a scoring system would help beginners who don’t understand life & death, and it would also provide a fair result for victims of score cheating. Still, it would avoid giving tactical hints to the players, violating OGS’ ToS.
Game resumption is uncontroversial in the absence of AI assistance. However, in a discussion where we consider using the AI to score, it makes sense to reevaluate.
That’s my understanding from this thread and the other threads where this was discussed (at length) in the past. Granted, it’s just a feeling though, so at some point there should be a vote to clarify (or a decision by authority of Anoek).
I would accept a manual marking of stones if that’s the community’s decision. Something like @gennan’s proposal would work.
If we are split over such change, then I believe retaining the current system but using @Feijoa’s algorythm would still be an improvement.
In any case I agree that the current system is not satisfactory as is.
Is it easy technically to put the result of autoscore v3 as the initial view in scoring phase? With still a clear notification that players should mark dead stones as they see it, and in case of dispute either resume play or call the mods or both.
So perhaps it’s time to have a poll?
Interface of choosing status of groups should become PERFECT. It should be clear that users need to click on groups in score mode, and it should be easy. Then marking all stones as alive by default may be good idea.
But before that is done, currently, there are more problems because users did something wrong, not because AI choosing too perfect.
With the poll option of pre-marking dead stones, I intended voters to assume that the pre-marking is perfected to be correct in more than 99.99% of the cases. So the players would basically only need to chose to accept the autoscore or resume playing. The UI that allows the players to change the markings might then even be phased out, leaving only the options “accept” or “resume”. That would make score cheating impossible.
Measures to prevent score cheating may also be included in the other poll options, so I didn’t include that concern in the poll.
I intended it to be a poll only about the information given to the players during the marking phase, which they would then be allowed to use in a resumption.
I think the second and third options should be choices in the game settings, and I suspect we’d soon need manual restricted somehow to limit score cheating.
I think the concern of score cheating is mostly independent from this poll.
For example, if a near-perfect autoscore algorithm is available, OGS could decide to use it bindingly when a game is resumed and then both players pass without playing any move.
