Another persistent issue with the auto-score is that under Japanese (or Korean) rules an uncompleted ko will be miscored. eg: Q12 in:
The best approach here is either to resume and ask your opponent to complete the ko, “Please complete the ko.” or manually adjust the score to mark that one point neutral. This problem only occurs with territory scoring.
That used to happen even in the old system 
Yes I know. It happened to me, twice. Both times a half-point game where it mattered^.
^ Mattered = Anything to do with Go.
Yeah back in my day, we hated the idea of losing an extra point resolving a ko we knew we could win.
Iwamoto knew what was what, and Go Seigen was just causing trouble, but nothing a few drinks couldn’t sort out, as was the style at the time. That and to be fair to him, he did bring about a bit of change.
Of course I’m talking nonsense, and I meant the old auto scoring as opposed to the current katago one.
All this talk of auto-scoring seems a bit silly to me 
I do not think it wise or even desirable to have a super accurate AI-based auto-scorer. Ultimately, this is a game between two players and it is up to the players to agree upon the outcome of the game. I quote from the Japanese rules:
Therefore:
Players should pass when they think the game is over, not when AI thinks the game is over.
The result should be what the players agree it to be, not what AI tells them it should be.
If both players agree on the result, then that is the result regardless of any counting mistakes that are spotted later by them or anyone else.
Any auto-scoring is just a suggestion to take advantage of the fact that we are playing on a computer and thus save manual counting. It is still the responsibility of the players to check the score and confirm their agreement or otherwise.
In the case of BobStevenage vs. Klaiddom1 highlighted by @Kosh, they both passed so the game stopped. It is correct that the one point of black “territory” next to the ko stone is not territory. It can even be confirmed by playing it out if necessary. But white needs to say so before accepting the auto-score. But in reality, both players confirmed the result so that is the result; and thus, as far as I can see, it was wrong for the game to be annulled (unless there were other circumstances I am not aware of).
In the case of a beginner vs a beginner, they may well score the game “incorrectly” according to how more experienced players would see it, but if they agree then that is the result. Such things are unlikely to persist for very long as they play more opponents.
In the case of bots, there may well be an issue. I am not too familiar with bot games so I don’t know how the scoring works but I can see the potential for issues when there isn’t a second human to judge and agree to the result.
Although I agree with the basic spirit of what you say in regard to minor scoring issues, the trouble with the Japanese rules is that they have nothing to say about some of the problems unique to online play. Undeserved wins from score cheating, stalling, and sandbagging should be annulled. These are fraudulent games, and the creeps who perpetrate them frequently prey on beginners who don’t know any better and don’t know how to counteract the cheating. Fraudulent competitions are not considered legitimate in any sport I know of. People used to get killed over thrown prize fights (maybe they still do), and it is still dangerous to cheat in a backroom poker game. I don’t think a cheated result should be considered legitimate under any circumstances.
Agreed 100% 
Maybe these issues are covered by the first line of the Japanese rules:
From this, I would say that if someone takes advantage of another player’s lesser experience to win by score cheating, they are violating the mutual trust. Therefore, they are subject to the the standard disciplinary actions of the place where they are playing. In the case of OGS, that is game annulled and a warning / ban, etc. Plus it’s against OGS terms of service so seems totally fair that they get a ban or whatever.
As for BobStevenage vs. Klaiddom1, that does not look like score cheating to me. That looks like an honest mistake that black did not understand that aspect of the rules and genuinely thought that was one point of territory, while white did not check carefully enough and accepted the auto-score to only realise after that they should have won. In that case, white should have checked more carefully before accepting the score and the result as agreed should stand. JMHO. 
I don’t see anything fraudulent about “stalling”. Rude, bad sportsmanship/etiquette, annoying douchbaggery sure. But not cheating and not fraud as no deception (I was once falsely accused of fraud by a train ticket inspector when I had mistakenly, with no mens rea, traveled with an expired discount card, and went to court and they dropped they case). Also my threshold for stalling is higher than some others: e.g heartbreak game 2 bad moves was not stalling.
Pointless restarting, one form of stalling, certainly is cheating. I have seen a game where the cheater restarted over 100 times. And I have seen several others where it ran into dozens. There is also a very common pattern of behavior in which the cheater first tries to win by stalling, and if that doesn’t work, resorts to score cheating. I have seen this many times. The motivation in these cases can hardly be disputed.
The result is fraudulent because it looks like a win but isn’t (i.e., the result is deceptive). Do you really need to resort to linguistic hair-splitting like this?
I think it’s fine to ban or annul in the case of people being annoying douchebags, but I think it’s important to not exaggerate the crimes you acuse them of, or the moral standing and legitimacy of the moderators is reduced.
I didn’t accuse anyone of crimes, which is your exaggeration. Score cheating, stalling, and sandbagging are violations of site rules. The illegitimate wins that result from them should be annulled.
Do you think any of those apply to the BobStevenage game? Do you think it should have been annulled (for those or other reasons)?
That game does not appear to fall within any of those categories if this was a Katago mistake in scoring. You have now digressed into a new offshoot of what I was discussing. If I were moderating, I would not take a report on that game; I would leave it for a more knowledgeable mod like mark5000.
However, unlike many people today, I do not dishonestly dodge legitimate questions, so I will try to answer you. If it is true that the point should not have been scored (I haven’t looked up the rule), then I am inclined to say that the game should be annulled. The real fault lies with Katago, and I am sympathetic to the plight of the player who may have assumed that the machine knew what it was doing and therefore did not check as diligently as he might have without an autoscore. I, for one, would not want to win a miscored game. Allowing a wrongly scored game to stand serves neither the interest of go nor of the players, in my opinion.
Now I have a question for you. Would you annul a score cheated game where the cheated player accepted the score? This is not hypothetical. It actually happens a lot.
Please elaborate on the nature of the score cheating. E.g is it some stones marked with incorrect status, do we have reason to believe it was done in ignorant good faith or malicious bad faith? Were they repeatedly mismarked against others wishes? Did one player try to communicate with the other their disagreement? Was a moderator called at the time or later? Was there time pressure to confirm the score? Were players aware they could escape or pause rather than confirm?
P.S. does OGS suffer from a similar race condition flaw like KGS where if your opponent changes the status of some stones a brief moment before you click done you end up confirming that altered state rather then what was showing on your screen when you started the action is clicking done? I’ve never quite understood the mechanics of the confirm phase with timers here. It often seems laggy, but a delay could be a way of avoiding the KGS style problem. If such a flaw exists, it makes me more sympathetic to those who appeared to accept the wrong score, because they may not have thought they were accepting it but the prior one.
Whole board, stones and territory, wrongly claimed for the cheater. This is the most common type. In any case, I am specifying an actual incident of cheating (if it’s really necessary, let us say the cheater admits it by jeering at his victim in the chat–this does sometimes happen), since I am trying to find out whether a cheating case (not a doubtful case) would be accepted as legitimate.
The real fault lies with Katago, and I am sympathetic to the plight of the player who may have assumed that the machine knew what it was doing and therefore did not check as diligently as he might have without an autoscore.
Even if we wanted AI scoring of games, I’m guessing that this is a bug in OGS’s use of KataGo, not in KataGo itself. Indeed, the AI analysis correctly has white winning even after Black’s pass, so KataGo “knows” the right result. I’m guessing OGS’s logic is overly simplistic - it is told in the analysis that Black’s stone in the ko lives (because Black to move first can connect and keep that stone alive), and then presumably OGS hasn’t implemented the better scoring algorithm like KGS did where false-eye shapes surrounded by living stones are still reliably excluded from being points.
Of course.
Then yes, I would annul the game if I were an OGS mod today where annulling is the done thing for such cases. I can see this is inconsistent with the approach dragon-devourer advocated which I liked in A compendium of OGS's terrible scoring system confusing beginners - #66 by dragon-devourer of “once both players agree, score doesn’t change”. However, I would justify this as the player probably didn’t willingly agree, but didn’t want to waste more time with the griefer, so just accepted and got on with their life and maybe called a mod. Perhaps there should be 2 done buttons: “Confirm score, it’s correct” and “Begrudgingly accept to stop wasting time and call a mod” to distinguish these cases.
Some more confusion caused by conflating score estimator assuming continued play with score tool of terminal position. Is anyone else having scoring issues - #3 by Uberdude