Stone removal and scoring updates

[gennan]

I think that black group is also 100% dead under area scoring rules with regular resumptions in case of a disagreement.

[hoctaph]

That’s as a result of Black having no ko threats for it.

(At least for groups, rather than for the whole board, I think ​ settled vs unsettled
is a local thing, and thus is not affected by Black’s lack of threats.)

[gennan]

I’m not sure I understand what you mean.

Would you say that the game isn’t actually finished in that position, because black’s group is unsettled?

Would you dispute it if you were black and your opponent marks the group dead? Or would you not mark the group dead if you were white?

[anoek]

I’ve pushed out an update that helps with the prior cases

@gennan

Though perhaps autoscore is aware that black is unconditionally dead there, because when stopping before black B3, it apparently doesn’t see a way for black to survive:

For this one though, as far as I can tell you’re right Black is unconditionally dead. KataGo estimates that there’s only a 50% chance it’s dead if black plays first, even with a fair number of estimates, which is why it’s not getting cleared out. I can’t explain why that is, seems pretty dead to me, but I’ll chalk that up as something that will probably work itself out as the networks evolve. (Heck, it might already behave better in the current network, I’m a bit behind in updating which network we’re on.)

[gennan]

I’m going to sleep now. Tomorrow I may try it out again to see to what changed by the update.

[yebellz]

I think @hoctaph is just saying that the particular shape (in general) is not necessarily dead, since Black could play at A2 or B1 to start a ko fight (but White gets to take the ko first) and Black would need a very large ko threat advantage in order to win that fight.

However, in this particular whole game position, Black is dead, since Black does not have any effective ko threats.

[gennan]

In other positions black may have a chance to live if he plays it out before the game stops. Still, when the game is stopped without black attempting to live by ko, I think black’s group should be considered dead as is, similar to how the top left corner group in Upcoming stone removal and scoring updates - #90 by hexahedron is considered alive as is (with territory) in a stopped game.

[hoctaph]

“Would you say that the game isn’t actually finished
in that position, because black’s group is unsettled?”
​
​
Yes, I would say that. ​ Under area scoring, I’d compare this to a minimal ko which White hadn’t filled.

Under territory-scoring rulesets where either minimal kos can’t give territory, or
one would need infinitely many ko threats to make them do so, I’d still compare it
to a minimal ko that White hadn’t filled, although in this case it would be Black’s
fault for not having thrown in. ​ (since white would need 2 moves to prevent the ko)

​

​

“Would you dispute it if you were black and your opponent marks the
group dead? Or would you not mark the group dead if you were white?”

​

Under area scoring, as White I would mark the group dead as Black I
would not dispute that. ​ Under territory scoring, these would depend on

Does the scoring system let me mark one of A3,C1 as neutral?
and
if no, then Does the 1-point difference change who wins?

.

[gennan]

@gennan

Still, when the game is stopped without black attempting to live by ko, I think black’s group should be considered dead as is, similar to how the top left corner group in Upcoming stone removal and scoring updates - #90 by hexahedron is considered alive as is (with territory) in a stopped game.

On 2nd thought, I think it’s more like the 3rd case in Upcoming stone removal and scoring updates - #90 by hexahedron. Black can’t live, but under Japanese rules optimal play for black is still to start the ko and keep resuming until white is forced to add another defensive at 4, whereby white loses a point:

Still, if black doesn’t start the ko before the game is scored, black is dead without white adding another move.

@hoctaph

Under territory-scoring rulesets where either minimal kos can’t give territory, or
one would need infinitely many ko threats to make them do so, I’d still compare it
to a minimal ko that White hadn’t filled, although in this case it would be Black’s
fault for not having thrown in.

Yes, I think you’re right. Under Japanese rules it’s in black’s interest to throw-in before stopping the game if there is no other means to remove a point from white by agreement.
With the upcoming seki and teire detector I don’t think it will subtract a point from white if black fails to throw-in before the game stops, so black should do it before scoring.
If black doesn’t throw-in, white should not add a defensive move at a 2-1 point before stopping the game, because then black can throw-in on the other side and white would eventually lose 2 points.

[Feijoa]

@gennan

optimal play for black

If I’m counting correctly, doesn’t Black lose the overall game no matter what happens with the lower-left? Could it be that KataGo doesn’t want to drag out the game with a pointless ko fight?

(If you could link to the board that would help.)

[gennan]

@anoek

I’ve pushed out an update that helps with the prior cases

My test results with that updated version:

^^ It now seems to think this situation is unsettled/seki.
In the previous version it considered black dead, which I think was the correct adjudication, because even after black’s best defence at B3, white still kills with A1.

^^ It now seems to think this situation is unsettled/seki. I think this is correct because white can kill with A1 or black can live in seki with either atari.
In the previous version it adjudicated black alive with territory, which was incorrect.

^^ It now seems to think this situation is unsettled/seki. I think this is incorrect. If the game is stopped at this point (without black throwing in at a 2-1 point), black should be considered dead as is.
Still, it is better than the previous version where it adjudicated black alive with territory.

^^ It now takes 2 white teire to convince autoscore that black is dead.
In the previous version it only took 1 white teire to convince autoscore that black is dead. So I think the previous version was slightly better in that regard.
But in actuality, white should not be required to play any teire, leaving it up to black to throw-in to gain a point.

So I’d say that overall the update is an improvement, but not quite fully correct yet. It is easier now to convince autoscore that white B2 is alive, but also a bit more difficult to convince autoscore when black is dead.

[gennan]

@Feijoa

(If you could link to the board that would help.)

I’m using this game to test: Vriendschappelijke Wedstrijd

But I’m updating that game regularly, so it’s ongoing. You probably need to fork it from move 7 to test these positions yourself.

[sidxartxa]

[Bunburyist]

Thanks so much for your continued efforts to make this platform better!

[Groin]

@stone.defender

“Let’s AIs finish our game” would come in my mind first

What about never scoring unfinished game?
If both opponents passed too early, then they would be encouraged to return or click “Let AIs finish the game”.
if both players click it instead of return, ghost black and white stones appear and game is always properly scored. No need to invent new scoring rules that are able to score unfinished game in not weird way.

What a highjack of my post from another thread! I was suggesting something very différent as scoring for lazy players to finish a game.

[Sadaharu]

Not sure if it’s been mentioned but why the red cross for dead stones? Red cross makes it seem like they are wrong moves but they are just innocent prisoners

[qnpnpmqppnp]

I assume it’s to make it more visible (and thus lowering the chance you may accept a result you disagree with).

[feritayl]

Thank you for all of your effort!

[döskee]

The red lines over dead stones hurt my eyes… I found it horrible !
Is there a way (option) to do not see them ?

[siimphh]

Bot games can look a bit funny, eg: here the UI says W+13.5, but also that black has 176 and white has 165.5 points. I guess that’s because we show the new scoring result but the game was actually scored with the score estimator or something?

jeffrey.z vs. amybot-beginner