I agree. Another benefit is that I expect this method to be the most tractable by humans, avoiding fuzzyness that might be difficult to explain in detail when asked about.
And it’s probably easy to implement as the beta version already seems to use floodfilling to mark viable areas with unclosed borders.
Interesting, so players won’t even be able to score a game with open borders, but will be required not just suggested to close them. That’s stricter than I would have gone for (as you can follow the scoring rules on that open-border position and get a load of dame and a “silly” score), but it’s definitely better than an AI closing the borders for you, which is akin to dropping out half-way through a marathon and then getting a finisher medal with double your half-way time as your overall time.
I don’t really see the issue with this though.
Unlike life and death situations, open borders only occur with real beginners, and the priority is mostly to teach them what’s happening rather than to be impartial and not give away the solution. Highlighting areas where there is no issue is not the most helpful.
I still prefer either the initial version, or this one:
In any case I’m glad we quickly found another way to have strong disagreements.
I lean towards either flood fill or the original, but a fuzzier original is close behind, though it loses points for ambiguity regarding how deep the fuzz should be
I forked the game from here
and I think it’s better but still not quite right by default in Japanese rules.
The auto score does
but I would expect
which aligns better with the no points in seki.
It’s just normally you do have to play to remove these kinds of dead stones in a seki to gain points.
However I think, since it doesn’t count the underlying territory, maybe in this specific case it’s ok, because taking the one stone off the board by playing and taking it off in this way without giving anymore territory would score the same I think.
Possibly it would get a funny in between result when there’s two dead stones, which normally you can capture for three points - maybe I’ll try cook up a game board. ( Also just generally look at other game boards too 
)
I think the Chinese rules game looks good though
And the AGA area scoring one
For example
Normally you wouldn’t count the two white stones as prisoners because there’s no points in seki.
It should be
If Black wants to prove the stones are dead they’ll actually gain points by doing so anyway (I split it into two games)
So anyway that’s kind of what I mean as an in between result (kind of an edge case). 3 points for black if there’s no marked dead stones, 5 points for just marking the two stones as dead, 6 points if black actually captures the two stones and then the extra stone needed to keep it a seki.
Also I left in the 8 point shape for fun, since there’s likely a seki in there, but the players might just want to score it as it. Seems ok.
You’re right. Black should lose a point if they don’t capture that white stone at G7 before scoring the game. But I don’t really mind if in a situation like this, OGS version of Japanese rules is slightly friendlier than the official rules, as long as black doesn’t get territory points for G7 and G6.
Yeah, I don’t really have a preference on this point, it seems easy enough to capture the stones if you want the points, but I also don’t have a problem with it counting them as captured
I think we should also consider positions like this (all stones alive):
I expect the “initial” (current beta) version to show red marks only at the D column, intersections that may seem to have little to do with the gap at G5, so it won’t be very helpful, and potentially confusing.
I don’t think a “helpful friend Clippy actively explaining the marked squares indicate a border that needs sealing.” would make it any clearer.
The “slightly fuzzy” version might mark the D, E and F columns and G5 red. Not sure if it would be very helpful and clear what the problem is, but at least G5 is included.
The “realistic” version would mark only G5 red, so it might be the most helpful how to fix the issue, but perhaps it would give too much assistence, and players may still miss that single red mark if they don’t pay attention.
The “fully covered” version would mark all empty intersections from the D column to the J column red. It should be very clear that something is wrong, but the player still needs to figure out by themselves how to fix it.
I don’t think the clear button clears X’s though.
Also at the moment this button doesn’t seem to work for me in analyse mode in a particular old test game.
If I take the above game, and just play out some moves in analysis until I want to score it say:
^ doesn’t seem to show a score or automark anything.
^ if I manually marked some captured stones, still no.
^ If I manually paint the areas.
(And similarly if I try again after two passes (just in case that was a requirement)
It does work here though 囲碁囲碁 vs. shinuito if I go back before the two passes, it does the same thing as the game of just counting the surrounded areas.
Actually it generally works much better in the analysis mode of that game. Unless that’s something else that’s counting those areas (some other mechanism with no dead stones).
One thing that’d be useful, might be that when you click the calculator/score button, that it gives you that total somewhere?
I think for instance, clicking the scorecard does the old counting of just solidly surrounded area
so the 6 territory is the ones that I’ve drawn around, but even marking the single stones dead, when it paints that as territory it doesn’t seem to be reflected anywhere in the counting that I can see?
Here you go
I don’t know if I would think of it as too much assistance necessarily. In theory it should just be there to help players score a game, and you would think that beyond a certain level it’s not going to make a difference. Though maybe in special circumstances or rules it might be the difference of 1 point or so, which may in even more special circumstances affect the result.
Just for sake of testing as well, Poll: how would you score if you were the auto-score algorithm? @Vsotvep had a collection of games and votes on how to score certain positions, which seems like a fun one to test
I didn’t finish the games by resuming or doing anything - just demoing the current autoscore and then the games end as is.
Maybe instead of a pink square, we could use some other sort of symbol that makes the ambiguity unclear, like a carefully chosen emoji
I vote for 
(but of course it should be a user setting 
)
Game 5 is actually a strange one. If you repeatedly click the autoscore button it might cycle between different kinds of markings.
I tried forking again and then clicking the autoscore and it would cycle between things like
So maybe something that’s more deterministic and doesn’t depend on the number of times the autoscore button is clicked might be useful?
(Not sure how the first version came up - maybe it’s depends on who passes last? But I’m not sure exactly. More testing needed)
I suppose the marking will get even weirder (and the accompanying text more confusing) if black also has a gap, like at C8.
If I understood correctly, @Feijoa’s autoscore algorithm compares KataGo evaluations where either black passes once or white passes once and adjudicates groups alive if they don’t die in at least one of those two variations.
That algorithm should be good enough for positions requiring only one additional play by either player to settle the status of all groups on the board to finish the game “properly”, but it can remain ambiguous when multiple unsettled groups are present on the board, or situations where group status depends on some multi-move trade or some sort of (multistep) ko.
I suppose that in such cases the algorithm would need to do some complicated tree search, potentially comparing several pass results deeper in the game tree, and/or several localised tree searches to deterministically adjudicate which groups should/might be considered definitely dead in the position that is being scored.
Your example has 3 unsettled situations (1 of which could turn into a ko), so I can understand how even after inserting one pass for either player and comparing KataGo’s evaluation afterwards, the algorithm’s overall group status adjudication still depends on whose turn it is.
One might attempt to make a more advanced algorithm for very unsettled positions like this, but I suppose it can still become untractable when a large number of unsettled situations are present on the board. I suppose that in such cases, the players should be involved to indicate which groups they consider dead and/or we should just be OK with it that autoscore is not deterministic for some very unsettled positions.
That being said, I think that the question of autoscore (automatic group status adjudication) in positions with unsettled groups is distinct from the current topic about the handling of open borders and scoring seki and (some) teire correctly, both of which should still be done even when the players mark dead groups manually and agree on those.
I think the OGS behavior for “dead” stones in a seki is reasonable and I addressed that situation here: Auto-score improvements - #31 by hexahedron
M19 area, T19 - these are also eyes in seki and are not points in territory. As a deliberate choice, M19 and N19 are still counted as points for captures because they are marked as dead even though the territory underneath the stones is not (and should never be) counted. This is a slight departure from the Japanese rules that mandate that you physically capture such stones during play before ending the game. Given the realities of server play and interfaces it’s a deliberate choice from the scorer:
Behaving more strictly could require careful corresponding UI work, since if stones that users explicitly mark to be dead are sometimes not counted as points, it may be confusing for users, and it may not be easy to see or know when they are and are not.
(this was written from the perspective of the territory scorer code that is being handed a list of stones being marked dead by users. If the marks are AI-supplied, then it’s not an explicit choice of the users so this is less of an argument. But the users could still click to mark them and then we’d be back in this situation. Plus also it would be weird and result in weird complications if the scorer code was in the business of disputing the life and death markings it’s being given or choosing to confusingly disregard them)
This choice is also more “AI-resistant” in that AI ownership markings will likely always mark such stones as dead (because future play will trivially be able to kill them), and so we should just count them as points rather than confusing users with a magically tricky exception that in this one special case, marked-dead stones somehow don’t count as points.
Obviously, in this particular case black should capture them and force white to throw in once more, gaining an additional point, but it’s not the scorer’s job to determine that, so we deliberately score only 2 points here for the marked stones, not 3.
I think I agree, that dead stones the user marks should count as points. My point was that whether the autoscore should be suggesting them to be marked as dead or not is a bit debatable. In the simplest case it doesn’t matter, and in slightly more complicated situations it might be a bit weirder as I mentioned above.
Do you not see this is as very big departure from Japanese rules? The confusion of automarking them as dead, which is kind of half correct in a way: They’re dead because they can be captured, but they’re not points because it’s seki.
At the end of the day you want people to play with the rules they’ll be faced with in an in-person game also, not a strange hybrid of “well this was convenient for the server” and “close enough to japanese rules” ?
It’s kind of like the Torazu sanmoku at Sensei's Library and other strange examples, where you make a ruling that such a shape is worth X amount of points. It would be strange to suggest something like this on users.
I suppose speaking of… what is going on here with the autoscore markings?
Basically I think something should be looked at in these cases for the autoscore, before accepting some strange rulings as “a convenient departure from Japanese rules”?
We kind of have that though don’t we now?
