https://www.reddit.com/r/baduk/comments/1ez20qz/why_is_this_group_not_dead/
I just played a 9x9 match on OGS as black. At the end of the match, OGS marked the stone on H1 as alive. When I passed, I thought it was dead, so I manually marked it as such. How would black be able to kill white?
Your quote suggests Black figured this out - maybe even immediately on seeing the markings -
and even if Black didn’t, someone on that reddit page probably pointed this out,
but just in case, I’m mentioning that indeed, that corner was not settled:
After blackJ2, White capturing the 2 black stones would require White giving up E2+E3.
black can play F3, white is forced to may respond with F2. Then black can play J2. H1 would become not dead as part of seki. But currently that is not created yet. Auto-scoring should not mark H1 as alive.
White would not be forced to respond with F2: White could instead respond with H2.
and H1 would be dead.
If black begins with J2, white H2, H1 is dead
its not possible to save H1 if white wishes to capture it, but OGS marked is as alive.
… that’s a good point.
If white plays first, there is no problem if H1 dies.
If Black plays first, White’s best response should be to make it seki.
Because if it becomes seki, white gets 0 points and black gets -1 point.
And if white insists on killing H1, then black will get 4 extra points, white only gets 3 points in this corner, losing 2 extra point.
In other words, killing H1 or not killing H1 depends on how white plays.
Therefore, you can’t say H1 is dead because it hasn’t been played yet, it’s undecided.
In undecided cases, the system should not make any marks.
If no mark is given, which is the default situation, it should be alive.
It dies only if it is marked dead.
after we passed, territory around D7 stone was not marked as white territory. I had to click here manually to fix it.
https://online-go.com/game/68498388
interesting that if I click Score Estimate before passes, that stone has small white square
I just forked the game with a bot
it finished incorrectly
That’s an example of the Autoscore Bug, which has been around, off and on, since August 2020.
This looks different from the bug where autoscore just doesn’t run. And the fact that it’s reproducible makes it sound much more like a basic flaw in the new algorithm.
That small white square is a good clue too. Maybe Katago has a weird blind spot here?
The score estimator has always scored Autoscore Bug positions correctly (i.e., the SE is superior to the autoscore in such cases), I don’t believe the autoscore not running is the explanation for the bug. The bug often strikes some but not all of the dead stones.
However, to take up the idea of a basic flaw in the new algorithm, it is interesting that the space in this case is a Bulky Five. So, we might hypothesize that the AI is seeing only the Bulky Five shape and is ignoring the other connected stones. And since the Black stone is on the vital point of a pseudo-Bulky Five, the confused AI is judging it as alive. Just an idea to play with.
another bug
in original game I had to mark this group as dead
(then, opponent resumed game and captured it)
