Re: [gnugo-devel] owl testcase

[Gunnar Farneback]

SP Lee wrote:
> I have tried to simplify the problem as attached, on 9x9 board.
> Surprisingly, GG can solve the problem correctly. The situation is:
> 
>    A B C D E F G H J
>  9 . . . . X . . . . 9
>  8 X X . X X O . . . 8
>  7 . . X O O O + . . 7
>  6 O O X . . . . . . 6
>  5 . . O O + . . . . 5
>  4 . . . . . . . . . 4
>  3 . . + . . . + . . 3
>  2 . . . . . . . . . 2
>  1 . . . . . . . . . 1
>    A B C D E F G H J
> While analyzing the problem "owl_attack A8", GG tried C9 and find a
> following answer directly at B9. While in the real game, GG answers R1
> with R2 during owl analysis due to pattern D1002. I have also tried to
> change the positions a little, but still can't repeat the flaw as in the
> real game. I hope someone with more knowledge of the static eye analysis
> and patterns can find the reason of this. Thus, why is the pattern D1002
> not matched on simplified board? And, why is the vital move not found in
> the real game?

The main difference is that you have an extra liberty at D6. The
cutting point at B5 might also matter, but that seems less likely.
With the extra liberty black has five liberties instead of four after
white C9 - black B7 and is thus considered tactically safe. This
causes C8 to be considered a halfeye instead of a false eye and the
right (or at least the most relevant) eyeshape is generated.

/Gunnar