terrible game, material for testcases

[gunnar]

 http://lists.gnu.org/archive/html/gnugo-devel/2005-04/msg00006.html

Respective developers,

A serious problem possibly exist in GNU go's utility function. Please view the
attached .sgf file for details. Near the end of the game, GNU acted almost 
blindly
to achieve NO benefits while there did exist simple strategies to keep GNU go 
from
losing up.

The attached file is my play with GNU go with 9 handicaps in a machine with
128M/Pentium 233. Finally, GNU go got NO living stones on the chessboard! I
suppose it be a serious problem to mention.

Best wishes,

Xuli Le
Dept. of Phys.
Univ. of Sci. & Tech. of China

This is indeed a very terrible game and well worth generating some testcases from.

[gunnar]

game record

[arend]

Tests

[arend]

The highlight of the game is a full-board semeai involving 8 groups around moves 360-364. GNU Go of course fails to understand the relations between the different races between different pairs of dragons.

It is worth pointing out that while this would be very difficult to fix in the existing semeai code, the position might be accessible to even a primitive full-board search code (that understands only direct captures).