Re: [gnugo-devel] yet more testcases

[bump]

Gunnar wrote:

> Evan wrote (on February 27):
> > Gnu Go plays to shorted its own liberties, substantially worsening the ko
> > situation.  As a result, a w group lives that shouldn't.
> > 
> > Also, Gnu Go shouldn't leave behind the adjacent ko threat in the first
> > place.
> > 
> > +loadsgf games/nngs/looknglass-ggl-200302271041.sgf 89
> > +420 gg_genmove black
> > +#? [L4]
> 
> It's not clear to me that M5 is worse than L4. Another alternative is R7.

R7 may be best but also L3 is locally stronger than L4. It's not
clear that move 89 is worth regressing over. At move 91 B
can play R7. (This allows W to live small on the edge but avoids
all complications and leaves B a won game.)

> > +loadsgf games/nngs/looknglass-ggl-200302271041.sgf 99
> > +430 gg_genmove black
> > +#? [R9]
> 
> This is a systematic problem that GNU Go often plays internal ko
> threats before taking the ko. This should be looked into.

See century:60 for an example of this.

Dan