Re: [gnugo-devel] bug report

[Paul Pogonyshev]

Dan Rosen wrote:
> I'm not sure whether I can call this a "bug," precisely, but it is
> certainly a defect.

Well, we prefer not to call such things "bugs", otherwise we wouldn't
be able to ever produce a stable release ;)

> I was playing a bot on KGS which claimed to be running 
> GNU Go v3.5.2, when in the endgame it made a fairly significant blunder.
> On move 234 in the attached SGF, it prefers to save 3 stones, preventing a
> 6 point loss, which results in a seki. If the 3 stones had been
> sacrificed, there would have been no seki: it would have captured 9 stones
> for a local gain of 18 points, or a net gain of 12 points including the
> sacrifice.

GNU Go is not very good with sekis.  Recently, some decent improvements
have been made, but it this position it doesn't even see A19/B17.  I'll
try to see if i can fix this particular issue.

Btw, GNU Go on KGS is not run by GNU Go team -- AFAIK it is run by
someone from the KGS team, perhaps William Shubert.  We are interested
in receiving some tests from the games it plays there.  However, no one
has to time to actually extract them from the games.  If you think you
may be willing to produce a small set of tests from time to time
(perhaps from your own games), i can explain how to do that.

regards,
  Paul Pogonyshev


P.S.  and what does "FSGCBot" mean?