Re: [gnugo-devel] regress endgame:603

[Evan Berggren Daniel]

On Fri, 11 Apr 2003 address@hidden wrote:

>
> > I'm agreed here.  My reference for miai counting:
> > http://senseis.xmp.net/?MiaiCounting
> >
> > The idea is that since there are two net plays between B:G6 and W:G6, each
> > of those plays is worth half of the 5 point difference, or 2.5 points.
> >
> > Because the D1 hane is sente for w, there are 0 net stones if w makes the
> > hane, and 1 net stone played if b plays the revers sente.  So there is one
> > stone difference, and the reverse sente is valued at three points.
> >
> > The problem here is that if b takes C1, then w gets tedomari (the last
> > play) with G6.  If b G6, then b gets the last play when w plays the D1
> > sente hane.
> >
> > For a good explanation:
> > http://senseis.xmp.net/?tedomari
> >
> > It looks to me like black should make a sacrifice of half a point in order
> > to get tedomari in this position.
> >
> > Adding a 3 point gote to the board would definitely be a good solution.
> > Perhaps also leaving this problem but marking it as a difficult problem.
>
> If I understand you correctly you agree now that G6 is the best
> move for both players.

I think that G6 is best for b, and D1 is best for w.  If w D1, then w gets
that sente sequence and then the G6 play; if w G6, then b gets the reverse
sente play at C1.  D1 is therefore the play for w, and G6 the play for b.
The fact that the players make different plays suggests that tedomari is
involved.

>
> I think we could have two problems, the existing one (with
> changed answer) and another one in which there is a 3 point
> gote added somewhere on the board (with C1 the correct answer).
>
> GNU Go would give the same answer in both tests. One would be
> right, the other wrong. But a real combinatatorial endgame engine
> would get them both right.

I agree.

Evan Daniel