Semeai with ko

[sanxiyn]

White A1 gives a good ko, but GNU Go plays C1 which gives a bad ko.

[gunnar]

This is also a nice testcase. It is known that GNU Go doesn't understand the idea of not taking the ko until late in the semeai and this might shed some light on why. It can probably be fixed in the semeai reading, but the position raises some interesting questions about the tactical reading as well. At the start we have

and attack A4 (left black string) returns good ko. Now it's not really expected that the tactical reading would understand a 3 vs 3 liberty semeai, although it may need to in the future. However, after playing some moves we reach

Now after black takes the ko white doesn't have any local move except retaking the ko. However, white can tenuki instead of making a ko threat, but I don't think the tactical reading has any provisions for letting the attacker (when reading on the A4 string) make a tenuki.

[gunnar]

Testcases.