Making a seki

[sanxiyn]

Simplified from a real game. White can turn the corner into a seki, but GNU Go finds out only after the exchange.

Try:

gnugo -l seki.sgf -L 1
gnugo -l seki.sgf -L 3

[gunnar]

This a known weakness in the semeai analysis, which can only work on pre-existing stones. Solving it would involve rewriting owl_analyze_semeai_after_move() so it can handle one of the dragons being the just played stone, writing an autohelper to verify whether playing a stone would result in seki, adding some patterns to indicate where it would be worthwhile trying to get a seki, and finding out how to compute the value of the move to make seki.

Some standard seki shapes, e.g. the one in seki:101,

are already detected but only by using very explicit patterns like S3 at source:patterns/patterns.db@2339#L13472.

[gunnar]

The semeai analysis has improved significantly with respect to my previous comment since then. The attached position would probably still fail but it should at least be added as a testcase.