Readers of this site are expected to have read my site on computer analysis of the game Kim vs Shikshin. before. Here four cases are shown, where analysis by Many Faces of Go fails three times and gives the right indications only in one of the four cases.
The four games were quick play by the U.S. youngsters Andy Liu
(alias BigBadWolf) and Curtis Tang (alias cjheaven)
on KGS on August 6 and 7, 2010.
For each of the games we give sgf (including kiebitz chat) and
a combined diagram for
% of success (= Win rate; MC-based) and
score estimates (Score; traditional approach).
Only for game 4, both curves (%-values and scores) reflect well
what really happened. For game 2, %-values are completely
wrong at the end: the indicate a win for White, although
Black has been winning.
Currently, my policy is to believe
in the computer analysis only when both curves look smooth
and point in the same direction
and when they do not contradict my feeling (as a high kyu player
with lots of experience in computer go).