Does AI review also look at non-played moves?

I already got the impression that the AI review that I get after playing a game misses my worst moves. In one of my recent matches, the top 3 game changing moves were 13.6, 12.2 and 10.4 percent. All these moves were relatively early in the game (move 25, 14 and 15 resp.).
At move #189 my opponent (fortunately for me) failed to kill a crucial group of mine. This move would have changed my winrate from an estimated 95% to around 5%. In other words, it would have been game over for me. How come this move did not show up in the list of top 3 moves?
Does the review only consider the moves that were made? How come it missed this rather simple killing move. The move was simply to atari a group of 6. I then have to connect to another group. The resulting group has only 2 liberties and is already surrounded by my opponent. Two ataris later the group is dead. Very easy to see, once you start with the initial atari.
This is the URL of the game:

3 Likes

I don’t know how the AI missed that, especially since you’ve got a higher tier AI reviewer.

I ran a full review—moves 188 and 189 were indeed the biggest game-changing moves.

Since you’re asking how the AI missed them, you should know how the AI chooses the top 3 moves. It does NOT secretly run a full review and pull the top 3 moves from that. Rather, it scans the whole game at few playouts, chooses the top 3 moves from that, and then reevaluates just those 3 moves at full playouts. This lets you see the result almost instantly, but it also means that the AI will miss things it would have found had it run a full review.

5 Likes