Scoring bug

My last game is shown on my profile as winning by 3.5 points but the in-game score is 2.5 points. When I run it in Katrain the score says win by 2.6 points but the end result says 3.5 points. Is the official score 2.6 points but it is rounding up? What is the final official score?

Game: IGA Season1 R3

The official score, as is marked on the board is B + 3.5, but White could have claimed that it should be B + 2.5 with legitimate reason, since C1 is marked as a Black territory, while it actually is not territory (White should’ve played A1, after which Black connects C1 and the point of territory disappears).

This is reflected in the AI evaluation of 2.6 points. Bots like Katrain tend to give scoring evaluations with smaller distinctions than whole points, since it helps them make better distinctions between good and bad moves (e.g., the AI may consider 10 moves to collectively result in the loss of a single point, but individually none of the moves being bad enough to lose the point). It’s an artefact that makes AI better at making decisions, but doesn’t make much sense from the perspective of how humans think about Go. And of course smaller distinctions than whole points are mathematically impossible and irrelevant for the scoring (although they are mathematically relevant for determining the best order of endgame moves, through the use of combinatorial game theory, see also this).

Usually, the AI evaluation should be rounded to the nearest half-integer, so 2.6 becomes 2.5 points for Black.


Ah, I totally missed that point in the bottom left. Thank you for clarifying the situation.


The AI doesn’t score your game position. It is an estimate of what the score would be assuming continued high level play. The differences can be even greater when the players leave big situations unresolved.