AFAIK, As in chess analysis, moves are determined from an equilibrium because under perfect play by both sides, a “draw” is what you got. Then, you really don’t gain estimated advantage by your moves, but because your opponent mistakes
Then moves usually are 0 points for “perfect” move, and worst moves are negative points. What I saw today are positive points, and not evaluated as the fav move.
In this game, movement 193. I’m white and is my turn to play, and as you can see, AI fav move is B2 with a 0 score, and next movement I played was K2, and estimated as a 19 point gain.
Perfect play is unknown, so the AI can just say “I think this is best”. AFAIK it means that the AI changed it’s mind after the move has been played. It says: “I think this move is 19 points better than what I would have played”. Usually that only happens if the AI does just a quick search (but there are exceptions).
Look at both evaluations: score and winning percentage. The AI thinks your move is better for points but decreases winning percentage. In other words you were leading, and playing B2 would guarantee a safe win by 2.5 points. Whereas your game move K2 makes more points but is more risky, you could have lost the game if your followup was wrong.
AI don’t think about all possible 361 moves. It thinks only about few moves that neural net guess are probably best. So, rarely, it is possible to find move that is better by both % and score.
If your PC is very strong, you can activate special mode which would analyze all possible 361 moves fast.
Yes it can happen that AI gives positive points. It means it didn’t really consider that move and that move turned out to be better than it expected. This is more likely to happen in a situation that needs deep reading, while the AI’s reading is limited by a low number of visits/playouts.