Is the percentage bad for analysing games?

I finished one hell of a match today
But the egnine bugs e on one move
My oponent had an ad of 97% and then made a move which was not that mutch costly and then i was 97 adv,.
My questions is as the move was at jose, is the perentage reliable here or should i change it to points

1 Like

points make more sense
perfect AI would always just say 100% or 0%


You can switch the AI analysis to show the estimated score rather than the winning percentage.

Toggle the switch here:

Maybe your opponent only made a small mistake, but if it was a close game, then even a small swing in points could easily change the winning percentage from 97% for one player to 97% to the other.


Thanks to the answers
At first i did look for any opoturnity for killing a group that would explain this big swap. I looked for minutes. Then i thought the engine was broken

1 Like

Points are a better metric for analysis, particularly for weaker players.

It’s possible the 97 swing is accurate, or it could be an artifact from insufficiently deep analysis. Link it and we can help diagnose which.


i dont think it is a misstake of the engine. Or does the horizon effect take in earlier in go engines than in chess engines?
I was only a bit surpised that it was this mutch of an change with only this amount if points lost

I don’t know how the horizon effect in go engines compares to chess ones, but it definitely exists and the default OGS analysis is shallow (under 1000 playouts as that’s what the first paid level is). Also 1 playout of a neural network bot takes a lot more time that of a chess alpha-beta type engine, so I would guess a similarly resource intensive analysis like you might get for free on lichess or of a go vs chess bot would be shallower so more prone to reading errors if the intuition is wrong.