Large difference between game review and local analysis eval values - Chess Forums

Oct 24, 2025

0

#1

In the following game, black move 12. .... Nh5 looses their queen in exchange for a bishop - blunder.

https://www.chess.com/live/game/144616649472

Game review calls this blunder a mere mistake and shows eval change of only 2 points, which seems very low for loosing a Queen in exchange for a Bishop. I would expect eval to drop by about 6 points and indeed this what analysis shows if I copy / paste PGN into a new analysis window. Why is there such large discrepancy between game review and local analysis eval??? I could understand 0.5 diff, perhaps 1.0 but not 3!

Thank you for your help!

justbefair

Oct 24, 2025

0

#2

The basic problem is that the move classifications use a different model than you do. They look at changes in winning chances. not just material gains and losses. They call it the "expected points model." https://support.chess.com/en/articles/8572705-how-are-moves-classified-what-is-a-blunder-or-brilliant-etc

Expected Points Model

Expected Points uses data science to determine a player’s winning chances based on their rating and the engine evaluation, where 1.00 is always winning, 0.00 is always losing, and 0.50 is even.

At 1.00, you have a 100% chance of winning, and at 0.00, you have a 0% chance of winning. After you make a move, we evaluate how your expected points—likely game outcome—have changed and classify the move accordingly.

The table below shows the expected points cutoffs for various move classifications. If the expected points lost by a move is between a set of upper and lower limits, then the corresponding classification is used:

Classification	Lower Limit	Upper Limit
Best	0.00	0.00
Excellent	0.00	0.02
Good	0.02	0.05
Inaccuracy	0.05	0.10
Mistake	0.10	0.20
Blunder	0.20	1.00

Black was already down by a piece and was considered to have a very strong chance of losing. So losing the queen for a bishop didn't really change the losing chances by that much- hence it was called a "mistake" and not a "blunder."

Hedgehog_InTheFog

Oct 24, 2025

0

#3

Thank you. I understand chess.com approach to move classification. Still think that classifying queen-bishop exchange as a mistake (assuming no other compensation) is terribly misleading, especially to beginners. However chess.com have their rationale.

Hedgehog_InTheFog

Oct 24, 2025

0

#4

The real big question is a huge difference in evaluation produced by local analysis engine and game review cloud engine - well beyond engine version differences. Does chess.com 'normalize' the eval value too instead of presenting raw engine evaluation??

justbefair

Oct 24, 2025

0

#5

Hedgehog_InTheFog wrote:

The real big question is a huge difference in evaluation produced by local analysis engine and game review cloud engine - well beyond engine version differences. Does chess.com 'normalize' the eval value too instead of presenting raw engine evaluation??

I don't think they do "normalize" the evaluations. I think most of the differences are usually due to differences in engine depth. (But I have to leave now and can't think about it any more.)

Hedgehog_InTheFog

Oct 24, 2025

0

#6

Thank you for responding. I tried and could not come up with an example where depth differences (depths achieved in practice in under 1 sec - not theoretical constructs such as 2-3 ply depth) would result in such significant eval differences. I agree that this seems the only reasonable explanation I can see too but it just doesn't connect with my understanding of how the engines work.