The Chess.com analysis labels each move with either "Best, Excellent, Good, Inaccuracy, Mistake, Blunder, Missed Win". I'm curious about the specifics regarding how the analysis qualifies each move.
I've seen moves labeled as "Good" when they're evaluated almost exactly the same as the "Best" move.
In the above example, the engine evaluates both moves with a difference of 0.01; yet, one is qualified as "Best", and the other merely "Good".
In contrast, I have also seen moves qualified as "Alternative" when they are far inferior to the "Best" move in evaluation, as per the following example.
The discrepancy here is clear.
Even more disturbing is the following image, in which the "Best" move is actually worse than the "Excellent" move according to the evaluation.
I suspect that these examples may be influenced by the evaluation's updating as the engine's calculation deepens while the label given to the move remains static and thus becomes outdated. Why the label doesn't update with the new evaluation is beyond me. Most likely that could just be an oversight.
All this got me thinking about how the analysis assigns its labels; that is, what is the exact relation between the move to be labeled and the best move in the situation? What formulae are used to determine the quality of a given move? If anyone has insight into chess.com's code, I would love to know where exactly these labels are coming from and what they really mean in terms of mathematical value.
Sometimes the computer does seem to give "odd" evaluations. I thought I had an answer to your third diagram (excellent v best), but as I looked at it a second time, I realized my thinking was wrong. With Black moving, Black would want the "score" to be lower and that's what the diagram shows for the excellent move, but the BEST move should be lower (closer to zero or below). So, like you, I'm puzzled.
The Chess.com analysis labels each move with either "Best, Excellent, Good, Inaccuracy, Mistake, Blunder, Missed Win". I'm curious about the specifics regarding how the analysis qualifies each move.
I've seen moves labeled as "Good" when they're evaluated almost exactly the same as the "Best" move.
In the above example, the engine evaluates both moves with a difference of 0.01; yet, one is qualified as "Best", and the other merely "Good".
In contrast, I have also seen moves qualified as "Alternative" when they are far inferior to the "Best" move in evaluation, as per the following example.
The discrepancy here is clear.
Even more disturbing is the following image, in which the "Best" move is actually worse than the "Excellent" move according to the evaluation.
I suspect that these examples may be influenced by the evaluation's updating as the engine's calculation deepens while the label given to the move remains static and thus becomes outdated. Why the label doesn't update with the new evaluation is beyond me. Most likely that could just be an oversight.
All this got me thinking about how the analysis assigns its labels; that is, what is the exact relation between the move to be labeled and the best move in the situation? What formulae are used to determine the quality of a given move? If anyone has insight into chess.com's code, I would love to know where exactly these labels are coming from and what they really mean in terms of mathematical value.