Would anyone explain how exacly is a move evaluated in Analysis?

Hello,

I don't understand how the analysis decided which mark to put on your or opponent's move. Is this based exactly on score difference computed from situation before and after your move?

I think I understand when my move gets marked with "good", "excellent" or "the best". I am intestered what is the difference between "a mistake" and "a big mistake".

And most of all, I am puzzled how to understand "inaccuracy". Of course it means I overlooked a better move. Atm I understand it as a move that is relatively neutral - not gaining an advantage, neither losing a position.

Btw I would like to decrease the Analysis depth. As a weak player, I usually think with depth <= 2. (I don't even know what depth on chess.com means, if it counts only your moves or opponent's too).

Thank you

The terms inaccuracy, mistake, and blunder are in general use well beyond chess.com, and are used to refer to moves that lead to a numerical difference in the evaluation between the move you or your opponent played and the best move.  Generally speaking, inaccuracy is less than +/- 1.0, mistake is greater than +/- 1.0 but less than 2.something and blunder is greater than that.  The terms get a little fuzzy because some programs try to take into account the game situation, and not call a move that changes the evaluation from -17 to -20 a blunder, for example.  There is no good reason to call that a blunder, because you were dead lost before and are not more dead lost in any real sense.  

All other terms that chess.com uses are opaque and not often independently meaningful.  I would not pay any attention to any of them.  Simply refer to the three "hard" terms that have a defined meaning.  Decide whether you want to only look at your blunders (if it was a blitz game and you are in a hurry) or also your mistakes (good practice, if you have time and took the game semi seriously) or also your inaccuracies (can be a challenge to figure them out - really only worth it if you want to invest real time in analyzing the game and squeezing every insight out of it).

Bottom line is the information is proprietary software and not open sourced. Chess.com will not reveal publicly the why's and how's the evaluations are derived. Much to do with potentially being copied and possibly cheat detection methods. It's a tool provided, albeit a very limited one (depth of search as an example), for use by members in reviewing their games. Making heads or tails of the Accuracy evaluations often appears to be arbitrary, often contradictory from game to game.  The tool should not be used in drawing conclusions regarding fair play. 98% - 99% evals can be quite normal. 70% eval could mean nothing to cheat detection software. Certainly, performances of 10 bullet wins in a row all with 99% is a Red Flag. Use the data as but another study tool for your own play. It's not meant to be used to draw conclusions about opponents play. I've noticed the evals can change over time as new data is entered into the program. Example: In the Smith Morra , the software used to give 3. c3 as an inaccuracy, the final eval would suffer. Now it is given as a book move. Long sequences of only moves in an endgame can dramatically increase the final eval, even though blunders were made early on.

I agree that chess.com's computer is sometimes off. Check out Black's move 23,   Ne5+, forking the king and queen.  White resigned.  End of Game. 

The analysis said an alternative was fX36+. which does  keep up the pressure.  But as the computer notes, the response Bf4 protects the King, and White's queen fights on. Maybe White would lose eventually..  though Black has plenty of weaknesses, such as the H3 pawn(!) backed up by the H Rook , 

I think chess.com puts in these "alternatives" sometimes to justify the yearly platinumum fee I pay . . I mean, the more advice i see, the less likely I will defect to lichess . . .  

Does anyone have an example of "brilliant" move? I've asked before but haven't seen one. Just curious.

Thanks all for kind responses

Before computers, humans analyzed games and gave them marks such as brilliant, good, interesting, dubious, etc.  All of these marks are judgemental.  Today analyses are performed by computers that have been programmed by humans.  The evaluations are based on criteria input by humans.  Again, they are judgemental.

As an aside, the evaluations sometimes leave a lot to be desired.  I remember reading about a grandmaster who made a serious blunder, but managed to win the game.  At the end of the tournament, he was awarded a brilliancy prize for his "deep sacrifice."

An evaluation by a comp is always based on material advantage. The material weight can be higher or lower, depending on positional consideration. Programmers are doing a mighty fine job of it to recognize how position effects the relative value of the pieces. Bottom line , superior strength is the path to victory.  Computers win by brute force. Can finesse be programmed?  That's what chess AI is about.

An excellent question. One that I'm not qualified to fully answer. 

After 23...fxe6 24.Bf4,

Ne5+ forking the king and the queen is still on the table. I haven't doubled checked it, but on this line, black should get the Queen for free whereas in the game, it was traded for the knight.

Do note Bf4 does not defend against Ne5+ as the bishop is pinned.

Also note that the only other sensible move 24.Ke4 loses to 24...Qc2+ 25.Rd3 Qg2+ with mate to follow.

@xlastovi Sorry if I make you more confused.  But I think your question concerns exactly how the determinations are made about the various gradations (Best, Excellent, Good, Inaccuracy, etc...) These determinations are a combination of chess culture, computer analysis, and some complicated (but meaningful) programming definitions.  The chess culture has already been referenced.  But I will opine...

When kids learn chess, they learn the basic scoring system: a pawn is worth 1 point, bishop 3, knight 3, etc...  This gives a player a rudimentary ability to judge the strength of any position or any possible complicated exchange.  Additionally, pre-software game evaluations did not include a numerical calculation of strength. Rather, notation was developed to indicate that a move was unexpected or poorly judged, etc...  

When we started  programming computers to play chess the only pathway to do so was through brute force (look at every possible move as deeply as reasonably possible and evaluate the relative outcomes based on material), then assign a numerical value to the current position to reflect best play.  This is actually a little misleading, but we'll get back to that.  Combine that with a few programming definitions and you get the current notations that we see in our computer analysis.  As the programming became more sophisticated, so too did the scoring as the computer was able to reflect the possibility of future gains in any position.  It has developed now to the point where the evaluation is much less about material, and much more about the strength of the current position on the board.  If you are five points down, but in your current position you can execute a series of forced moves that will gain 10 points, the evaluation will reflect this possibility as your actual strength in that position and show you as being up approximately 5 points.

As concerns the notations, @ArtNJ gives a decent approximation of the numerical definitions behind the terms.  But the exact definition is actually quite complicated. (I wouldn't worry to much about the exact numerical demarcations and the programing definitions.  But just to be thorough I will explain).  

Early in a complicated game, a move that reduces the numerical value of a position by up to 1.5 points will be noted as an inaccuracy.  1.5 or greater would be called a mistake.  Over 2.5 and your heading into blunder territory. But as the game progresses and material is taken off the board, the numerical values that denote these notational descriptions actually shrinks.  There are many moves available in these positions, so a computer's numerical evaluation isn't quite as telling.  But later in the game, when a player has less resources to recover from a bad position, the loss of a point may well be counted as a mistake, or even a blunder, depending on the position and the moves that were available.  

Look at it this way. When the computer analysis gives you these different notations what you should hear the programmer/computer saying is: Blunder = you really should have noticed that, Mistake = check this out, maybe learn from the error, Inaccuracy = you can't play like a computer.  Don't worry about the inaccuracies.  Concentrate most of your efforts on not making blunders.  Avoiding mistakes will come with practice.

Before the advent of AI like Alpha Zero and the like, the chess culture had become so engrained in engine analysis that there was a thought we should learn to play like computers.  Thus inaccuracies were seen by some as marking possible learning opportunities.  Less so now.  So the differences between Brilliant, Best, Excellent, and Good: Brilliant is the top move of an engine specifically when that move is extremely difficult to find (I don't know how a computer would judge such a thing.  I just assume they figured out how to program it to do so  ). Best equals engine's top move.   Excellent are moves that are extremely close, numerically, to the best move (usually only differing by a tenth or less).  Good is a move that is within half a point or so to the best move.

Earlier I mentioned how the numerical value assigned to any position reflects best play.  The thing to understand is that evaluation is always FORWARD.  Therefore, it is ALMOST impossible to make a play and substantively increase one's own evaluation.  (The strange exception to this are some very unique and very rare positions that somehow leave the computer blind.  This has to do with decision trees and heuristics that tell the computer to stop calculating particular lines as worthwhile and place it's resources on other lines.  But the error is very rare.) If your current evaluation sits at 3.52, typically there is no move that can be made to bring your evaluation to 4 or 5.  The computer is basically saying that IF you play the best move, your position will have a numerical value of 3.52.  (If you play against an engine, you will see it "think" and you may see that evaluation change in the decimals as it calculates deeper. But) for analysis in an actual game against mere mortals, any positional value is the computer's evaluation and score for the best move that is possible on the board.  Generally speaking your score will not be increased by your good moves; rather your score will be increased by your opponents bad moves.

Here is why this is misleading. It is quite often the case that a particular position is judged as strong for one player because there is ONE good move that an individual can play.  That move may be the only good move and it may be hard to find.  But it's there.  So the computer reflects the strength of that position, calculating what it WOULD be IF that move were made.  The opponent, on the other hand, may have a myriad of decent options, yet all of the possibilities are relatively the same in strength.  During this kind of a game, we would say that the individual who only has only ONE good move, but it is a move that is hard to find, while his opponent has many good moves unless that one move is played... we would tend to say that this is player is limited and in a poor position.  The computer evaluation just isn't capable of reflecting this.  You can not decipher that this is the case by looking at a number.  Yet isn't this the very thing that we hope to achieve??

I hope, during my games, to play in such a way that I have many good moves available, and that my opponent has very few.  

The conclusion to all my ranting is that the whole 'computer notation' question is a rabbit hole that can easily move you away from the fundamentals.  Generally speaking, computer analysis is a valuable tool to spot mistakes and blunders.  That's about it.  But for the curious, I hope that helps.  Also, I wouldn't limit the depth of analysis.  That won't really help you to understand the game.  Concentrating on Blunders first, Mistakes after you begin to limit your Blunders.

It appears that chess.com has changed the way the analysis function works, at some point during the past several months. It used to evaluate my 18. … Ne5! as "Brilliant", but now it just marks it as "Best".

@jobishcw Thank you very much for your thourough explanation

thanks Blueemu