lichess.org
Donate

Change in evaluation of blunders and mistakes

I like the feature of running computer analysis after my game, but I propose that the criteria for blunders, mistakes and inaccuracies be changed. Take for example a game I played just now:

http://en.lichess.org/lfXf08oX#0

On 43. move, pacal plays 43...Kd4, which the computer considers a "blunder" as its evaluation goes from 6.8 to 15.5. However, any human can see that this is a move as reasonable as any in a lost position; it's hardly a blunder in the usual sense of the word. Furthermore, more intensive analysis (or maybe an 8-piece tablebase, whenever it is out) would surely show that Black will be mated in maybe 15-25 moves with best play, which shows that calling a move like that a blunder isn't really correct.

I think the computer analysis is a good feature and I would like to see this aspect improved. So how about changing the criteria like this; a move which changes a position evaluation that is already at -3 or worse isn't considered a blunder or a mistake, but only an inaccuracy at worst?

An additional point; it sometimes happens in my games that either me or my opponent is a few moves short of a mate, but because they picked a line that is slightly longer than optional (but still perfectly reasonable), their move is called an inaccuracy. I think this is unnecessary, how about changing that completely?

- Thanks for running the website, I love this site.
That would only serve to be more unhelpful and an unnecessary adjustment to the engine evaluation (this is Stockfish's domain, not lichess) just to not hurt somebody's feelings. The point of engine analysis is to supposedly highlight mistakes that deviate considerably from what would otherwise be perfect play. Fates can also change considerably and the winning side can make a colossal blunder that costs them the game. I wouldn't want that labelled as an 'inaccuracy.'

And why not point out shorter mate sequences? The point of engine analysis is to show you the mistakes you made and how you could've played better. Checkmating in fewer moves is playing better.

The engine's not here to stroke anyone's ego, and these proposed adjustments would only serve to inhibit learning from one's games.
Hey Admiral!

It's hard to understand why you think I'm talking about improving players' feelings and stroking their ego. It feels as if you just made that assumption out of thin air, since I do not mention anything like that in my post. Additionally, I am not talking about making changes to the engine evaluation but how the data from the engine is interpreted. But no worries, we all misunderstand things we read sometimes - you might just want to read my post a second time and comment again. :-)

I think that the purpose of game analysis should be to learn from it. But in order to learn from your chess games and become a better chess player, it is not a top priority to learn how to mate a little faster in a position where the outcome is already certain. In times with longer time control especially, such things makes little difference. Therefore, I believe it is not only inaccurate to call a move like 43...Kd4 a blunder in the game linked above, but also unhelpful from a pedagogical point of view.

To conclude, I can only repeat myself; mating in fewer moves is indeed playing better chess, but selecting a mate line that is a few moves longer is not usually a blunder in the common understanding of the word and calling it such is not helpful for the ordinary player. Therefore, I stick to my original suggestion of only calling such moves an inaccuracy at worst.

Now thanks still for your comment AdmiralA. I would like to see more people weigh in on the issue, which I believe could make a great site just a little better.
"An additional point; it sometimes happens in my games that either me or my opponent is a few moves short of a mate, but because they picked a line that is slightly longer than optional (but still perfectly reasonable), their move is called an inaccuracy. I think this is unnecessary, how about changing that completely?"

"To conclude, I can only repeat myself; mating in fewer moves is indeed playing better chess, but selecting a mate line that is a few moves longer is not usually a blunder in the common understanding of the word and calling it such is not helpful for the ordinary player. Therefore, I stick to my original suggestion of only calling such moves an inaccuracy at worst."

These two are entirely contradictory.

I do understand your sentiments when it comes to longer time control, considering how much I enjoy that myself, but considering how most games played on lichess tend to be blitz and bullet, being able to checkmate in fewer moves is fairly critical.
I don't see the contradiction. In my view, such moves should not be considered inaccuracies, but I think it still better than to call them mistakes or blunders. That's why I think they should be called inaccuracies at worst, which I think is not entirely accurate but still better than the current system. In fact, I think that calling them inaccuracies or not might be a matter of opinion, but surely most people would agree that those types of moves are not blunders or mistakes.

In all honesty, I did not expect someone to come here and vigorously defend the current system - which, as I understand it, is still experimental (see discussion here: http://en.lichess.org/forum/game-analysis/quickest-possible-checkmate). So thanks for saying your opinion, but it would be interesting to hear some other developers' point of view too.
@erindreki The contradiction is that first you say they should not be called anything at all and then you say they should be called inaccuracies.

I'd probably guess you meant the first one, anyway.
A good idea would be to increase the bounds for returning blunders as the evaluation increases in numerical value.

For example if the eval goes from 0.8 -> 3.8 (+3.0), this is likely a game changing blunder, whereas if the eval goes from 10.2 -> 30.2 (+20.0), this could be better termed as an inaccuracy.

Another example, 5.5 -> Mate-in-9 would probably be a blunder though...

I think this would be a good idea.

A further thought from my personal experience: When I analyze my games on lichess, I tend to judge my performance based on the number of inaccuracies, mistakes, blunders that are reported by the analysis. Sometimes this gets skewed because in completely winning positions, it reports many blunders because I missed the "shortest possible path to victory". This is in contrast to games where I make 1 key blunder and none after that defending a lost position. This is frustrating because the former was technically a much better played game than the latter, where just 1 blunder was made but it was critical.
Adding to my above post:

The reason why this makes sense in a human setting is that we don't strive for perfect play all the time. Sometimes, we play the most practical variation, instead of the most accurate. Thus, the analysis should be interpreted to reflect human psychology correctly.
That does make sense @erindreki, however. The point raised by @AdmiralA that reflecting the human psychology only really would be functional in games that last a considerable amount of time. Most games on lichess are played in a short timeframe in which time-management is a critical factor. Therefore SF points out the better move. I do agree on the point raised by @runpawnrun that blunders should be considered as gamechanging mistakes. Not merely judged on how many eval are lost. This seems to boil down to one fair suggestion towards SF (again, not Lichess) that only gamechanging bad moves should be considered blunders. This would be useful from a didactical perspective.

This topic has been archived and can no longer be replied to.