Nice to read you again. I suggest to use the lichess approach for mistakes. It relies on the winning probability decrease, rather than the evaluation change. As the goal in chess is winning the game, their approach makes a lot of sense.
Hi, I'm using the Lichess approach. I have defined the expected score (which is the same concept as Lichess' winning probability) based on GM games and use this to define mistakes.
I just like to call it expected score because there are also draws and Lichess gives an equal position a 50% winning probability even though it will most likely be a draw, so a player expects to score 0.5 points.
That last graph is so interesting! I'm trying to develop better intuition about this. Basically, the point is that from winning position you can go to equal, worse, or even losing, but from worse position you can only go to losing?
This makes a lot of sense mathematically, though I feel like it unduly overshadows your point about there being more mistakes from slightly better than from slightly worse positions, which is much more rooted in "popular imagination" / folk science. WDYT?
Yes, the mathematical reason is that a bad positions can't be messed up too badly so there are fewer mistakes from them. With my definition where a mistake is a move that reduces the expected score by more than 10%, it's even impossible to make a mistake in a position where one has winning chances lower than 10%.
I'm actually unsure what most people would intuitively say about when mistakes happen. It's also often said that players often "crack" in slightly worse positions and make mistakes there. But then again, when someone is worse the position often deteriorates little by little whereas losing a better position more often than not involves big mistakes.
Hi Julian,
Nice to read you again. I suggest to use the lichess approach for mistakes. It relies on the winning probability decrease, rather than the evaluation change. As the goal in chess is winning the game, their approach makes a lot of sense.
Hi, I'm using the Lichess approach. I have defined the expected score (which is the same concept as Lichess' winning probability) based on GM games and use this to define mistakes.
I just like to call it expected score because there are also draws and Lichess gives an equal position a 50% winning probability even though it will most likely be a draw, so a player expects to score 0.5 points.
That last graph is so interesting! I'm trying to develop better intuition about this. Basically, the point is that from winning position you can go to equal, worse, or even losing, but from worse position you can only go to losing?
This makes a lot of sense mathematically, though I feel like it unduly overshadows your point about there being more mistakes from slightly better than from slightly worse positions, which is much more rooted in "popular imagination" / folk science. WDYT?
Yes, the mathematical reason is that a bad positions can't be messed up too badly so there are fewer mistakes from them. With my definition where a mistake is a move that reduces the expected score by more than 10%, it's even impossible to make a mistake in a position where one has winning chances lower than 10%.
I'm actually unsure what most people would intuitively say about when mistakes happen. It's also often said that players often "crack" in slightly worse positions and make mistakes there. But then again, when someone is worse the position often deteriorates little by little whereas losing a better position more often than not involves big mistakes.
> the sample size gets smaller and it’s more difficult to make conclusions.
sigh!