Please, take a look at my bullet game:
http://en.lichess.org/oMVC8fqH/white#728 moves
4 innacuracies 1 mistake and avarege centipawn loss 3!?
Is this possible? Or just an engine mistake?
What do you people think?
Its not one engine mistake:
1) You are not making mistakes, or inaccuracies because you have already many advantatge and its hard to be wrong
2) Once you have many advatatge, most of times engine doesnt care what you do, because you are already winning.
Once, I remember I was playing and I had like 2 full rocks and 1 queen more than my opponent, I sacrified my queen just to simplify the situation (killing one pawn) and Stockfish didnt detect it like inaccuracy or mistake (when for sure it was a mistake, because I had more easy ways to win)
So you mean if I have so much advantage whatever I do my moves don't lower my chances - thus my centipawn loss is extremely small for many moves and that's why average centipawn is so low?
Yes I think so, but only if these moves doesn't lower a lot your chances; but I didn't analyse your game to much; maybe you make an amazing bullet game :D
I think it must be a stockfish mistake, bug or something. Take a look:
mistake = - 1 pawn in a move which is - 100 centipawn
innacuracie = - 0,5 of a pawn which is - 50 centipawn
1 mistake + 4 innacuracies / 28 moves = - 100 centipawn + (4 x - 50 centipawns) = - 100 centipawn + (- 200 centpawns) / 28 moves = - 100 centipawns - 200 centipawns / 28 moves = - 300 centipawns / 28 moves = around - 11 / 28 moves
Simple math is proving stockfish mistake :) So even if I made all of the rest 23 moves perfect which is impossible, the average centipawn loss should be 11.
Cheer up boys - even stockfish makes simple mistakes!
#5 Stockfish does only the following:
* Evaluate every position which occurred in the game
* Provide principal variations (best moves from Stockfish's perspective) for every position in the game
Stockfish doesn't do calculation of average CP loss or labeling of inaccuracies, etc.
AVERAGE centipawn loss is SUM over MOVES. Your mid- to end-game was pretty much perfect, drastically lowering the few losses.
#7 Yeah, dragging out a game as long as possible tends to quickly reduce average CP loss!
I should make a userscript plugin to display statistics about CP loss... (interquartile range, median, standard deviation)
