lichess.org
Donate

Winter Marathon 0 inaccuracies game

#11
@achja: Well, keep in mind that the threshold for an "Inaccuracy" is pretty high, at 0.5 pawns.

It's not like SF actually thinks a 0/0/0 game is perfect; it just means no single move dropped the evaluation by at least 0.5.

In fact, in theory, you could lose 0.45 every move and end up down a queen after 20 moves, and you'd get a 0/0/0 from SF :)

The inaccuracy/mistake/blunder is a very rough guide, nothing more.
#12
@OneOfTheQ #11

Very interesting observation.
Most insightful comment I've come across this week.

Thanks for sharing ! :)

#15
Actually, I sometimes go from -2.4 to -4.5 (losing), but it's not even a inaccuracy. Not sure how these are calculated.
#16
Example in the game you play Nf3
Engine analysis for Nf3 = -4.5

Engine recommended move is g3
Engine analysis for g3 = -2.4

CPL = centipawn loss
PL = pawn loss
PL = -2.4 - (-4.5) = -2.4 + 4.5 = 2.1
PL = +2.1
CPL = +2.1 x 100
CPL = +210

if CPL >= 300 then it is a "Blunder"
else if CPL >= 100 then it is a "Mistake"
else if CPL >= 50 then it is an "Inaccuracy"

Since CPL is 210, then it is a Mistake.

The Blunder, Mistake and Inaccuracy are only used to define the score difference between
the engine move and the player move.

#17
#14 Interesting. I'll give some thought to that win expectancy distribution...

In my experience CPL >= 250 / CPL >= 100 / CPL >= 40 is a more intuitive classification than the current classification system (300 / 100 / 50).
#18
@achja;
I see your point with the time losing manover in the opening but that's how I played this for a while and it always was sort of okay. I think Bxe2 might be a bit strange but not that bad since B pair isn't worth much here. Btw, I agree that the later position is a (little) bit better for black exactly because the N isn't worse than the B in that position. Though on Qa5 I would have just blitzed out c4, not that nice but somehow ok.

In the other game after Na4 Qc7 I would probably have played c4.
And yes, 0 0 0 doesn't mean perfect but still quite decent after 19h exhausting play. ;) (though as I mentioned I don't even view them as that well played; all very obvious straightfoward moves)
#19
@RealKool: I'm aware of the correlation, and the truth behind it is exactly the reason for my claim.

My point is that the inaccuracy/mistake/blunder count from the analysis is a rough guide, because it excludes moves with centipawn loss less than 50.

Those mistakes can be quite significant even in isolation (going from a stable -0.4 to -0.8 can mean a serious worsening of the position, in some positions that can even be going from a draw to a loss), and in combination can obviously hurt very much. Just a handful of such mistakes can leave you down a piece.

The inaccuracy/mistake/blunder count is important, but in determining the quality of a game it is only a rough guide, since it excludes so many potentially serious mistakes.

That's why it's awesome that we have average centipawn loss here; you can get a reasonable handle on the quality of play with a metric that doesn't throw away the information from those other mistakes.

Of course, it's also not a perfect metric, especially for human play (a human will do better playing a -0.3 position with a clear plan than a +0.3 position that can only be justified with SF-level tactics), but it's still a better metric for overall quality of play than the inaccuracy/mistake/blunder count.

That latter measure is quite useful for other reasons, like finding your biggest mistakes, but for overall quality of play it is a very rough guide.
#20
@moistvon: Yes, 0/0/0 is still quite nice. Goodness knows I don't achieve such games :)

My response was not meant to belittle your achievement, but to explain why a game with moves that appear suboptimal might still get 0/0/0 :)

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