What is the real value of accuracy metrics?

Reading the lichess.org/page/accuracy and curious, what is the real value of accuracy metrics if for blunderfest:



with 5 blunders I still have 89% accuracy?

How is that possible? I am trying to dig into the lichess.org/page/accuracy , but can't still understand what accuracy really reflects in case in 42 moves I had 5 blunders and still have 89% accuracy?

My "personal feeling" of accuracy here is not more than 60%.

Can it really makes any sense if blunderfest has "almost" 100% accuracy? If so, we have very narrow range of percentage between blunderfest and "good" games. Like, these first 89% almost doesn't work.

How do you fell about it?

Accuracy% = 103.1668 * exp(-0.04354 * (winPercentBefore - winPercentAfter)) - 3.1669

@Toadofsky said in #2:
> Accuracy% = 103.1668 * exp(-0.04354 * (winPercentBefore - winPercentAfter)) - 3.1669

Yes, I read the lichess.org/page/accuracy.

My question, is if it makes real sense? See my arguments above. I see the formula, but I see that it was just made-up and tunes to work somehow. My point it due to the logarithmic nature of human perception it just confuses.

When I see that the game is 89% accurate I assume that it is really good played and never think that this is a blunderfest.

Just played another blunderfest



and with the 4 blunders and 1 mistake in gives 62% of accuracy which makes total sense for me.

@Motroskin said in #3:
> My question, is if it makes real sense?
How would you design it?

Stop blundering?

@Motroskin said in #3:
> and with the 4 blunders and 1 mistake in gives 62% of accuracy which makes total sense for me.

... because your moves were 62% accurate. 0% would represent "every move is a blunder" and you're nowhere near that.

@Toadofsky said in #6:
> 0% would represent
Surprisingly (to me), I was unable to get 0% accuracy; presumably because winPercent is based on the regression against lichess games, rather than the engine-supplied win percent.

It really has absolutely no value and it upsets me. #removeAccuracy

Long ago, I read a nice quote in a book: "The only problem with dictionaries is that people believe them." (Meaning that people tend to believe blindly to what they find in them.) The same could be said about many other sources of information, accuracy numbers (and actually all the engine output) being among them.

@corvusmellori said in #7:
> Surprisingly (to me), I was unable to get 0% accuracy; presumably because winPercent is based on the regression against lichess games, rather than the engine-supplied win percent.
Or lichess uses rounding up, together with the fact that a win percentage will never be exactly 0% or 100%, as your example proves.