Your conclusion seems right to me. The computer's estimates of the likelihood of a draw at move 0 certainly overstates the odds of a draw between two humans. It only makes sense that the WDL percentages at move 0 be set to some more realistic numbers. That being said, I think we could increase the accuracy of the computer's WDL estimates further by factoring in the sharpness of a position. If the WDL estimates are based solely on evaluation, then the computer will predict that a draw is just as likely to result from a very sharp Sicilian position with an evaluation of +0.00 as in a Berlin endgame with the same evaluation. While that may be true of computer games, it is clearly not true that humans are equally likely to draw these two positions. The Berlin endgame has a draw rate greater than 90%, while that's not true of the sicilian.
The idea behind the WDL graph is that you see the draw rate in the graph, so it should show if the position is a "boring" draw (the win and loss percentages are very low) or if it's sharp (the win and loss percentages are quite high and roughly level).
Right. In those terms, my suggestion is that we distinguish between the probability a position generates a decisive result from its sharpness, since these two are not synonyms. A positions sharpness, it seems to me, reflects a broader variety of factors than merely the likelihood of a decisive result. For example, players use the word sharp to describe positions where it is very easy for both players to go wrong, and in which both players have some possibilities for active play. It seems to me that these features, among others, distinguish Sicilians from Berlins.
Now ,you might think that sharpness picks out some other set of features, but no matter. The more general point I want to establish is merely that sharpness goes beyond, but of course correlates with, mere decisiveness.
Your conclusion seems right to me. The computer's estimates of the likelihood of a draw at move 0 certainly overstates the odds of a draw between two humans. It only makes sense that the WDL percentages at move 0 be set to some more realistic numbers. That being said, I think we could increase the accuracy of the computer's WDL estimates further by factoring in the sharpness of a position. If the WDL estimates are based solely on evaluation, then the computer will predict that a draw is just as likely to result from a very sharp Sicilian position with an evaluation of +0.00 as in a Berlin endgame with the same evaluation. While that may be true of computer games, it is clearly not true that humans are equally likely to draw these two positions. The Berlin endgame has a draw rate greater than 90%, while that's not true of the sicilian.
The idea behind the WDL graph is that you see the draw rate in the graph, so it should show if the position is a "boring" draw (the win and loss percentages are very low) or if it's sharp (the win and loss percentages are quite high and roughly level).
Right. In those terms, my suggestion is that we distinguish between the probability a position generates a decisive result from its sharpness, since these two are not synonyms. A positions sharpness, it seems to me, reflects a broader variety of factors than merely the likelihood of a decisive result. For example, players use the word sharp to describe positions where it is very easy for both players to go wrong, and in which both players have some possibilities for active play. It seems to me that these features, among others, distinguish Sicilians from Berlins.
Now ,you might think that sharpness picks out some other set of features, but no matter. The more general point I want to establish is merely that sharpness goes beyond, but of course correlates with, mere decisiveness.