When a position becomes completely winning, the winning path is a purely subjective choice. Do I win the queen for a bishop or do I execute a complicated checkmate? If I go for the checkmate, I might mess things up, taking the queen is often practical. Computers have no right to critique such a choice.
This doesn't seem to be factored in at the analysis board, where a jump in the eval from +8 to +5 could be seen as a "blunder" or a "mistake". This is not very useful. Humans care about whether they made tactical mistakes that could've changed the outcome of the game.
There won't a perfect solution though as there is always the possibility that a drop from +6 to +3 could actually change from winning to draw, as computers often mis-evaluate drawn positions. But I think it would be extremely rare for a computer to evaluate a drawn position as a +5 or more.
I think It would be better if it ignored differences over circa +5 positions. You could also think about creating alternatives to the "centipawn" that also ignored such irrelevant differences, because when you have wild tactical games there can be frequent swings even when one side is always winning. The "centipawn" becomes kinda useless in those scenarios.
It is "analysis", not the score. After game is won, centipawn numbers don't matter.
Of course it s very helpful to see quickest winning combination (beauty of chess) rather than grabing a lot of pieces and going for a winning endgame.
Long variations take long time in the last seconds of a blitz game. One need to learn shortcuts.
There are good choices, and better choices. It doesn't tell you that you made the best choice if you didn't. The fact is that going for the queen simply isn't as effective as a checkmate sequence.
And besides, it'll still display a huge advantage for you if you win the queen for a bishop. You're basically asking the computer to lie about the numbers to make yourself feel better... which is silly.
So after the fact, you want the computer not to tell you what the best move or sequence of moves was, because it should assume you might have messed it up? This is what you're saying?
It should still show it, but to classify a drop from +9 to +5 as a blunder is simply wrong. It should arguably be classified as an "inaccuracy".
> There are good choices, and better choices. It doesn't tell you that you made the best choice if you didn't. The fact is that going for the queen simply isn't as effective as a checkmate sequence.
It wins in fewer moves, yes, but it doesn't change the outcome of the game. So to say that it's better or more desirable to win in fewer moves is a subjective judgment. What if there's a more beautiful way to win with more moves? Or a win with a simpler, more reliable plan?
> You're basically asking the computer to lie about the numbers to make yourself feel better... which is silly.
No, I'm asking the computer not to make subjective judgments about which wins are more desirable.
It's not subjective at all.
The objective of chess is to checkmate your opponent's king.
If you are faced with a choice between guaranteed checkmate or taking a piece to increase your prospective chances of achieving checkmate later on... the objective choice would obviously be the first one.
Sure... you would PROBABLY win the game, assuming you didn't make any mistakes, but one choice is clearly superior to the other.
Following your logic, you'd also consider taking a bishop for free instead of a checkmate in one to be only a trivial inaccuracy, because you'd probably win either way.
As a side-note - Computers are incapable of making subjective judgments.
> If you are faced with a choice between guaranteed checkmate or taking a piece to increase your prospective chances of achieving checkmate later on... the objective choice would obviously be the first one.
My whole point is that the checkmate might not be "guaranteed". The player could see that "ok there's probably a checkmate in here, but I'll rather win the queen" *exactly because* that is more likely to give a win. It's a rational, objective choice by a player who knows his own limitations. When the computer says it's a "blunder", it's just not helpful. These points apply even more forcefully when we're talking about a jump from +8 to +6 and there's no forced checkmate.
> As a side-note - Computers are incapable of making subjective judgments.
This is wrong. Computers make subjective judgments if programmers tells it to.
That, my friend, is literally the opposite of an objective choice. the player has decided that they're not confident enough to play accurately, so they chicken out based on what they feel like they'd prefer.
By the way... what exactly is it that you want to be changed? Do you want mistakes or blunders to be reclassified as inaccuracies when there's still a significant advantage for the same side? What exactly does that achieve?
It actually does do something similar... in that imperfect checkmate sequences or missing a forced mate to fall back to a +10 advantage is classified as an inaccuracy.
And, finally... your final statement makes absolutely no sense. If you program in a set of rules... the judgments made by the computer are based solely on these rules. This is pure objectivity.
> you program in a set of rules... the judgments made by the computer are based solely on these rules. This is pure objectivity.
I get your point, but it's just wrong: The rules contain subjective judgments made by the programmer. For the user it doesn't matter where these rules came from; it's part of the program.
> It actually does do something similar... in that imperfect checkmate sequences or missing a forced mate to fall back to a +10 advantage is classified as an inaccuracy.
I checked that before I posted, look at this game:
http://en.lichess.org/OcIdSapm/black#72I counted the mistakes, so the lost checkmate at the end must've been the blunder. (probably a bad example as that was genuinely a blunder, it was a bullet game though)
> Do you want mistakes or blunders to be reclassified as inaccuracies when there's still a significant advantage for the same side? What exactly does that achieve?
Well that was kind of an attempt at compromise from my part, I still think it's wrong to call those rational choices inaccuracies. But it's less wrong.
It achieves a more accurate description of how well a side played. The "mistakes" and "blunders", from my human perspective aren't accurate, because it ignores pragmatic choices, choices which the player can look back at and think "that was a good choice".
How else would you define inaccurate play other than a distinct drop in the evaluation? Large drops are classified blunders, moderate drops are classified mistakes, and small drops are classified as inaccuracies. It's not a matter of whether or not you'll still probably win - if the evaluation drops a lot... it's a blunder by definition. Sure, the definitions may have been thought up by humans... but this is essentially a universally accepted principle.
I looked at the game, and I'd say that it's a pretty huge mistake to go from guaranteed checkmate to 5 pawns up... I mean, you dropped a queen for god's sake.
Finally, accurate chess is not based on pragmatic choices, it's based on what the actual best moves are. Sure, human players may choose to go with safety rather than the ideal move combinations, but that doesn't make it accurate.
It's analogous to looking at a 100 m sprint, and saying that all of the runners ran just as well because they all crossed the finish line eventually.