s
Sort:
Mar 23, 2023
0
#2
CalculatedPuzzle wrote:
Why chess.com puzzle ratings mean nothing 2418 rated puzzle
This puzzle, rated 2418 is literally just a one move queen blunder. The puzzle played Qh4, and you can quickly calculate that taking on d6 doesn't work after the simple kg7. This puzzle could be solved by a true beginner, and this proves chess.com puzzle ratings are so inflated. Chess.com should really consider fixing the puzzle ratings, because there are also people rated 65,000 on the top of the leaderboards. The puzzle ratings like other websites like chess tempo, or lichess match the actual rating of the puzzle.
It also gives a false sense to 1400 players, thinking "I'm 2000 puzzles, and 1400 on blitz, so therefore I should be 2000. The puzzle rating of a player should match the level of their play. Chess.com should not give them a false sense of rating and give these super inflated ratings.
Puzzle ratings change based on members solving them. Apparently enough high rated players are getting it wrong to keep its rating high.
CalculatedPuzzle wrote:
I am just suggesting that they re evaluate the ratings of the puzzles because they don't seem to be too accurate.
"Hypnotic" blunders are not common but not rare either. You make a blunder and your opponent fails to notice it because he "trusts" you. The likelyhood is greater under time pressure. Puzzle solvers take risks when time consumption is heavily punished.
This is part of the domain of "strategic solving" very familiar to professional solvers. As a solver you "know" that a puzzle can have only 1 solution and conclude strategically that you need not calculate Qh1+ when Qh2+ clearly has the same effect. They can't both be right. The strategic solver would never consider 1. QxQh4 in the diagram as key moves to professional puzzles may capture at most a pawn. And there are many, many more strategic considerations. An obvious one is that you assume a highly rated puzzle to be hard and not simple.
One way to solve this is to have an engine determine the puzzle rating. For that you will need AI-tuning such that it can mimic human strengths and weaknesses, including strategic and psychological methods. Commonly, engines only aim at objective scoring - which is to say the scoring against their own kind, 4000 rated beasts.