Lately did a lot of puzzles and what I noticed is that some very hard puzzles have a low rating and some easy puzzles have a high rating, could have been just me if it happens a few times, but it actually seems to be pretty common.
For example? (Can you provide links?)
They just added thousands of new puzzles, so there are bound to be some misrated ones.
here some extreme examples:
https://www.chess.com/puzzles/problem/979006/practice
https://www.chess.com/puzzles/problem/775710
I think I know why that happens. I read somewhere they extract the puzzles automatically from real games. Once a puzzle has been tried by a bunch of rated users, they should have an accurate rating estimate for a puzzle. But before they’d need to rely on computer generated ratings which could be wildly inaccurate because computers solve problems differently than humans.
So my guess is the ones that seem misrated are new puzzles that haven’t been tried many times yet. The problem probably gets worse as you move up in rating because there are less highly rated players to try the more advanced puzzles.
In Machine learning, this is called the cold start problem.
This is a realistic scenario though it doesn't appear to cover all cases. Some ratings are simply insane. I disagree that the puzzles in this phase seem misrated. I'd say they are misrated because the proper rating is measured by the human solver as the puzzles only target humans. Engines have their own secret rating list
I'd say chess.com ought to hire some personnel to set initial puzzle ratings. This can be done quickly because no great accuracy is required to beat the cold start. Then again it shouldn't be all that hard to develop some AI software to do the same job. When engines can match imperfect human player ratings then why shouldn't they be capable of matching human solving ratings?
This happens to me when I play a few dozen puzzles in a row. The first 30 or so will be business as usual, then suddenly all hell breaks loose. I'll run into a smothered mate in 2 sequence that's easily spotted in under 10 seconds and it's a 2177 puzzle. The next one will be a 7 move sequence leading to a trapped queen that's nearly impossible to spot in under a minute, and it's a 1539 rating. Three things could be happening: One, like was mentioned, the engine sees things far more differently than you or I would. Two, the player in the games that the puzzle was extrapolated from was using an engine themselves and has completely knackered the puzzle. Three, low-rated people are using engines to solve puzzles (completely impractical, but I know it happens).
Lately did a lot of puzzles and what I noticed is that some very hard puzzles have a low rating and some easy puzzles have a high rating, could have been just me if it happens a few times, but it actually seems to be pretty common.