lichess.org
Donate

Do puzzle's ratings deflate over time?

I'm speaking of the ratings of puzzles, not of the players solving them.

My thought is that players mostly improve over time. Their rating sees a net increase, which means there is a net decrease in the ratings of puzzles. Meanwhile there is no balancing force to equalize.

I don't know if the math works, I just wanted to see someone else's thoughts.
The puzzle loses rating when the player solves it and vice versa, so I think it's pretty balanced.
@Deadban If every player goes from 1500 to 2300 and quits, then the overall rating of puzzles will decrease because there is nobody going from 2300 to 1500 and then quitting. That's what I mean by no balancing force.
@Alice_ex Your theory sounds very plausible. A counter theory would be that at any time the distribution of player's ratings would be the same. There are always new players coming in, and people leaving. Even if the graph is much higher at the lower end its shape stays about the same. So then the puzzles' rating and the players' rating would reach a dynamic equilibrium.
If you think about it it makes sense because do a thought experiment. If a puzzle's rating got crazy high for some reason like a computer error, say it was 3500, well then it would not gain points when people failed it but it would lose big points when people solved it. So there's always going to be that equilibrium point where it stabilizes even in an environment where people's skill is rising.
Of course all that relies on the premise that the rating distribution is staying the same. But right now would be a good example of when that is NOT the case since we just got so many new people learning chess in a short time. So I expect there was a lot of puzzle rating inflation recently.
@Alice_ex said in #3:
> @Deadban If every player goes from 1500 to 2300 and quits, then the overall rating of puzzles will decrease because there is nobody going from 2300 to 1500 and then quitting. That's what I mean by no balancing force.

the balancing force is player rating, which is based around a median

also if what i heard about puzzle algorhythm is true low rated players failing a hard puzzle wont influence the puzzles as much as higher rated players who fail the puzzle.

even if thats not even the case, ill take your specific example here: player goes to 2300 in puzzles and quits. he was also quite low rated but solved the puzzles well so he decreased the rating of the puzzles -> puzzles are harder -> people higher rated fail them -> rating increases again -> problem solved.

thats the beauty of a median based progression with diminishing returns, you dont really have to screw around much, it will just balance itself out. and for puzzles specifically there are also several voting buttons which might trigger a review/recategorisation as an additional balancing factor, even tho thats almost overkill if your looking at a mere training tool.
@Linspiring I like your thoughts. My idea assumes that the player rating distribution is always shifting upwards. Another way to look at it is to compare it to a traditional player vs. player rating. Consider a rating system with only two players. If they both gain skill equally, their rating does not change. In contrast, in puzzle vs. player, the player is gaining skill while the puzzle remains static.
<Comment deleted by user>
@Alice_ex said in #6:
> @Linspiring I like your thoughts. My idea assumes that the player rating distribution is always shifting upwards. Another way to look at it is to compare it to a traditional player vs. player rating. Consider a rating system with only two players. If they both gain skill equally, their rating does not change. In contrast, in puzzle vs. player, the player is gaining skill while the puzzle remains static.

if the puzzle remained static there wouldnt be a rating change.

ill make another example like yours with an oversimplification, but to the point where it still works (with multiple people):
you have 2 players with the same chess rating.
player 1 fails every puzzle
player 2 solves every puzzle
the puzzle rating stays the same, but its not static.
@Rookitiki Let's say everyone suddenly stops playing puzzles. Then a single new player starts playing puzzles. They start at 1500 and slowly improve to 2300. Because they gained rating, overall the pool of puzzles has lost rating, no?
you didnt get my point. if there is only 1 person doing it, rating has become irrelevant, as there is nobody to compare to.

This topic has been archived and can no longer be replied to.