I wonder whether accounts that have solved a fairly large number of puzzles and solve puzzles reasonably frequently — that is, accounts with a rating deviation (RD) ranging from moderate to low — and have a puzzle rating above 3050 tend to encounter puzzles that, when solved correctly, yield between 6 and 12 rating points. Meanwhile, they do not encounter puzzles rated above 3000 Elo very frequently, and when they encounter puzzles rated >3000, these accounts typically gain more than 12 rating points. Furthermore, as an account’s puzzle rating increases, do the typical rating gains it receives for solving a puzzle correctly, the number of rating points it gains for correctly solving puzzles rated above 3000 Elo, and the likelihood of encountering puzzles rated above 3000 Elo all tend to decrease?
gravereaper in particular, and other Lichess accounts with puzzle ratings above 3050 that are not suspected of cheating or manipulating their Elo ratings in general, seem to fit this hypothesis, and these accounts tend to gain relatively few Elo points when solving a puzzle correctly.
The puzzle ratings are not something I pay any attention to.