The puzzle ratings on chess.com don't correspond in any way to USCF ratings. I complained about this to chess.com, but they responded that their puzzle ratings are where they want to them to be. (BTW, the upper limit on puzzle ratings is ridiculously high at around 32768. Some people have actually reached this limit. For computer nerds like me, this matches the upper limit on a 16-bit signed number. This tells me that they are using 16 bits to store ratings in their database.)
I wanted to see if I could push my Chess.com puzzle rating up to 3000. I've been there before, but it is a hard rating to maintain.
My puzzle rating averaged around 2935. At this level, I am almost as likely to fail to solve a puzzle as I am to succeed.
I had a theory that if I did enough puzzles I could reach 3000 through a "random walk". The idea was that if I bounced up and down enough I would eventually hit 3000 through random variation. This wouldn't mean that I deserve to be at 3000, but got lucky.
It appears to me that chess.com will present puzzles with a sizeable range of difficulty. This is where luck plays a factor. However, every time my rating would creep up, I would face problems that seemed too difficult. This definitely took me out of my comfort zone.
It took me about 2.5 hours to reach 3000. However, to get there I had to analyze at a deeper level than I am used to. Whereas the simple chess problems on my website are designed to build pattern recognition, it seems to me that the puzzles on chess.com are more of a measure of how well a person can analyze. However, pattern recognition is still a factor.
