OK, ratings seem genuinely weird right now

But... hmm.

I haven't been tracking the active player stat, which is too bad.

It's important because there's an inflationary effect too. When a new player joins, loses 2 or 3 games, then quits the site forever deciding chess isn't for them, they've injected some points into the pool. Most new players quit... but each of them inflate the ratings very little. The few who stay and don't improve much have a negligible effect I'd say. And then those who stay and improve a lot of a big deflationary effect... but at the same time they're rare.

So overall is there inflation or deflation due to the new players? I don't know.

But the average is really simple... the average goes down when the players who are joining aren't as good as your current average. Simple

The average has been going down though.

If you google old topics, you'll find it used to be about 1100, 4 or 5 years ago.
At the beginning of this year it was 990.
But recently it's dropped much faster

So that might be why I'm up about 60-100 pts the last couple of days? I had put it down to variation in alcohol input

True.

Sometimes I like to check Hikaru's graph. He's been, arguably, the best blitz player since the early 2000s (before chess.com existed) so his rating graph gives a good idea of what the ceiling rating was here through time... and it's gone up.

But now that I think about it, FIDE ratings inflate. I wonder if the causes are the same and why it always seems to be inflation and not deflation.

I assume that average is counting non-active players.

I've had some basic statistics, so I know what you mean (sigma and mu to define a standard distribution) but I don't know enough about applying it to real world things... obviously the graph is bell-curve-like... but how much should it match, and what does it mean when it doesn't? I don't know.

Also the rating system is indeed not correctly predicting ratings when the players are ~400 points apart. Chessbase had some article about it years ago, and some statistician suggesting some improvements.

But it should only mirror if we randomly sample the population right? In reality, very bad players quit (otherwise I'm sure some ratings would be below zero, because some new players are truly that bad).

Hmm, ok, so in your own analysis, what are the drawbacks? Would players ratings have to be adjusted outside of games periodically to make sure there is a good fit to the distribution? That might be a reason they dislike it.