How accurate are chess engine ratings?

I read this in a forum and it was censored the next day (i don't know why, they censor anything nowadays), and i don't know if he was right. Does someone agree with him??

"Engines played a lot of games against humans in official FIDE tournaments, like Hiarcs in Mercosur, for example. Not anymore, but they did a lot in the past.

Some engine developers have FIDE rating themselves, for example Rodent iv developer has 1800 FIDE rating. They can calibrate their engines vs themselves. Although I think they use a more accurate method than playing vs their own engine, but at least they should notice if they can't beat it at lower rating, or if they are winning at ratings they shouldn't. I think they test rating by playing other engines with known rating and also checking how long the engine takes to solve certain special test positions.

Also many engines played thousands of games online in ICC. For example Madchess 1.4 in ICC got 2100 rating in classical and 2400 in blitz (it's 2200 rating in CCRL list).

According to what JV Merlino said in forums, Chessmaster engine was tested vs people in chess clubs. The rest of personalities got rating playing vs the personalities that were tested vs humans. I think that was for Chessmaster 9000.

I could give plenty of examples engine vs humans. But you can find for yourselves.

By simple logic, I don't think engine developers would choose to calibrate their engines with a rating that doesn't correspond to FIDE.

That's why i think they should be well calibrated with human FIDE rating.

Besides, ELO formula should work for those engines that didn't play humans using engines that played humans and got FIDE rating. The same way it works for humans in FIDE, it must work for engines. In fact, it should be more accurate for engines, since computers don't vary strength for being tired, afraid, nervous, etc..."