And also... who cares? there are no stake, just playing against random people, who cares if they cheat as long as you have fun playing. It's less fun for the cheater, but doesn't impact non-cheater.
I would be upset if someone was inconsistently cheating, and so I'm playing someone nominally at 1400 but for this game they're actually playing 2100.
Of course for true tournament play, the rules are the rules, whatever they may be, and they must be followed for it to have any meaning.
They have almost the entirety of the lichess system source code publicly available, but from what I can gather they have a private set of weights on inputs - I assume to prevent cheaters from fully reverse-engineering the system.
How true is that at the top levels of play?
What do you think the success rate is for online players telling an engine from a human? 1% 99%
In middling ratings, if I was allowed to look at the engine (in a Lichess replay for instance) I'd guess I can tell a player using the engine with 99+% accuracy. People _always_ lose on tactics (or misplayed opening lines, misplayed endgames, etc). Engines never do. At high ratings I assume this goes down, especially if the players are playing 'boring' draw-ish lines, in which the best moves are fairly obvious most of the time.