kaladin VS irwin

For Lichess, you can get some sort of idea from reading the code. If I remember correctly, there's some threshold for the site suspecting you of cheating based on a variety of simple metrics (accuracy, blurring, etc) and then it gets sent to machine learning tools to analyze (here is one of them, and here's another).

There is also this repo named kaladin

Kaladin(https://github.com/lichess-org/kaladin) is the newer model.

Yeah Irwin is the old lichess model - here (https://github.com/clarkerubber/irwin) as well as the more maintained fork (https://github.com/lakinwecker/irwin/commits/master)

https://github.com/lakinwecker/irwin has more recent commits and I think is what actually runs on Lichess, or at least close to it. Not sure much fundamentally changed though. This isn't "the whole of Lichess' anticheat". It's a tool doing one specific thing which it mostly has been doing that way since it was created. There are a bunch of other parts to cheat detection, some of which are much more recent developments. For example kaladin, which has also been linked in the other thread about Lichess. But there's a fair bit more. Understandably though, Lichess doesn't really talk much about all of them, even though pretty much all of it is open source if you know where to look and how to use it.