Suggestion to improve anticheating operations: the futile mistakes

I'm very glad to see that lichess (differently by other chess platforms) tried really to fight cheating in the online chess, anyway, in spite of this your effort, there are yet a lot of cheaters (imho) on lichess platform, too.
I have a suggestion to get more visible their cheating.

The smartest cheaters make intentionally mistakes (and some blunder) to get less suspicious their game's analysis prospects. In this way they will get some "mistakes" and "blunders" and also centipawn lost will raise a bit.
A good way to remove this their "front" is to introduce a new category of "mistakes" that we could call "futile mistakes" (and f-mistakes in following text).
Pratically, when a player a mistake but he, also after that mistake, is still with a good advantage to win easily (we could say +1.5 like a reasonable value), that mistake will be stored as f-mistake and not a real "mistake" or "blunder". Also updating of centipawn value should be excluded for above kind of f-mistakes.
The result of this new method could be to expose better the typical "smarter" cheater profile: some Inaccuracy, no "mistakes", no "blunders" and a lot of "f-mistakes". Also lostcentipawn value will be always very low for those guys.

This is actually how "learn from your mistakes" work. If the mistake doesn't change the outcome of the come it isn't counted as a mistake.

I see in analyse of any game all mistakes that let the player with huge advantage counted as mistakes or blunders. I don't know about counting of lost centipawns but I suppose it was counted also that. My idea was remove from these computation these not meaningful mistakes to expose better some suspect games. Now I have to perform manually this job

Thanks, that's one of many factors we consider:
github.com/clarkerubber/irwin

as announced at:
www.reddit.com/r/chess/comments/6qy6kt/how_does_lichess_detect_cheaters/?st=jealx6ak&sh=e9964e72

@lovlas
You sure about that?
How about 'not the best checkmate sequence'

Doesnt that show up a a mistake?

Or how about in anti; where you can make a mistake but still keep people in forced lines?

@breakreign lovlas is correct.

I hope when you say a mistake if it still leaves an eval of +1.5 is not counted as a mistake, does not go for the totals at the end of the game ie. inaccuracies, mistakes, blunders and it would still be counted for that. Because if you're up a piece and a pawn and drop the piece, that's an big mistake and I sure as heck want to learn from that! I hope this counting system you guys are talking about only applies to cheat detection.

Develop a neural network to spot cheaters :)

Why develop what we already have?

@Toadofsky

Because people like re-inventing the wheel.

Remember that time the Mythbusters tried square wheels? That's pretty much how everybody thinks most of the time (backwards).

In technology, re-inventing the wheel is a must. For example: Docker containers. They're just fancy chroot environments (like a BSD Jail). But its some big, new thing. And neural networks weren't invented in the 1980's during the first dry-run of AI (Lisp, Prolog, etc).

So, yes, now that @DVRazor has suggested a neural network to spot cheaters -- we should now re-invent the wheel.

;p