The solution to anti-cheat is to cancel matches without necessarily banning players

Idea in image form

The problem Valve have with anti cheat is they want to be wrong almost never. I think Valve should cancel matches far more often without necessarily banning the suspicious players.

I say this because simply looking at CS Stats, Leetify, or seeing 5 scout wallbang headshots in a round is enough evidence to be right about whether someone is cheating very often. But not often enough to ban them per se. What if I ran into the next Zywoo or someone had a very good day?

When players are false banned it ruins their steam account, leads to bad publicity and disgruntled customers. Valve have to decide how sure their anti-cheat must be to accept this risk. Let's say Valve want to be 99.99% sure a player is cheating before they ban them.

If Valve banned players they were 99% sure are cheating then 1 in every 100 banned players would be legitimate players. That's hundreds of players per week. Lots of players who have spent money, time and effort on the game and are very good at the game. This is an unacceptable risk, obviously.

However there is a middle ground.

Extreme luck or standard deviation from expected outcomes should lead to cancelled matches.

This would require finding a global average or expected average for things like: kill death ratio, average damage per round, bombs defused, clutches won, wallbang headshots, accuracy, time to damage, proper counter strafing, average score over last 1:10:30 games, first kill in round.

Then a standard deviation. E.g. 98% of all players have a KD between 0.5 and 1.5. Anyone at 1.6 consistently in premier has one red flag. 95% of all players achieve <3 wallbang / through smoke kills per game, anyone >3 has another red flag.

Do this across many variables. Calculate a cumulative overall "Z score" for the player. E.g. out of 1 million legitimate players, 999,000 players have a Z score <7. Any player who achieves a Z score over 8.5 in a game, several games, is given a cooldown. They might just be very lucky in many different ways, they might be the 1 in a 1000 player, but they probably won't be the 1 in a 1000 player very often so a cooldown and match cancel here and there is an acceptable risk. If they are so good they pull this off frequently, their gameplay could be sent to a human to categorise this player in either "not cheating" where their Z threshold is made higher (pros could be given this obviously) or "likely cheating."

Equally this would help Valve insulate very probably legitimate players from cheaters. If your Z score is always low, you can be placed into a "very likely not cheating" category of matchmaking.