One match can no longer put a verdict on a player
A player’s verdict now comes from how many of their analysed matches tripped a review signal, compared with a typical analysed player. We need 10 matches with at least 12 kills before we say anything, and the aim model’s trigger is stricter.
Until Sep 5, a player’s profile showed the worst single match we had analysed for them. Play enough games and one of them will look odd, so the more someone played, the likelier they were to read SUSPECT. Four regular players we had checked and knew to be clean scored between 82 and 97 out of 100 that way. Only a manual allowlist kept those numbers off their profiles.
The per-match flags weren’t much better. 12.9% of analysed player-matches with 12 or more kills carried one, and when we checked them against Steam bans they told us almost nothing: 1.7% of flagged players had a ban, against 1.2% of everyone we had analysed.
How a verdict works now
- We take your distinct analysed matches where you got at least 12 kills. The same match uploaded twice counts once.
- We count how many of them tripped a review signal from the wall model or the aim model.
- We compare that count with a typical analysed player, who trips up to about 6 matches in 100.
- Under 10 matches the profile says NOT ENOUGH DATA, whatever the count.
- From 10 matches up, the count decides between LEGIT, WATCH and SUSPECT. At 10 matches WATCH takes 4 signals and SUSPECT takes 6. At 20 matches it’s 7 and 9.
The aim model got stricter at the same time. It used to flag a match at a score of 70. It now needs 88, the first cutoff where it fires on our players about as rarely as its own test results promised: around 3% of clean players.
What you see on the site
Player pages show the verdict word, the count (for example 2 of 47) and one sentence that spells out what it means. Report pages show that match’s signal in a neutral colour, next to the profile verdict. A flag on a single match stays a review signal. It never turns into a red stamp, and it never ranks anyone on a leaderboard.
Rule v2 removes accusations our data couldn’t back up. It doesn’t make detection any sharper. When we re-scored all 4,056 analysed player-matches on Sep 5, no profile reached WATCH or SUSPECT under the new rule, because nobody had enough rated matches with enough signals. Sharper detection needs labelled kills from real matches, which is what Community Overwatch collects.
How to use it
- STEP 1Open a player page. The ring shows how many rated matches tripped a signal, out of how many we rated.
- STEP 2Read the sentence under it. It tells you how many signals a typical player would have after that many matches.
- STEP 3On a report page, the signal from that one match is shown apart from the profile verdict.
- STEP 4If you think a report about you is wrong, use the appeal link on it. Appeals go to our moderation queue.
Known limits
- Most profiles say NOT ENOUGH DATA. Few players have 10 rated matches with us yet.
- Some older analyses can’t be re-scored by the new aim model, because we didn’t keep the numbers it needs. The sentence lists them as matches we could not re-check. Their aim flags don’t count as signals; wall flags still do.
- A verdict says the data is unusual enough to look at. It doesn’t prove anyone cheats, and LEGIT doesn’t prove anyone is clean.