What the measurements can and cannot tell you, and the rules Overshoot follows so they are used fairly.
By definition about 5 in 100 ordinary kills sit in the most extreme 5% of any measurement. Every player has some unusual kills. What matters is how many, across how many matches, and whether a person who watches agrees.
A score is ranked within one lobby, never against a global threshold.
Kills, K/D, headshot percentage and rank never feed a score. Being good is not evidence.
Subtle assistance and newer methods may look different from what the measurements and the model were built on. External AI aimbots are not covered by the current benchmark.
The public dataset the sequence model was trained on uses Valve bans as labels, which skew toward obvious cases.
Lobby skill, weapon, settings and hardware all change what is typical. The reference is rebuilt as it grows, and a check for how often ordinary players look unusual comes before any comparison becomes a rule.
Overshoot never shows aim scores publicly. You can ask to be removed from rankings and player pages, and the privacy page explains what is collected and why.