01In the app
Mouse-to-crosshair fit
Did the crosshair move because the hand moved it?
Quick explanation
CS2 demos record the raw mouse input next to where the view pointed. Fit asks how much of the crosshair's turning the mouse explains.
02 · Does the hand match?
How Overshoot reads an aim. Evidence you can check.
Every measurement, what it means, and what it cannot tell you. Overshoot points to the moments worth watching. People decide what they see.
One kill
From a kill to something a person can check
Every kill goes through the same five steps.
Open the window
Each kill becomes about four seconds of movement, read from the demo the server recorded.
Line up hand and crosshair
The raw mouse input is set beside where the crosshair actually turned.
Measure
Fit, jerk, corrections and more are measured for that one kill.
Compare with ordinary play
Each number is put in context against real Premier kills made with the same weapon.
A person checks
Overshoot points to the round and tick. A person opens the demo and decides.
The measurements
Nine ways Overshoot looks at an aim
Each one has its own page. Nothing here is a verdict, and no single measurement decides anything.
In the app available today Research preview built in our research tool, coming to the app
Full explanation
02In the app
Jerk
How abruptly does the aim change how it is speeding up or slowing down?
Quick explanation
Arms and wrists ramp up and ease off. Jerk measures the sharpest sudden change, compared with the flick's own top speed.
03In the app
Corrective submovements
Humans flick, overshoot or undershoot, then correct.
Quick explanation
After a big flick, a hand usually makes a tiny correction to land on target. Flicks that land dead-on with no correction stand out.
04In the app
Sequence model
A small neural network reads each window of movement as a whole.
Quick explanation
A research signal trained on labelled matches. It is shown next to the other measurements and never added into the score.
05Research preview
Peak crosshair speed
The fastest the crosshair turned during the kill.
Quick explanation
Compared only with kills made with the same kind of weapon, because a sniper flick is not a rifle tap.
06Research preview
Chance check
How many unusual kills would we expect anyway?
Quick explanation
By definition about 5 in 100 ordinary kills sit in the most extreme 5% of any measurement. The check compares what a player shows with that.
07Research preview
Ordinary-player baseline
What do ordinary Premier kills look like?
Quick explanation
Built from real Premier matches and compared within weapon class, so a number means something.
08In the app
Checkable moments
Every unusual kill comes with the round and tick, so anyone can watch it.
Quick explanation
Overshoot points to the moment. You open the demo there and decide for yourself.
09Research preview
Human verdicts
A person watches the moment and labels it. The label describes the moment, not the player.
Quick explanation
After watching, a reviewer picks one of four labels. Labels are stored without names or IDs and used to check the measurements.
Context
The bigger picture
The rules
What Overshoot will never do
- Label anyone a cheater. It shows measurements and people decide.
- Show aim scores publicly or rank suspected players.
- Let kills, K/D or rank count as evidence. Being good is not evidence.
- Show a number without the tick behind it, so anyone can watch the moment.
Keep going