Sequence model
The measurements above look at single ideas. The sequence model reads a whole window of movement at once and gives a separate research signal.
What does the whole window of movement look like to a model trained on labelled matches?
What it is
Each kill becomes a window of about four seconds. For every moment in the window the model sees how the view turned, how fast, how far off target, the mouse input and whether the player was firing. A small neural network reads the window as a whole and produces a signal.
Why it is separate
It finds more than the single measurements do, with more false alarms. So it is shown next to them and is never added into the score. Its accuracy and limits are on the How it works page.
What it was trained on
A public research dataset of labelled matches. Labels come from game bans, which lean toward obvious cases, so the model says little about subtler or newer methods.
What this cannot tell you
- It is a research signal, not a verdict.
- It was trained on a public dataset, not on your lobbies.
- Newer methods may look different from the cases it was trained on.