How it works
The model that
wins your veto
MapAssistant is powered by a purpose-built machine-learning engine trained on tens of thousands of real FACEIT CS2 matches — a dedicated model for every map, calibrated to give you an honest win probability before a single map is banned.
The lineup
A trained model for every map
Each map plays differently, so inside MapAssistant each gets its own trained meta-model — with a global model as a fallback for anything new. No generic one-size-fits-all guessing.
Under the hood
How the prediction is built
Real match data, continuously harvested
Every analysed match feeds back into the dataset. The model learns from tens of thousands of real FACEIT games and 400k+ player records — not theory, not a tiny sample.
Skill-relative feature engineering
Each player's recent form, map history and skill are weighed relative to the lobby — so a smurf in a low lobby and a struggling player in a high lobby are both read correctly.
Per-map meta-models
Ten players' features are combined into team scores, then a map-specific model turns them into a win probability that captures how that particular map actually plays.
Calibrated, honest probabilities
Outputs are calibrated confidence — a 70% means roughly 7-in-10, not marketing fluff. You see exactly how sure the model is, so you know which calls to trust.
Running on the v5 model family · model version is shown live inside MapAssistant. See the full model family →
Put the model to work
Paste your match room and get a win probability for every map before the veto starts. Free to try.
Open MapAssistant