Flashbangs win rounds. UTILITY measures who controls them.
About this Model
UTILITY predicts round-level win probability from utility usage data extracted from CS2 demos: flash assists, smoke screens deployed, grenade damage, and molotov/incendiary usage. AUC 0.649 on held-out rounds. The v1→v2 improvement (+0.12 AUC) came from a single critical fix: using team_elo_history.elo_before (per-match Elo, no leakage) instead of the static current Elo snapshot. UTILITY produces a round-win-lift estimate — how many additional rounds does superior utility win per map?
Top Feature Drivers
Baseline quality; fixed v1 leakage with elo_before
Flash-facilitated kills gap — key utility metric
HE + molotov chip damage edge
Screens per round — site control proxy
Damage + assists per dollar of utility spent
Feature importances reflect normalized split gain across the training set. Values are approximate — the model uses all features simultaneously.
How to Interpret
UTILITY team rating is a 0–100 percentile score relative to all teams in the DB. A score of 75 means this team uses utility better than 75% of CS2 pro teams. The round-win-lift estimate shows the expected additional rounds won per map purely from utility advantage — typically 0.5–2.5 rounds per map for significant edges.
Live Performance
Selector Role
Sub-model — utility edge feeds into TITAN ensemble.
Version History
AUC 0.529 — leaked static Elo as feature
Fixed to team_elo_history.elo_before; AUC 0.649 (+0.12)