PISTOL v1 DEPLOYED
Pistol & Force-buy Rounds · Gradient Boosting (sequential)

Pistol rounds decide maps. AUC 0.9053.

AUC
0.9053
held-out test set

About this Model

PISTOL specialises in pistol and force-buy round prediction, trained on 500,000+ CS2 rounds from pro demos. With AUC 0.9053, it is the highest-accuracy domain model in the GESUS suite — a testament to how predictable pistol rounds are once you know each team's pistol win rate on CT and T sides, their eco conversion history, and current match state. PISTOL win probability directly predicts round 1, round 13, and force-buy outcomes.

Top Feature Drivers

CT pistol win rate (rolling) 35%

Historical CT-side pistol round conversion

T pistol win rate (rolling) 30%

Historical T-side pistol round conversion

Current score state 18%

Pressure context — 0-0 vs 14-14 pistol is very different

Equipment value differential 12%

Small differences (full pistol vs half) matter a lot

Eco chain momentum 5%

Prior round outcome creates pistol momentum

Feature importances reflect normalized split gain across the training set. Values are approximate — the model uses all features simultaneously.

How to Interpret

PISTOL probability is expressed as the CT side's win probability for a pistol round. Most teams are in the 45–55% range; anything outside 40–60% is exceptional. A team with CT pistol win rate >60% is a pistol monster — they break eco chains and dominate the anti-eco. Combine with PHANTOM's eco win rate for full economy picture.

No settled predictions recorded for this model yet.

Live Performance

No live predictions recorded yet.

Selector Role

Live sub-model — predicts pistol/force-buy rounds; feeds into SPECTER and PHANTOM.

Version History

v1 2026-Q2

Launch on 500k rounds; AUC 0.9053; clean of leakage

All Models