About Project Leggy

Project Leggy is a CS2 match predictor. Every night after the day's matches conclude, a model trained on historical HLTV data predicts the next day's slate of matches. Predictions are published before matches start; results and ✓/✗ marks are added the following day.

How the model works

The model is an XGBoost classifier+regressor pair trained on every CS2 match in our HLTV scrape. For each map, it considers per-team stats (round win rate, side balance, player ratings, recent form) and per-player stats (ADR, KAST, opening duels, clutch). Map-pick context (who picked, decider, permabans) is fed in too. The series outcome is then simulated 100,000 times across the predicted maps.

All-time accuracy

• Match winners: 23.5% (35 of 149)
• Map winners: 87.8% (216 of 246)

Credit

Match data is scraped from HLTV.org. This site is fan-built and not affiliated with HLTV or Valve.