Under the Hood: How We Build Match Probabilities

August 7, 2025

Our pipeline has four stages:

1. Input modeling

• Player ratings updated with minutes, recency and position.
• Injury & suspension status, plus likely rotation given congestion.
• Team styles: pressing intensity, directness, width, set‑piece threat.

2. Expected production

• Translate inputs into xG/xGA expectations with opponent adjustments.
• Account for rest days and travel distance (fatigue tax).

3. Simulation

• Thousands of match simulations using correlated goal models.
• Read off derived markets: 1X2, totals, both teams to score, and modal scores.

4. Calibration

• Isotonic/Platt scaling per league and per month to align predicted and realized frequencies.

We do not attempt to “out‑guess” the market news cycle. We systematize how new information flows into the model so that we’re early enough, often enough.

← Back to Blog