Project brief

PitchPrice

Honest, reproducible soccer probability diagnostics — distributions first, picks second.

Method

Modeling + validation

Dixon–Coles score models, chronological feature engineering, expanding / holdout / rolling backtests, calibration and proper scoring, bootstrap intervals, and model-versus-market comparisons using a complete no-vig bookmaker triplet.

Python Streamlit scipy / sklearn

Integrity

Synthetic demo data by design

The bundled dataset and metrics are deterministically generated for offline demos. They do not claim real-league predictive performance. The product never retrains on a UI rerun; artifacts are persisted with provenance checks.