MultiForecast

MultiForecast predicts the 2026 World Cup and the 2026/27 seasons in the Premier League, Serie A, Ligue 1, La Liga, and Bundesliga. It ships as two applications backed by one Python forecasting system:

• web/: the complete public Next.js product, ready for Vercel.
• app.py: the Streamlit analytics application for live model runs and deeper exploration.

Both surfaces use the same versioned JSON forecast contract.

Product coverage

• World Cup group advancement, knockout progression, title probabilities, and projected bracket.
• Domestic title, UEFA qualification, relegation, expected points, and full position distributions.
• Elo, Elo-Poisson, attack/defense Poisson, Skellam, Dixon-Coles, and Bradley-Terry/Davidson models.
• Walk-forward model comparison with log loss, Brier score, RPS, and calibration.
• Match previews with W/D/L probabilities, expected goals, likely scorelines, O/U 2.5, BTTS, clean sheets, engine comparison, and season-impact scenarios.
• Daily last-known-good data refreshes and static artifact publication.

Repository map

app.py                    Streamlit entry point
export_forecasts.py       Builds all Vercel forecast artifacts
refresh_all.py            Daily data and artifact refresh
domestic/                 Shared Big Five engine, UI, and pipeline
worldcup/                 World Cup models, simulation, UI, and export
premier_league/           Legacy compatibility package
web/                      Next.js/Vercel application
data/domestic/processed/  Canonical domestic match data
data/wc/                  World Cup inputs and fitted strengths
tests/                    Python unit and integration tests
docs/                     Architecture, data, modeling, and deployment guides

See domestic/README.md for the domestic package map and web/README.md for frontend routes.

Run locally

Python 3.11 is the deployed runtime.

python -m pip install -r requirements.txt
streamlit run app.py

Generate the shared artifacts and run the Vercel app:

python export_forecasts.py
cd web
npm install
npm run dev

The frontend is available at http://localhost:3000; Streamlit defaults to http://localhost:8501.

Refresh and verify

python -m domestic.refresh
python refresh_all.py
python -m pytest -q

cd web
npm run lint
npm run typecheck
npm run build

refresh_all.py updates current schedules and results, refreshes the World Cup inputs, reruns model selection and simulations, and writes web/public/data/manifest.json plus one artifact per competition.

Documentation

• Architecture
• Data and models
• Deployment
• Domestic engine
• World Cup engine
• Vercel frontend

Forecasts are probabilistic estimates, not betting advice.