Predict26

World Cup 2026 match estimator and tournament simulator using ELO ratings.

Live: https://sail.zo.space/predict26 Repo layout + deploy workflow: see AGENTS.md.

Features

• Single-match estimation — W/D/L probabilities + expected goals (Poisson) for any two teams
• Manual ELO overrides — scenario testing with custom ratings
• Group-stage simulation — full standings with FIFA 2026 tie-break rules (9 criteria)
• Third-place ranking — coefficient formula from the FIFA reference, best 8 of 12 advance
• Knockout bracket — projected bracket with real team names: confirmed results shown in bold, ELO-predicted slots in italics; exact slot mapping from the official AssignThird table (495 entries)
• Full tournament Monte Carlo — 10,000 simulations by default, adjustable
• Streamlit UI — browser-accessible, upload your own CSV/JSON, charts and tables
• Locked real results — matches 1–20 use real scores; remaining matches are simulated

Tournament format

Stage	Matches
Group stage (12 groups × 4 teams, 3 rounds)	72
Round of 32 (top 2 per group + best 8 thirds)	16
Round of 16	8
Quarter-finals	4
Semi-finals	2
Third place + Final	2
Total	104

Model

• ELO → win probability: P = 1 / (1 + 10^(-ΔR/400))
• Draw model: peaks at ~30% for even match, derived from ELO difference
• Goals: Poisson with λ = 1.2 × exp(ΔR/1000) per team
• Home advantage: +100 ELO for USA, Canada, Mexico in group stage
• Penalties: 50/50 coin-flip

Setup

pip install -r requirements.txt

Run

streamlit run app.py

Run tests

python -m pytest tests/ -v

Project structure

WC26-engine/
  data/
    teams.csv           # 48 teams, ELO, group, host flag
    schedule.json       # 104 matches with dates, venues, scores (played=true for known results)
    assign_third.json   # 495-entry bracket mapping table from FIFA reference
  src/
    data_loader.py      # load/validate teams and schedule
    elo.py              # ELO probability + Poisson simulation engine
    rules.py            # standings, tie-breaks, third-place ranking, bracket resolution
    simulator.py        # Monte Carlo tournament simulator
  tests/
    test_elo.py
    test_rules.py
    test_simulator.py
  app.py                # Streamlit UI
  requirements.txt
  README.md

Data source

ELO ratings from World Football Elo Ratings as of June 18, 2026. Group assignments and bracket mapping from the WCup_2026_4.2.9 Excel reference model.

Architecture note

The project is split into four pure modules:

1. data_loader — I/O only, no business logic
2. elo — probability math and match simulation, no tournament state
3. rules — FIFA rules and bracket logic, no simulation state
4. simulator — orchestrates the above to run a full tournament

app.py only calls these modules — it contains no logic of its own.

License

GNU General Public License v3.0 — see the LICENSE file for the full text. Copyright © 2026 Alberto Jiménez Bákit.

Support

If you find this useful, you can buy me a coffee at ko-fi.com/kekojb.