⚠️ This is an EDUCATIONAL EXPERIMENT about predictive models and AI. It is NOT a betting tool. We are against gambling. We accept no responsibility for any use of this material. Read the full DISCLAIMER.
Open project by NoMa Studio AI: how well does a simple statistical model based on national-team Elo ratings predict the 2026 World Cup? We built it, measured it against the real results, and document honestly what works and what doesn't.
The point of the case study isn't "beating the World Cup": it's showing, with data, where the ceiling of a model like this is, and why.
🇦🇷 ¿Preferís leerlo en español? Ver README.es.md.
Last auto-update: 2026-06-28. Backtest: the model gets the 1X2 right in 61.1% of 72 matches played (random ≈ 33%).
🏆 Projected champion (most-likely bracket): Spain. Projected final: Spain 2-1 Argentina.
Most likely finalists (Monte Carlo): Argentina vs Spain (25.9% of simulations).
| Team | Reaches final | Champion |
|---|---|---|
| Spain | 57.5% | 40.0% |
| Argentina | 45.1% | 24.0% |
| England | 26.4% | 12.4% |
| France | 18.0% | 8.1% |
| Colombia | 9.1% | 3.1% |
| Netherlands | 6.0% | 2.0% |
Full detail (our group-stage predictions vs the results, the match-by-match bracket) in predicciones/PREDICCIONES.md.
Post-mortem generated on 2026-07-18. Every match was predicted with the same model (Elo pre-tournament + momentum, host advantage only). The group figure is the backtest; the knockout figure applies the identical model to each real tie.
| Stage | Matches | We got right | Accuracy | Reference |
|---|---|---|---|---|
| Group stage (1X2) | 72 | 44 | 61.1% | random ≈ 33% · "higher Elo wins" ≈ 61% |
| Knockouts (who advances) | 30 | 24 | 80.0% | coin-flip = 50% |
| Whole tournament so far | 102 | 68 | 66.7% |
The two metrics are not the same thing. In the group stage a match has three outcomes (win / draw / loss), so 33% is the random baseline. In the knockouts there are no draws: someone always advances, so the honest baseline is a 50/50 coin-flip. The knockout number is measured as "did we call the team that advanced", running the same Elo model on each real tie.
The model called the final right. Its most-likely projected final was Spain vs Argentina, and that is exactly the real final (19/07). It also had both eventual finalists as the two most likely champions in the Monte Carlo (Spain 32.3%, Argentina 30.9%).
1X2 pick before kick-off vs the real result. Full match-by-match table (72 games) in predicciones/PREDICCIONES.md.
| Group | Right | Matches | Exact score |
|---|---|---|---|
| A | 4 | 6 | 2 |
| B | 4 | 6 | 1 |
| C | 5 | 6 | 1 |
| D | 1 | 6 | 0 |
| E | 4 | 6 | 1 |
| F | 4 | 6 | 1 |
| G | 2 | 6 | 2 |
| H | 2 | 6 | 0 |
| I | 6 | 6 | 0 |
| J | 5 | 6 | 1 |
| K | 3 | 6 | 0 |
| L | 4 | 6 | 1 |
| Total | 44 | 72 | 10 |
| Round | Right | Matches |
|---|---|---|
| Round of 32 | 13 | 16 |
| Round of 16 | 5 | 8 |
| Quarter-finals | 4 | 4 |
| Semi-finals | 2 | 2 |
| Total | 24 | 30 |
Our pick = the team the model made favourite. Pred. score = the model's most-likely decisive scoreline. ✓ = we called who advanced, ✗ = miss. (a.e.t.) extra time, (pens) penalties.
| Round | Match | Real result | Advanced | Our pick | Pred. score | Hit |
|---|---|---|---|---|---|---|
| Round of 32 | South Africa vs Canada | 0-1 | Canada | Canada | 0-2 | ✓ |
| Round of 32 | Germany vs Paraguay | 1-1 (pens 3-4) | Paraguay | Germany | 2-1 | ✗ |
| Round of 32 | Netherlands vs Morocco | 1-1 (pens 2-3) | Morocco | Netherlands | 1-0 | ✗ |
| Round of 32 | Brazil vs Japan | 2-1 | Brazil | Brazil | 2-1 | ✓ |
| Round of 32 | France vs Sweden | 3-0 | France | France | 2-0 | ✓ |
| Round of 32 | Ivory Coast vs Norway | 1-2 | Norway | Norway | 0-1 | ✓ |
| Round of 32 | Mexico vs Ecuador | 2-0 | Mexico | Mexico | 2-1 | ✓ |
| Round of 32 | England vs DR Congo | 2-1 | England | England | 2-0 | ✓ |
| Round of 32 | United States vs Bosnia and Herzegovina | 2-0 | United States | United States | 1-0 | ✓ |
| Round of 32 | Belgium vs Senegal | 3-2 (a.e.t.) | Belgium | Belgium | 2-1 | ✓ |
| Round of 32 | Portugal vs Croatia | 2-1 | Portugal | Portugal | 2-1 | ✓ |
| Round of 32 | Spain vs Austria | 3-0 | Spain | Spain | 2-0 | ✓ |
| Round of 32 | Switzerland vs Algeria | 2-0 | Switzerland | Switzerland | 1-0 | ✓ |
| Round of 32 | Argentina vs Cape Verde | 3-2 (a.e.t.) | Argentina | Argentina | 2-0 | ✓ |
| Round of 32 | Colombia vs Ghana | 1-0 | Colombia | Colombia | 2-0 | ✓ |
| Round of 32 | Australia vs Egypt | 1-1 (pens 2-4) | Egypt | Australia | 2-1 | ✗ |
| Round of 16 | Paraguay vs France | 0-1 | France | France | 0-1 | ✓ |
| Round of 16 | Canada vs Morocco | 0-3 | Morocco | Canada | 2-1 | ✗ |
| Round of 16 | Brazil vs Norway | 1-2 | Norway | Brazil | 2-1 | ✗ |
| Round of 16 | Mexico vs England | 2-3 | England | England | 1-2 | ✓ |
| Round of 16 | Portugal vs Spain | 0-1 | Spain | Spain | 0-1 | ✓ |
| Round of 16 | United States vs Belgium | 1-4 | Belgium | Belgium | 0-1 | ✓ |
| Round of 16 | Argentina vs Egypt | 3-2 (a.e.t.) | Argentina | Argentina | 2-0 | ✓ |
| Round of 16 | Switzerland vs Colombia | 0-0 (pens 4-3) | Switzerland | Colombia | 1-2 | ✗ |
| Quarter-finals | France vs Morocco | 2-0 | France | France | 2-0 | ✓ |
| Quarter-finals | Spain vs Belgium | 2-1 | Spain | Spain | 2-0 | ✓ |
| Quarter-finals | Norway vs England | 1-2 | England | England | 1-2 | ✓ |
| Quarter-finals | Argentina vs Switzerland | 3-1 | Argentina | Argentina | 1-0 | ✓ |
| Semi-finals | France vs Spain | 0-2 | Spain | Spain | 1-2 | ✓ |
| Semi-finals | England vs Argentina | 1-2 | Argentina | Argentina | 1-2 | ✓ |
Where the model missed in the knockouts (6): Germany was favoured but Paraguay advanced (Germany vs Paraguay); Netherlands was favoured but Morocco advanced (Netherlands vs Morocco); Australia was favoured but Egypt advanced (Australia vs Egypt); Canada was favoured but Morocco advanced (Canada vs Morocco); Brazil was favoured but Norway advanced (Brazil vs Norway); Colombia was favoured but Switzerland advanced (Switzerland vs Colombia).
These two matches had not been played when this post-mortem was generated (third place 2026-07-18, final 2026-07-19). No real result yet, here is only the model's forecast, to be checked afterwards. Forecast made the forward way (live eloratings.net Elo, same as the projected bracket), with a 50,000-run Monte Carlo resolving draws by penalties: "1/X/2" is the 90-minute result, "Wins the tie" includes extra time and shoot-outs.
| Match | Round | Date | Favourite | Pred. score | p(1/X/2) at 90' | Wins the tie (MC) |
|---|---|---|---|---|---|---|
| France vs England | Third place play-off | 2026-07-18 | England | 1-2 | 30/29/40 | England 57.5% |
| Spain vs Argentina | Final | 2026-07-19 | Spain | 2-1 | 43/28/27 | Spain 61.8% |
- Consistent, not clairvoyant. ~61.1% in groups and 80.0% in knockouts is roughly "the higher-Elo team wins": solid, but it lives or dies with the favourites. It cannot see an upset coming (Paraguay over Germany, Morocco over the Netherlands, Norway over Brazil), which is exactly where a pure-Elo index hits its ceiling.
- Penalties are a coin-flip the model doesn't model. Several ties it "lost" were 90-minute draws decided from the spot, and the Elo index has nothing to say about a shootout.
- It nailed the big picture. The projected bracket's final (Spain vs Argentina) is the real final, and the two finalists were the model's top-2 title favourites. The signal is strongest exactly where there is the most data (elite teams, large Elo gaps) and weakest in the close, one-off games.
Full detail in predicciones/POSTMORTEM.md.
We started from a base model (Elo → expected goals → Poisson) and evaluated it with a backtest against the matches already played. Then we tried to improve it on three fronts. Honest result:
| Improvement we tried | Did accuracy improve? | Did probability quality improve? |
|---|---|---|
| #1 Model draws (Dixon-Coles) | No | Yes (exact score and well-calibrated draw prob.) |
| #2 In-tournament form (fresh match-by-match Elo) | No | No |
| #3 Reduce overconfidence (temper / shrink) | No | No |
Key findings:
The model never predicts a draw, and that CANNOT be "fixed" by forcing it. In the backtest, draws happened when favorites slipped up (Spain 0-0 Cape Verde, England 0-0 Ghana), not in evenly-matched games. Forcing draws lowers accuracy without catching the real draws. What does help (Dixon-Coles) is making the draw probability well-calibrated (~24.5% predicted vs 25% real).
In-tournament form doesn't help. A walk-forward backtest (no data leakage) showed +0.0 pp of accuracy. Even using the "perfectly fresh" live Elo from eloratings (an optimistic, leaky upper bound) the ceiling barely rises to ~66%.
The model is already well-calibrated. Tempering confidence worsens the Brier score: the apparent overconfidence on big favorites was offset by underconfidence on close games.
Conclusion and why we leave the model as-is: the predictor is at the ceiling of what the Elo index can deliver for the group stage (~61-62% 1X2 accuracy, the same as "the higher-Elo team wins"), with already-calibrated probabilities. The only improvement that added real value was Dixon-Coles (probability quality), so that's the one we kept. Genuinely raising accuracy would need data outside the index (squad value, injuries, rest, etc.), which is another project.
📄 Backtest detail: .claude/skills/prediccion-mundial-2026/analisis-backtest.md
- Index: national-team Elo rating (World Football Elo Ratings, eloratings.net), taken from the El Atlas chart.
- Result probability: the Elo difference (plus home advantage for the hosts Mexico/Canada/USA) is mapped to an expected goal difference, which feeds a Poisson model per team → 1X2 probabilities and scorelines. With a Dixon-Coles correction for low scores.
- Recent form: a bounded adjustment from index momentum or recent results (optional). Tested: does not improve accuracy (see above).
- Tournament simulation: Monte Carlo (50,000 runs) combining real results with simulations of the remaining matches, resolving groups, best thirds and the official knockout bracket up to the final.
- Projected bracket: a single most-likely path that predicts every knockout tie (score + winner) from the Round of 32 to the final.
Backtest (pre-tournament, no leakage): uses the Elo from before the World Cup. Forward predictions: use the live Elo from eloratings (most current).
All parameters live in
model.mjs and are tunable.
Requires Node.js (no external dependencies).
cd .claude/skills/prediccion-mundial-2026
node predict.mjs "Spain""Argentina" --neutral # one match
node predict.mjs USA "Mexico" --host A # with host advantage
node backtest.mjs --md # backtest vs reality
node simular.mjs 50000 --json # simulate the tournament
node experiment-forma.mjs # form experimentThe Elo index and the predictions update themselves with:
bash scripts/actualizar.shThis pulls the live Elo from eloratings.net, refreshes everything and regenerates
the predictions and this README. A GitHub Action
(.github/workflows/actualizar.yml) runs it
every day during the World Cup and commits the changes.
New match results: add group games to
data/grupos-resultados-2026.json
(played_matches field), moving them out of remaining_fixtures; add knockout
games to
data/knockout-resultados-2026.json
(move a tie from pending to played with its score and who advanced). That's the
only manual step; the rest is automatic.
- Elo: eloratings.net (via El Atlas). 46 of 48 teams come from the Atlas extract; Scotland and Curacao are filled in with the live Elo from eloratings.
- Results and fixtures: collected and cross-checked across multiple public sources (FIFA match centre, Wikipedia, ESPN, Sky Sports, Yahoo, FOX, CBS, Al Jazeera). Nothing is invented: results not confirmed by two sources are kept as "pending" (the third-place match and the final had not been played when the post-mortem was written, so they carry only the model's forecast, not a result).
- Knockout bracket: official FIFA structure; the fine pairings after the Round of 32 are the least certain part (see notes in the data).
README.md · this case study (English)
README.es.md · Spanish version
DISCLAIMER.md · legal notice (EN/ES)
LICENSE · MIT (code) + data note
scripts/actualizar.sh · automatic update
.github/workflows/ · daily Action
predicciones/PREDICCIONES.md · full predictions (auto-generated)
predicciones/POSTMORTEM.md · tournament post-mortem (auto-generated)
.claude/skills/prediccion-mundial-2026/
├── model.mjs · the model (Elo + Poisson + Dixon-Coles)
├── predict.mjs · predict one match
├── backtest.mjs · validation against real results
├── simular.mjs · tournament Monte Carlo
├── proyeccion.mjs · deterministic bracket projection
├── experiment-forma.mjs · in-tournament form experiment
├── actualizar-elo.mjs · live Elo refresh
├── genera-predicciones.mjs · generates the public documents
├── postmortem.mjs · groups + knockouts prediction vs reality
└── data/ · Elo index, groups, results, bracket
Built by NoMa Studio AI for educational and research purposes about AI and predictive models. We do not promote gambling. The predictions carry a high, documented margin of error. Use at your own sole risk. Read the DISCLAIMER.