Right on 68 of 99 published calls.
69% direction accuracy across every World Cup 2026 call — hits and misses in the open. The three-lens model behind those numbers is explained below, alongside the calibration that tells you whether to trust them.
No cherry-picked highlights. Every settled match is below — the calls, the calibration, and the misses.
Live track record
Every settled match, published — hits and misses. 99 played.
Calibration — claimed vs actual
When the model says a team is X% likely, how often does it actually happen? A well-calibrated model tracks the diagonal; over-confident models fall short.
One match is one data point. The value is the distribution across all fixtures, not any single call. Probabilities, not promises.
How the model works
Three lenses, each with a clearly defined job.
- 01 Structural
Structural simulation — the headline
The primary output. A Monte Carlo simulation runs across squad depth, FIFA/Elo ratings, historical head-to-head records, tournament context, rest days, travel distances, and venue factors. The result is a set of home/draw/away probabilities for every fixture. This is what you see as the headline prediction.
Gaffer API — OddsMetrica's in-house engine
- 02 Live form
Live form — the modifier
A modifier, not a separate prediction. Recent results shift the structural output: a team on a sustained winning run gets a momentum nudge; a run of poor results pulls it back. Form is always labelled as a secondary lens. It can adjust, but never override, structural quality or tournament context.
- 03 Market
Market consensus — the sanity check
De-vigged bookmaker odds are the benchmark. The market aggregates an enormous amount of information efficiently. When the model and the market agree, there is no edge — and we say so explicitly.
De-vigged market consensus
The only gap worth calling an “edge”
Edge is the difference between the structural model and the de-vigged market, on team outcomes only. We show it only when that gap reaches 5 percentage points or more — below that, it’s noise. When the lenses disagree sharply we label the match high variance instead of forcing a value call. No guaranteed-profit claims here; these are probabilities.
Powered by Gaffer One in-house engine runs every prediction, live score, and settled result.
Gaffer is OddsMetrica’s proprietary prediction and data engine. It ingests and cleans football data from multiple feeds, runs the structural simulation, applies the live-form and market lenses, and exports the probabilities you see — then serves live scores during matches and settles results the moment they go final.
No third-party tipsters. No white-label prediction feeds. The model and the data layer are both ours, which is why the track record is reproducible from the same pipeline rather than back-filled to look good after the fact.
The limits
No model is perfectly calibrated. A 70% probability means the outcome fails to happen 30% of the time. Injuries, red cards, weather, and tournament-day variance aren’t fully capturable. The track record above is the honest measure of how often the model is actually right — not a marketing number. Use these probabilities as one input, not a betting strategy.
For the full prediction set, see World Cup 2026 predictions.