Method and track record
How getSmartBet calculates its predictions, and how they did on every match played, with nothing filtered out.
Track record
Figures for the current month, reset on the 1st. Every prediction published before kickoff is counted, right or wrong: none is removed or corrected afterwards.
- of predictions correct842 settled matches in October 2026
- 54.3%
- right calls when the model is at least 65% confident133 matches in October 2026
- 75.9%
Accuracy by confidence
A well-calibrated model is right about as often as it says: when it gives an outcome 70%, that outcome should happen roughly seven times out of ten. Here are the announced confidence and the observed hit rate, band by band.
| Announced confidence | Matches | Average confidence | Observed hit rate |
|---|---|---|---|
| Below 45% | 516 | 40% | 41% |
| 45% to 55% | 452 | 50% | 51% |
| 55% to 65% | 254 | 60% | 65% |
| 65% and above | 227 | 74% | 80% |
Latest settled predictions
The last 12 matches played, with the prediction published before kickoff. Each row links to its match page.
| Match | Prediction | Score | Verdict |
|---|---|---|---|
| Toluca – TijuanaOct 22 | 164% | 2-2 | Missed prediction |
| Puebla – MonterreyOct 22 | 246% | 1-4 | Correct prediction |
| Atlas – AméricaOct 22 | 247% | 2-5 | Correct prediction |
| Palmeiras – FluminenseOct 22 | 166% | 0-1 | Missed prediction |
| FC Cincinnati – D.C. UnitedOct 22 | 162% | 2-1 | Correct prediction |
| Montevideo City Torque – Atlético-MGOct 22 | 239% | 2-0 | Missed prediction |
| Sporting CP – LASK LinzOct 21 | 180% | 2-3 | Missed prediction |
| Real Madrid – RB LeipzigOct 21 | 168% | 4-1 | Correct prediction |
| Real Betis – Feyenoord RotterdamOct 21 | 160% | 1-0 | Correct prediction |
| Internazionale – Shakhtar DonetskOct 21 | 182% | 1-0 | Correct prediction |
| Club Brugge – LensOct 21 | 154% | 3-0 | Correct prediction |
| Bayern Munich – ArsenalOct 21 | 158% | 2-2 | Missed prediction |
How the model works
The data
Results and match statistics (goals, shots, shots on target, corners, possession) come from ESPN's public API, over several seasons of every competition covered. The whole history is replayed in chronological order: each match is described only with what was known before it.
What the model looks at
Team strength (Elo and pi-ratings, updated after every match), recent form, goal and shot averages adjusted for the strength of the opponents faced, home and away results, head-to-head record, rest days, and the profile of the competition: home advantage, draw frequency, number of goals.
How it calculates
Three statistical models are combined: a logistic regression, a gradient boosting model, and two Poisson regressions with the Dixon-Coles correction, which give the expected goals and the probability of every score. That is where the 1/X/2 probabilities, the most likely score, “over 2.5 goals” and “both teams to score” come from. Their weighted average beats each of them on its own, because they don't get the same matches wrong.
No odds as input
The model never uses bookmaker odds, or anything derived from them. They are only used to compare it with the market, on each match page. A model fed with odds would only copy them.
Tested in time order
The model is evaluated the way it is used: each period is predicted by a version trained only on the matches before it. The final measurement covers the last months of data, never used to tune it: it is the closest estimate of how it performs in real conditions.
Locked at kickoff
Predictions are published the day before or on the morning of the match, and never changed once the match has started. That prediction is the one that gets judged, and it is shown in full on the match page once the game is over, even for a members-only match.
Its limits
Football remains unpredictable: even when the model is very confident, it is regularly wrong. It knows neither line-ups nor last-minute injuries, which the odds on the day often already reflect. Its probabilities are statistical estimates, not certainties or betting advice.
Who publishes getSmartBet
getSmartBet is published by BNT Enterprise, 6 rue d'armaillé, 75017 Paris, France. A question about the method or a figure on this page? Write to us. Full publisher details are in the legal notice.