The Model
How we generate, evaluate and improve quantitative probabilities.
Current version: poisson_v1.0
Probability
How we estimate.
Every probability published by The Edge House is the output of a Poisson statistical model of goals per team that combines:
- ›Team goal history: goals for and against per match across recent seasons of the same league, with less weight on older seasons.
- ›Opponent's defensive strength: each team's attack is combined with how permissive the opposing defense is.
- ›Home advantage and league level: each league uses its own average goal rate, plus an advantage for the home team.
- ›Shrinkage toward the mean: when a team has few matches, its estimate is pulled toward the league average to avoid overfitting.
The result is a calibrated probability between 0 and 1 that answers the question: how likely is this event to occur?
Evidence
How we evaluate.
Every published prediction is permanently stored along with:
- ›Match (teams and date).
- ›Market evaluated (1X2, over/under goals or both teams to score).
- ›Published probability with timestamp.
- ›Model version that generated it.
- ›Actual outcome of the match once played.
- ›Performance metrics derived from the outcome.
This history grows with every matchday. It is the platform's most valuable asset: it lets us measure calibration (when we say 70%, do we hit 70% of the time?) and detect systematic model biases.
All metrics are public on the Track Record page.
Evolution
How we improve the model.
The model does not learn automatically after each match. It evolves through versions using accumulated evidence:
Each new version requires explicit human approval and rigorous backtesting against the history. No automatic decisions. No manual adjustments without statistical evidence.
The philosophy is simple: quantitative probabilities are a long-term product. They are refined with accumulated evidence, not with impulsive adjustments.
Follow the model's evolution
Sign up for free to receive every update, including version changes and performance analysis.