About the model
A plain-English description of how Oddrail estimates probabilities, what the classifications mean, and where the model's limits are.
Score matrix
For every match the model estimates how many goals each team is likely to score. From those goal expectations it builds a score matrix: a table of probabilities for each possible scoreline (0–0, 1–0, 0–1, and so on).
Market probabilities such as home win, over 2.5 goals or both teams to score are then read directly off this matrix by summing the relevant scorelines.
Dixon-Coles adjustment
Simple goal models treat the two teams' goal counts as independent, which is known to misprice low-scoring results such as 0–0 and 1–1.
The model applies the Dixon-Coles correction, a small dependency adjustment for those low-score outcomes, and weights recent matches more heavily than older ones so that current form matters more than distant history.
Removing the bookmaker margin
Quoted odds contain a margin (overround), so the raw implied probabilities of all outcomes add up to more than 100%.
Before comparing the model with the market, the margin is removed with a normalisation method to obtain fair market-implied probabilities. The method in use is shown on the Settings page.
Edge classification
The edge is the difference between the model probability and the market-implied probability, shown in percentage points.
When the edge exceeds configured thresholds, a selection is labelled 'Potential value'; when it is clearly negative, 'Low value / inflated'; otherwise 'At market level'. These labels are statistical observations about disagreement between model and market — they are not betting advice.
Confidence and insufficient data
Each market carries a confidence band (low, medium, high) reflecting how much reliable data supports the estimate — sample size, data recency, and league coverage all matter.
When confidence falls below a minimum, or inputs are missing, the market is reported as 'Insufficient data' and no classification is shown. When incoming data is outdated, classification is paused rather than shown on stale inputs.
Limitations
All probabilities are estimates with uncertainty. The model does not know about late team news, referee decisions, weather, motivation, or anything outside its input data.
Past data does not predict outcomes. Even a well-calibrated 70% probability loses 3 times out of 10. Nothing on this site should be read as betting, financial, or investment advice.