Independently reviewed educational arithmetic, not a current recommendation or measured model advantage. All prices and probabilities are hypothetical; expected returns depend on the assumed probability and exclude costs.
By Edge93. Prepared with AI assistance; mathematical examples and publication copy reviewed before release.
A football screen says "5% edge." Before deciding what that means, ask: five percent of what? A difference between probabilities and an expected return at an offered price are different calculations. Neither number, on its own, establishes a profitable bet.
This is an educational guide with hypothetical numbers, not a current match recommendation or a report of Edge93 model performance. At publication on 27 September 2026, Prematch, Live and Insights remain paused. The public website supplies no current match prices or independent Live forecasts.
Start with three separate quantities
- Model probability: an independent estimate of an outcome, with a known model, inputs and decision time. It can be wrong.
- Bookmaker probability: a probability derived from prices. Raw implied probability is 1 divided by decimal odds; it still includes the effect of pricing margin.
- Offered odds: the price actually available for that selection at the decision time. Decimal odds include the returned stake when the bet wins.
For a complete Home/Draw/Away market from one bookmaker, one simple margin-adjustment method divides each raw implied probability by their total. That makes the triplet sum to one. It is a proportional adjustment, not proof of the true probabilities or an independent model. Do not combine the best Home, Draw and Away prices from different bookmakers and label the result one coherent bookmaker market.
Probability gap measures disagreement
Suppose a model estimates a Home win at 60%, while a margin-adjusted bookmaker estimate is 55%. The probability gap is 60% minus 55%: +5 percentage points. That is not a 5% expected return. It is also not a statement that the model is more accurate than the market.
Expected return also needs the offered price
For an ordinary win-or-lose bet with one unit staked, estimated net expected return is p × decimal odds − 1. Here p is the probability as a fraction. This follows from p × (odds − 1) + (1 − p) × (−1). The calculation below excludes commission, taxes, voids and other settlement rules.
| Assumed model probability | Hypothetical offered odds | Estimated net expected return |
|---|---|---|
| 60% | 1.60 | 0.60 × 1.60 − 1 = −4% |
| 60% | 1.75 | 0.60 × 1.75 − 1 = +5% |
| 60% | 1.90 | 0.60 × 1.90 − 1 = +14% |
These are separate price scenarios, not three simultaneous quotes or a single coherent market. The estimated chance stays the same; changing the price changes the estimated return. A 60% favourite can therefore have a negative estimated return. Expected return is not realised ROI: one losing bet still loses its stake.
A small change in the estimate can reverse the conclusion
At odds of 1.75, the break-even probability before the excluded costs is 1 / 1.75, approximately 57.14%. If the true chance were 55% instead of the assumed 60%, expected return would be 0.55 × 1.75 − 1 = −3.75%, not +5%. The price calculation cannot validate the probability fed into it.
Calibration asks whether events assigned similar probabilities occur at roughly those rates across a suitable evaluation sample. It does not certify any individual match. Missing inputs, a different competition or an unfamiliar match state should qualify how an estimate is interpreted. A precise-looking percentage is not a measure of certainty.
Why there is no universal 5% cutoff
Reporting bands such as 0–2%, 2–5% and above 5% can organize a comparison. First declare whether they describe probability gaps in percentage points or expected returns in percent. Keep negative and zero values visible too. A band boundary is not evidence that every selection above it is good or every selection below it is useless.
A larger apparent gap can come from a better estimate, a stale price or a mistaken input. A smaller gap can still be informative, but there is no automatic right-to-bet implied by either. In particular, there is no abrupt improvement in evidence between otherwise similar readings of 4.9 and 5.1 percentage points.
What to check before interpreting an edge
- Which probability is independent, and which comes from bookmaker odds?
- Does the complete market come from the same bookmaker and coherent observation?
- Was the offered price available by the decision time, for the same score and market rules?
- Are inputs missing or delayed, and has probability quality been evaluated separately from betting results?
- Does the displayed number mean percentage points, expected return or realised ROI?
Useful match analysis does not have to promise profit: score, time, supported statistics and changes in play can clarify the situation while uncertainty remains visible. A betting signal is decision support, not a guarantee. You can use the analysis without placing a bet. If you choose to gamble, never stake money you cannot afford to lose. 18+.
Probability gap describes disagreement. Estimated expected return combines an assumed probability with an offered price. Realised ROI describes what actually happened. Keep all three separate, and never let a 5% label substitute for checking the evidence.