Odds and probability
Try the calculator, understand expected value and check a model’s claims. All numerical examples are illustrative.
Try the calculation
An illustrative binary bet with no push or split settlement. The probability is your assumption, not a verified forecast.
- Break-even probability
- 74.07 %
- Expected return (EV)
- -8.20 %
- Edge, percentage points
- -6.07
01
Why accuracy needs context
Win rate alone ignores the odds. In an illustrative simulation of ten selections with a one-unit stake each, seven wins at odds 1.35 return 9.45 units. Against ten units staked, the result is −0.55 units and ROI of −5.5%, despite a 70% win rate. This is an arithmetic example, not measured Slyx performance. Read success rates with the odds, simulation rules and sample size.
02
What the odds actually tell you
In a binary outcome scenario, odds of 2.00 give a two-unit payout per unit staked on a winning selection. The break-even probability is 1 / 2.00 = 50%. This is not the measured chance of winning: a bookmaker quote also reflects margin and market conditions. One price cannot remove the margin of the whole market. A higher price changes the simulated payout, not the match probability.
03
Expected value depends on the probability assumption
For a binary win-or-lose settlement, expected value per unit is probability × decimal odds − 1. In the illustrative default above, 68% at odds 1.35 gives −8.2%. A model estimate of 54% at odds 2.00 would give +8%, but only if that probability assumption were accurate. A difference of a few percentage points can change the sign. Positive calculated EV is a claim to test, not a promise about the next match. Integer totals with a push and split Asian lines require a different payout calculation.
04
How to read the evidence behind a prediction
Check the source, capture time and exact definition of the predicted outcome. Distinguish confirmed injuries and lineups from missing information. An analysis describes the information available at publication. If you change the odds in a model scenario, its EV changes too. Missing source times or probabilities without an evaluation sample are reasons to question the evidence.
05
What it means for a model to learn
Predictions must be recorded before the event and kept unchanged when results arrive. Fit a correction using earlier outcomes, then score later matches that were not used to choose that correction. Brier score compares each probability with the actual binary outcome; lower is better on the same sample. Compare against simple baselines and check individual sports separately. One successful week and repeated adjustments to the same test set do not establish general reliability. We also retain candidates with negative EV, so research is not limited to published analyses.
06
A useful community makes reasoning visible
Bring a source, timestamp and specific question. Separate confirmed information from speculation and keep incorrect predictions in the record. Slyx has rooms by sport and for model feedback, with controls for reporting misleading or abusive posts. Discussion should help people understand uncertainty. The platform is for adults 18+; its financial examples are simulations, not personal recommendations.
Common questions
Can a high win rate still produce a loss?
Yes. Seven wins at odds of 1.35 and three losses, with one unit staked on each selection, lose 0.55 units overall. The odds and stake sizes matter alongside the win rate.
What is the difference between ROI and bankroll change?
Betting ROI divides net profit by settled stakes. Bankroll change divides net profit by the starting bankroll. The same profit can therefore produce different percentages. Slyx shows both measures in the weekly breakdown.
Is positive expected value a guarantee?
No. Expected value depends on the probability estimate being accurate. Even with an accurate estimate, a single result and a short series can lose.