From bookmaker odds to a clearer decision.

Decimal odds show the total return for each unit staked. Odds of 2.00 imply 50% before margin because 1 ÷ 2.00 = 0.50. Odds of 4.00 imply 25%.

That first calculation is the raw implied probability. It is useful, but it is not yet a clean estimate of what the market believes.

In a complete market, the raw implied probabilities usually add to more than 100%. The excess is the bookmaker margin, also called the overround. Clauseground currently removes it proportionally from a complete designated reference market.

Clauseground’s football model estimates scoring distributions using team strengths, competition scoring levels, venue context, and supported contextual inputs. A Dixon-Coles low-score adjustment helps account for correlation in common low-scoring results.

The output is labeled model probability. It is an estimate with assumptions and missing information — not an upgraded version of the market probability and not a statement of certainty.

The difference is the model probability minus the relevant no-vig market probability. A large difference can be worth investigating, but it does not automatically mean the model has found an edge. The market may know something the model does not, or either input may be stale.

Clauseground also calculates a combined estimate using fixed, market-specific weights. The current implementation leans more heavily on the reference market for major markets. Combined estimates are labeled separately from market and model probabilities.

A likely outcome is not automatically a good bet. If an outcome is estimated at 60% but offered at 1.50, the displayed price implies 66.7%; winning may be likely, but the price can still be poor. Price determines the trade-off between risk and return.

Expected value compares an estimate with the return offered at a specific price. Clauseground uses the term potential valuebecause fees, limits, freshness, uncertainty, and model error can all change the decision.

Books do not agree. The same match, priced at the same moment by different venues, carries a different margin depending on which quote you take — and the difference is larger than any model disagreement this product has been able to demonstrate.

Measured across 27,323 completed matches (11 August 2026): at the market average the books held 5.62% of every 1X2 market. Taking each outcome at whichever surveyed book priced it highest left −0.14% — 5.77 points of margin. Per individual selection, the best surveyed price sat 6.83% above the market average.

That is a property of the football market, not a record of returns. It is also an upper bound rather than a target: the best price is taken across every surveyed book, including books that limit winning accounts and quotes that were never simultaneously obtainable. Clauseground does not currently claim a proven historical edge. This is why every fixture shows the best price we collected with the venue it came from and the time it was seen. The method, the per-competition split and the limits have a page of their own.

The Kelly formula relates estimated advantage and price to bankroll size. Full Kelly is highly sensitive to estimation error, so Clauseground uses conservative fractional Kelly calculations and caps internally to order the price-comparison list. No stake size is displayed anywhere in the product. Any size you calculate yourself is mathematical context, not a personal recommendation and never a reason to exceed a pre-set limit.

Clauseground does not use one vague confidence score. Instead, it shows the evidence that affects reliability: source, update time, market coverage, model inputs, lineup status where supported, and sample size for historical metrics.

A correct prediction can still have been a poor decision at the available price. A losing outcome can still have been a reasonable decision if the price fairly compensated for the risk. Process is evaluated across many recorded decisions, not from one result.