Methodology
By David Adams
Honest documentation of what we do, how we do it, and where the limits are. The date of the last substantive update is at the bottom of the page.
Probability sources
Fixture probabilities come from the mapped prediction feed where the provider marks a fixture as predictable. SportSignals connects those probabilities to the right teams, matches and market outcomes. Player-prop probabilities are separate, versioned predictions. Every published prop names the model version and generation time used for that player, market and line.
Book-implied probability
We turn each price into a probability by taking its reciprocal, so a price of 5.25 implies 19.0%. That figure still contains the bookmaker's margin, and we compare the model probability against it exactly as displayed. We do not strip the margin out first, because the implied percentage printed beside the price is the number the comparison has to agree with. Stripping it would lower the reference probability and widen every published edge, so measuring against the displayed price is the more conservative of the two readings. The price stays visible so you can check the comparison for yourself.
Edge percentage
Public edge is the percentage-point difference between the displayed model probability and the raw implied probability of the displayed bookmaker price. A model probability of 67.6% against a 61.7% implied probability is a 5.9-point edge. Rows that do not satisfy the publication contract are withheld rather than rounded into a persuasive-looking claim.
What Sharp Lean is, and is not
Sharp Lean is a 0 to 100 research score derived from observed price action across the eligible bookmakers available for that market. The actual comparison count varies by fixture and is shown with the evidence. It is not a licensed bets-vs-money feed. We do not have access to public bet share data. The score is a price-action signal, and we say so.
Public publication contract
- The fixture must be scheduled or live and must not be in the past.
- Public player props are limited to the next 72 hours.
- The exact displayed price observation must be no more than 10 minutes old.
- The model run must be no more than 24 hours old.
- Market, outcome, line, player and fixture participants must agree across the prediction and price records.
- One eligible bookmaker is enough to publish. Where a single bookmaker prices the exact selection and line, the card names that bookmaker and calls the figure an available price rather than a best price. Zero eligible bookmakers withholds the row.
- A player prop requires an index-ready player, a full-confidence signal and at least five verified relevant matches.
Provisional or conflicted rows may remain useful inside research tools, but they do not appear on public team, player, article or prop pages.
Freshness and corrections
Reference facts and volatile markets have separate clocks. Odds movements never change a team or player profile's reference modification date. Each market card exposes its price observation, model generation time, evidence sample and public validity window. Identity or source corrections are applied to the relevant team, player and match records, after which affected public context is regenerated.
Methodology updated .
Recommended price against the closing market
Shortly before kickoff we store one closing row per market, outcome, line, period and player. That row holds the best price visible across the eligible bookmakers at that moment, and it records which bookmaker offered it as a label only. Comparing a price we published against that row therefore measures it against the market's best close, not against the same bookmaker's own close.
That is not closing line value, and we do not report it as one. A same-bookmaker comparison would need the bookmaker in the stored closing key, which it deliberately is not. Read the figure as a directional check on our pricing rather than as evidence of an edge.
What we do not have
Licensed bets-vs-money splits. Same-bookmaker closing prices. Insider injury news. Crystal balls. We are honest about the gaps. Where the market is more accurate than the model, the public surface publishes no qualified positive-edge opportunity rather than manufacturing one.