How it works
From bookmaker prices to the most likely bets
- Step 1
Collect prices
Match-result and over/under prices from 20+ European bookmakers, several times a day, through a licensed odds feed.
- Step 2
Remove the margin
Each bookmaker's built-in margin is stripped out, so their prices become fair probabilities.
- Step 3
Build a consensus
All bookmakers are combined into one probability, with the sharpest books weighted most.
- Step 4
Model every market
A model per sport spreads that consensus across the popular markets: BTTS, handicaps, sets, totals…
- Step 5
Rank the bets
The most probable bets are highlighted, and prices above fair odds become value signals.
Worked example
Why bookmaker odds are not probabilities
Turn odds into percentages (1 ÷ odds) and they add up to more than 100%. The extra — here 4.2% — is the bookmaker's margin. We remove it with Shin's method, which also corrects the habit of shortening long shots more than favourites.
We do this for every bookmaker, then average them in log-odds space. Exchanges and low-margin bookmakers, whose prices track results most closely, count two to three times as much. Prices older than 6 hours and books with margins above 20% are ignored.
Fair odds for Home: 1.96. A bookmaker offering more than that is a potential value price.
Markets by sport
The bets people actually place, in every sport
Bookmakers price the match result and the main over/under line in our feed. A model per sport is fitted to those prices and spreads the same view over the other popular markets. Those show the model's chance and fair odds; they are marked “model” and never become value signals.
Soccer
Poisson goals for each team, fitted to the 1X2 and over/under prices.
- 1X2, double chance, draw no bet
- Total goals 0.5–4.5
- Both teams to score
- Handicaps ±1.5, ±2.5
- Team goals
- Result & goals combos
- 1st half result & goals
- Correct score
Tennis
A constant chance of winning each set, solved from the match-winner price.
- Match winner
- 1st set winner
- Set betting (2–0, 2–1…)
- Set handicap ±1.5
- Total sets
- Total games (bookmaker priced)
Basketball
Normal distributions for the winning margin and total points.
- Moneyline (incl. overtime)
- Point spread ladder
- Total points ladder
- Team totals
- Winning margin bands
Ice hockey
Poisson goals in regulation, adjusted for hockey's frequent ties, plus overtime.
- 60-minute result (3-way)
- Moneyline (incl. overtime)
- Double chance
- Puck line ±1.5
- Total goals
- Team goals
- Overtime yes/no
Tennis tournaments (ATP and WTA) are added automatically while they run. The models know nothing beyond prices — no team news, injuries, weather or surface — and simplify reality, so treat model-only markets as rough guides.
Two ways to read a match
Most likely bets and value signals are different things
Most likely bets
For every match, the most probable selections across the sport's popular markets — one per bet type, between 50% and 90% (shorter odds say little).
Answers: what is most likely to happen?
Watch out: likely bets pay little and still lose regularly — a 78% bet loses about one time in five.
Value signals
When a bookmaker's price is higher than the fair odds, the price is better than the market thinks it should be. That is where any long-term edge would come from.
Answers: where is the price too generous?
Watch out: a value bet on an outsider still loses most of the time. Value is about price, not likelihood.
Value signals
Strict rules, frozen forever
Signals are published automatically when every rule holds — nobody picks them by hand. Once published, the price, probability and time are locked in the database: they can't be edited or deleted, so the track record shows every loss.
Confidence ≠ chance of winning
“High”, “medium” or “low” confidence describes how much the bookmakers agree and how much data there is. A high-confidence signal on an outsider will still lose most of the time.
- At least 5 bookmakers price the match
- The best price is 2% to 15% above fair odds — bigger gaps are almost always stale or wrong prices
- Prices more than 12% above the market median are ignored as errors
- The best price comes from a bookmaker licensed in Europe (offshore books count towards the consensus only)
- Exchange prices are reduced by the exchange's commission first
- Odds between 1.25 and 8, model probability at least 12%
- Kick-off less than 48 hours away
- At most one signal per market per match: match result and main over/under line
The role of AI
Probabilities, most likely bets and signals come only from the statistical method above. For Pro members, an AI model (Claude, by Anthropic) writes a short analysis of each match from those numbers — once, the first time the match is opened, and then kept. It is told not to add facts such as team news, promise outcomes or encourage betting, and its text is checked against banned phrases. AI text is always labelled AI analysis.
Settlement
Signals settle on the final result from the data provider. In two-way markets a draw is void; a whole-line over/under is void when the total lands on the line. Ice hockey 60-minute bets are void when a one-goal final margin could hide overtime. Matches without a result seven days after kick-off are void. The track record assumes a 1-unit stake at the published price and shows closing-line value — the published price against the consensus just before kick-off.
Limits — please read
- The consensus is itself an estimate. Markets are efficient, so real edges are small and easily outweighed by noise.
- The models know nothing beyond prices: no team news, injuries, weather or motivation.
- Prices move. The price shown may no longer be available when you look.
- Many bookmakers limit or close accounts that consistently take value prices.
- Hundreds of results are needed to tell skill from luck. Short-term profit or loss says very little.
- Most likely does not mean safe. Every bet can lose; only bet what you can afford to lose.
See it on today's matches
Model version PROBABLE_CONSENSUS_V2