How the scanner works

Most betting products claim to predict football. We tested that road properly — sophisticated statistical models, expected-goals data, seasons of walk-forward backtesting — and the conclusion matched the academic literature: public-data prediction models do not reliably beat the market's closing price. So the scanner does something more defensible. It never forms an opinion about any match. It continuously compares two sets of numbers: the sharpest price in the market, and every other bookmaker's price on the same outcome. Sharp price = best available estimate of the truth. Soft price meaningfully better than truth = flag. That's the entire philosophy — the edge comes from bookmakers disagreeing with the sharp market, not from us out-guessing anyone.

For each match the scanner takes the price from the sharp side of the market — Pinnacle, the professionals' bookmaker, with the exchanges as reference. Why trust it? Structure, not reputation: Pinnacle runs tiny margins and, uniquely, welcomes winners rather than banning them. Every sharp bettor on earth is therefore incentivised to punish any error in their line, which means errors get corrected in minutes. Decades of research keep finding the same thing — the sharp closing price is the most accurate public probability estimate that exists for football. The scanner treats that price as the benchmark. Not because it's perfect — because it's the best available, and crucially, better than any soft bookmaker's lagging line.

Even Pinnacle's prices contain a small margin, so the raw price slightly overstates every outcome's probability. The scanner removes it: convert each outcome's odds to implied probability (1 ÷ odds), sum them (you'll get ~102% at a sharp book), then rescale so they total exactly 100%. Example: sharp 1X2 prices of 1.56 / 4.21 / 5.58 imply 64.1% / 23.8% / 17.9% — summing to 105.8%. Rescaled: 60.6% / 22.5% / 16.9%. Those are the true odds — the fairest available estimate of the match, margin-free. This exact de-vig is the same maths used in our 44,051-bet historical validation, so what the scanner flags live is what the evidence was built on — not a lookalike.

Now the comparison: the scanner sweeps 35+ bookmakers' prices on the same outcomes and computes each one's expected value against the true probability: EV = (true probability × odds) − 1. True chance 60.6% and a soft book still up at 1.72? EV = (0.606 × 1.72) − 1 = +4.2% — flagged, with the bookmaker named and the EV shown. Everything at +2% and above makes the board, because that's the threshold the historical record supports: in validation, the +2–5% band alone returned +3.8% ROI over thousands of bets, with returns stepping up as flagged EV rose. Why do these gaps exist at all? Soft books lag the sharp line on purpose and by nature — promotional pricing, slow trading desks, recreational-market priorities. The scanner just catches the lag, continuously.

A naive scanner would be dangerous, because the most spectacular-looking 'edges' are usually broken data. So flags pass filters before you see them: • Stale/suspended-line guard — an EV above 15% against Pinnacle is almost never real value; it's a price a book failed to take down after team news, or a suspended market still echoing through a feed. Filtered. • Pre-match only — in-play prices move too fast for feed-latency comparison to be trustworthy. Any started match is dropped. • Arb sanity cap — 'guaranteed profits' above 10% are treated as dead lines, not opportunities. This is the unglamorous majority of the engineering, and it's exactly what free odds-comparison surfaces don't do. Showing you fewer, realer numbers is the product.

Sometimes the disagreement isn't book-vs-sharp but book-vs-book: the best available price on EVERY outcome of a match, combined, implies less than 100% total probability. Back all outcomes in the right proportions and the maths locks the same small profit whatever happens. The scanner detects these automatically and does the awkward part for you — the exact stake split per outcome, per bookmaker, and the locked percentage. Real-world notes: genuine arbs are typically 0.5–3%, they close within minutes as prices correct, execution has to be quick, and bookmakers limit accounts that arb them repeatedly. They're best treated as an occasional bonus the scanner hands you, not the main course. (The main course is volume +EV betting at sane stakes.)

Every flagged bet ships with a suggested stake — the Kelly criterion, computed from the edge and the odds: stake fraction = edge ÷ (odds − 1), capped at 25% of full Kelly. Why pre-capped: full Kelly assumes your edge estimate is exact. It never is — so full Kelly systematically over-stakes, and over-staking is how people with genuine edges still lose bankrolls. Fractional Kelly keeps virtually all the long-run growth while cutting drawdowns dramatically; it's what professionals actually use, and the scanner bakes it in so the number on screen is one you can stake directly. Bigger edge → bigger suggested stake; longer odds → smaller suggested stake at the same edge. The full reasoning lives in the bankroll guide.

It is: a live comparison engine that finds prices mathematically out of line with the sharp market, filters the illusions, and sizes the stakes — the tool for exactly one job: never taking a bad price again. It is not: a tipster or a prediction machine. The scanner doesn't know who'll win tonight and doesn't pretend to. Two real-world limits, stated plainly: soft bookmakers limit accounts that beat them persistently (breadth of accounts is your counter), and edges must keep earning their keep — markets sharpen, and we keep re-validating against the data rather than resting on one good backtest. The maths puts the percentages on your side. What you back, and at what stake, stays your call — as it should.