We don't predict who wins. We estimate how likely each outcome truly is — and only bet when a bookmaker has mispriced that probability. Here's the approach in plain language.
For every match, the model estimates the true probability of each outcome. When a bookmaker's odds imply a lower probability than ours, that gap is the value (positive expected value). That's the only time we bet.
That's the whole difference from ordinary tipping: it's not about whether a favourite wins, it's about whether the price under- or over-rates them. Across hundreds of these small edges, the long-term advantage compounds.
Two market inefficiencies. First, public-money bias — popular teams and 'over' bets get over-played, which distorts soft bookmakers' prices. Second, information asymmetry — a disciplined data model beats the average odds compiler.
The sharpest market (Pinnacle) is our truth anchor. We hunt the gap between that sharp closing line and the higher odds soft books offer — and bet into it.
Under the hood: gradient-boosted trees for match outcomes, Poisson goal models for totals, Elo ratings for team strength, and a hundred-plus features (form, expected goals, injuries, rest, venue, weather…). Probabilities are calibrated against the sharp market — if the model says 60%, about 60% should land.
We don't publish the exact feature set, the weights, or the individual edges — that's the moat. What we do publish is the approach and the honest record.
Staking is fractional Kelly (25%, max 5% of bankroll per bet) — bankroll ruin is permanent, so that's non-negotiable. The most honest leading indicator is closing line value: consistently beating the closing line is long-term profitability, regardless of short-term luck.
We track every settled bet in the open — wins and losses. Past performance is not a guarantee. We sell a model, not gambling advice.
Every number — win rate, ROI, CLV, Brier and calibration across every sport — is in the open and continuously updated on the track record.
See the track record