Track Record
Every pick from the EdgeLab 4.1 model, locked in before the match at the Bet365 odds of the moment. Results published as they come, nothing is ever rewritten.
If the market unanimously turns one of our picks around before the match (say 1.53 to 2.20 everywhere within a few hours, as with Martineau at the Bonn Challenger), the signal is automatically voided and stays displayed as such. A move like that almost always reflects information the stats cannot see yet: injury, poor fitness, withdrawal. Rather than let you follow a polluted bet, we void it before the match, in the open. Nothing is deleted, everything stays visible.
August 2026-0.39u
July 2026+11.69u
BACKTEST · SIMULATIONThe model, validated over ten years (2016-2026)
The current model replayed over 55,317 priced ATP matches (real historical odds), under the exact rules of the site: official scope (strong + moderate · odds < 2.00) and stakes 2 u / 1.5 u. No pick was published during that period: this is a retrospective simulation, reproducible from the code, that validates the method; the proof under real conditions is the track record above, which applies those same rules live.
BACKTEST · SIMULATIONThe model on the WTA, validated over ten years (2016-2026)
The same model, replayed over 55,589 priced WTA matches (real historical odds). Scope specific to the women's tour: strong + moderate + slight at odds < 2.00, stakes 2 u / 1.5 u / 1 u — the backtest shows that slight signals, loss-making on the ATP, are profitable on the WTA. No pick was published during that period: a retrospective simulation, reproducible from the code. The site's WTA picks are under observation ahead of their public launch.
Going further: CLV and official scope
CLV compares our odds at the time of the pick to the market's closing odds. Beating it consistently is the most reliable sign of a model that sees clearly, long before ROI says so. It is our #1 KPI, detailed in the methodology.
Edge: the gap between the model's probability and the one contained in the odds, bookmaker margin removed. A pick is only issued when that gap clears our thresholds: value has to stay rare.
Official scope (strong + moderate · odds < 2.00): over ten years of real odds (55,317 matches), the model wins on favourites and slight underdogs, and loses beyond 2.00. The official ROI therefore counts only those picks. Since 03/08/2026, no pick is published at odds ≥ 2.00 anymore (those published before remain displayed as they were); slight edges below 2.00 are still tracked under observation: if a pocket proves a lasting edge, it will join the scope.
Staking: the track record is staked at 2 u (strong) / 1.5 u (moderate) / 1 u (slight, observation), the backtest-validated sizing. The quarter-Kelly variant (stake proportional to the size of the edge) does even better in the 2016-2026 simulation: +3.72% per unit staked versus +2.6%, with a maximum drawdown of 29 u versus 56.2 u. Variable stakes are unreadable for a public record, so the track record keeps fixed stakes; the quarter-Kelly advised stake is shown to Pro members on every official-scope pick, as a % of bankroll, for guidance only.
Voided (market reversal): when the match favourite flips at our reference bookmaker before the first ball, with a move of at least 8 probability points, the signal is automatically voided and timestamped before the match. Such a reversal almost always reflects information the stats cannot see yet: injury, poor fitness, withdrawal. The line stays visible, excluded from win rate, ROI and CLV. The rule is symmetric: it also fires when the market moves towards our pick.
Today's picks are for members
Everything already settled stays public, right here, untouched. What Pro adds: the picks BEFORE the match, the model's probability and the analysis behind it.
Discover EdgeLab ProThe evening recap
Every evening, the day’s settled bets — wins and losses alike — and the Arena standings. Nothing else.
One email per evening, one-click unsubscribe. Your address is used for this and nothing else, and never shared.
EdgeLab AI