Methodology

1. What the model looks at

Our EdgeLab 4.1 model combines several dimensions computed on our own database: level by surface, the matchup between playing profiles, recent form, head-to-head history and tournament conditions. The components, their weights and their thresholds remain private — they are the result of years of testing, and the one thing we do not publish.

2. What we publish instead: the protocol

An idea only enters the engine after clearing the same path, written before the numbers come in:

• strict chronological replay, no data from the future;
• a significant accuracy gain across the whole corpus;
• then a test against real odds — a lab improvement that does not beat the market is not shipped;
• year-by-year stability, never a rule built on a few dozen bets.

This protocol rejects more ideas than it accepts, appealing ones included. That is the point of it.

3. Value stays rare

Clearing the threshold is not enough: several guards can still cancel a signal — odds quality and consistency, how young a rating is, scope. When a guard fires, the match page says so.

A captured signal can even be cancelled before the first ball if the market turns sharply: a move like that almost always carries information no statistic has seen yet. The line stays visible in the track record, marked “Cancelled”, out of the results.

4. CLV, our yardstick

Over a few dozen bets, results prove nothing — either way. Closing Line Value, our odds at signal time against the market's closing odds, converges far faster: beating the close over time remains the most reliable predictor of profitability known. It is shown in the track record.

5. The data

• over 300,000 matches across ATP, WTA and Challenger since 2013, updated continuously;
• around 13,000 players profiled;
• odds captured several times a day, from opening to close;
• an EdgeLab court-speed reference, tournament by tournament.