Method
From raw data to a qualified signal
Five stages. A signal is published only when it survives all of them — and most candidates do not.
- 01
Collect
Team and player metrics, injuries, rest, travel, schedule context and sportsbook lines across major books.
Advanced team and player metrics, lineup and injury reports, rest days, travel distance and time-zone changes, schedule density, referee and venue context, and live odds from both retail and market-setting sportsbooks. Where sources disagree, the discrepancy is recorded rather than silently resolved.
- 02
Analyze
Matchup-level modelling produces a projection and a fair line independent of the posted market.
Each sport has its own feature set, because what predicts an NHL result is not what predicts an NBA one. The model produces a projected outcome distribution and converts it to a fair price. Critically, it does this without looking at the sportsbook number first, so the posted line cannot anchor the estimate.
- 03
Detect
The fair line is compared against live prices. The difference is the raw edge.
Fair price is compared against every tracked book. The gap, after removing the vig, is the raw edge. We also compute the minimum price at which that edge still exists — the number published on every signal as the minimum acceptable line.
- 04
Validate
Edges are screened for injury risk, stale lines, key numbers and confidence before publishing.
A raw edge is not a signal. It is screened for unconfirmed lineups, thin or stale markets, proximity to key numbers such as 3 and 7 in football, weather, and model confidence. Most candidates are discarded here. What survives is published with its risk factors attached rather than hidden.
- 05
Monitor
Prices are tracked after release. Signals are re-rated as the market moves.
Publishing is not the end. Every signal is tracked until it settles: prices are recorded as they move, status changes to Line Moving or Entry Lost when the number passes the minimum, and the closing line is stored so closing-line value can be measured honestly.
Reading a signal
Every signal card carries the same fields, in the same order, whatever the sport. Three of them do most of the work:
- Estimated edge — how far the posted price sits from our fair price, after removing the vig. It is an estimate, not a promise.
- Minimum acceptable odds — the number at which the edge disappears. Past this, the signal is void. This is the field most people ignore and should not.
- Risk factors — what could break the projection. An unconfirmed lineup, a key number, weather, a market that has already moved.
What we deliberately do not do
- Call anything a lock. No guaranteed winners, no promises of profit. A model that produced certainties would not need a confidence score.
- Hide the losses. Results settle against the closing line and the full history stays published, drawdown included.
- Tell you what to stake. Sizing and bankroll management are yours, and they matter more than any single signal.
- Rewrite history. The line and odds available at release are stored once and never overwritten.
An edge is not a guarantee