Publication Gate
Market comparisons are withheld unless their data, model, and risk checks are all traceable.
No browser-side edge calculation
An old score or an uncalibrated probability can make a gap look meaningful when it is not. SAM only displays a market comparison after the Python pipeline records a time-stamped odds snapshot, a feature cutoff, a calibration artifact, and a passed risk gate.
Data freshness
No verified signal
Model & calibration
No verified signal
Risk gate
No verified signal
What must be true before this page publishes an analytic comparison
- • A licensed odds quote is preserved with its provider ID, bookmaker, and capture time.
- • Every model feature was available before the game and is bound to a versioned dataset manifest.
- • The winning model is evaluated out-of-sample and calibrated before release.
- • Exposure, data-quality, and staleness limits pass; otherwise the decision remains on hold.
