Methodology
We tell you WHAT we look at, not HOW we compute it. The model architecture is proprietary — public calibration is our public proof of honesty.
Signals feeding the model
Recent team performance weighted by match importance.
Head-to-head history between teams, adjusted for context.
Confirmed and doubtful absences. Impact weighted by player relevance.
Probable XI when available. Rotations and tactical changes.
League, matchday, importance (relegation, cup, international), rest days.
Closing odds from top books as calibration reference. We do NOT copy odds.
6 markets covered
Calibrated probabilities for 1X2 and double chance. Specialized models by league level.
Goal markets with independent calibration. Robust on leagues with small samples.
Adaptive lines based on match style of play and referee characteristics.
What we always publish
- →Every pick with UTC timestamp before kickoff. Never edited after.
- →Final outcome (WIN/LOSS/PUSH) at the same font size as the hits.
- →Calibration curve per market — accessible to Pro subscribers.
- →Auditable Track Record with global stats — public, no login.
What we don't publish
To protect the model's edge, we keep private:
- The specific ML architecture
- The weight of each signal in the computation
- The thresholds that trigger a pick publication
- The calibration techniques applied
- Periodic retraining details
What you can verify: published probabilities hold up over the long run. That's the only promise that matters.