How Every Verified Pick Is Made
The track record above is the output of a single, repeatable process. Three core principles, one five-step pipeline and a confidence formula that is fully disclosed — no black boxes, no inflated picks.
Data-first
Every verified pick on this page started as raw facts — fixture data, xG numbers, live odds from the bookmakers we use and the latest squad news — collected before any model runs.
Quantitative core
A Poisson probability model, expected-value calculations and a conservative Kelly staking formula form the mathematical backbone. Numbers drive the output, not intuition.
AI synthesis
AI layers context on top of the numbers — flagging conflicting signals and squad absences. Where data conflicts, confidence drops; we never smooth over uncertainty.
Data Collection
API-Football (700+ leagues), Understat xG, live odds from six bookmakers. No AI yet — facts only.
AI Analysis
Attack vs defence strength, recent form, H2H record and squad news processed together.
Math + Value Engine
Poisson model and Value Engine run in parallel. Any conflict between tracks reduces confidence.
Final Review
Internal consistency check. Confidence clamped 5–95% — we never claim certainty.
Publication
Card locked pre-match, then auto-verified against official results — exactly the bets you see above.
Confidence Formula
Confidence = Data Quality ×40% + Market Consensus ×30% + Historical Stability ×20% + Context ×10%
Clamped 5–95% · conflicts between data tracks reduce the score automatically