How Probalyx Predictions Work
Transparent statistical modelling, current-model validation and clear confidence rules.
Probalyx is an information and entertainment platform, not a betting-tips service. Predictions are probabilities, not guarantees.
Our approach
Probalyx estimates football match probabilities using a hybrid model rather than a single formula. The system combines a statistical football base with a learned outcome layer so each prediction reflects both match structure and patterns learned from historical results.
What the model analyses
The current pipeline uses historical results, team strength, home advantage, recent form, attacking and defensive performance, head-to-head context where useful, and other validated match features available before kickoff.
Current model · v4 hybrid ensemble
The v4 model starts from the v3.5 football base built around Dixon–Coles/Poisson goal modelling, history, Elo-style team strength and form. A learned multinomial outcome layer then refines the home/draw/away probabilities. The final blend is 80% learned outcome layer and 20% Dixon–Coles base, followed by calibrated probability scaling.
Best Prediction gate
A match is only labelled a Best Prediction when the current v4 model is high confidence, the leading 1X2 probability is at least 60%, and the lead over the second-most-likely outcome is at least 10 percentage points. If the gate is not met, Probalyx does not force a top pick.
How performance is measured
We evaluate settled predictions using outcome accuracy, Brier score, log loss and calibration. Rolling holdout tests are kept separate from live performance so historical validation is not presented as a guarantee of future accuracy.
Transparency rules
Predictions are timestamped before kickoff, settled against final results and retained for performance review. Premium history can include misses as well as correct calls. Public proof is calculated server-side from settled results rather than hand-picked examples.
Odds and markets
Bookmaker odds are currently disabled as a model input. Probalyx generates its probabilities from its own statistical and learned signals. Secondary markets such as goals and BTTS are analytical outputs, not guarantees.
Important limitations
Football remains uncertain. Injuries, red cards, tactical changes, motivation, data gaps and unexpected events can change a match after a prediction is generated. No model can remove that uncertainty.
Questions about the methodology can be sent through the Contact page.
Probalyx Pro
Unlock complete match analysis
Advanced markets, model explanations and performance analytics.
