QAFLW
Performance
How our tipping heuristics are performing.
Current Season 2026
20 / 20 rounds completed
How profit is calculated: each tip carries a simulated $10 stake. Games with bookmaker odds (The Odds API) settle at the real decimal price for the tipped team; games without an odds snapshot settle at a representative $1.90; drawn games refund the stake. Odds coverage shows the share of tips settled at real prices.
Best Bet
Current SeasonWeighted Tip
Current SeasonBoosted Tip
Current SeasonWeighted Tip Model
v77The Weighted Tip model combines all 5 ML model predictions using a linear regression trained on 86 historical games. Each model contributes a margin weight (influence on score margin) and a confidence weight (influence on confidence). The current model (v77) has R² = 0.463 with intercept -0.62. Most influential models: elo rating diff, form wins diff, home advantage home constant. The model is retrained weekly with updated coefficients.
Each model contributes a margin weight (× its predicted margin toward home) and a confidence weight. Models with larger absolute weights have more influence on the final tip.
Margin Weights
Active Boosted Model (XGBoost)
v63The Boosted Tip model is a gradient-boosted tree ensemble (XGBoost) trained on 86 historical games to predict the home-team-signed margin from the same model predictions the Weighted Tip model uses. The current model (v63) reaches R² = 0.767 with MAE = 16.7 points and is retrained weekly. Feature importances below are mean |SHAP| values — how much each model's margin and confidence signal moves the prediction, with elo rating diff, form wins diff, value win rate diff the most influential.