How Accurate Is Alphr?
Every prediction we make is tracked and verified. Here's how our models have performed across 2287 resolved and held-out model-tested AFL, NRL and Super League predictions.
H2H Accuracy
All Models
Brier Score
📊Round-by-Round H2H Accuracy
Bars = round accuracy · Lines = cumulative accuracy · Dashed = 50 % baseline
💰H2H P&L, 1 Unit Flat
Simulated profit / loss betting 1 unit on every Alphr H2H tip
Accuracy by Model
Strike rate and Brier score per model across AFL, NRL and Super League.
Head to Head
Win/loss predictionsMargin Band
Victory margin rangeOver/Under
Total score predictionsEdge-Filtered Picks
Predictions where our model found ≥7 % edge over market odds.
173/230
86/163
odds pending
Expected vs Actual
How closely predicted probabilities align with outcomes. A perfectly calibrated model sits on the diagonal.
50-60%
60-70%
70-80%
80%+
Brier Score by Model
Lower is better. 0 = perfect, 0.25 = coin flip.
Head to Head
Margin Band
Over/Under
Result Composition
Breakdown of resolved predictions by outcome.
Methods
All metrics are calculated from 648 resolved AFL predictions and 637 resolved NRL predictions from the 2025 season, plus 1002 Super League held-out model-tested predictions. AFL predictions are generated by XGBoost models trained on 18+ years of match data using 49 engineered features. NRL predictions use XGBoost models trained on 18+ years of data with ~160 features including referee bias modelling. Super League uses 183 engineered features including team ELO, player lineup ELO, rolling form, venue, H2H and rest-day signals.
Super League metrics use held-out 2024-2025 official match data and the same H2H, margin and total prediction primitives as AFL and NRL. Odds ingestion is pending, so the H2H P&L chart uses a flat 1.91 display benchmark for Super League.
Accuracy is the proportion of predictions where the model's selected outcome matched reality. Brier score measures the mean squared error between predicted probability and actual outcome (0 = perfect, 0.25 = coin flip).
Edge picks are filtered to require ≥7 % edge between model probability and market-implied probability. Calibration is assessed by grouping predictions into probability buckets and comparing expected win rate to actual win rate.
All 2288 predictions are transparently tracked and verifiable on this site. Super League public settlement will be added once the odds and market-tip layer is live.
Frequently Asked Questions
Common questions about how we measure prediction accuracy.