A statistical NRL model: try-scorer, goal-kicking and player-points projections priced against live multi-bookmaker odds to surface value. Educational — not betting advice.
Every figure here is out-of-sample: the model is trained on earlier seasons and scored on 2023, 2024, 2025 it never saw. Numbers it hasn’t earned aren’t shown — see how it performs before trusting it.
0.730
Try-scorer AUC
vs 0.651 recent-form
0.8%
Calibration error
mean |pred − actual|
0.141
Brier score
lower is better
21,671
Test sample
player-games
Try-scorer calibration
Each point bins players by predicted anytime-try chance (x) against how often they actually scored (y). On the diagonal = perfectly calibrated.
Per-stat error (mean absolute error)
| Stat | Model MAE | Baseline | Better by | Sample |
|---|---|---|---|---|
| Hit-ups | 1.14 | 2.11 | 46% | 21,671 |
| Runs | 2.15 | 2.95 | 27% | 21,671 |
| Run metres | 22.56 | 29.93 | 25% | 21,671 |
| Post-contact m | 9.3 | 12.12 | 23% | 21,671 |
| Tackles | 3.8 | 5.89 | 35% | 21,671 |
| Performance pts | 9.9 | 12.76 | 22% | 21,671 |
“Better by” is the reduction in mean absolute error versus predicting each player’s last-5-game average. Lower MAE = closer to the real result. Educational only — not betting advice.