How the model works
We believe a tipping service should show its working — including the parts that don't flatter us. Here's exactly what you're getting, and what you're not.
1. The prediction model
A dynamic Elo rating model (team strength updated game by game, with home advantage) is blended with gradient-boosted trees and a Poisson scoring model to estimate each team's win probability, expected margin and total points. A Monte-Carlo simulation of the scoreline turns those into coherent line and totals probabilities. The model is trained on real historical NRL results and closing odds — it learns from team form, not from the bookmaker's price.
2. What each pick shows
For each market we publish the model's probability and the odds available, so you can see the model's view of the game and the maths behind it. It is not a profit signal, and we do not present it as one. We publish picks as analysis and do not recommend a stake on any of them.
3. The LLM council (Council tier)
The Council tier can add a panel-of-language-models commentary on a pick: variance, late mail, context the model can't see. The council layer adds expert-style commentary and risk framing. It does not increase the statistical edge of any pick. It's a second opinion and a risk lens — explicitly not a higher win rate.
4. What we don't claim
We don't claim guaranteed profit, a fixed strike rate, or that any single pick will win. Sports betting is high-variance. Treat this as transparent analysis and entertainment — one input at most — stake responsibly if you choose to bet elsewhere, and never bet money you can't afford to lose.
Our picks are statistical model estimates, not guarantees. Past performance does not predict future results. No tipping service can promise a profit. Betting involves real risk of loss. Only ever stake money you can afford to lose.