Ryno's Edge — NRL Edition

Nexus Hub Active

Command Dashboard

Quantum-weighted algorithms have processed the latest fixture shards. Targeting high-probability outcomes for Round ....

Model Precision

...

Active Database

...

Round Schedule

Power Rankings

The Edge Pick

AI High Confidence

Model Intelligence Live

Scouting Matrix

Player X-Ray

Detailed performance metrics & trends →

League Meta

Aggregated season-wide stat trends →

Last Matrix Sync

05 MAR 2026 09:13

Season

Stat

Aggregation

Select Teams to Display

Load a season first…

Type a player name above to load their profile.

Season

Round

Current Rankings

Historical Performance

Season Performance (Round by Round)

Select Season:

* ELO ratings are calculated based on match results, opponent strength, and point margins. The history chart shows the progression of each team's rating over the seasons.

Season

Round

Model Weight Distribution

XGBoost 70%

CatBoost 0%

LightGBM 5%

Random Forest 20%

Current Mix: 70% XGBoost / 5% Logistic Regression / 0% CatBoost / 5% LightGBM / 20% Random Forest

Open Source Acknowledgements

OPEN SOURCE ACKNOWLEDGEMENTS

Ryno's Edge relies on several incredible open-source libraries and mathematical models to power our predictive analytics. We gratefully acknowledge the developers and contributors of the following software:

XGBoost

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at: http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

CatBoost

LightGBM

Licensed under the MIT License.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.

scikit-learn (Random Forest & Logistic Regression)

Licensed under the 3-Clause BSD License.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.

Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

KNOWLEDGE HUB

Betting Guide

Master the basics of H2H, Line, and Over/Under markets in the NRL.

Model Insights

Understand the math behind the Blended Model and ELO ratings.

NRL Rules

Essential rules, interchange laws, and scoring mechanics explained.

Edge Index Model

Statistical Normalization & Probability

1. The Concept of Normalization

The Edge Index model doesn't just look at raw numbers; it looks at Relative Performance. We compare every team's stats against the rolling League Average to account for seasonal trends (e.g., higher scoring in dry tracks).

Normalized Score

Team Stat / League Average

Recent Form Window

Rolling average of the last $N$ games

2. Pythagorean Expectation (Exponent 8)

To convert normalized scores into a win probability, we use a modified Pythagorean Expectation formula. While MLB uses an exponent of 1.83, the high scoring and variability of the NRL require an exponent of 8 to achieve maximum predictive accuracy.

Win % = Home⁸ / (Home⁸ + Away⁸)

Where "Home" and "Away" are normalized power scores

3. The 4-Factor Advanced Formula

To reach 62% predictive accuracy, the model balances four distinct aspects of the game:

Core Performance (Slider Weight) Custom

Territorial Pressure (FDO, Kick Metres) 15%

Discipline & Efficiency (Errors, Penalties) 15%

Positional Influence (Spine, Engine Room) 10%

Ensemble Integration

The Edge Index provides the "Base Logic" for the Ryno's Edge system. It is blended with XGBoost and LightGBM outputs to form the final prediction you see on the dashboard.

Global Ensemble

Blending Independent Models into a Single Prediction

1. What is an Ensemble?

No single machine learning model is perfect. Each algorithm sees the data through a different lens, finding patterns the others might miss. An Ensemble combines the outputs of multiple independent models into a final, shared prediction — similar to the concept of a "wisdom of the crowd."

Reduces Variance

When models disagree, the average is more stable than any single model's guess.

Reduces Bias

Different algorithms capture different patterns, limiting the collective blind spots.

2. The Model Stack

The Ryno's Edge Blended Model uses five independent classifiers, each trained on historical match data and player performance stats:

XGBoost Gradient boosted trees. The workhorse — 70% Blended Weight due to its superior 91.78% training accuracy.

Rand Forest Aggregates many uncorrelated decision trees. Great at detecting non-linear feature interactions. 20% Blended Weight.

LGBM Leaf-wise gradient boosting. Fast and efficient on high-dimensional data. 5% Blended Weight.

CatBoost Symmetric trees with ordered boosting. Robust to categorical features and noise. 0% Blended Weight (used for Margin/Total only).

Log. Reg. Logistic Regression. A simple linear baseline that helps prevent overfitting. 5% Blended Weight.

3. Weighted Average Blending

Each model outputs a win probability between 0–100%. These are combined using a weighted average. Weights were determined via a grid search to maximize back-tested historical accuracy:

Final % = (XGB × 0.70) + (RF × 0.20) + (LR × 0.05) + (LGBM × 0.05)

Grid-search optimized weights — back-tested to 91.78% accuracy

4. Margin & Total Predictions

Beyond win probability, the ensemble also provides separate regression models that predict the winning margin in points and the total match score. These use the same features but are trained as regression tasks (not classification), with their own ensemble of models including CatBoost and LightGBM regressors.

Pro Tip

You can adjust the weight of each model using the sliders in the Match Predictor tab. If you believe XGBoost is overfit for the current NRL season, reduce its weight and redistribute to Random Forest or LGBM.

Betting Guide

Mastering the NRL Markets

1. Understanding the 'Edge'

"The Edge" is the core metric of this system. It represents the difference between the Model's Estimated Probability and the Market's Implied Probability.

Implied Probability = 1 / Decimal Odds

If the market offers odds of $2.00 (50% implied), but our model predicts a 65% chance of winning, you have a 15% Edge. This is where value is found.

2. Market Types Explained

Head-to-Head (H2H)

Simply picking the winner of the match. Our primary focus for probability analysis.

Line / Handicap

Betting on a team with a points head-start. E.g., Panthers -6.5 means they must win by 7+.

Over/Under

Predicting whether the total points scored by both teams will be above or below a set number.

3. Strategy & Bankroll Management

Never bet more than you can afford to lose. We recommend using a Unit System where 1 unit equals 1% of your total bankroll.

The Banker (85%+)

4-5 Units

Highest probability logic. Peak confidence. These represent the strongest value propositions found by the blended models.

High Conviction (75-84%)

2-3 Units

Strong statistical evidence. High historical win rate. Reliable core plays for a long-term strategy.

Strong Play (65-74%)

1-2 Units

Reliable edge with consistent value. These plays form the foundation of a disciplined betting approach.

Solid Lean (55-64%)

0.5-1 Unit

Favorable matchup with higher expected variance. Good for filling out a portfolio with smaller stakes.

Coin Flip (<55%)

Caution / Pass

Speculative or high-risk scenarios. Low statistical edge. Best used for small speculative plays or total avoids.

Pro Tip: The Kelly Criterion Explained

The **Kelly Criterion** is a mathematical formula used by professional bettors to determine the optimal size of a bet. Its goal is simple: maximize the long-term growth of your bankroll while minimizing the risk of losing everything.

The "Edge" Concept

To use Kelly, you first need to find your **Edge**. This is the difference between what you think the probability is vs. what the bookmaker thinks.

Example: If bookie odds are $2.00 (implied 50% chance), but our model predicts a **55%** chance, you have a **5% Edge**.

Real-World Scenario ($1,000 Bankroll)

Using our **5% Edge** example from above, here is how a professional would determine their bet size:

$50.00

Full Kelly (5%)

Mathematically optimal but very aggressive. High risk of large swings.

$25.00

Half Kelly (2.5%)

The "Aggressive Standard". Balances growth and safety.

SAFE

$12.50

Quarter Kelly (1.25%)

The "Professional Standard". Highly safe for long-term survival.

The ratings in this dashboard (Banker, High Conviction, etc.) use a simplified version of this logic to suggest unit sizes that keep your bankroll safe while still capitalizing on high-edge opportunities.

XGBoost Model Deep Dive

Technical Documentation & Insights

How XGBoost Works

XGBoost (Extreme Gradient Boosting) is an ensemble learning method that combines the predictions of multiple "weak" decision trees to form a single "strong" predictor. It works through a process called Boosting, where each new tree is specifically designed to correct the prediction errors of all previous trees.

The Process

1. Starting with a base prediction (usually 0.5 win probability).
2. Building a decision tree to predict the "Residuals" (errors).
3. Adding the tree's output to the tally with a "Learning Rate" penalty.
4. Repeating this 300+ times to refine the final probability.

Statistical Inputs (The "Features")

The model analyzes Rolling 5-Game Averages for both teams. This ensures it focuses on current form and momentum rather than historical reputation.

Attack Stats

Tries, Linebreaks, Tackle Busts, Offloads, Run Metres, Post-Contact Metres, Kick Return Metres.

Defence & Discipline

Tackles Made, Missed Tackles, Errors, Penalties Conceded, Tries Conceded.

Model Config

Total Trees 300

Max Depth 5

Learning Rate 0.05

Subsample 80%

Did You Know?

XGBoost handles "missing data" automatically. If a team's missed tackles aren't recorded, it uses its learned historical patterns to estimate the impact.

Random Forest Analysis

Ensemble Learning & Bagging Logic

How Random Forest Predicts

Random Forest is an ensemble learning method that serves as a "Wisdom of the Crowd." Unlike a single decision tree which can be biased, Random Forest grows hundreds of different trees and lets them "vote" on the outcome. This process is called Bootstrap Aggregating (Bagging).

The Logic

1. Sampling: Creating random subsets of the match data.
2. Feature Shuffling: Selecting random stats for each tree to consider.
3. Voting: Each of the 300 trees makes a win/loss prediction.
4. Consensus: The final percentage is the ratio of trees that picked the winner.

Classification vs Regression

While Random Forest is elite at predicting Win Probabilities (Classification), it can sometimes produce inconsistent results for continuous variables like scores (Regression).

Optimization Note

To ensure the highest accuracy in our Blended Model, we have suppressed the Margin and Game Total outputs from the Random Forest model. This forces the consensus to lean on more robust regressors like XGBoost and LightGBM for score predictions, while still utilizing RF's strength in picking winners.

Model Config

Estimators 300

Max Depth 10

Random State 42

Min Samples 2

Key Features

The model heavily weights ELO Difference and Post-Contact Metres. It excels at identifying when a dominant forward pack will eventually break down the opposition.

LightGBM Deep Dive

Gradient Boosting & Leaf-wise Efficiency

How LightGBM Works

LightGBM is a high-performance gradient boosting framework developed by Microsoft. It is designed to be fast, distributed, and efficient. While it shares the "Boosting" DNA with XGBoost, it introduces several innovations that make it ideal for the high-dimensional data used in NRL prediction.

Technical Innovations

1. Leaf-wise Growth: Growing trees by splitting the leaf that reduces loss most, rather than level-by-level.
2. GOSS: Retaining training instances with large gradients while randomly sampling those with small gradients.
3. EFB: Bundling mutually exclusive features (like team IDs) to reduce the number of features.
4. Histogram Learning: Bucketing continuous stats into discrete bins for near-instant split finding.

LGBM in the Blended Model

In our NRL predictor, LightGBM acts as the "Speed Demon." It processes the thousands of data points from our Rolling 5-Game Averages faster than any other model, making it the primary model we use for high-frequency updates or real-time odds simulations.

Handling Complexity

LightGBM excels at finding subtle interactions between stats (e.g., how "Post-Contact Metres" impact "Missing Tackles" later in the game).

Memory Efficiency

Its histogram-based approach uses less memory, allowing us to train deeper models on more seasons of historical data.

Model Config

Total Trees 500

Num Leaves 31

Learning Rate 0.05

Subsample 80%

Speed Advantage

While XGBoost takes roughly 15-20 seconds to train on our full dataset, LightGBM often completes the same task in under 5 seconds with comparable accuracy.

CatBoost Model Deep Dive

Categorical Boosting & Symmetric Trees

The CatBoost Edge

Developed by Yandex, CatBoost (Categorical Boosting) is a powerhouse specifically optimized for datasets with many categories—like NRL team names, match venues, and weather conditions. It is the only model in our ensemble that handles these labels natively without losing information.

Core Principles

1. Symmetric Trees: Building balanced decision trees where every node at a level uses the same split condition.
2. Ordered Boosting: A unique training method that prevents "target leakage" by strictly respecting the data sequence.
3. Native Categorical Support: Converting team and venue names into high-precision vectors using advanced Target Statistics.

Why We Use MAE for CatBoost

For our score predictions, we configured CatBoost to optimize for Mean Absolute Error (MAE). Unlike standard models that try to minimize distance to "perfect" scores, MAE models prioritize being "roughly right" most of the time.

Performance Note

By using MAE, CatBoost doesn't get "overwhelmed" by massive 50-point scorelines that occur once or twice a season. Instead, it maintains its focus on predicting the competitive 1-12 point margins common in top-tier NRL matches.

Model Config

Iterations 500

Tree Depth 6

Loss Func MAE

Random Seed 42

Inference Speed

Due to its symmetric tree structure, CatBoost can generate predictions for a full NRL round in under 10 milliseconds, making it the fastest live-inferencing model in the group.

ADMIN PORTAL

User Management & System Access

Authentication

RYNO'S EDGE NRL EDITION

Command Dashboard

Recent Results (Round ...) WRAPPED UP

Round Schedule VIEW ALL

Power Rankings ELO DATA

The Edge Pick

Model Intelligence Live

Scouting Matrix

Current Rankings

Historical Performance

Season Performance (Round by Round)

Stat Weights Remaining: 100%

OPEN SOURCE ACKNOWLEDGEMENTS

XGBoost

CatBoost

LightGBM

scikit-learn (Random Forest & Logistic Regression)

KNOWLEDGE HUB

Betting Guide

Model Insights

NRL Rules

Edge Index Model

1. The Concept of Normalization

2. Pythagorean Expectation (Exponent 8)

3. The 4-Factor Advanced Formula

Global Ensemble

1. What is an Ensemble?

2. The Model Stack

3. Weighted Average Blending

4. Margin & Total Predictions

Betting Guide

1. Understanding the 'Edge'

2. Market Types Explained

3. Strategy & Bankroll Management

The Banker (85%+)

High Conviction (75-84%)

Strong Play (65-74%)

Solid Lean (55-64%)

Coin Flip (<55%)

The "Edge" Concept

Real-World Scenario ($1,000 Bankroll)

XGBoost Model Deep Dive

How XGBoost Works

The Process

Statistical Inputs (The "Features")

Attack Stats

Defence & Discipline

Model Config

Did You Know?

Random Forest Analysis

How Random Forest Predicts

The Logic

Classification vs Regression

Optimization Note

Model Config

Key Features

LightGBM Deep Dive

How LightGBM Works

Technical Innovations

LGBM in the Blended Model

Handling Complexity

Memory Efficiency

Model Config

Speed Advantage

CatBoost Model Deep Dive

The CatBoost Edge

Core Principles

Why We Use MAE for CatBoost

Performance Note

Model Config

Inference Speed

ADMIN PORTAL

ACCESS RESTRICTED

Sign In

Recent Results (Round ...)

Round Schedule

Power Rankings