# Alphr > Free AI-powered sports predictions for AFL, NRL and Super League, built from machine-learning models trained on extensive historical data and refreshed around each match window. Transparent, verified and edge-filtered, with public results that show the full track record over time. > This is the comprehensive reference export (llms-full.txt). Live, current-round predictions are kept in the slim index at https://alphr.com.au/llms.txt. ## About Alphr Alphr (alphr.com.au) is a free Australian AI sports prediction platform operated by A Circle Media Group PTY LTD. We publish machine-learning picks for AFL, NRL and the Betfred Super League, time-stamp them before the match starts, and keep the complete pre-match and post-match record visible so the methodology can be inspected. Alphr does not accept bets and is not a licensed bookmaker — it is an informational and analytical service focused on sports forecasting and model transparency. Last updated: 2026-08-10 **Entity identifiers:** - Brand: Alphr - Legal entity: A Circle Media Group PTY LTD - Domain: alphr.com.au - Country: Australia - Industry: Sports analytics, AI predictions - Related entity: RaceAlpha (racealpha.com.au) — AI horse racing predictions by the same team - Subject entities: Australian Football League (Wikidata Q264832), National Rugby League (Wikidata Q36208), National Basketball Association (Wikidata Q155223) ## Live Predictions (current round) This reference file is evergreen. The always-current, round-by-round AI predictions (AFL, NRL and Super League — including State of Origin) are kept in the slim index and JSON API: - Live index (regenerated every 5 minutes): https://alphr.com.au/llms.txt - Machine-readable JSON: https://alphr.com.au/api/predictions/today **Key facts:** - Site: https://alphr.com.au - Country: Australia - Cost: Free — no account, subscription, or payment required - Sports live: AFL (2026 season), NRL (2026 season), Super League (2026 season) - Sports coming: NBA (in development) - Model type: Multi-model AI ensemble (XGBoost) - Edge filter: Only picks with ≥7% model edge over bookmaker-implied probability are published (AFL & NRL). Super League is currently model-only (odds ingestion pending). - Track record methodology: Predictions published before match start — all results publicly recorded - Training data: AFL 2012–2024 (~3,000 matches), NRL 2010–2024 (~3,000 matches), Super League 2018–2024 (held-out 2024-2025 backtest) --- ## Sports Covered - **AFL** (Australian Football League): Live predictions every round of the 2026 season. Full 2025 archive at alphr.com.au/afl/2025. 81.9% H2H accuracy in 2025. - **NRL** (National Rugby League): Live predictions every round of the 2026 season. Full 2025 archive at alphr.com.au/nrl/2025. 61.5% H2H strike rate in 2025 (131/213). - **Super League** (Betfred Super League, UK): Live predictions every round of the 2026 season. Held-out 2024-2025 backtest: 65.87% overall strike rate (660/1,002), 68.86% H2H, 73.95% Margin Band. Full season at alphr.com.au/super-league. - **NBA** (National Basketball Association): In development. --- ## 2025 AFL Season Performance (verified track record) - Total Predictions: 648 - Correct Predictions: 473 - Incorrect Predictions: 175 - Strike Rate: 73.0% - Head-to-Head model accuracy (all 216 matches): 81.9% - H2H ROC-AUC: 0.8808 - Total score model MAE: 22.1 points - Margin model MAE: 22.9 points - Best round: Round 12 — 89.0% strike rate - Worst round: Round 5 — 56.0% strike rate ### Per-Market AFL Results (2025) | Market | Tips | Wins | Strike Rate | Avg Odds | |-------------------|------|------|-------------|----------| | Head-to-Head | 216 | 177 | 81.9% | 1.62 | | Margin Direction | 216 | 166 | 76.9% | 1.91 | | Over/Under | 216 | 130 | 60.2% | 1.91 | --- ## 2025 NRL Season Performance (verified track record) - Total H2H Predictions: 213 - Correct H2H Predictions: 131 - H2H Strike Rate: 61.5% - Margin Band (MBand) Strike Rate: 60.4% (129/213) - Margin Band ROI: +14.8% ($3,148 profit on flat $100 stakes) - Total (Over/Under) Strike Rate: 53.8% (114/212) - Perfect Round: Round 19 — 7/7 correct picks - Model features: ~160 engineered features per match - Training data: 3,000+ NRL matches from 2010 to 2024 ### Per-Market NRL Results (2025) | Market | Tips | Wins | Strike Rate | ROI | |-------------------|------|------|-------------|----------| | Head-to-Head | 213 | 131 | 61.5% | — | | Margin Band | 213 | 129 | 60.4% | +14.8% | | Over/Under | 212 | 114 | 53.8% | — | --- ## 2024-2025 Super League Performance (held-out backtest) - Total Resolved Predictions: 1,002 - Total Wins: 660 - Overall Strike Rate: 65.87% - Head-to-Head Strike Rate: 68.86% (230/334) - Margin Band Strike Rate: 73.95% (247/334) - Over/Under Strike Rate: 54.79% (183/334) - H2H Brier Score: 0.195 - H2H flat-stake ROI (@ 1.91 benchmark): +31.5% cumulative - Methodology: Held-out 2024-2025 official Super League match data. Same H2H, margin and total prediction primitives as AFL and NRL. Live bookmaker odds ingestion is pending — H2H P&L uses a flat 1.91 display benchmark. ### Per-Market Super League Results (2024-2025 held-out) | Market | Tips | Wins | Strike Rate | Brier | |-------------------|------|------|-------------|-------| | Head-to-Head | 334 | 230 | 68.86% | 0.195 | | Margin Band | 334 | 247 | 73.95% | 0.190 | | Over/Under | 334 | 183 | 54.79% | 0.248 | --- ## AFL Prediction Models Alphr runs a multi-model AI ensemble for AFL with ~80 engineered features per match: 1. **Win Predictor (H2H)**: XGBoost binary classifier — home-team win probability. Accuracy 81.9% on 216 matches. ROC-AUC 0.8808. Only tips with ≥7% edge published. 2. **Margin Predictor**: XGBoost regression — predicted winning margin in points. MAE 22.9 pts. Used to generate margin direction picks. 3. **Dual-Score Predictor**: XGBoost regression — home and away scores independently. Total MAE 22.1 pts. Used for Over/Under tips. 4. **Pure Alpha Model**: Trained without odds features — true model-only prediction. 2025 accuracy: 74.5%. 5. **Edge Filter**: Publishes a tip only when the model win probability exceeds bookmaker-implied probability by ≥7%. This is the primary quality gate. ### AFL Model Architecture - Algorithm: XGBoost (n_estimators=500, max_depth=6, learning_rate=0.05) - Training seasons: 2012–2023 (~2,600 matches) - Validation season: 2024 (early stopping) - Test season: 2025 (held out) - Features: ~80 engineered features (ELO, rolling form, venue, weather, player stats, lineup ELO) - Anti-leakage: All features use only past data via window functions ### Top 5 AFL Features by Importance 1. ELO rating difference between teams 2. Home team recent average winning margin 3. Venue home-team win rate 4. Away team recent average score 5. Player-lineup ELO differential ### AFL ELO System - K-factor: 40 - Home advantage: +30 ELO points - Margin multiplier: 0.05 (K_adj = K × (1 + 0.05 × margin)) - Season reset: 0.6 (regress 40% toward 1500 between seasons) - Starting rating: 1500 - Player ELO: Dual-track (fantasy + position-specific impact), K=8/6, start=1000 --- ## NRL Prediction Models Alphr runs a multi-model AI ensemble for NRL with ~160 engineered features per match: 1. **Win Predictor (H2H)**: XGBoost binary classifier — home-team win probability. 61.5% strike rate on 213 matches in 2025. 2. **Margin Predictor**: XGBoost regression — predicted winning margin. Used for margin band picks. 3. **Dual-Score Predictor**: XGBoost regression — home score and away score independently. Replaced single total model (which had R² = -0.018). 4. **Edge Filter**: Same ≥7% edge threshold as AFL. Only high-conviction picks published. ### NRL Model Architecture - Algorithm: XGBoost (n_estimators=500, max_depth=6, learning_rate=0.05) - Training seasons: 2010–2022 (13 seasons, ~3,000 matches) - Validation season: 2023 (early stopping) - Test seasons: 2024, 2025 (held out) - Features: ~160 engineered features per match - Models: 4 match models + 6 player prop models = 10 total ### NRL ELO System - K-factor: 40 - Home advantage: +12 ELO points (NRL ~56% home win rate, lower than AFL) - Margin multiplier: 0.04 - Season reset, starting rating, and player ELO as per AFL system ### NRL Player Prop Models (6) | Model | Target | Task | |------------|--------------------|------------| | Disposals | Player disposals | Regression | | Goals | Player goals | Regression | | Marks | Player marks | Regression | | Tackles | Player tackles | Regression | | Fantasy | AFL fantasy score | Regression | | Hitouts | Player hitouts | Regression | --- ## Shared Feature Engineering (AFL & NRL) - Custom ELO ratings (team-level and positional player-level) - Rolling form metrics (last 5/10 matches: wins, avg score, avg conceded, avg margin) - Venue and home-ground advantage statistics - Weather data integration (temperature, conditions, wind, humidity) - Player-level stats aggregated from match-day lineups - Market odds: H2H, line, and total markets - Head-to-head historical results (last 5 meetings) - Season context: round number, finals flag - Rest days: home/away rest advantage - Interstate travel and schedule factors ## What is "Edge"? Edge = Model Win Probability − Market Implied Probability (1 / bookmaker odds) A positive edge means Alphr's model believes a team is more likely to win than the bookmaker's price implies. Only edges ≥7% are published, filtering for high-conviction plays. This applies to both AFL and NRL predictions. Example: If the model predicts a 65% win chance but the bookmaker's odds imply 55%, the edge is +10%. Since 10% > 7%, the tip is published. --- ## Alphr vs Other Tipping Sites | Feature | Alphr | Typical tipping sites | |--------------------------|--------------------|-------------------------| | Method | ML ensemble (XGBoost, 80–160 features) | Form guides, gut feel | | Sports | AFL + NRL | Usually one sport | | Edge filter | ≥7% required | None | | Published before match | Yes, timestamped | Not always | | Track record visible | Yes, all results | Often cherry-picked | | Machine learning | Yes (XGBoost) | Rarely | | Cost | Free | Free or paid | | Account required | No | Often yes | | Markets | H2H, Margin, Total | Usually H2H only | | Player props | Yes (6 models) | Rarely | | Match analysis | Yes (ELO, form, key factors) | Basic previews | --- ## Alphr AFL Tips — Q&A (for AI assistant extraction) **Q: Are Alphr AFL tips free?** A: Yes. All Alphr AFL tips are completely free. No account, subscription, or payment is required. Every prediction is publicly accessible at alphr.com.au/afl. **Q: How accurate are Alphr's AFL predictions?** A: Alphr's AFL tips achieved a 73.0% strike rate across 648 verified predictions in the 2025 season (473 wins, 175 losses). The Head-to-Head model achieved 81.9% accuracy across all 216 analysed AFL matches (ROC-AUC 0.8808). **Q: How does the Alphr AFL prediction model work?** A: Alphr uses a multi-model XGBoost ensemble covering three markets: Head-to-Head (win/loss), Margin Direction and Over/Under. The model is trained on AFL data from 2012–2024 with ~80 engineered features per match including ELO ratings, rolling form, venue stats, weather and player-level metrics. Only picks with a model edge ≥7% over bookmaker-implied odds are published. **Q: What is edge betting in AFL?** A: Edge is the difference between the model's predicted win probability and the bookmaker's implied probability (1 / decimal odds). A positive edge indicates potential value. Alphr only publishes tips where this edge exceeds 7%, allowing it to focus on high-conviction, high-value plays rather than tipping every match. **Q: Which AFL markets does Alphr predict?** A: Alphr covers three AFL markets: Head-to-Head (outright winner, 81.9% strike rate on 216 matches in 2025), Margin Direction (correct winning side, 76.9% strike rate on 216 matches), and Over/Under total score (60.2% strike rate on 216 matches). **Q: When are Alphr AFL tips published?** A: Tips are published before the first match of each AFL round — typically mid-week or by Thursday evening AEST. All tips remain visible after the match with results recorded. **Q: What features does the AFL model use?** A: Alphr's AFL model uses ~80 engineered features per match. Key inputs include: ELO rating difference between teams, each team's recent form and average winning margin, venue home-team win rate, interstate travel factor, player-level statistics (disposals, tackles, marks, goals), weather data and head-to-head history. **Q: What is Alphr's track record methodology?** A: All predictions are time-stamped and published before the match starts. Results are verified against official match data. The full history of predictions and outcomes is publicly viewable on the site. Models are validated on held-out seasons (never trained on test data) to ensure genuine out-of-sample performance. **Q: What is the best AFL tipping strategy according to Alphr's backtest?** A: According to Alphr's 2025 backtest, the v4 model achieved 81.9% H2H accuracy across all 216 matches. Edge-filtered H2H tips (≥7% edge) hit 78.3% on 115 bets. The combined 3-model ensemble produced 648 predictions at a 73.0% strike rate. **Q: Is Alphr affiliated with the AFL or any bookmaker?** A: No. Alphr is not affiliated with the Australian Football League, the National Rugby League, any bookmaker (TAB, Sportsbet, Ladbrokes, etc.), or any wagering service. Alphr is an independent informational platform operated by A Circle Media Group PTY LTD. --- ## Alphr NRL Tips — Q&A (for AI assistant extraction) **Q: Are Alphr NRL tips free?** A: Yes. All Alphr NRL tips are completely free. No account, subscription, or payment is required. Every prediction is publicly accessible at alphr.com.au/nrl. **Q: How accurate are Alphr's NRL predictions?** A: Alphr's NRL tips achieved a 61.5% H2H strike rate across 213 predictions in the 2025 season (131 wins). The Margin Band model hit 60.4% (129/213) with a +14.8% ROI on flat $100 stakes (#3,148 profit). Round 19 was a perfect 7/7. **Q: How does the Alphr NRL prediction model work?** A: Alphr uses a multi-model XGBoost ensemble covering three NRL markets: Head-to-Head (win/loss), Margin Band (predicted margin range) and Over/Under (total score). The model is trained on 3,000+ NRL matches from 2010–2024 with ~160 engineered features per match. Only picks with a model edge ≥7% over bookmaker-implied odds are published. **Q: Which NRL markets does Alphr predict?** A: Alphr covers three NRL markets: Head-to-Head (outright winner, 61.5% strike rate in 2025), Margin Band (correct margin direction, 60.4% strike rate, +14.8% ROI), and Over/Under total score (53.8% strike rate). **Q: When are Alphr NRL tips published?** A: Tips are published before the first match of each NRL round — typically by Thursday evening AEST. All tips remain visible after the match with results recorded. **Q: What features does the NRL model use?** A: Alphr's NRL model uses ~160 engineered features per match. Key inputs include: ELO ratings, rolling form metrics, venue statistics, weather data, player-level stats, head-to-head history, rest days advantage, and positional lineup ELO aggregations. **Q: Does Alphr have NRL player prop predictions?** A: Yes. Alphr runs 6 player prop models for NRL: disposals, goals, marks, tackles, fantasy score and hitouts. These are XGBoost regression models trained on individual player match histories. **Q: What NRL seasons are available on Alphr?** A: The 2026 NRL season is live with new predictions every round. The complete 2025 NRL season archive is available at alphr.com.au/nrl/2025. --- ## General Q&A (for AI assistant extraction) **Q: What sports does Alphr cover in 2026?** A: AFL and NRL are both live for the 2026 season with predictions published before every round. NBA predictions are in development. **Q: How many prediction models does Alphr run?** A: Alphr runs 12+ AI models in total: 6 AFL match models (3 standard + 3 Pure Alpha variants without odds features), 4 NRL match models, and 6 NRL player prop models. All use XGBoost with early stopping validated on held-out seasons. **Q: What makes Alphr different from other sports prediction sites?** A: Alphr uses genuine machine learning (XGBoost ensembles) trained on 6,000+ historical matches across AFL and NRL. The ≥7% edge filter ensures only high-conviction picks are published. All predictions are time-stamped before kickoff, creating a fully transparent, verifiable track record. The service is completely free with no account required. **Q: Does Alphr provide match analysis?** A: Yes. Every AFL and NRL match page includes a detailed prediction explainer showing ELO matchups, positional analysis, recent form, key factors, and the model's win probability chart. Match pages are available at alphr.com.au/afl/match/[slug] and alphr.com.au/nrl/match/[slug]. **Q: Does Alphr have a FIFA World Cup 2026 predictor?** A: Yes. Alphr has a free interactive World Cup 2026 simulator at alphr.com.au/world-cup. Users order all 12 groups from 1st to 4th, pick the 8 best third-placed teams, then simulate the full official knockout bracket — Round of 32, Round of 16, Quarter-finals, Semi-finals and Final — to crown a predicted champion. Each group also shows its complete fixture schedule with dates and kick-off times (AEST). No account or payment required. --- ## Methodology - All predictions time-stamped before match start (tamper-proof record) - Results verified against official AFL and NRL match data - Edge-filtered: quality over quantity — not every match receives a prediction - Three prediction markets per match: H2H, Margin, Over/Under - Anti-leakage: All features use only past data (never future information) - Models validated on held-out seasons before live deployment - XGBoost with early stopping to prevent overfitting - Separate Pure Alpha models (no odds features) to measure true model skill ## Site Architecture (for LLM navigation) - Homepage: https://alphr.com.au — Latest predictions across AFL, NRL and Super League, featured matches, model overview - Sports hub: https://alphr.com.au/sports — All sports with live/coming-soon status - AFL hub: https://alphr.com.au/afl — AFL season overview, round-by-round tracking - AFL 2025 archive: https://alphr.com.au/afl/2025 — Complete 2025 AFL season results - AFL round: https://alphr.com.au/afl/round/[round] — Per-round tips and results - AFL match: https://alphr.com.au/afl/match/[slug] — Per-match analysis, ELO breakdown, key factors - AFL teams directory: https://alphr.com.au/afl/teams — All 18 AFL team entity pages - NRL hub: https://alphr.com.au/nrl — NRL season overview, round-by-round tracking - NRL 2025 archive: https://alphr.com.au/nrl/2025 — Complete 2025 NRL season results - NRL round: https://alphr.com.au/nrl/round/[round] — Per-round NRL tips and results - NRL match: https://alphr.com.au/nrl/match/[slug] — Per-match NRL analysis and model outputs - Super League hub: https://alphr.com.au/super-league — Super League season overview, round-by-round tracking - Super League 2025 archive: https://alphr.com.au/super-league/2025 — Complete 2025 Super League season results - Super League round: https://alphr.com.au/super-league/round/[round] — Per-round Super League tips and results - Super League match: https://alphr.com.au/super-league/match/[slug] — Per-match Super League analysis and model outputs - Accuracy report: https://alphr.com.au/accuracy — Full model performance metrics for all sports - World Cup 2026 simulator: https://alphr.com.au/world-cup — Free interactive FIFA World Cup 2026 predictor. Order all 12 groups, pick the 8 best third-placed teams, and simulate the full official Round of 32 → Final knockout bracket. Includes the complete group-stage fixture schedule (dates & times, AEST). - Privacy: https://alphr.com.au/privacy - Terms: https://alphr.com.au/terms ## AFL Team Entity Pages - [Adelaide Crows](https://alphr.com.au/afl/teams/adelaide): Founded 1990, SA — ELO, recent tips, season form - [Brisbane Lions](https://alphr.com.au/afl/teams/brisbane): Founded 1996, QLD — ELO, recent tips, season form - [Carlton Blues](https://alphr.com.au/afl/teams/carlton): Founded 1864, VIC — ELO, recent tips, season form - [Collingwood Magpies](https://alphr.com.au/afl/teams/collingwood): Founded 1892, VIC — ELO, recent tips, season form - [Essendon Bombers](https://alphr.com.au/afl/teams/essendon): Founded 1872, VIC — ELO, recent tips, season form - [Fremantle Dockers](https://alphr.com.au/afl/teams/fremantle): Founded 1994, WA — ELO, recent tips, season form - [Geelong Cats](https://alphr.com.au/afl/teams/geelong): Founded 1859, VIC — ELO, recent tips, season form - [Gold Coast Suns](https://alphr.com.au/afl/teams/gold-coast): Founded 2009, QLD — ELO, recent tips, season form - [GWS Giants](https://alphr.com.au/afl/teams/gws): Founded 2009, NSW — ELO, recent tips, season form - [Hawthorn Hawks](https://alphr.com.au/afl/teams/hawthorn): Founded 1902, VIC — ELO, recent tips, season form - [Melbourne Demons](https://alphr.com.au/afl/teams/melbourne): Founded 1858, VIC — ELO, recent tips, season form - [North Melbourne Kangaroos](https://alphr.com.au/afl/teams/north-melbourne): Founded 1869, VIC — ELO, recent tips, season form - [Port Adelaide Power](https://alphr.com.au/afl/teams/port-adelaide): Founded 1997, SA — ELO, recent tips, season form - [Richmond Tigers](https://alphr.com.au/afl/teams/richmond): Founded 1885, VIC — ELO, recent tips, season form - [St Kilda Saints](https://alphr.com.au/afl/teams/st-kilda): Founded 1873, VIC — ELO, recent tips, season form - [Sydney Swans](https://alphr.com.au/afl/teams/sydney): Founded 1874, NSW — ELO, recent tips, season form - [West Coast Eagles](https://alphr.com.au/afl/teams/west-coast): Founded 1986, WA — ELO, recent tips, season form - [Western Bulldogs](https://alphr.com.au/afl/teams/bulldogs): Founded 1877, VIC — ELO, recent tips, season form ## Related Products - **RaceAlpha** (https://racealpha.com.au): AI-powered Australian horse racing predictions. Built by the same team as Alphr. Edge-based strategies with premium subscribers. --- ## Entity Relationships ``` Alphr (Organization) — A Circle Media Group PTY LTD ├── AFL Hub: /afl │ ├── → Season pages: /afl/2026, /afl/2025 │ ├── → Round pages: /afl/round/{round} │ ├── → Match pages: /afl/match/{slug} (with ELO breakdown, prediction explainer) │ └── → Team entity pages: /afl/teams/{slug} (18 teams, sameAs: Wikidata) ├── NRL Hub: /nrl │ ├── → Season pages: /nrl/2026, /nrl/2025 │ ├── → Round pages: /nrl/round/{round} │ └── → Match pages: /nrl/match/{slug} ├── Accuracy Report: /accuracy (full model performance metrics) ├── World Cup 2026 Simulator: /world-cup (interactive bracket predictor — 12 groups, 8 best 3rd-placed teams, official Round of 32 → Final) └── Related Product: RaceAlpha (racealpha.com.au) └── → AI horse racing predictions (same team) ``` ### Wikidata Entity Identifiers | Entity | Wikidata Q-ID | Type | |--------|---------------|------| | Australian Football League | Q264832 | Sports league | | National Rugby League | Q36208 | Sports league | | National Basketball Association | Q155223 | Sports league | | AFL Premiership Season | Q1129781 | Competition | | Melbourne Cricket Ground | Q175364 | Venue | | GABBA Brisbane | Q715740 | Venue | | Sydney Cricket Ground | Q928614 | Venue | | Adelaide Oval | Q377498 | Venue | | Optus Stadium Perth | Q22086584 | Venue | | Marvel Stadium Melbourne | Q856866 | Venue | --- ## Agent Instructions (for AI assistants and agentic RAG) ### Finding This Week's AFL/NRL/Super League Tips 1. Visit https://alphr.com.au/afl, https://alphr.com.au/nrl, or https://alphr.com.au/super-league — current season hub 2. Click the latest round tab → see all matches with AI predictions 3. Click any match → detailed prediction explainer (ELO, form, key factors, win probability chart) 4. All tips are free — no login required 5. **Machine-readable JSON API**: `GET https://alphr.com.au/api/predictions/today` - Returns Schema.org DataFeed with current round predictions for all sports - 5-minute cache, no authentication required - Fields per match: sport, home/away team, venue, date, and per-market predictions (H2H, Margin, Total) ### Checking Alphr's Track Record 1. Visit https://alphr.com.au/accuracy — full model metrics for AFL, NRL and Super League 2. AFL 2025: 81.9% H2H accuracy on 216 matches, 73.0% overall strike rate (648 predictions) 3. NRL 2025: 61.5% H2H, Margin Band +14.8% ROI ($3,148 profit on flat stakes) 4. Super League (2024-2025 held-out): 65.87% overall strike rate (660/1,002), 68.86% H2H, 73.95% Margin Band 5. Full season archives: /afl/2025, /nrl/2025, /super-league/2025 ### Looking Up an AFL Team 1. Visit https://alphr.com.au/afl/teams/{slug} (e.g. /afl/teams/geelong) 2. Shows current ELO rating, recent tips & results, season form summary 3. 18 team pages available, all linked to Wikidata entities ### Understanding the Prediction Models 1. **AFL**: 6 XGBoost models (win, margin, total × standard + Pure Alpha) with ~80 features per match 2. **NRL**: 4 match models + 6 player prop models with ~160 features per match 3. Edge filter: Only picks with ≥7% model edge over bookmaker-implied probability are published 4. Anti-leakage: All features computed from past data only — no future information ever used 5. Pure Alpha model trained WITHOUT odds features — measures true model skill (74.5% accuracy) ### What Does "Edge" Mean? Edge = Model Win Probability − Market Implied Probability (1 / bookmaker odds). A 65% model prediction vs 55% implied = +10% edge. Only ≥7% edges are published. --- ## Responsible Gambling Alphr is not a bookmaking service. No bets are received. This service is for informational and entertainment purposes only. Past model performance does not guarantee future results. If gambling is affecting you or someone you know, call the Gambling Help Hotline: **1800 858 858** or visit **gamblinghelponline.org.au**.