Better Betting Intelligence

AI that actually beats the bookies

Alphr is a free sports prediction platform powered by machine learning. We publish edge-filtered AFL and NRL picks before every game, verify the results, and show you exactly how the model works. No tipsters. No paywalls. Just transparent signal.

81.9%
AFL H2H accuracy
2025 season
73.0%
Overall strike rate
648 verified picks
≥7%
Minimum edge filter
before publishing
0.738
ROC-AUC (H2H)
model discrimination
What is Alphr?

A smarter way to look at sport

We built Alphr because the sports tipping industry is broken. Pundits cherry-pick wins, paid tipsters hide losses, and bookmakers shade odds away from sharp money. Alphr does the opposite: every prediction is public, every result is verified, and every model is calibrated.

Machine learning, not gut feel

XGBoost models trained on 13+ seasons of AFL and NRL data. Team ELO, positional player ELO, rolling form, venue effects, rest days, and lineup-level features — over 80 inputs per match.

Edge-filtered picks only

We compare model probability to market-implied probability. A pick only goes out when our edge is ≥7%. That's the difference between a tip and an actual bet worth making.

Anti-leakage by design

Every feature uses only data available before kick-off. Rolling stats are computed with strict point-in-time window functions. No peeking at the future means accuracy you can trust.

Radical transparency

Every prediction is timestamped before the game. Every result is verified after. The full track record lives on the Accuracy page — wins, losses, calibration curves, and Brier scores.

How it works

From raw data to published edge

Five steps, fully automated, run before every round.

01

Ingest

Match data, player stats, lineups, weather, and venue history are scraped daily from official AFL and NRL feeds.

02

Engineer

80+ features per match — team ELO, positional player ELO (mid/fwd/def/ruck), rolling form, rest advantage, head-to-head history, and venue effects.

03

Predict

XGBoost models output win probability, predicted margin, and total score. A Pure Alpha variant trains without odds to confirm the model finds true signal.

04

Filter

We calculate edge: model probability minus market-implied probability. Only picks with ≥7% edge make it to the site.

05

Verify

Every pick is timestamped before kick-off and settled automatically post-match. The track record is public and permanent.

Under the hood

The model, plainly explained

Team ELO

Strength ratings that adapt

Every team carries an ELO rating updated after each match. Margin of victory amplifies the update. Home advantage is baked in (+30 ELO equivalent). Seasons partially reset (40% regression to 1500) so dynasties don't dominate forever.

Player ELO (positional)

Lineups matter, individuals matter more

Every player gets a dual rating: a fantasy ELO and a position-specific impact ELO. Midfielders are scored on disposals and clearances, forwards on goals and inside-50 marks, defenders on intercepts and rebounds. Aggregated across the match-day 22.

Pure Alpha

Model without market crutches

We train a parallel model with no odds features at all. It still hits 74.5% H2H accuracy on AFL. That proves the model finds real signal — not just regression to the bookmaker line.

Edge filter

Only the value plays publish

Edge = model probability − market-implied probability. We require ≥7% before a pick is published. On the 2025 season this filter produced a 62.8% strike rate across 191 bets — well above the breakeven needed at typical odds.

Training data

12 seasons of training (2012–2023), 1 season for validation (2024), and 2025 held out entirely for honest backtest results.

~3,000 AFL matches · ~130,000 lineup rows · ~120,000 player stat rows

Why it's epic

Honest edge, freely shared

No other Australian sports site combines accuracy this high with transparency this complete — and gives it away.

Free, forever

No paywalls, no tipster subscriptions, no upsells. Every pick, every accuracy metric, every model breakdown — open to anyone.

Beats the market, provably

Our Pure Alpha model — trained without any odds features — hits 74.5% on AFL H2H. That's mathematical proof we're finding signal the bookies miss.

Published before games start

Tips go live before kick-off, not retrofitted afterwards. No 'we said this all along' nonsense. The timestamp is the receipt.

Calibrated, not just accurate

A 70% confidence pick wins ~70% of the time. We track Brier score and calibration curves so our probabilities mean what they say.

Ready to see the edge?

Live AFL picks for round 2026 are already published. NRL tips and accuracy data are one click away.