🔞 18+ Only  |  Gambling can be addictive — please gamble responsibly  |  Gambling Help: 1800 858 858  |  GambleAware
Model-driven selection · Transparent tracking · Known limitations stated

How the Model Works

PuntersEdge uses sport-specific data, machine-learning confidence bands and fixed staking tiers to publish selections. The model can find patterns — it cannot guarantee outcomes.

Sport-specific model inputs

Different sports behave differently, so the input set changes by code, league and market maturity.

SportInputs considered
AFLRecent form, team statistics, venue/weather conditions, injury reports and head-to-head history.
NRLRecent form, team statistics, venue/weather conditions, injury reports and head-to-head history.
NBATeam form, rest days, travel spots, home/away splits, roster availability and matchup context.
TennisServe and return statistics, Elo ratings, surface type, recent form and player matchup history.
Soccer/EPLExpected goals (xG), recent form, injury/news context, fixture congestion and weather where relevant.
Cricket/Darts/SnookerElo-based strength estimates, player/team form and event-specific context where reliable data exists.

How confidence thresholds are chosen

The model outputs an estimated probability band. PuntersEdge groups published selections into three confidence tiers. These bands are model outputs, not guarantees and not promises of profit.

VALUE65–74% model confidence
STRONG75–84% model confidence
BANKER85%+ model confidence

How staking tiers are assigned

Staking tiers are standardised to keep the record consistent: VALUE = 1u, STRONG = 2u, BANKER = 3u. One unit is defined as 1% of the starting bank. This is a record-keeping convention, not personal financial advice.

What XGBoost does and doesn't prove

XGBoost is useful for pattern detection across structured sports data. It can weight interacting variables better than simple rules. It does not prove a pick will win, cannot predict random events, and cannot remove market efficiency. Past patterns do not guarantee future results.

How backtesting avoids overfitting

Backtests use out-of-sample testing, multiple time windows and fixed rules. We avoid cherry-picked periods and do not retroactively change thresholds to make old results look better.

Why BANKER is a tier, not a guarantee

BANKER does not mean certainty. It means the selection sits in the 85%+ model confidence band. BANKER tips still lose. Historical BANKER strike rate shown on the record page: 63%. That historical strike rate is not a prediction or guarantee of future results.

Model risk warnings

  • Variance can create losing runs even when a model has edge.
  • Line movement may reduce or remove value before execution.
  • Execution timing, liquidity and user staking affect outcomes.
  • Exchange commission can reduce real returns.
  • All models have known and unknown limitations.

When the model may be wrong

  • Delayed injury or team news.
  • Weather changes after tips are published.
  • Short-notice lineup or tactical changes.
  • Illiquid markets with unreliable prices.
  • Blind spots in historical data or unusual match context.

Useful links

This site contains wagering-related analysis and is intended for Australian users aged 18+. Gambling involves risk. Please gamble responsibly.