EDGE Bet Model dashboard

NRL Edge Goal 10

This layer is for paper trading and analytics only. It must not be described as
a real betting edge until validation proves it with closing-line value (CLV) and
out-of-sample results.

Definition

A qualifying "10% edge" means:

  • model expected value is at least +10% at placement time
  • the recommendation uses only information available before placement
  • the bet later shows positive CLV versus the closing market

EV is calculated from the model probability and the taken decimal price:

EV = model_prob * taken_price - 1

ROI alone is not enough. Short-run ROI can be noise, fixture mix, or price
luck. A model that cannot beat the closing market should not be promoted even
when its paper ROI is temporarily positive.

Markets

The canonical paper-trading schema supports separate market reporting for:

  • h2h
  • line
  • total

Raw odds snapshots are the source of truth for CLV. Averaged odds can be used
for display, but not as the only retained market record.

Promotion Gates

A model/strategy can be promoted only when all gates pass:

  • 100-200+ qualifying paper bets
  • positive CLV
  • positive ROI
  • recent-season robustness
  • beats the market-only baseline
  • no closing-line leakage in actionable features

Models with negative CLV get zero vote weight in promoted recommendations.
Voting is evidence-weighted by historical CLV, calibration, recent-season
performance, and market-specific reliability; naive majority voting is not a
promotion path.

Leakage Rules

Actionable models can use only placement-time information: current price, current
line, current team/form/lineup state, and pre-kickoff schedule context.

Closing prices, closing lines, line movement, CLV, and settled result fields are
diagnostics only. They can be used for scoring, reporting, and model promotion
decisions after the fact, but not as features in a recommendation that could
have been placed before kickoff.

Model Bench

The current bench is:

  • market_baseline: market-implied comparator, not an edge claim
  • elo_core: H2H Elo baseline
  • logit_h2h: leakage-safe H2H logistic model
  • lineup_margin_ridge: spread model using predicted home margin
  • total_points_ridge: total-points model for over/under
  • gbm_candidate: optional candidate only after baselines are stable

Until the promotion gates pass out of sample, all outputs remain paper-trading
recommendations, not real-money betting advice.