TIPBLITZ Postmortem

We Tried To Beat The Bookies. Here Is Exactly What Happened.

Mission: build an NRL model with real edge and test whether we could beat market prices over time. We ran multi-year walk-forward tests, rebuilt the models without odds leakage, and measured market-by-market performance with strict validation.

H2H forward ROI: -8.6% to -17.8%
Line forward ROI: -9.0% to -0.5%
Totals forward ROI: +5.5% to +6.0%

Original Goal

Beat Market

Build honest, evidence-led models that could win on price.

Data Scale

79,583

player-game records + odds + fixtures.

Current State

Totals Only

Only market with repeat positive forward ROI.

What The Data Shows

Year-by-year diagnostics and confidence intervals from walk-forward testing. The strongest repeat signal is totals, while other markets remain below benchmark.

All Markets: Win Rate by Year

Walk-forward accuracy across H2H, totals, and line markets vs coin flip baseline

45%50%55%60%65%70%75%2018202020222024

Interactive snapshot · 2024

H2H Accuracy: 60.1%

Bookie Baseline: 60.1%

Totals Win Rate: 52.7%

Line Win Rate: 60.4%

Coin Flip (50%): 50.0%

Model Breakdown So Far

We tested multiple model families and benchmarked each one against market reality.

  1. 1. Market baseline (bookie favorite)

    Hard benchmark: 66.7% H2H average accuracy from 2018-2024.

  2. 2. Odds-heavy baseline models

    Good headline hit rates, but edge was too similar to market and faded after vig.

  3. 3. No-odds player-history models

    Cleaner and leakage-resistant, but forward ROI stayed negative in H2H and line markets.

  4. 4. Market-specific deployment tests

    H2H -8.6% to -17.8%, line -9.0% to -0.5%, totals +5.5% to +6.0% with confidence intervals still crossing zero.

2018-2024 H2H average: Model 65.3% vs market baseline 66.7%.

Totals ROI by Year (Diagnostic)

Totals v2 at fixed 0.15 threshold, showing how volatile yearly ROI is

-5%0%10%20%30%40%20202021202220232024

Interactive snapshot · 2024

Open Snapshot ROI: 2.4%

Close Snapshot ROI: 0.7%

Totals ROI Confidence Intervals

Point estimates are positive and encouraging, but every 95% ROI interval still crosses 0. This keeps totals in promising research-phase territory rather than proven edge.

Hardened Totals (Open)

+6.00%

CI: -2.91% to +14.91%

Hardened Totals (Close)

+5.51%

CI: -3.53% to +14.54%

Totals V2 (Open)

+5.77%

CI: -3.25% to +14.78%

Totals V2 (Close)

+5.93%

CI: -2.97% to +14.97%

What Improved

  • Validation quality is much stronger than early versions, with strict walk-forward testing.
  • Data engineering now supports repeatable market snapshots and cleaner leakage controls.
  • Totals continues to show positive forward ROI estimates worth further live tracking.
  • The project now has a clear benchmark-first process for truth-checking new ideas.

Model Scoreboard

Market Baseline

Bookie favorite H2H accuracy: 66.7% (2018-2024).

H2H Models

Forward ROI range: -8.6% to -17.8%.

Line Models

Forward ROI range: -9.0% to -0.5%.

Totals Models

Forward ROI range: +5.5% to +6.0%, with confidence intervals still crossing 0.

Research postmortem only. Gamble responsibly. For help call 1800 858 858 or visit gamblinghelponline.org.au.