Four ingredients only: historical results, upcoming fixtures with bookmaker odds, a published model, and public grading. No bookmaker API, no Telegram, no payment system required for free testing.
football-data.co.uk CSVs — two seasons of results for 12 leagues, plus upcoming fixtures with prices. Free, no API key.
Attack and defence strengths with time decay and shrinkage so early-season flukes do not become “facts”.
Dixon-Coles scoreline model, then blend 35% model / 65% market so we do not publish fantasy edges.
Singles need ≥75% model chance and price ≥1.25. Accumulators use efficiency ranking, one leg per match.
Free pick + 2 / 5 / 10 / 20 / 50 / 100 odds boards. Each board shows combined odds and honest landing chance.
Every free pick and board leg is logged before kick-off and graded from full-time scores. Losses stay published.
Walk-forward on 8,107 completed matches: high-confidence picks (≥75%) hit about 86%, but still roughly break even after bookmaker margin (−2.3 per 100 staked). Chasing “value” did worse.
A 100 odds board is a long shot — about 1–3% chance on our numbers. Big tickets exist because people want them; we publish the real probability so nobody is misled.
Free testing is for proving process and transparency — not for claiming guaranteed ROI.
Re-run anytime: node scripts/backtest.js