How the model works
No black box and no hype. Here is exactly how we pick, how we stake, and how we decide whether the edge is real — with the numbers updated live from our own settled results.
We only call an edge "proven" once the 95% Wilson lower bound on win rate clears break-even over a large enough sample of real, settled single bets. See the full public record.
1. A model of how goals are scored
We use a Dixon-Coles model — the standard model in football analytics. Every team gets a latent attack and defence rating, fitted by time-weighted maximum likelihood on matches that happened before each fixture (never after — that would be cheating). From those ratings we compute the full distribution of scorelines and read off calibrated probabilities for Home / Draw / Away, Over/Under 2.5 goals and Both-Teams-To-Score.
2. We only bet value at fair odds
A pick qualifies only when our probability beats the market price — i.e. the expected value P × odds − 1 is positive — at odds of 2.00 or higher, where the break-even win rate is just 50%. Singles only: one wrong leg never sinks an accumulator.
3. Disciplined, edge-sized stakes
Stakes follow fractional (quarter) Kelly on the model's edge — bigger edge, bigger stake — capped for safety, with an automatic stop-loss after a drawdown. No martingale, no chasing.
4. We only call it "proven" when the maths says so
A hot week proves nothing. We require the 95% Wilson lower bound on our win rate to clear break-even over a large enough sample of settled bets before we ever call a strategy proven — and we publish wins and losses either way. Until then the badge above honestly reads "building track record".
Model diary
Our own honest scorecard over time — the verdict, win rate, statistical floor and calibration on every settled single. This is the model's diary, kept in public.
| Date | Verdict | Settled | Win rate | 95% floor | Brier |
|---|---|---|---|---|---|
| 09 Oct 2026 | RED | 62 | 53.2% | 41.0% | 0.251 |
| 08 Oct 2026 | RED | 60 | 53.3% | 40.9% | 0.249 |
| 07 Oct 2026 | RED | 60 | 53.3% | 40.9% | 0.249 |
| 06 Oct 2026 | RED | 60 | 53.3% | 40.9% | 0.249 |
| 05 Oct 2026 | RED | 60 | 53.3% | 40.9% | 0.249 |
| 03 Oct 2026 | RED | 59 | 54.2% | 41.7% | 0.245 |
| 02 Oct 2026 | RED | 58 | 55.2% | 42.4% | 0.242 |
| 01 Oct 2026 | RED | 58 | 55.2% | 42.4% | 0.242 |
| 30 Sep 2026 | RED | 58 | 55.2% | 42.4% | 0.242 |
| 29 Sep 2026 | RED | 58 | 55.2% | 42.4% | 0.242 |
| 28 Sep 2026 | RED | 58 | 55.2% | 42.4% | 0.242 |
| 27 Sep 2026 | RED | 57 | 54.4% | 41.6% | 0.241 |
| 26 Sep 2026 | RED | 57 | 54.4% | 41.6% | 0.241 |
| 24 Sep 2026 | RED | 54 | 55.6% | 42.4% | 0.243 |
| 23 Sep 2026 | RED | 54 | 55.6% | 42.4% | 0.243 |
| 22 Sep 2026 | RED | 54 | 55.6% | 42.4% | 0.243 |
| 21 Sep 2026 | RED | 54 | 55.6% | 42.4% | 0.243 |
| 19 Sep 2026 | RED | 51 | 54.9% | 41.4% | 0.241 |
| 18 Sep 2026 | RED | 50 | 54.0% | 40.4% | 0.243 |
| 17 Sep 2026 | RED | 50 | 54.0% | 40.4% | 0.243 |
| 16 Sep 2026 | RED | 50 | 54.0% | 40.4% | 0.243 |
| 14 Sep 2026 | RED | 50 | 54.0% | 40.4% | 0.243 |
| 13 Sep 2026 | RED | 50 | 54.0% | 40.4% | 0.243 |
| 12 Sep 2026 | AMBER | 47 | 51.1% | 37.2% | 0.253 |
| 11 Sep 2026 | AMBER | 45 | 53.3% | 39.1% | 0.252 |
| 09 Sep 2026 | AMBER | 44 | 52.3% | 37.9% | 0.252 |
| 08 Sep 2026 | AMBER | 44 | 52.3% | 37.9% | 0.252 |
| 07 Sep 2026 | AMBER | 44 | 52.3% | 37.9% | 0.252 |
| 06 Sep 2026 | AMBER | 43 | 51.2% | 36.8% | 0.254 |
| 05 Sep 2026 | AMBER | 37 | 51.3% | 35.9% | 0.249 |