There is a whole industry built on not showing the math. You pay $25 a month, someone yells “lock” at you, and when the lock loses there is always another lock tomorrow. The record is self-reported. The method is a secret. The secret is usually that there is no method.
We built NRFI Edge the other way around:
- The formulas are printed in the product. Both models on the board are documented right under the games: what they average, what they weight, what they multiply. The Full Count blends pitcher form, the top of the order, park and weather; The Heater decays each start exponentially so last week counts most. The three research models we tested them against — Log5, Poisson, a plain rate average — are published in the methodology with their full-season results. None of it is proprietary magic, and we are not pretending otherwise.
- Backtesting is a button, not a favor. Pick any past date and the board shows every lean beside the actual first-inning result, with the day’s hit rate tallied.
- We publish the misses. Our full-season backtest shows our best model hitting 73% of top picks in April and 52% in May, and it says plainly that betting the whole board loses money at -130. If a tool only shows you its good weeks, it is a pick-seller with a UI.
- We publish the corrections. A seven-week study had a different model on top. A full season disagreed, so we rewrote the conclusion instead of quietly leaving the flattering version up.
The honest pitch is modest: the first inning is modelable, our best model beat the league base rate by about 12 points on its top pick over 1,819 games, and you can check every number we just said. That is the whole product. If someone promises you more than that, ask to see their math.