METHODOLOGY

How the model actually works

Not a black box, and not a "secret formula" pitch. Here's the real discipline behind every pick on this site.

01

Proprietary, purpose-built

Trifecta isn't a public formula or a licensed algorithm — it's a proprietary model built and refined specifically for this. The mechanics are ours; the results are the part we show you.

02

Zero market data as a training input

Spreads, moneylines, and line movement are never fed into the model that generates a prediction. The model forms an independent view of the game first. The closing line only enters the picture afterward, to define what "covered" means for grading and as a benchmark to compare against — never as something the model learns from.

03

Point-in-time correctness, enforced mechanically

Every statistic that feeds a prediction is exactly what was knowable before that game kicked off — nothing from later in the season leaks backward. This isn't a promise; it's checked automatically on every build with an independent leakage-detection pass.

04

Selective, not exhaustive

The model doesn't try to have an opinion on every matchup. It's built to only surface a pick when its own view diverges sharply from the market — that selectivity is the product, not a filter applied after the fact.

05

Confidence, ranked honestly

Every week, each game is ranked by how far the model's own view diverges from the market's. The single largest divergence is that week's most confident pick; the five largest make up the Top 5. Those tiers are fixed in advance by the model's own math — never chosen after the results are in.

06

Every pick, graded in public

Every Top 5 pick — including each week's single most confident call — across 5 NFL seasons is on the Track Record page, wins and losses both. Nothing is filtered out after the fact to make a number look better.

No lock of the week, no guaranteed win rate — a validated edge, shown with its full history.