How the Gridiron King model works
No black boxes and no guru gut-feel. We publish a projection and a probability for every player and game, show the math behind them, and then hold ourselves accountable by checking those probabilities against what actually happened. Here is exactly how the numbers are built.
1. Player projections
Every projection starts from a player's own game-by-game log, not a season average. We weight recent games more heavily with an exponentially-weighted moving average (EWMA), so a hot or cold stretch shows through while older games fade rather than vanish. That form estimate is then regressed toward a stable baseline for the player's role and position — small samples get pulled hardest toward the baseline, which stops one big or empty game from dominating a projection.
On top of that we apply matchup context:
- Opponent defense vs. position — how the defense has performed against the position the player fills (e.g. yards allowed to slot receivers).
- Role & usage — snap share, target share and red-zone involvement, which drive volume-based props more than raw talent does.
- Game environment — the projected game total and pace, so expected volume scales with how many plays a game should produce.
2. From a projection to a probability
A single number is not enough to price a prop — outcomes are a distribution, not a point. We wrap each projection in a distribution calibrated to how that stat actually varies (its historical spread and skew), then read the probability of clearing the posted line directly off that distribution.
- Projection
- Our expected value for the stat — the center of the distribution.
- Model probability
- The chance the outcome lands over (or under) the posted line.
- Line
- The number the sportsbook or PrizePicks posted.
3. Game lines & moneylines
Team-level markets use power ratings that blend season-long efficiency with recent form, adjusted for home field and rest. Those ratings produce a projected margin and total, which convert to win probabilities and a fair spread the same distributional way the player props do.
4. What “edge” means here
We compare our model probability against the market's own no-vig (de-vigged) probability — the market price with the bookmaker margin removed — not the raw price with the juice still in it. The gap between the two is the edge:
edge = model probability − no-vig market probability
A positive edge means our model thinks a side is more likely than the fair market price implies. That is an information signal about disagreement with the market — it is not a promise the bet wins.
The honest part: NFL betting markets are highly efficient, and we have not CLV-validated a repeatable edge over the closing line. Treat every projection and edge as a research view, not a guaranteed money-maker. If a number ever looks too good to be true, it usually is — and we would rather tell you that than sell you a fantasy.
5. How we prove the probabilities are honest
Anyone can post confident picks. The real test is calibration: when we say a prop hits 60% of the time, does it? We grade every published probability against the actual result out-of-sample — no hindsight, no cherry-picking — and plot predicted vs. realized frequency on the Model Accuracy page, alongside our Brier score versus the market's. That page is a calibration record, never a betting record: no units, ROI or profit are claimed there.
What this is — and isn't
- It is a transparent set of projections, probabilities and de-vigged edges you can check against results.
- It is not betting advice, a tout service, or a claim of guaranteed profit.
- Markets move and edges erode — a number that had value at open may not by kickoff.
- For entertainment and information only. 21+. Please bet responsibly. Not affiliated with the NFL.