How Huffalytics works

Every team carries an offense rating and a defense rating in expected points added (EPA) per play, adjusted for the opponents faced. Each rating blends a preseason projection with the season's plays, weighted by how many plays have been run, so an early-season number leans on the prior and a late-season number on the games. Net rating is offense minus defense allowed.

The team rankings show the same inputs for all 32 teams entering each week: EPA per play, success rate, explosive plays, pressure and sacks, third downs and the red zone, with week-over-week change.

The engine's own number for a game is the model margin: one straight line fit on every completed game with a closing line since 2006 turns the gap between the two teams' net ratings into points, plus a home-field constant. Beside it sits the market's line and the difference between the two.

An adjusted margin folds in the injury report, but only the terms a backtest found worth pricing: a starting quarterback out or questionable, and a skill player's share of his offense lost or in doubt. Tap any game on the week page to see the ratings with their league ranks, the arithmetic, and the injuries that moved the number.

For each game the engine projects a box score: team points anchored on the market total and spread, then passing, rushing and receiving lines for every player from his share of the offense, the opponent's defense, forecast wind and the injury report.

The projected leaders rank those lines as PPR fantasy points by position, and once the week is played the actual sits beside each projection. The player outlook blends each offense with how soft the defenses left on its schedule are.

Before kickoff the engine writes a single-game parlay plan for each game: a set of player prop calls, each an over or an under on a line, with the reasoning behind it. From the 2026 season the market those calls are read against is Kalshi, the regulated prediction exchange, which lists NFL player markets as ladders of rungs such as 55+ receiving yards or fewer than 250 passing yards.

Every call is graded off the box score at three cushions down that ladder:

• Normal is the line the plan called.
• Edge is the listed Kalshi rung nearest halfway between the plan's line and the ladder's last rung on the easy side.
• Edge+ is that last rung: the lowest N+ on an over, the highest <N on an under.

The week page shows overs and unders separately with the record at each cushion, and every call can be opened to see its rungs, Kalshi's last pregame price for each, and what happened. Huffalytics is not affiliated with Kalshi.

Kickers and defenses are ranked by walk-forward models fit on 2013 through 2025: implied team points carry most of a kicker's week, and a low opponent implied total plus being the favorite carries most of a defense's. Quarterback, running back, receiver and tight end streamers outside the starter line are ranked by the game projection engine.

The top five in each tab are the week's stream picks, and once played a pick is a hit when it finishes in the top ten.

Scores, schedules, closing lines, rosters, injury reports and play-by-play come from the open nflverse project. Live injury designations and depth charts come from Sleeper. Market prices are Kalshi's public API. Weather is a forecast feed read in the hours before kickoff. Team logos are served by ESPN.

The engine, the projections, the plans and every grade are Huffalytics' own. The numbers behind each page are published as JSON beside it, for example week.json and rankings.json.