How we pick the booms

The whole thing, in one line

1. 1

   Pull the data

2. 2

   Build ~50 ideas

3. 3

   Test on unseen seasons

4. 4

   Keep what works

5. 5

   Score all 180 starters

6. 6

   Turn it into a %

7. 7

   Lock at kickoff

8. 8

   Grade every pick

The data

Refreshed before every board. Here's what we pull and what we do with it.

• 🎬 Play-by-play

  Every snap of every game since 2017: down, distance, score, who threw, who caught, how far downfield.

  We roll it up per team, per game: how often they throw when the game is close (so blowouts don't fake it), dropbacks, and how good the QB is per dropback.

• 📊 Box scores

  Every player's yards, targets, air yards and carries, game by game.

  Recent yards (last few games blended with the season), share of the team's targets and air yards, share of the carries, and how swingy a QB is week to week.

• 📈 Vegas lines

  The spread and over/under for every game.

  Turned into implied team points: a team expected to score 30 throws and runs for a lot more than one expected to score 17. It quietly prices in domes, cold and backup QBs too.

• 🌬️ Weather

  Kickoff wind: game-time readings for history, a forecast for this week's outdoor games. Domes count as calm.

  Wind is the one weather factor the lines don't fully price. It only moves QBs, and only down.

• 🚑 Injury reports

  Official Out, Doubtful and Questionable tags for both teams.

  When a top receiver or back is ruled out, we show whose targets or carries just opened up. Questionable players get their chance cut to ×0.9.

• 📋 Snaps + depth charts

  Who's on the field and who's starting this week, including backups filling in.

  Only starters get scored. For QBs we use the confirmed starter, so a relief appearance never counts as a start.

• 🛡️ Matchups

  Yards each defense allows to each position, built from the same box scores.

  A soft run defense matters a lot for RBs; for pass catchers, their own role matters more than the matchup.

One hard rule across all of it: only what was known before kickoff. A model that peeks at the future looks brilliant in testing and useless on Sunday, so we test for leaks automatically.

How we chose what matters

Then we pitted it against a black-box model allowed to use all 50 ideas. The simple one won or tied at every position, and it can tell you why.

What each position is judged on

Straight from the live model (v2.0). Bar = how much of the score each factor carries.

The honesty rules