Glossary

EPA per play (Expected Points Added)

Before the snap, every spot on the field is worth some number of points. First-and-10 at your own 25 is worth about +1. First-and-goal at the 1 is worth about +5.5. EPA is how much one play changed that number. An 8-yard gain on 3rd-and-7 is a large positive EPA. The same 8 yards on 3rd-and-12 is negative, because you still have to punt.

In college: CFBD's PPA is the same idea, with a model trained on college games.

What's good: An offense at +0.10 per play is top tier. At −0.10 it is bottom tier. For defense, lower is better.

Success rate

Success rate asks whether a play kept the offense on schedule. A play counts as a success if it gains 40% of the needed yards on 1st down, 60% on 2nd, or all of them on 3rd or 4th. nflverse counts any play with EPA above 0 instead, and the two line up closely.

Why use both: EPA rewards big plays. Success rate rewards staying on schedule. A team with high EPA and a low success rate depends on big plays.

Explosive play rate

The share of plays that gain 20+ yards passing or 15+ rushing. Big plays swing a lot from week to week, so the model relies more on EPA and success rate.

Early-down EPA

EPA on 1st and 2nd down only. Third-down results are noisy. Early downs predict future scoring better.

Pass rate over expected (PROE)

How often a team throws compared with what an average team would do in the same down, distance, score and clock. A positive number means the team passes more than expected.

QB EPA and CPOE

EPA per dropback is what a quarterback's plays were worth. CPOE (completion percentage over expected) measures whether he completed throws that were harder than average. Together they are the best single QB summary in public data.

Opponent-adjusted rating

Raw EPA, corrected for who a team played. Wins over three bad defenses count for less.

The spread and the gap

The spread is the Vegas guess at the final margin. The total is the Vegas guess at combined points. The gap is our model's number minus the Vegas number. Bettors call this the edge. NFL Lab only makes a pick when the gap is 2+ points on the spread or 3+ points on the total.

Break-even is 52.4%

A standard bet costs $110 to win $100. Bettors call that -110 odds. At that price you have to win 52.4% of bets to break even. Below that, you lose money.

Closing line value (CLV)

CLV asks whether the Vegas line moved toward our pick between the time we picked and kickoff. A line that keeps moving our way is an early sign the model sees something real. It shows up long before the win-loss record means anything.

How the record is scored

• Picks are saved with a timestamp before kickoff and never edited.
• We grade the record once, at the end of the regular season, on 100+ picks.
• It must pass every test: above 52.4% against the spread, the line moving our way on more than 55% of picks, and an average closing-line move in our favor.
• We graded the 2022 to 2025 seasons once. The model won 48.8% on spreads and 51.0% on totals, below the 52.4% break-even.

NFL equivalents of your college ratings