How it works

Methodology

The model is calibrated on 1,087 regular-season games of NFL play-by-play, 2022–2025. These are the numbers underneath it.

From 1,087 games of play-by-play

68.6%

of first touchdowns are scored in the first quarter

41.7%

of first touchdowns are scored from inside the 5-yard line

61%

when the goal-line back isn't the volume back, the goal-line back scores

19.1%

of carries inside the 5 go to quarterbacks, converting at 45%

65.6%

of the time, a 7+ point favorite scores the game's first touchdown

30.3%

how often the team's leading scorer actually gets the first one

The decomposition

Nobody scores enough first touchdowns to model them directly — even a star scores a handful per season. So the model splits the question into parts it can actually estimate: the probability the game has a touchdown at all (99.3%), the probability the first one is scored by an offense rather than a defense or return unit (94.3%), which team scores it, and finally which player scores it given that their team does.

Which team scores first

The point spread is the strongest single predictor. Fit against every game from 2022 through 2025, a one-point move in the spread shifts first-touchdown probability by just under two percentage points. A pick'em home team scores first 51.7% of the time; a seven-point favorite, 64.4%.

Which player scores it

Opportunity beats talent, and the right opportunity metric depends on how a player scores. For rushing touchdowns, carries inside the 5-yard line beat total carries decisively — when a team's goal-line back and volume back are different players, the goal-line back scores the first rushing touchdown 61% of the time. For tight ends, red-zone targets are roughly twice as predictive as overall targets. For wide receivers it's a blend: nearly a quarter of first touchdowns are 20+ yard plays, so volume still matters.

Usage shares blend last season with the current one, weighting in-season data more heavily as the year goes on.

Removing the vig

First touchdown markets carry roughly 19% vigorish — about four times a normal side. We take the consensus across five sportsbooks, remove the vig with the power method (which handles longshots correctly where proportional devigging doesn't), and compare the model's number to the true market price. A value play requires at least a 10% edge at the best available price, quoted at three or more books.

What the model won't do

It suppresses picks it can't stand behind. Rookies and players without meaningful prior usage are flagged, and when they hold a meaningful share of a team's market, the whole team is marked reduced-confidence and its picks are withheld. Edges above +100% are treated as data artifacts, not free money. Early-season weeks run on last season's roles and are labeled accordingly.

Every pick is published before kickoff and graded automatically from official play-by-play. Nothing is edited after the fact.