· the record

A number is only worth its record.

When the NFL board carries a Signal, it comes from one fixed formula: tonight's workload and matchup, adjusted by effects measured from settled games, turned into a chance the prop lands. Every night it is replayed over every settled prop and judged on a week it never saw. It goes on the board only if it beats a coin, and beats what the board would otherwise show. This page is that judgement, unedited.

Tonight

formula-1: not on the board

On the held-out week the formula scored 0.723 log loss against a coin's 0.693: no better than a coin.

Judged Sep 27, 11:03 AM ET · 1,282 real props and 11,415 from last season replayed.

The week it never saw

Week of 2026-09-15: 637 props the books actually posted, priced the morning of each game. Lower log loss is better; a coin scores 0.693.

SourcePropsLog loss
The formula6370.723
A coin6370.693
The Signal the board published, same props450.747 (formula 0.790)

Where it was surer, it landed on its side 51% of the time; where it was less sure, 55%.

What it said, and what landed

It said overPropsAverage saidLanded over
0-10%56%60%
10-20%1016%60%
20-30%3227%50%
30-40%12136%43%
40-50%31345%46%
50-60%12753%50%
60-70%2164%38%
70-80%774%43%
80-90%183%100%

A calibrated formula lands about as often as it says. Bins with a handful of props swing.

The effects it runs on

Counted from settled games before 2026-09-27. An effect measured on fewer than ten cases is not used.

EffectMeasuredCases
Lead back picks up a missing back's carries67%36
Other backs pick up a missing back's carries1%21
Top receiver picks up a missing receiver's targets23%63
Other receivers pick up a missing receiver's targets17%100
Top tight end picks up a missing tight end's targets20%15
Back's yards per carry with the heavier load×1.0336
Receivers' targets with the passer out×0.97123
Tight ends' targets with the passer out×1.2354
Back's carries when playing while Questionable×0.8142
Receiver's targets when playing while Questionable×0.9078

Plus a yards-allowed factor for every defence against every position, and each position's game-to-game spread. How they combine is on the Signal page.

The history-based fits

  • Pro football · off · On the held-out week the fit scored 0.693 log loss against a coin's 0.693: no better than a coin, so no Signal.
  • College football · on the board · On the held-out week the fit scored 0.678 log loss against a coin's 0.693, and the props it rated higher landed 61% against 53% for the rest.

Every judgement

  1. Sep 27, 11:03 AM ETformula-1off637 props · 0.723

Statistics, not advice. DailyPropBoard takes no wagers and links to no sportsbook.