The NFL is the sports section's founding league and its most filed-in one: it is the only league with totals of any kind, the only league with situational pick types, and the only league whose picks file by the week instead of by the day. Everything in this guide that goes beyond the spread pick — the wind under, the revenge spot, primetime unders, the week window — is NFL-only. None of it applies to the other three leagues.
The rating
The rating behind every spread pick in this section starts as a season scoring differential per game: points scored minus points allowed, divided by games played, computed from completed games only — ESPN final scores, season to date, read at filing time. Only regular-season and postseason games count toward it; preseason and exhibition games never enter the window. Since Rule v1.4 (November 8, 2026) the pick no longer decides on that raw average: each game's margin is adjusted by the opponent's own average margin over two passes — in this league each team's last 8 games are double-weighted in every pass — and the pick decides on the adjusted figure. It is still not blended with any published rating. The team with the better adjusted differential is the pick, taken at the filed spread; when the two adjusted differentials sit within 1.0 point of each other, the pick is the home team. The power-ratings table on the league's report page ranks teams on the same adjusted figure and says so.
Lines come from The Odds API's US board as captured at filing. The anchor book is DraftKings where it carries the game, otherwise the first US book carrying all three markets. A game files only when the anchor book carries all three markets and both teams have completed games this season; otherwise it is excluded and the reason is recorded. Every pick settles from ESPN finals, in exactly the markets its record carries, under the rule version that filed it — earlier filings stand as filed and are never restated. Every pick is UNPROVEN; the section is a paper record and claims no edge.
What files under Rule v1.3
| Pick type | What files | Record |
|---|---|---|
| Differential spread | Every game in the week window: the better-differential team at the filed spread, home team inside 1.0. | The record's spread pick; the ledger's spread rows. |
| Wind under (Rule v1.2 WIND) | An outdoor game whose forecast kickoff wind is 15 mph or more files the Under, labeled with the forecast, the venue and the source. No forecast, a roofed venue, or less wind files nothing. | The record's total pick, flagged wind_play. |
| Revenge (Rule v1.3 REVENGE) | In a same-season rematch, the loser of the pair's first meeting, against the spread at the filed line. A tied first meeting files nothing; a third meeting still keys on the first. | Its own record field and its own ledger — it never changes the differential pick, and when both fire they book as separate pick types. |
| Primetime unders (Rule v1.3 PRIMETIME) | The Under at the filed total when the game falls on Thursday, Sunday or Monday and kicks off at 8:00 PM ET or later. | Its own record field and its own ledger. A game can carry a wind under and a primetime under at once; each books separately. |
The moneyline does not file in the NFL. No other total files. The v1.3 types are never parlay legs.
The week window
NFL picks file by the week: the window runs Thursday 00:00 ET to the next Thursday 00:00 ET, with the week number counted from Thursday, September 3, 2026. Every not-yet-filed game inside the current window whose kickoff is still ahead files in the daily cycle; a game whose kickoff has passed never files late.
The wind rule
The wind under is the one supported positive in the lab's corrected study — annotation-grade support, from history: outdoor NFL games with kickoff wind of 15 mph or more went under 56.97% of the time, +10.05% ROI on 684 bets (nflverse 1999–2026, closing totals; the corpus's 27.8% line-noise flag applies). The forecast comes from Open-Meteo, read by stadium coordinates from the lab's 32-venue table. Retractable-roof venues count as roofed — roof state is not knowable at filing time — and an international-series home game reads its home venue's forecast, a limitation the rule file records. Cold alone is the published null and files nothing.
The two adopted spots, with their flags
Rule v1.3 (adopted November 8, 2026) came out of the lab's harvest process: published angles are frozen word-for-word in a preregistration and tested on the lab's own history before anything files. The revenge spot tested at 1,521 bets and a 52.50% cover rate against the 52.38% break-even — +1.51% ROI at recorded prices, +0.22% at flat −110. Primetime unders tested at 1,065 bets and 52.53% — +2.33% recorded, +0.28% flat. Both passed the lab's gates, and both carry its concentration flags: the single season 2003 contributes 59.78% of the revenge spot's total profit; 2022 contributes 57.14% of the primetime profit, and primetime's first-half sample is negative. Under the lab's standards an edge concentrated like that is presumed a spike until explained — so both spots file as UNPROVEN, in their own ledgers, and each gets a checkpoint report after 30 settled picks.
The consensus that used to run here
For one day the NFL spread pick came from a five-factor points consensus (Rule v1, November 6, 2026): the market; an opponent-adjusted power rating (two adjustment passes, the last eight games double-weighted, plus 2.2 home points); key numbers; rest; and venue-split form — with a matching totals machine whose weather factor sat inactive for want of a forecast feed. Rule v1.1 added public opinion as a sixth factor. Then the lab found its own baseline had graded away against-the-spread picks with the spread applied backwards: corrected, 13 of 16 against-the-spread verdicts across the sports program flipped to NOT SUPPORTED and no corrected spread cell reached the 52.38% break-even. Rule v1.2 pruned the machines, the moneyline and the totals — the wind under excepted — and spreads reverted to the plain differential: on the corrected record the least-negative against-the-spread machine the lab has tested, a weaker claim than "supported," recorded as one. The public-opinion snapshot still runs every morning and its tallies still publish; since the prune it selects nothing in this league.
Data and surfaces
Form and settlement read ESPN finals. Lines read The Odds API. Wind reads Open-Meteo. Tests read the lab's own NFL corpus of scores and lines back to 1999. The league page carries the filed picks, a line-movement block — the first captured line against the latest, from a ledger whose history begins November 5–6, 2026, published as a display and an input to nothing — and a moneyline board captured from eight books by Betting Data API, which is also display only: no moneyline pick files from it. The NFL is the one league with a weekly per-league report (season-to-date market splits), linked from the league page; its settled record also rides the sports results page with the other leagues'.
Keep reading
How the Sports Lab Files a Pick — the section overview →
Three Legs or No Parlay: How the Lab Builds Its Combos →
The NFL page — picks, board and movement →
Sources: the NFL's governing rule file (Sports Rule v1.4, with the v1.1 public-opinion amendment, the v1.2 prune, the WIND rule, the v1.3 adopted spots and the v1.4 adjusted differential), the lab's daily cycle and its corrected backtest record. Every figure in this guide comes from those files.