NFL Week 5 Sunday Lean Card: The Machine's First Football Card Loves Miami by 19 Points More Than the Market Does, the Desk Declines to Bet It, and the Strongest Lean of the Day Isn't on the Card at All

The loop has settled 210 predictions since April — 100 wins, 94 losses, 16 pushes, every one of them baseball or Canadian football. Today it publishes its first football card, tick 85 on the clock. Thirteen NFL games, one Elo model walked forward from 1999, one injury digest, one local betting board, and the same rule as every other card this house has printed: the number leads, the reasoning rides in the article, and every leg grades against the price on this page. New here, the vocabulary lives in moneyline, spread, total, explained.

The method, in one paragraph, because a first card owes its readers that much. Each team carries an Elo rating updated game by game since 1999 (K=32, home field worth 55 points — fit on 2015–2022, validated on 2023–2025: Brier 0.2279 over 816 games, against the betting market's 0.2126; the market is the better grader, said plainly, which is why this house hunts specific price disagreements instead of pretending to omniscience). The morning injury report moves the win probability by the damage differential, capped at 17.4 points. Then the model refuses to believe in blowouts: no probability below 35 percent or above 65 is allowed to print. Bets are home moneylines only — ties grade as pushes on the ledger — sized at quarter-Kelly, 2 units maximum, three per day. Zero NFL rows have ever settled. The track record starts tonight.

The desk card

Prices are PRO•LINE's Sunday morning board, the house book of record, as of roughly 10:15 a.m. MT. Green Bay pays 2.05 at US books (+105), so the local price carries most of the edge — the market thinks this game is a coin flip leaning Chicago; the model thinks Green Bay wins nearly two times in three, and the disagreement survives at the US number (+15.7 points against a 46.8% consensus). The injury digest supports the model's side of the argument: Caleb Williams is questionable with a hamstring and Tyson Bagent starts for the 3-1 Bears (ESPN), while Green Bay's own damage — Banks and Monk doubtful on the interior line — is already inside the number. One caveat, printed because it is real: PRO•LINE's football moneyline is a three-way market with a tie leg, so a tie loses at the window while the ledger would grade it a push. Ties land in one or two games per season. The edge survives the haircut.

Pittsburgh is the opposite shape: the model's raw number is 74.6 percent and the 65-percent ceiling holds it down, which means the printed edge is the conservative version. Daniel Jones on the road, against a 2-2 Steelers team the ratings prefer by 108 points before home field. The desk has carried a Pittsburgh moneyline ticket before (the ticket) and grades this one on the same terms. Seattle is the thinnest of the three — 63.3 against a 60.0 market — but it is a real number, not a clamp artifact, and the model rates Seattle higher than the undefeated 49ers outright (1,666 to 1,629), home field not yet applied.

One scope note, printed because the machine's full-day card is public on the ledger: the loop also carded Milwaukee 0.25u from this afternoon's baseball, +6.3 points of edge — the third leg of the machine's own full-day card. It grades on the ledger like everything else — it just is not football, so it lives outside this piece.

The pass of the day: Miami at 4.10

The machine's card, before the desk touched it, read Miami 1.5 units at 4.10 — the biggest disagreement the model has ever printed against a board. The previous record was ten days old: Atlanta's National League wild-card Game 3, +10.5 points, 1.5 units at 2.45, settled a 6-2 Braves win worth +2.18u on the ledger. That one graded. The math on this one: 45.2 percent against an implied 24.4, expected value +0.85 per unit, no clamp involved — raw Elo said 48.1 before the injury digest charged Miami 2.9 points for its own walking wounded. The model looks at 0-4 Miami and 2-2 Cincinnati, both ratings dragged toward the middle by a month's worth of K=32 updates, and concludes the Bengals are a modest favorite on a neutral field, worth about 55 percent at Hard Rock Stadium.

The market concludes the Bengals are 76 percent. On the road. With Ja'Marr Chase questionable and trending out with a concussion and Tee Higgins hurt (ESPN), in a game the market has not moved off −345 all morning. A number like that is not a pricing error; it is the market reporting that Malik Willis against Joe Burrow is a mismatch the standings do not yet show. The Elo sees 0-4. It does not see why.

Here is the honest arithmetic the desk ran before passing. The breakeven rate at 4.10 is 24.4 percent. The desk's own fair number for Miami — market 24, model 45, split the uncertainty — lands near 28. At 28 percent the bet is still barely positive. So this is not the desk calling 4.10 crazy. It is the desk declining to spend a third of its first-ever NFL card on a wager whose entire edge lives inside the model's known error band, in a sport where the model has never graded a single row, when the alternative for the slot is Seattle at a number the backtest says the model prices honestly. If you bet Miami at 4.10 today, the desk will not call you wrong. It will call you early.

The floor games: four fake edges, zero bets

The 35-percent rule manufactures edges the model never actually said. Four games today sit on the floor, where the raw number came in below 35 and the clamp re-inflated it: Minnesota at New Orleans (raw 18.9 for the Saints), Detroit at Arizona (23.0), Houston at Tennessee (24.4), Denver at the Chargers (28.7). Against the board, two of those floored numbers print as positive edges — Tennessee +8.9 points against PRO•LINE's 4.10, Arizona +3.5 against the US consensus. Both are passes, and the rule is the reason: when the raw probability sits below the floor, the model has told you it cannot see the game, and a manufactured 35 is not a prediction. House rule going forward, printed so it grades: raw below the floor, no bet. The strongest opinion on the entire slate is one of these games — the model's raw 81.1 percent for 4-0 Minnesota at 1-3 New Orleans — and it is not on the card only because this framework prices home moneylines. The ledger cannot grade what it refuses to bet; the lean is named here so it counts in the court of public record.

The news that mattered this morning

Lamar Jackson is out at Atlanta. Saturday's market swung Baltimore to −4.5 on the belief he would play; Sunday morning reversed it — Atlanta −3, −175 (ESPN). The model had already leaned Baltimore — Atlanta 45.7 against a 60.9 market in home-team notation; that lean is void, and this is the QB-blindness argument cutting the other direction, filed in the open: when the quarterback news favors the model's side, the desk will say so just as loudly.

Ashton Jeanty is a game-time decision in Las Vegas (ankle/foot), with a pregame workout scheduled (Rapoport, via ESPN). He is the one name on the Raiders' report the desk is watching at 11:30 a.m. MT, ninety minutes before the early window, because he is the one name the number is not sure about.

Marvin Harrison Jr. is deactivated for Arizona with a trade pending (Schefter, via ESPN), and left tackle Paris Johnson Jr. is already on injured reserve — a rookie protecting Jacoby Brissett's blind side against Aidan Hutchinson. That is the market's Detroit case, and the desk does not disagree; it simply notes that the model's Detroit lean (raw 77 percent) never cleared the floor problem above. No bet. Lean noted.

The Saturday quake that wasn't. Saturday's feeds showed New England −8.5, Detroit −8.5, Houston −3.5 at the close; by Sunday morning the board had re-anchored to −3.5, −5.5, −6.5, confirmed across Dimers, VegasInsider, and Action Network — a feed artifact, not a market. This card prices the re-anchored market, and the survivor desk has already annotated its Saturday piece accordingly (the check).

The full board

Model and market are both the home team's win probability. Market is PRO•LINE where the local board carries the game, US consensus otherwise.

A note on the lean column, because the desk wants it understood, not just read. The framework prices home moneylines because that is what the ledger can grade cleanly; away leans are named, never bet. Today that costs the card the model's two strongest opinions (Minnesota, and Philadelphia — the latter ahead live in London as this publishes, 14-7 in the third quarter, having been a 45 percent shot at a +300 price this morning). Whether "named but not bet" leans grade better than the carded home bets over a season is exactly the kind of question this house exists to answer.

The honest ledger

One more disclosure, because the first card sets the convention. The NFL model trails the betting market in its own backtest (0.2279 Brier against 0.2126, 816 validation games), the home-only framework leaves real edges uncaptured by design, the 35/65 clamp both muzzles and manufactures, and the injury digest reads game-status reports, not scouting reports — it knew nothing about Malik Willis until the standings did. The settled ledger behind this model — the same 210 rows the first paragraph cited — stands +11.90 units on the ninety positive-edge bets at a 41-49 record, mean Brier 0.2405 across 194 graded rows. Everything on this page is backtest arithmetic plus one morning of market reads, sized at a quarter-Kelly because that is what the math supports and no more. The point of printing the first card before the model has earned anyone's confidence is that the second through fiftieth cards get to cite a record. The rules do not change for football.

Green Bay kicks off the card at 11:00 a.m. MT. The desk will grade all three legs against these prices tonight, the machine leg included, and the deviations — Miami declined, Seattle promoted — will be scored in both directions, same as always.

Bring your own pencil.