Week 3 Model Picks

Week 2 results, if you had bet the posted card: 2-2 on spreads, 0-3 on parlays, -4.11 units. You didn’t bet it. Here is Week 3, on the same rules. A spread has to clear a raw edge of 4.0. Totals stay on the page and off the card. Three 1-unit moneyline parlays. The budget is eight units. This week spends seven.

Our Bets

Spreads (4 units)

• 2 units on Bears +4.5 (-105). The Model has this as a pick’em the other way, Chicago by 0.7, against Philadelphia -4.5. That is a 5.2-point edge, the biggest on the board. The caveat matters. Caleb Williams is week-to-week with a hamstring, Tyson Bagent is in concussion protocol, and the depth chart we ingested has Case Keenum. Keenum has no QBR history in the file, so The Model gives him a 45. That is replacement level, not a scouting report. This bet is “4.5 is too many.” It is not “Case Keenum is the answer.” The game is Monday night. If the starter changes, this number is stale.
• 2 units on Dolphins +10.5 (-108). The Model has Kansas City by 5.6, not 10.5. Mahomes’ QBR is 68.9. Malik Willis is at 44.6. Ten and a half is a lot of points.

Parlays (1 unit each)

• Rams ML (-135) / Panthers ML (-148) / Seahawks ML (-345). Pays +276.
• Giants ML (-130) / Saints ML (-185) / Bills ML (-345). Pays +252.
• Texans ML (-125) / Packers ML (-245) / Lions ML (-290). Pays +241. Green Bay plays tonight. That ticket has to be in before 8:15 if you are betting it.

No Thursday spread. Atlanta at Green Bay is GB -4.5 and The Model has Green Bay by 4.6. A tenth of a point. One unit of the budget is unspent. We are not forcing it onto a 3.5-point edge or a total.

The Model

Same ensemble as the first two weeks: Ridge, LightGBM, XGBoost, and a small neural net, refit on games through Week 2, using team stats, EPA, and starter QBR. It is not a money printer. Through two weeks it is 17-15 straight up, 14-18 against the spread, and behind the market on margin error, 12.6 to 11.6.

One yellow cell: Pittsburgh over Cincinnati. The Model calls that game even, and the market has Cincinnati -3.5. The edge is 3.5, so it stays off the card. Chicago is not an outright pick. It is a “the favorite wins, just not by that many” bet, and with Keenum at a 45 the favorite might not win at all. Miami is the same shape at a bigger number.

Over/Under Picks

Week 2 totals went 4-12. Week 1 was 5-11. We are showing the numbers and not betting them. The biggest total edge is Seahawks at Commanders Over 40.5. The Model says 45.7. Still not a bet.

Where we disagree with Vegas

• PHI at CHI: PHI -4.5; model MOV +0.7. On the card for 2 units. Hurts is at 71.7. The Chicago quarterback in the file is Case Keenum at 45, because he has no matched QBR history. Read the caveat above before you bet Monday.
• KC at MIA: KC -10.5; model MOV -5.6. On the card for 2 units. Kansas City is better. Ten and a half is a lot of points.
• LV at NO: NO -3; model MOV +7.0. Off the card. The raw edge is 3.98, which the table rounds to 4.0. Week 2 we left Atlanta off at 3.9. Same rule. Tyler Shough’s QBR is 59.8, Kirk Cousins’ is 51.7, and the Saints are on parlay 2 to win, not to cover.
• CIN at PIT: CIN -3.5; model MOV +0.0. Off the card. Joe Burrow is at 64.1 and Aaron Rodgers is at 31.8, and The Model still only calls it even. That is not a 4-point edge.
• ARI at SF: SF -8.5; model MOV +11.7. Off the card at 3.2. Brock Purdy is at 90.1. The win pick is on the board, not in a parlay, because -455 was too short to put on a ticket.
• BAL at DAL: BAL -3.5; model MOV -0.4. Off the card at 3.1. Lamar is at 74.2, Dak at 86.3. The Model thinks Baltimore is a field goal better, not a 3.5-point favorite.

Two starter notes that do not move the card. Seattle’s row is Drew Lock, which matches the reported starter, and the edge against Washington is 0.1. Washington’s row still says Jayden Daniels. Marcus Mariota is the reported starter while Daniels is out with a dislocated elbow. Do not read that row as a Daniels game. Minnesota is Kyler Murray, back from the concussion, which is what the file has.

Remember, the best way to use this is as one input into your own thinking. The Model does not know who is holding out, and it only knows the quarterback nflverse listed.

These are our personal bets and opinions, not betting advice. Wager at your own risk and within your means.

Good luck everyone.