Five books post a first-quarter touchdown price; nine post the full game, Pinnacle both ways. The model turns the full-game price into expected touchdowns and keeps the share scored in the first quarter (18.9% on 2021–25).

When and where to bet

Each timing bets the price with the best positive model EV (Yes, or No where posted), at DraftKings alone or the best book. Pinnacle posts full-game prices only in the last few hours, so earlier timings use 2+ Yes-only books with their margin removed.

Players

Weeks 8+ and anything still to grade: 876 of 4172 lines. Show the whole season

How it's priced

Full-game P(TD) is Pinnacle's devigged Yes/No price at the latest full-game poll before the quote; if Pinnacle hasn't posted, the median of any book posting both sides; failing that (italic, "est.") two or more Yes-only books with their margin removed, using each book's median ratio of Pinnacle's fair price to its own Yes price on past closes (DraftKings 0.954, FanDuel 0.954). Expected touchdowns = −ln(1 − P), and P(Q1 TD) = 1 − e^{−0.189 × expected}. The share was fitted on 19,751 player-games, 2021–25; the 2025 holdout predicted 5.8% against 6.1% actual. Frozen 2026-10-10, before any line was graded.

DraftKings record. DraftKings' last quote before kickoff. DK P is devigged when it also posts No, otherwise its raw Yes price (vig included, so a Yes edge is understated). The model's side is Yes when its probability is higher, No when lower and DraftKings posts a No; otherwise no bet. Shaded rows have an edge of 5+ points. A player who didn't take a snap is void. Move = whether DraftKings' price moved toward the model's view at its open.

Capture. Polled when first posted, at 48h and 24h, then every 90 minutes inside 6h and every 30 minutes inside 2h. Captured from 2026-10-10. Times are UTC.