Skip to main content
4DFourth & Data
Fri 01:15ATLGB70%Sun 18:00CAR64%CLESun 18:00CIN55%PITSun 18:00HOUIND53%Sun 18:00KC59%MIASun 18:00LACBUF83%Sun 18:00NEJAX69%Sun 18:00NYJDET53%Sun 18:00SEA71%WASSun 18:00TEN53%NYGSun 21:05ARISF67%Sun 21:05MIN52%TBSun 21:25BALDAL54%Sun 21:25LVNO57%Mon 01:20LA65%DENTue 01:15PHI52%CHI
The record

Every pick, published before kickoff, graded in public

Model18-1456.3% of picks right
Market favourite23-971.9%
Brier score0.275lower is better, 0.250 is a coin flip
Published late0of 32 picks after kickoff

When we say 65%, does it happen 65% of the time?

Picks grouped by the model's confidence, 252 picks. Lines show the 95% range; the dashed diagonal is perfect calibration. (pick-side probability, Wilson 95% CI; all graded games 2025-2026)

30%40%50%60%70%80%90%100%50%60%70%80%90%100%50%–60%: predicted 54.8%, actual 58.9% (n=124)n=12460%–70%: predicted 64.7%, actual 65.6% (n=90)n=9070%–80%: predicted 74.1%, actual 66.7% (n=30)n=3080%–90%: predicted 82.9%, actual 75.0% (n=8)n=8Model confidence in its pick

Where we disagreed with the market

Each 2026 game: the market's home-win chance across, the model's up. Green is a right pick, red a miss.

0%0%50%50%100%100%Wk 1 ARI at LAC: model 59%, market 81% homeWk 1 ATL at PIT: model 62%, market 71% homeWk 1 BAL at IND: model 53%, market 39% homeWk 1 BUF at HOU: model 35%, market 50% homeWk 1 CHI at CAR: model 31%, market 40% homeWk 1 CLE at JAX: model 75%, market 79% homeWk 1 DAL at NYG: model 26%, market 41% homeWk 1 DEN at KC: model 46%, market 55% homeWk 1 GB at MIN: model 36%, market 54% homeWk 1 MIA at LV: model 25%, market 60% homeWk 1 NE at SEA: model 54%, market 60% homeWk 1 NO at DET: model 63%, market 73% homeWk 1 NYJ at TEN: model 42%, market 52% homeWk 1 SF at LA: model 67%, market 64% homeWk 1 TB at CIN: model 51%, market 64% homeWk 1 WAS at PHI: model 55%, market 68% homeWk 2 CAR at ATL: model 39%, market 42% homeWk 2 CIN at HOU: model 45%, market 57% homeWk 2 CLE at TB: model 71%, market 78% homeWk 2 DET at BUF: model 74%, market 68% homeWk 2 GB at NYJ: model 63%, market 38% homeWk 2 IND at KC: model 68%, market 70% homeWk 2 JAX at DEN: model 34%, market 57% homeWk 2 LV at LAC: model 44%, market 73% homeWk 2 MIA at SF: model 61%, market 87% homeWk 2 MIN at CHI: model 67%, market 66% homeWk 2 NO at BAL: model 77%, market 76% homeWk 2 NYG at LA: model 44%, market 72% homeWk 2 PHI at TEN: model 42%, market 27% homeWk 2 PIT at NE: model 55%, market 67% homeWk 2 SEA at ARI: model 64%, market 36% homeWk 2 WAS at DAL: model 61%, market 66% homeMarket: home win chanceModel
The pick ledger · 2026

The receipts

WeekGameOur pickScoreResultPublished (UK)Before kickoff
Wk 2DET at BUF74%31-41Right17 Sept 10:0515.2h
Wk 2CAR at ATL61%34-3Right20 Sept 16:012.0h
Wk 2CIN at HOU55%20-6Right20 Sept 16:012.0h
Wk 2CLE at TB71%23-19Wrong20 Sept 16:012.0h
Wk 2GB at NYJ63%20-17Wrong20 Sept 16:012.0h
Wk 2MIN at CHI67%9-3Wrong20 Sept 16:012.0h
Wk 2NO at BAL77%24-17Wrong20 Sept 16:012.0h
Wk 2PHI at TEN58%24-20Right20 Sept 16:012.0h
Wk 2PIT at NE55%3-20Right20 Sept 16:012.0h
Wk 2JAX at DEN66%13-20Wrong20 Sept 16:015.1h
Wk 2LV at LAC56%26-14Right20 Sept 16:015.1h
Wk 2MIA at SF61%13-35Right20 Sept 16:015.4h
Wk 2SEA at ARI64%31-7Wrong20 Sept 16:015.4h
Wk 2WAS at DAL61%20-37Right20 Sept 16:015.4h
Wk 2IND at KC68%30-33Right20 Sept 16:019.3h
Wk 2NYG at LA56%6-28Wrong21 Sept 10:0915.1h
Wk 1NE at SEA54%10-13Right9 Sept 09:2315.9h
Wk 1SF at LA67%27-7Wrong10 Sept 19:026.5h
Wk 1ATL at PIT62%13-20Right13 Sept 10:057.9h
Wk 1BAL at IND53%41-23Wrong13 Sept 10:057.9h
Wk 1BUF at HOU65%36-31Right13 Sept 10:057.9h
Wk 1CHI at CAR69%59-37Right13 Sept 10:057.9h
Wk 1CLE at JAX75%10-34Right13 Sept 10:057.9h
Wk 1NO at DET63%30-31Right13 Sept 10:057.9h
Wk 1NYJ at TEN58%23-10Right13 Sept 10:057.9h
Wk 1TB at CIN51%27-33Right13 Sept 10:057.9h
Wk 1ARI at LAC59%26-14Wrong13 Sept 10:0511.3h
Wk 1GB at MIN64%22-39Wrong13 Sept 10:0511.3h
Wk 1MIA at LV75%13-27Wrong13 Sept 10:0511.3h
Wk 1WAS at PHI55%22-24Right13 Sept 10:0511.3h
Wk 1DAL at NYG74%20-28Wrong13 Sept 10:0515.2h
Wk 1DEN at KC54%10-31Wrong14 Sept 10:0815.1h

Each pick is graded from the last model run before its true kickoff, and every run is stored unchanged, so any row can be checked against the file we published at the time.