NFL 2026
LIVE20,000 simulations of the real 272-game 2026 schedule and the full playoff bracket, through to Super Bowl LXI - a ratings model built from three seasons of scoring margins, replaying real results as they land, and predicting every game with the market alongside.
points-v3 · blended (weight 0.45, futures+spreads, DraftKings) · updated 2026-09-15 · after 16 games
The race for Super Bowl LXI
Share of simulated seasons each team lifts the Lombardi Trophy.
The next games, called16 games
Win probability for the home side.
How this is measured
Show all 16 games
Games that matter this week16 games
Points of playoff probability each team stands to swing by winning versus losing, most consequential first.
Show all 16 games
Bubble watch8 teams
Teams most often sitting on the 7-seed line - in the field or the first team out - across the simulation.
How the model is doing10 games
Every team, every outcome8 divisions
Expected wins and the odds of each landing spot, division by division.
AFC
NFC
AFC
NFC
Get the data: season table as CSV · raw JSON on GitHub · the frozen call ledger · every book’s price on every game
Citizen of Nowhere Picks
Call every game of the week blind, rank your confidence, and take a side on the Upset Radar - the games where this model and the betting market disagree most. The model plays its own card, graded by the same rules.
Play Citizen of Nowhere PicksWhere these numbers come from
Ratings start from three seasons of NFL scoring margin (2025, 2024, 2023), blended with Super Bowl futures markets, then replayed forward 20,000 times against the real remaining 2026 schedule. Last generated 2026-09-15, after 16 games played. Weekly picks blend the model 50/50 with the posted line and are graded against final scores above; the Brier scores on this page and on the Ledger use that same graded record.
How the model works
Each team's rating starts from its scoring margin per game over the last three regular seasons (2025, 2024, 2023), recency-weighted and shrunk hard toward the mean (×0.55) - NFL form carries over far less than fans assume. On top of the stats sits the market: Super Bowl futures for all 32 teams, de-vigged and mapped onto the points scale through the model's own rating-to-title-odds curve, blended in at weight 0.45. The stats know what teams have done; the futures know what the market knows now - quarterback moves, rosters, a champion's continuity - that margins cannot see. This season's real results fold into the rating as they land. Every remaining game on the actual schedule is simulated as a point-spread probability (home advantage 1.6 points, margin sd 13.4), with each simulated season drawing every team's rating from a distribution (σ 2.5 points). Division winners, the seven seeds and the full bracket play out inside every simulation; division tie-breaks are approximated by record then head-to-head. Weekly picks blend the model with the posted line 50/50 and are graded against final scores above. Where the blend still disagrees with the futures alone, that gap is the interesting part.
Two tiers of this model run side by side: a stats-only "lite" rating and the full "classic" build that also blends in the market. Uncertainty shrinks as the season does - the spread of outcomes each simulated season draws from narrows week by week as fewer games remain, and widens back out for a team the stats and the market disagree about most. Each simulated season also draws one correlated error for the whole league, one for each division, and one for each team, rather than treating every game as its own coin flip - real seasons run hot or cold together, not independently. Every tier reuses the same random draws for the same simulated season, so a change from one build to the next reflects an actual change in the inputs rather than noise in which season got drawn.