| Team | Record | Rating | Proj. W-L | Win div | Playoffs | #1 seed | Win conf | Win SB |
|---|
What if…
Tap a team to lock in a win. The whole season re-simulates instantly. Percentages are the model’s pre-pick chances.
How it works. Each team gets an offense and defense rating, fit from every 2026 score so far plus the betting market’s spreads and totals (which carry injury and QB news the scores don’t). Games that already have a posted betting line use that line. Recent games count more than older ones, and early in the season last year’s results are blended in until about Week 5. Each remaining game is simulated with a typical NFL spread of outcomes (±13.5 points), and each simulated season also nudges every team’s true strength up or down a bit, since ratings after 8 weeks are uncertain.
Simplifications. Tiebreakers use record, head-to-head, division record, conference record, then a coin flip (the NFL’s full list goes deeper). Seven teams per conference make the playoffs; the higher seed hosts; ratings don’t change mid-season.
Drive log
How it works. The game is played possession by possession. Each drive ends in a touchdown, field goal, punt, turnover or turnover on downs, with odds set by the offense’s rating against the defense’s and by field position. Drives burn real clock, so a game runs about 11 possessions per side. Overtime follows current NFL rules: both teams get the ball, then sudden death; regular-season games can still end tied after 10 minutes.
How accurate is this? We tested the model on every regular-season game from 2012 to 2025 (about 3,700 games), fitting ratings using only what was known before each week. Its average miss on a game’s final margin was about 13.2 points when predicting the next week, and about 13.6 for games 4 to 8 weeks out. The closing betting line, the benchmark to beat, missed by about 13.0. Its win chances held up: favorites won slightly more often than the model said, not less.
What it can’t see yet. It doesn’t know who is hurt or sitting, except what the betting lines already show, so a backup QB or a missing star isn’t reflected in the ratings. It is least reliable in Weeks 1–4, when it has little data on this season’s teams. For games that already have a posted betting line, the simulator uses that line, since the line beat the model in testing. Last tested Nov 2026 (the season odds use the farther-out figure, and their spread of outcomes matches it); this note will change as the model improves. For entertainment and analysis, not betting advice.