How we judge
The Verdict is built only from the public nflverse play-by-play and our own models, refit on the current season as games are added. No outside ratings, no hand-picked numbers. Every constant below is read from the same code that draws the pages.
1 · Robbed & Lucky
Net scrimmage EPA. For each game we sum expected points added on passes and runs for each offense (kneels and spikes out). Every fumble counts as the league-average fumble (-2.56 EPA): offenses keep 52.6% of their own fumbles, and who falls on the ball is close to random. Kicks and returns are left out.
Deserved margin = 0.82 + 0.772 × (home net EPA − away net EPA), an ordinary least-squares fit on 136 games this season (R² 0.830, residual standard deviation 5.49 points). Deserved win probability = Φ(deserved margin / 5.49). Robbed: the loser's deserved win probability is 70% or more. Coin flip: the winner's is under 60%.
2 · Luck Index
Luck wins = real wins (ties count half) − deserved wins. Luck points = fumble-recovery points (recoveries above or below the league rate, times the gap in EPA between a kept and a lost fumble) + opponents' field-goal points below expectation (make probability by distance, logistic fit on this season's kicks: 6.3983 -0.1048 × distance) + opponents' extra points below the league rate (95.0%). One-score games: final margin of 8 or less. Pythagorean wins: exponent 2.37.
3 · Flag Ledger
Each accepted penalty is credited with the snap's win-probability change (nflverse WPA) in favor of the team that drew it. Declined and offsetting fouls are dropped. On live-ball fouls the swing includes the play, so treat it as an upper bound. Touchdowns erased = 6.95 minus the expected points of the offense after the flag.
4 · Coach on Trial
Conversion: logistic fit on third and fourth downs, coefficients [1, -0.847, -0.107, 0.143, -0.631] on [1, ln(yards to go), 4th down, 4th down × ln(yards to go), goal to go]. Success = expected points of a first down 6.3 yards past the marker, or 6.95 for a touchdown. Failure = minus the opponent's expected points at the spot. Punt = minus the opponent's expected points where opponents actually start after punts from that spot (quadratic fit, touchbacks included). Field goal = make probability × 3; a miss gives the opponent the ball at the spot of the kick (or the 20). First-down expected points come from nflverse's EP model, smoothed with a cubic in field position.
Graded only with the offense's win probability between 10% and 90%, outside the last 5 minutes of the fourth quarter, the last 2 minutes of the first half and overtime: 1,296 decisions so far. Clear call: best option ahead by 0.75 expected points or more; lean: 0.25 or more. Chicken Index = share of clear or leaning go calls answered with a punt or a kick.
5 · Garbage Time Tax
Garbage time = offense win probability before the snap below 10% or above 90%. Passing yards count completions; sacks are left out.
6 · Our Receipts
Our pick is the team-strength model behind the playoff odds (half point differential, half net EPA, regressed by games played, plus home field), without the betting line. From Week 10 every pick is frozen in our database before kickoff. Weeks 2-9 are rebuilt using only the results known before each week. The Brier score is compared with the closing line from nflverse.
Use it
Every table has a CSV (luck, games, flags, fourth downs, garbage time, receipts) and a JSON feed at /api/verdict. Free to use with a link to the page you took it from. The Luck Index is also a free widget.