Gridiron Odds

Pick your team

AFC

NFC

★ high confidence marks picks at ≥65% win probability — in an 11-season backtest those hit at 73%, while all picks together hit 65% (the Vegas closing line manages 66%). ⚠ upset watch flags a favorite missing 3+ starters against a mostly healthy opponent — favorites in that spot win only 62% historically; each game’s panel shows how many points the absences cost vs full strength.

Probabilities come from an Elo rating blended with an XGBoost model using team form, EPA, rest, weather forecasts, and injury reports weighted by each player’s individual rating — EPA per game for offense, box-score playmaking for defense. Numbers reflect data as of the date above and sharpen as kickoff nears — injury reports and forecasts only exist in game week.

Each game also shows an against-the-spread lean: the Vegas closing line (from nflverse, refreshed daily) vs the model’s expected margin, with a cover probability. In an 11-season backtest these leans covered 55.2% against verified closing spreads (breakeven at −110 juice is 52.4%; above breakeven in 10 of 11 seasons). Season-to-date ATS record is tracked live above — a backtest is evidence, not a guarantee, and lines for later weeks post as the season approaches.

Player props are quantile projections (P10–P90) from each player’s rolling usage and efficiency, opponent positional defense, and game context — validated walk-forward at 6–10% better than a player-history baseline. They feed a paper-trading ledger, not bets: a moneyline backtest against 11 seasons of closing lines lost money at every threshold, so props get no real stake unless the paper record clears the vig with a real sample.