Game Picks
Team efficiency (offense vs. defense EPA/play), blended from last season into this one, plus a few explainable situational adjustments — set against the real spread, not a black-box number.
2026 Season
Game Board
How to read these cards
The bar on each card is one number line from the away team (left) to the home team (right): the
dot is the model's projected margin, the tick is Vegas's spread, and
the shaded stretch between them is the gap the pick comes from. A bigger gap isn't a safer pick, though —
in last season's backtest, big-gap picks covered no more often than small ones, so no game here is
flagged as a sure thing. Team ranks (#1 = best) are the model's blended efficiency ratings, which
lean on last season early in the year.
Moneyline is a bet on who simply wins the game. A negative number is the favorite —
e.g. −148 means you'd bet $148 to win $100 (plus your $148 back). A positive number is the
underdog — e.g. +124 means a $100 bet wins $124. Spread handicaps the favorite by a
point margin so both sides are roughly equally likely to "cover" — the favorite (shown negative, e.g.
KC −3.0) must win by more than 3 to cover; the underdog (shown positive, e.g. DEN +3.0) covers by
losing by less than 3, or winning outright. Total (over/under) is a bet on the
combined score of both teams, over or under one number. Each Model pick below is the
side of the spread our model favors: if the model's projected margin beats the spread, it takes the
favorite; if it falls short, it takes the underdog with the points. When the two match exactly, there's
no lean.
Accountability
Track Record
Every game is graded two ways: did the spread pick cover, and did the team the model projected to win actually win? They can be different teams — the model can project KC to win by 1.5 yet pick DEN +3, because 1.5 falls short of KC's 3-point spread.
Educational research project, not betting advice. Picks come from one tunable model's output, not a guarantee of results, and this is not a recommendation to wager real money.
Football Forecast is a living project — models and methods get refined as the season teaches us things, and past results may be re-graded when a fix changes how they should be scored. What you see always reflects the current method.