Best Bets
Record
Every pick
Newest first. Graded at the line and price each pick was posted at.
Reading this page
Spread picks start from the gap between the model's line and the market's middle line (the median across US sportsbooks). The backtest showed most of that gap is the model's own error, so only part of it is trusted: about 17% in Power 4 games and less in Group of 5 games, after a quarterback change, and early in the season. Edge is the trusted part, and a pick needs at least 0.5 points. The pick shows the best line available for that side.
Underdog moneylines were recorded through October 11 as an unofficial record and then stopped. Once the model trusted only part of its gap from the market, the moneylines it still liked came from how it turns a spread into a win chance, not from anything it knew. The ones already posted are still graded.
Model v2 made the week 7 picks posted before the switch to v3 on October 11. They're kept and graded, but shown separately and not counted in v3's record.
Line move (CLV): how far the market moved toward the pick between when it was posted and kickoff. Beating the closing line is a faster, less noisy sign of a real edge than wins and losses. For moneylines it's the change in the market's implied chance, in percentage points. Until a game's closing line is in (usually a day or two after the game), the move shown is so far, up to the latest odds snapshot. After that it's the move to the close, from CollegeFootballData's closing lines, compared book by book with the same sportsbooks' lines when the pick was posted. The record counts only moves to the close; a pick with no sportsbook to compare is left out.
A pick locks the moment it's first posted (usually Sunday night), because that's when it would be bet. Later snapshots only track where the line went.