What We've Measured
Every number used elsewhere in this section comes from here — our own archive of FBS results joined to closing lines. The angles that didn't work are published too, because a page that only showed the wins would tell you nothing about whether we checked.
Where college games actually land
7,560 games. The spikes are why a half-point is not always a half-point: a line moving through 3 changes far more results than one moving through 9.
Median total in an FBS game: 54 points. The home side wins 58% of the time.
Is the market beatable on its face?
The first thing to check, and the answer is no. If either column sat far from 50% there would be free money in simply always taking one side. Sitting at 50% is also how we know the archive is joined correctly — a sign error would show up here as a fake edge.
| Check | Games | Hit |
|---|---|---|
| consensus — home covered | 6,062 | 48.9% |
| consensus — over hit | 6,062 | 48.9% |
| DraftKings — home covered | 2,378 | 50.6% |
| DraftKings — over hit | 2,378 | 50% |
Angles we tested and killed
Four popular theories, each measured against closing prices. None of them beat the number. We publish them because the alternative — quietly dropping what failed — is how a section ends up looking cleverer than it is.
Backing the side SP+ disagreed with the market on. Break-even is 52.4%. It failed in-sample too, and the −4.8% ROI is almost exactly the cost of paying the vig at a coin flip — so the disagreement carries essentially no information. The market has fully priced a free, public rating.
Windy games really do score about 4 points fewer than calm ones — the mechanism is real. But the market moves its total to match, and backing the under lost at 12 mph (−3.4%), 14 (−11.1%) and 16 (−6.8%). One band at 18+ mph looked profitable on 52 games; the bands either side of it failed, which is what noise looks like.
Games with 5mm+ of rain. Worse than a coin flip, −9.0% ROI.
Sea level 50.0%, 1,500m+ 48.6%. One middle band cleared break-even on 85 games, but the band above it — where thin air should matter most — failed. A real effect strengthens with the dose; this one doesn't.
Tested, but the spots are too rare to call
Not everything resolves. These are angles where the measurement came back inconclusive rather than negative — and saying so is the honest answer, because an underpowered test that lands below 50% is exactly how a confident-sounding system gets built out of nothing.
The most-repeated angle in college betting — a West Coast team at a noon Eastern kickoff. We mapped every archived game to its stadium, so we can measure the visitor's exact travel distance, timezone shift and what time their body clock reads at kickoff. Visitors travelling two or more timezones east into a body clock before 11am went 44.7% against the spread, which is the direction the story predicts. It is also 76 games. Detecting a true 45% edge at 80% power needs about 780 games, and 47% needs 2,175 — so a decade of closing lines is an order of magnitude short of being able to tell this apart from noise. The well-powered parts of the same question are flat: distance travelled is a coin flip in every band out to 2,000+ miles, and so is every timezone shift.
What survived
Not a model — the numbers above say this market is priced efficiently, and we won't publish picks we can't back. What survived is context that is true by construction: the key-number weighting on line movement, the hold on every futures market, and game-day conditions. If we ever do publish a college pick, it will have cleared the bar on this page first.
The calibration behind all of it — cover rates by spread size, over rates in blowouts, and how the public ratings compare to the closing line — is on The Numbers.