College Football / Picks

Why the confidence sits near 50%

We spent weeks building a college football spread model, published its leans and graded them in public. So we ran the test that settles the question. Over 1,017 games, we checked whether our number predicted the final margin once you already knew the closing line:

margin = 0.76 + (−0.084 × our model) + (1.059 × the market)

The market's weight is 1.06, which is what an efficient closing line looks like. Ours is negative. Our rating did not just fail to add anything on top of the closing line, it was very slightly worse than useless. We then swept every setting that could have caused it, six shrinkage values, three blowout caps, three recency weightings. Every single one landed between 46% and 50% against the spread. Not one cleared the 52.4% you need to break even.

So we print the real number. Every game still gets a spread pick, and its confidence is what our testing supports: a fraction of a point from 50%. Our totals number failed the same test, and its first weekend of published over/under picks went 18-31. So over/under picks are paused. We show each game's posted total and keep grading our totals number in private until a clean test says it earns a place back.

Want the thing that actually wins? See the MLB first-inning board →