Week 3, 2026 college football picks

Every game we have a call on this week — the side we pick to win and our probability. The games we ran and decided not to call are here too, marked, so you can see what we left alone and why. Kickoffs run Thursday, September 17 through Saturday, September 19.

games we call

we ran and left alone

settled so far

66-9

record, all season

88%

picks correct

The Sweat Meter

Every call carries a word for how nervous to be about it. The word is our number translated: Nap-safe at 95 or better, Cruise control at 85, One eye on it at 75, Full sweat at 65, Coin flip with a lean at 55. This week, by tier:

10Nap-safe
10Cruise control
12One eye on it
6Full sweat
11Coin flip with a lean

Where the line moved

We picked a side when the board published. Since then the betting line has moved toward some of these picks and away from others. Tap a number to see just those games. The number we published is still the number that gets graded.

49 of 49 games

Thursday, September 171game

Syracuse Orange at Pittsburgh PanthersNo call
We ran it and decided not to publish a call.

Friday, September 182games

Saturday, September 1954games

9 where the line moved our way, 15 against, and 25where it hasn’t. Those counts are the whole board, locked rows included. Pro opens the picks behind them before kickoff, and every pick opens to everyone once it grades.

How to read this board

Miami Hurricanes, 83% to win. Each row is one call: the team we pick to win and how likely we think that is. The probability is the pick — it is what gets graded, and a 68% call that loses is a 68% call that lost, not a bad model.

One number per row, and it is the probability. We do project a margin for each game, and it is on the game page where there is room to label it — but it is context, not a call, and nobody is graded on it. Where the line moved says whether a betting line came toward a pick or away from it. We never publish the line itself, or anyone else’s number.

A short board is the honest one. We publish only where our model clears its gates, so a week with a lot of close games has fewer picks on it — the games we leave out are the ones we would be guessing at. Those games are still listed, greyed out, and each one has a page saying which gate stopped it. A pick we did not make is part of the record too.