Upset Watch
Five games get flagged this week. Two are genuine crowd/market oddities — Oklahoma–Kentucky, where 54% of the public picked the underdog against a 6.5-point line, and UNLV–Air Force, a true coin flip where the second-opinion model and the moneyline both lean the other way from the crowd. The other three — Ohio State–Indiana, Duke–North Carolina and Wisconsin–UCLA — sit under a 60% model number while the crowd is 74–86% on the favorite; UCLA is the sharpest of those, with the public on a team that just lost its starting quarterback for the season.
Every Pick’em Matchup
Methodology
Same build as every week: the closing point spread as the primary input, not the moneyline. Decades of published accuracy studies on football betting markets find the point spread to be the single best-documented predictor of a game's outcome, consistently beating standalone power ratings and outperforming moneyline-implied probabilities, which are thinner markets that skew more with “favorite-longshot bias” (sportsbooks tend to shade extra vig onto lopsided favorites, which inflates their moneyline-implied win probability relative to what the spread says). We researched the actual closing or current spread for all 10 games; each one becomes the mean of a normal distribution of final margin, with a standard deviation of 16 points — the rough historical spread of FBS scoring margins, widened slightly for the largest lines to account for extra garbage-time variance — and we run 100,000 simulated outcomes per game. The result is a win probability, an upset probability, and a projected score built from that margin and a researched over/under total. The contest's own moneyline is still shown on every card, since that's the number Pick’em players actually saw.
Where we could find one, we added an independent cross-check — Dimers' published computer model for nine of the ten games, plus a prediction-market (Kalshi) read on Missouri–Ole Miss. Every game had one this week, and nine of ten agree on direction with the spread-based model. The exception is UNLV–Air Force, where Dimers has Air Force at 50.1% — a technical flip of the thinnest possible kind, which is why it carries a Trap tag even though the crowd and the spread both lean UNLV. Notre Dame–BYU is the only game where ESPN's FPI and SP+ were also available, and both side with Notre Dame by a wide margin.
Every card compares what the crowd actually picked against what the spread-based model says, and the gap between those two numbers is its own signal — a market can be efficient while a betting public is still overconfident in a name-brand favorite. The average gap this week is about 14 points, pulled down by the one game where the crowd sides with the underdog outright: Oklahoma–Kentucky (−20). Strip that one out and the other nine favorites are still being picked harder than the model supports, by roughly 18 points on average — the widest single gap is Wisconsin–UCLA (+30), the game where the public hasn't priced in a quarterback change.
Win probability and public sentiment don't explain why, so every card also carries 2026 season context researched separately: records through Week 6, head-coaching notes, quarterback performance, and each team's form and roster strengths and weaknesses entering the game — from UCLA's season-ending quarterback injury to Air Force starting a sophomore who ran for more than 110 yards and threw one pass in his first start.
Kickoff times and TV windows reflect the games as scheduled as of Oct 15, 2026. Lines and public picks move between now and kickoff — treat this as a snapshot, not a guarantee.