Methodology

One model. One stored run behind every number. Here's how it works and where it can be wrong.

Seasons simulated
10,000
Every remaining game, every run
Paired runs per swing
2,000
Same seeds, each side winning
Playoff field picked
89%
Committee model, hidden seasons
Brier · 2025 Week 13
0.029
Playoff odds, lower is better

Ratings

Every team gets one number: how many points better or worse it is than an average FBS team on a neutral field. A team rated +14 would be a two-touchdown favorite over an average opponent. The rating is fit from every final margin this season, so each result is judged against who it came against.

Home field is worth 2.5 points. Margins are capped at 28, so running up the score on a bad team doesn't buy extra credit. Before a team plays, its rating starts from last season's, pulled 25% back toward average. That starting point loses weight with every game and counts for only 20% by game 8, when this season's results do most of the talking.

Simulating the season

Every game left on the schedule, conference title games included, gets played out 10,000 times. In each game, the rating gap plus home field sets the win chance, and a random draw decides who wins. Your odds for any goal are the share of those 10,000 seasons where it happened. 4,300 out of 10,000 reads as 43%.

Ratings can be wrong, so each simulated season gives every team a hidden shock of about 6 points that it keeps all year. A team the model has rated too low gets a chance to be good across the rest of its schedule, not just in one game. That shock shrinks as a team plays more. Conferences get a shared shock too, 6.3 points, because when a league is overrated, it's usually the whole league. That number held up when we tested it on seasons the model hadn't seen.

Swing, lean, wash

Shown odds changes compare each result with your odds now. At 67.1% now, 72.6% after a win is Win +5.5 pts; 37.0% after a loss is Loss −30.1 pts. A point means a percentage point of goal odds, not a betting spread. The full gap between outcomes stays the internal sorting metric.

Stakes: Season-defining at 10+ pts, Big at 3–10, Matters at 1–3, Barely under 1.

Comparing two separate batches of random seasons would bury small swings in noise. Instead, we play 2,000 pairs of seasons from the same random seed, identical except for who wins this one game. Everything else cancels out, which leaves the effect of that game. Each swing also carries a margin of error from those pairs.

A game makes your Sweat Sheet only when its swing is at least 0.2 points and more than twice its margin of error. A lean is smaller but clearly points one way. A wash means the pairs can't tell, and we say that rather than guess. Games against FCS teams count only when the upset chance is at least 10%.

The committee

Making the playoff means getting picked, so the simulator needs a committee. Ours learned from 12 seasons of real rankings, 2014 through 2025, by studying which of two teams the committee put higher, thousands of times over. It looks at record strength, schedule strength, Q1 and Q2 wins, top-25 and top-10 wins, blowout losses, Group of Six status, conference title games and our rating. Recent seasons count more: 2025 three times, 2024 twice, and 2020, the COVID season, about a third.

Some rules sit on top: head-to-head can swap neighbors, an idle team tends to hold its spot, and conference champions get their automatic bids. There are separate versions for the first ranking of the year and for week-to-week moves. Late-season losses were tested as an extra penalty and dropped, because they made predictions worse.

Shown seasons it never trained on, it picks 89% of the playoff field and 92% of the top 12, missing by 2.0 spots per team on average. The AP poll released on Selection Day does better (94% and 97%), but that poll comes out after the committee's final ranking, so it's not a fair head-to-head. Each simulated season also adds about 2 spots of random committee noise, the size of the model's misses on held-out seasons.

Goals outside the playoff race

Not every team is chasing a bracket spot, so every simulated season also counts regular-season wins and conference title-game trips. Make the conference championship means a top-two finish in your league. Make a bowl means six or more wins. Outperform last year means more wins than in 2025.

Only your own games move these goals; another team's result doesn't change how many games you win. When your current goal falls under 2%, Our Goals switches to whichever remaining goal is most in doubt, so you always have something real to sweat.

Mood Tracker

Mood Tracker scores how much one Saturday changed each fanbase's season: playoff odds gained or lost, plus whatever rivals and enemies did. Zero means a wash of a weekend.

The six words run from Miserable to Ecstatic. Instead of fixed cutoffs, they're set from this season's actual scores: everything below zero is split into thirds, and everything above zero is too. A "Thrilled" means that weekend ranked among the better ones this season, not just that the number was positive.

How we tested it

We rewound to 2025 and ran the full model as of Weeks 5, 9 and 13, using only games, polls and ratings available at the time. Then we scored its playoff odds against the field the committee actually picked.

Wk 5

0.044

Wk 9

0.037

Wk 13

0.029

That's a Brier score: the average squared gap between the odds we gave and what happened, so lower is better. Giving every team the same 1-in-11 chance would score about 0.079. The model got sharper each checkpoint, from 0.044 in Week 5 to 0.029 by Week 13.

One weak spot: teams we had above 90% made the field 5 of 7 times. The top of the board can run a little hot, so treat a 95% as very likely, not a lock.

See the receipts →

Data and blind spots

Schedules, scores and polls come from public results data. We refresh every 6 hours during the week and every 10 minutes on Saturdays, and the simulations rerun when something changes. Betting market odds come only from FanDuel.

Conference tiebreakers follow each league's own order. Where the ACC and Big Ten fall back on an outside computer rating, we use ours instead.

What it can't know: injuries nobody reported, weather, a quarterback benched at halftime, or what a team looks like on film. The model works from results.

How the Two-Deep works

Three kinds of source. Listed is the school's own depth chart from its weekly game notes or depth chart page. Played is who actually got the work the last three games: carries, targets, tackles, sacks, breakups. Status comes from the official report (SEC, Big Ten, Big 12 and ACC availability reports), from injuries reported in the news (always marked as reported, never official), or from our editor on show days.

Who wins. Editor beats Listed, Listed beats Played. For status, Editor beats the official report, which beats news, which beats "missed the last game". When the chart says one thing and the stat sheet says another, tap the player to see both. A true rotation, where two players genuinely split the work, shows as ⇄ OR in the depth chart.

Next man up. A player who's out, doubtful or out for the season comes off the chart for that game, and the next man in his spot moves up in his place. If the spot runs dry, we take the best available player at that position. Questionable players stay put with an amber ring, and their card names who'd replace them.

Soft calls. A dashed ring means we're not confident: two players sharing the work, or not much evidence yet.

The O-line. There are no public snap counts, so when a school doesn't publish a chart, the offensive line comes from the roster and experience. Treat it as a soft call.

Refresh. Official reports: every 2 hours Wednesday to Friday, every 30 minutes on Saturday until an hour after the last kickoff, once Sunday. School depth charts: once a day per school Monday to Thursday until we find this week's. News: daily. Some schools don't publish a depth chart at all; those are built from who's actually playing, and we say so.

No headshots, ever: players are their jersey numbers in team colors. And no sportsbook other than FanDuel shows up anywhere.