FAQ

The basics

What is the v25?

A college football Top 25 set by the betting market instead of poll voters. We take the point spread on every game, work out the single power rating per team that best explains all of those spreads at once, and rank the FBS teams by it.

What does a rating like +28.8 mean?

It's how many points a team would be favored by against an exactly average FBS team on a neutral field. It is a margin of victory, not a score: +28.8 doesn't mean the team scores 28.8 points; it means the market would expect it to win by about 29.

Ratings are on one scale, so you can compare any two teams by subtracting. Notre Dame (+28.8) vs. Texas (+25.4) on a neutral field: Notre Dame by about 3.5. If the game is at Notre Dame, add the home edge of about 2.6: Notre Dame by about 6. The Matchup box on the home page does this for you.

What does “range #8–23” mean?

How sure the market is about where a team ranks. Lines don't pin a team's strength down exactly, so every rating has a margin of error (typically 2–3 points). We re-rank all 138 FBS teams 4,000 times, nudging each rating randomly within its margin, and report where the team landed in the middle 90% of those runs.

A narrow range (#1–4) means the market has a firm view. A wide one (#8–23) means the team is in a crowded pack and the lines can't separate it from its neighbors yet. Teams just outside the top 25 also show their chance of really belonging in it.

Why is the v25 so different from the AP poll?

They answer different questions. Poll voters mostly reward what a team has done: record, big wins, losses. The market prices how good a team is going forward. An unbeaten team that keeps winning close games against weak opponents can be a top-10 AP team and a #25 v25 team, and both can be “right” by their own standard.

Example: when the AP had BYU at #8 at 4–0, every line on BYU's games priced them at only +8 to +14, alongside teams like Washington and Iowa, and the v25 had them #26.

How often does it update, and what are the weeks?

Every 6 hours, and hourly on Saturdays to catch closing lines. Weeks are numbered like the AP poll: v25 Week 6 is the ranking after the same games as AP Week 6. Use the arrows on the home page to flip back through earlier weeks.

Betting lines & big bets

Which lines do you use?

Point spreads from 11 sportsbooks that take bets from U.S. customers (listed below), recorded all week long. For each game we take every book's last spread before kickoff (the closing line, typically about an hour before the game) and use the median across books, usually 9–11 of them. The books rarely disagree by much: the typical gap between the highest and lowest line on a game is 1 point.

Which sportsbooks?

The same 11 in every sport:

• State-licensed U.S. sportsbooks: DraftKings, FanDuel, BetMGM, Caesars Sportsbook, BetRivers and Fanatics Sportsbook.
• Offshore books with U.S. customers: Bovada, BetOnline.ag, MyBookie.ag, BetUS and LowVig.ag.

We read their lines through a commercial odds feed every hour. Not every book posts every game, so most games have 9–11 of them. Because we use the median, no single book (licensed or offshore) can move the consensus on its own. For baseball we use the same books' moneylines.

We aren't affiliated with any of them and don't take bets or link to them.

Why closing lines?

The closing line is the market's final answer, after all the news, injuries, weather, and money have come in. It's the standard every serious prediction model is measured against, and few manage to beat it for long.

Do you use the amount of money bet, or the percentage of bets on each side?

No. The line already is the result of that money: books move the number until they're comfortable with the action on both sides. Bet counts and money percentages measure public sentiment, not team strength. They're also only published by a few books, and inconsistently.

You can still see the money in the line movement. Spreads move about 1.2 points on average between opening and close, and about 1 game in 9 moves 3 points or more. We use the number after all of that movement.

What would a single $1 million bet do?

Almost certainly nothing to the v25, and it depends on whether the rest of the market agrees with the bet.

• Only one book moves. Very few books will take a single bet that size on a college game. The ones that do usually move their line in response. But we use the median of about 10 books, so one book moving doesn't change the consensus. We tested this on every game this season: moving any one book's line by 1, 3, or even 10 points left our consensus unchanged 92% of the time, and never moved it more than ¾ of a point.
• The whole market moves. If other books and bettors agree the bet was smart, the consensus moves, and the rating should move with it, because the market has changed its mind. Even then the effect is small. A full 1-point move on a team's most recent game shifts its rating by about 0.25 points. On a game from the start of the season it's only about 0.06, because every other line on that team still counts.

So no single bet, bettor, or book can push a team up or down the v25. Only a lasting, market-wide change of opinion can.

The model & its weights

How do lines become ratings?

Every game's line is treated as a measurement: home rating − away rating + home edge ≈ line. With about 450 games connecting 235 teams (FBS and FCS), there's one set of ratings that fits all of those equations best, and that's what we solve for. A team's rating comes from its own lines and from its opponents' lines, its opponents' opponents', and so on through the whole schedule.

Are all games weighted equally?

No. Three things set how much a line counts:

• Size of the line. A 3-point line is precise; a 45-point line on an FBS-vs-FCS mismatch is a rough guess. Lines on blowouts count less:

• Recency. Each team gets a rating for every week, and ratings are allowed to drift (about 1.6 points a week, estimated from the data). Old lines still count, but recent lines matter far more for where a team stands now: about four times as much as an opening-week line for a typical top-25 team. A bye week doesn't cost a team anything.
• A starting guess, early on. Before a team has many lines, its rating starts near its conference's average, and conference averages start near the FBS average. Once the lines come in they quickly take over. By midseason this starting guess barely matters.

How is home field handled?

One home-field value for every team, estimated from the lines themselves: currently 2.6 points. Neutral-site games (bowls, kickoff classics) get no home edge. We don't give teams individual home-field values; with about four home games a season there isn't enough data to tell a real difference from noise.

Do scores matter?

Not directly. The lines after a game already reflect how the market read that result: a blowout win over a good team moves next week's line, while a lucky one-point escape often doesn't. Feeding scores in again would count the same information twice and add noise.

Where do the margins of error come from?

From how well the lines fit. The model measures how far real lines land from what the ratings predict (about 2.7 points for a close game), works out how precisely that pins down each team, and runs the 4,000 simulations behind the ranges and top-25 chances. The simulations are seeded, so the same data always gives the same ranges.

Does it work?

How do you show it's accurate?

A rating system makes a testable claim: “Team A is N points better than Team B.” We test it two ways, always on games the model didn't see:

• Hide one game, predict its line from every other game. Typical miss: 4.5 points (4.0 in close games).
• Predict next week's lines using only games already played. Typical miss: 5.4 points (4.9 in close games).

For comparison, our first version missed by 9.0 and 12.5 points on the same tests, and the best of several simpler alternatives missed by 4.9 and 5.7. The current model beat them all, and the gaps are statistically significant, not luck.

Is it better than models that add more variables?

We aim at a different target. Most power ratings (SP+, FPI, Sagarin and others) build their own forecast from stats: yards per play, recruiting, returning starters, results. The betting market already absorbs all of that, plus injuries, weather, and what the sharpest bettors know, with real money behind every adjustment. That's why closing lines are the benchmark those models are judged against, and why beating them consistently is so rare.

The v25 doesn't try to out-predict the market. It shows you the market's own view, translated into a Top 25. Adding variables would make it our opinion instead of the market's. The honest claim is: if you want to know what the people putting money on the line think, this is it, with its uncertainty shown.

When the v25 and the AP disagree, who's right?

Early scorecard, through Week 5: in 72 FBS games where the AP ranked at least one team, the two polls agreed on the better team 67 times, and that team won 57 of them. They disagreed 5 times; the v25's pick won 3, the AP's won 2.

That's far too few games to prove anything yet; a fair verdict needs a full season or more. A single game is also a noisy test: even a 7-point favorite loses about a quarter of the time.

Other sports

Which sports are covered?

College football, the NFL, college basketball, the NBA and MLB. Pick one with the buttons at the top of the page; the site opens on whichever sport is in season, in that order. A sport's first ranking appears after its opening week or two, once enough games connect every team.

Why the v32 and the v30?

College sports have hundreds of teams and a Top 25 poll to compare with, so they get the v25. The pro leagues are ranked in full: all 32 NFL teams (the v32), and all 30 NBA and MLB teams (the v30).

What are pro teams compared with, if not the AP poll?

Their record. Each team's market rank is set against where its regular-season record ranks in the league. “Better than their record” lists teams the market rates well above what their wins and losses say, which usually means close losses or a hard schedule; “Worse than their record” is the reverse.

How is baseball different?

Baseball's spread is almost always 1.5 runs, so the opinion lives in the price instead. MLB uses moneylines: we strip out the bookmaker's cut to get each team's win chance, and the model works with those. An MLB rating is the chance of beating an average team on a neutral field: 60% is a strong team, because even the best baseball teams lose about 4 games in 10.

Starting pitchers move single-game odds a lot. A team's rating is its strength across its whole rotation, not for any one start.

Is it as well tested as college football?

Not yet. The model and the tests in this FAQ were built on college football. The other sports use the same method with settings chosen for each league (home edge is estimated from the lines; how quickly ratings drift is set by hand), and they'll get the same out-of-sample testing as their seasons build up data.

Limits

What doesn't the model know?

• Injuries, weather, suspensions: nothing separate. The lines already price them in, which is the point.
• The price on a line: we use the spread but not the juice. A −3 at −120 is a slightly stronger opinion than −3 at −105, worth a fraction of a point.
• Preseason: Week 1 has no v25, because there aren't enough lines yet to connect all the teams.
• Crowded middle: teams ranked about #16–30 are usually within a couple of points of each other. Treat their exact order as a near-tie, and check the ranges.

Is this betting advice?

No. The v25 describes what the market already thinks; by design it won't find you an edge against that market. It's an independent analysis for entertainment, not affiliated with any sportsbook or the Associated Press.