How to read a game card
- Line is the betting spread from ESPN. The small gray number is where it opened, so you can see how it moved.
- Model is the predicted final margin. Pick is the side the model thinks covers the line.
- Edge is how many points the model disagrees with the line by. It's labeled like a fourth-down call: Go for it is 3–5 points, Lean is 1.5–3, and Punt is less, meaning no bet. A gap of 5+ points says Check the news: in testing, gaps that big did no better than smaller ones, and they usually mean the market knows something the model can't see, like an injury or a quarterback change.
- Cover chance is the model's chance the pick covers. During a game it updates with the score and clock. The yellow tick on the bar marks 52.4%, the break-even point at standard -110 odds.
- The Track button saves a pick to My picks, where it's graded automatically when the game ends.
- Spread / Total switches every card between the model's pick against the spread and its over/under pick against the total. Totals use two projections: Market-anchored (the default) starts from the betting total and adjusts it a little toward what the teams' scoring says; Adjusted points adds each offense's points against the other's defense.
- At open shows edges of 1.5+ where the line hasn't moved more than half a point from where it opened. In testing, the model did best at the opening number; once the market moves toward a pick, most of the value is gone. Pick boxes say how far the line has moved toward or away from the pick. Next week's lines usually post on Sunday, and a notice appears when they do.
- Close shows live games within 4 points of the spread, where a field goal could flip who covers. They're sorted tightest first and tagged on every card.
- Tap a game for the full breakdown: what every projection says, how the line has moved, the matchup, weather, and how the game played out.
Player props enter your book's line to see the over/under chance
How these projections work, and how good they are
Each projection is the player's per-game average this season (games where he had a real role), using ESPN box scores. Rushing yards, passing TDs and receptions are nudged by half of how much this week's opponent allows compared with an average defense. Enter the line and odds from your sportsbook, and the page estimates the chance of going over or under, and leans only when that chance beats the odds' break-even by 3+ points.
Untested against real lines. Free historical prop lines don't exist, so these projections have never been graded against what sportsbooks actually offered. What we can check is how close they come to what players actually did, predicting each week from earlier games only (2026, weeks 4–8):
| Stat | Typical miss | Plain season average | Player-games |
|---|
Sportsbooks know about injuries, depth-chart changes and weather; this doesn't. A big gap between your line and the projection usually means the book knows something. For entertainment and research, not betting advice.
Updates everything we've changed, newest first
Model updates change how picks are made, so they're dated and explained. Picks already in the forward-test log are never changed afterward.
Team ratings the numbers every pick is built on
How to read the ratings
What each column means
- #
- The team's rank by FPI, from best (1) to worst.
- FPI
- ESPN's Football Power Index: how many points better (+) or worse (−) the team is than an average FBS team on a neutral field. A 10 would be expected to beat an average team by 10.
- Offense
- Points the team scores per game against an average defense. It's adjusted for who they played, so 40 points against a top defense counts for more than 40 against a weak one. Higher is better.
- Defense
- Points the team allows per game against an average offense, adjusted the same way. Lower is better.
- Net
- Offense minus defense: how many points per game the team outscores an average opponent by. Green is better than average, red is worse.
- Off EPA
- Expected points added per play on offense, against an average defense and adjusted for opponents. It measures how well a team moves the ball on every play, so a fluky defensive touchdown or a garbage-time score counts for little. +0.10 is good; higher is better. From CollegeFootballData, with garbage time left out.
- Def EPA
- The same, allowed per play on defense. Lower is better.
- Games
- Games played against FBS teams so far. Fewer games means less certain ratings, especially early in the season.
How the picks use them
- ESPN FPI projection: home team's FPI − away team's FPI + 3 points for home field (none at a neutral site).
- Adjusted points projection: (home offense + away defense) − (away offense + home defense) + 3.
- Efficiency projection: the EPA per play each offense should get against the other's defense, turned into points (about 67 points of margin per 1.0 EPA per play of difference) + 3.
- Blend: the average of FPI and Efficiency.
- Market-anchored (the default): starts from the betting line and shifts it toward what FPI and adjusted points say.
- The difference between a projection and the betting line is the edge. That's what the Games tab labels Go for it (3–5 points), Lean or Punt, or Check the news when the gap is 5+ points.
Tips
- Search a team before a game to see how it stacks up, or use Compare two teams.
- Sort by Net or FPI (click a column) to see the strongest teams. Click again to flip the order.
- When FPI and Net disagree a lot about a team, the projections will split on its games. That's often a reason to punt.
- Ratings refresh every morning from the latest results. FCS teams aren't rated.
Compare two teams
Click a column to sort; hover a column name for what it means.
Picks are saved in this browser only. Track a pick from the Games tab before kickoff; it's graded when the game ends.
Bets are saved in this browser only. Blank odds count as -110. ·
Live results picks recorded before kickoff
Why line movement is the scoreboard
A win-loss record needs hundreds of picks before it means much: after 50 picks, a 50% picker and a 55% picker look the same, and the margin of error is about ±14 points. Line movement settles much faster. Sportsbooks move a line when sharp bettors bet it, so if lines keep moving toward the model's picks after they're logged, the model is seeing something real. This is called closing line value, and professional bettors track it more closely than wins and losses.
How much the moves matter depends on their size. In the 2023-2026 backtest, picks made at the opening line won about 3 points more often (2.5 for totals) for every point the line later moved their way. So the estimate is 50% plus that much for the average move. To beat the 52.4% break-even, the line needs to move about +0.8 points toward the picks on average (+1 for totals).
Recent Market-anchored picks (edge 1.5+)
Model changes every adjustment to how picks are made, dated
Site and data updates are on the Updates tab.
Backtest past seasons, replayed week by week
By season (opening line, every game)
How this was tested
Every FBS game from week 3 on was predicted using only information available before that week, then graded against the line. At standard -110 odds you need 52.4% to break even. Line moved our way is how often the line moved toward the pick between the open and kickoff, which is the standard sign of a real edge. Market-anchored learns its weights from earlier seasons only, so its first season isn't graded. The ± after a hit rate is its 95% margin of error. Totals are graded the same way against the opening and closing over/under. This is analysis, not betting advice.