How it works
1. Ratings adjusted for schedule. Every FBS team gets an offense rating (points scored against an average defense) and a defense rating (points allowed against an average offense), worked out from every game this season and who it was against. Scoring 30 on a top defense counts for more than scoring 30 on a bad one, and a 56–3 loss to Ohio State counts as roughly expected. Ratings are pulled a little toward average, because six or seven games is a small sample.
2. Projected score. A team’s projected points are its offense rating plus how much more or less than average the other team’s defense allows, plus half the home field (2.5 points by default) for the home team.
3. The gap. The two projected scores give a model line and a model total. The gap is the model line minus the sportsbook line, in points. A gap of 7+ points is flagged as a bet to consider, on the side shown in that box, at the book’s number.
4. Line movement. The board can’t see injury reports or quarterback changes, because ESPN doesn’t publish usable ones for college football. Instead it compares the opening line with the current one. If the line has moved 2+ points toward the other team since it opened, that usually means news, so the pick is left out by default.
Plain averages (the clip’s method) are still available in the settings. Averaging what a team scores with what its opponent allows cuts every predicted margin in half, so that method leans toward big underdogs.
Not modeled: weather, motivation, rivalry and lookahead spots. A big gap is a reason to look closer, not a sure thing, and this board has no track record yet.
How it works
Gaps. Each row puts a player’s average per game next to what tonight’s opponent allows per game, and highlights the big differences. For quarterbacks it’s a straight comparison, like the clip (19.5 completions vs. 17 allowed). A single receiver or guard only gets part of what a defense allows to his whole position, so for everyone else the opponent’s number is scaled to his role: a defense that allows 20% more receiving yards to receivers than average shows up as a 20% gap. “Include last season” off uses this season’s games only, for both the player and the opponent. “Top-usage teammate out” (NBA) keeps players whose team is missing one of its top two by plays used per game; long absences don’t count. In past seasons, rebound and assist overs for players who moved up their team’s usage order because a teammate sat hit 50–53%. In a player’s page, drag the line under his last 20 games to see how often he cleared any number. NBA cards also tag tonight’s spread: Favored when his team is favored, and 7–13.5 spread when the game’s spread is 7 to 13.5 points either way.
Odds turns the same numbers into a projection and compares it with Kalshi’s prices:
1. His number. The player’s average over his last games, carrying over from last season early in the year.
2. The number allowed. For NFL quarterbacks, exactly the clip: the defense’s completions allowed per game and yards allowed per completion. For everyone else, how much this opponent gives up to his position compared with the league, applied to his average. A defense that allows 8% more points to guards turns a 25-point average into 27.
3. The projection. The average of the two, the same step as the clip. QB passing yards are averaged completions times the defense’s yards per completion (19.5 and 17 completions at 12.5 yards = 228). The matchup weight setting changes how much the opponent counts.
4. Model chance. His game-to-game swing turns the projection into a chance of clearing each line.
5. Edge. Model chance minus the price. Kalshi prices are already chances (62¢ = 62%), plus its trading fee. The sweet spot is each player’s line with the biggest edge.
Over score (NBA points, rebounds, assists). Built from a study of past props: 18,260 real 2022-23 lines, 6,563 US-book lines from 2023-26, and book-style lines for every prop in the last two seasons. The biggest piece is where the line sits against his season average. Until he has played 10 games this season, last season is blended in as 10 extra games, as long as he played 20+ games last season; rookies and players who missed most of last season use this season’s games alone. +2 when the line is 10%+ under and +1 when it’s 7–10% under (lines only 0–7% under went over no more often than lines at his average, so they earn nothing); −1 when it’s 3–10% over and −2 when it’s 10%+ over. Books lean on recent games, and players drift back to their season numbers. Then +1 for points when teammates averaging 20+ ppg are out (and no penalty for a raised points line then: books raise it for the extra shots, and those overs held up), +1 after a 15+ point loss, and −1 each for a starter in a 14+ point spread, a playoff game, a player back from missing 3+ games, or a cheap line that comes with fewer minutes. A second round of the study added +1 for assists when teammates averaging 5+ assists combined sit, +1 for rebounds and assists when he moves up his team’s usage order because a teammate sits, and +1 each for points and assists after a tough-run, soft-tonight schedule swing (his last 5 opponents were stingy to his position, tonight’s is generous) or a 3+ game losing streak. In past seasons, props scoring +4 or more hit 63% (58–69% by data set), +3 hit 54%, +2 hit 53%, +1 hit 50%, 0 hit 48%, and −1 or lower hit 44%. The signals were picked and tested on the same seasons, so expect live results a little lower. The line used is Kalshi’s 50/50 line when it has one, otherwise where books would likely set it; type your book’s line in a player’s page to score it exactly.
A+ Sweet Spots (NBA). Rebounds and assists: the line sits 10%+ under his season average. Points: his team’s top scorer is out and he isn’t just back from a 3+ game absence. Both also need him to be one of his team’s top two players by plays used per game (among teammates not listed Out), a spread under 14, and a regular-season game. These picks hit 56–58% in past seasons, about 8 a night (points picks 58–60%, even when the book raised the line 10%+ over his average). The A+ list is ranked by past hit rate: Tier 1 is an over score of +4 or more (72%), Tier 2 assists and points (58%), Tier 3 rebounds with the line 15%+ under his average (57%), and Tier 4 rebounds 10–15% under (53%, barely above break-even at −110). Recent hot streaks didn’t help: A+ picks that had cleared the line in 7+ of his last 10 hit 57%, no better than the rest. The old version (high total, close spread, beat his projection in 4 of his last 5) hit 46–47%, because hot streaks push lines up.
Spread spots and one lower (A+ cards). Each A+ card shows how past A+ picks did in the same spot on the spread: favored by 7–13.5 points hit 62% (677 picks), favored by under 7 hit 57%, underdogs hit 55% either way, and 14+ point spreads hit 47%, which is why A+ skips them. Close games did no better than the rest. The card also shows one number under the line: past A+ picks cleared it 75% of the time for assists, 72% for rebounds and 65% for points. One lower beats the main line at −110 when it’s priced −199 or better for assists (−217 rebounds, −136 points), and loses money past −292 (−251, −183).
Reading an A+ card. The big number is the likely hit rate: how often past A+ picks like this one hit (same stat, side of the spread and spread band, plus the matchup for rebounds), blended toward the stat’s average when the group is small. Tier 1 picks use the Tier 1 rate the same way. The dark bar is the bet. The five icons are, in order: a line 10%+ under his average (for points, his team’s top scorer out), top-two usage, his team favored, a 7–13.5 point spread, and the opponent allowing more to his position (rebounds only; it made no difference for assists or points). Green means yes, grey no, orange a warning. Tap a card for why it’s A+, his recent games against any line, one lower and the over score. Analysis shows how each filter moved the odds, his spot against the others, the over score’s pieces, and a price check for your book’s odds.
Late news (NBA). ESPN’s injury report puts game-time decisions on the cards, with tonight’s designation from ESPN’s note (probable, questionable, doubtful) and any minutes limit. In the hour before each tip the board checks every 5 minutes instead of every 20. The Late news box lists status changes for rotation players (15+ minutes a game) in games that haven’t tipped, and the A+ picks each one added or dropped, plus game-time decisions that would move A+ picks (if a team’s top scorer sits, his teammates’ points picks can turn A+). Changes that land while the board is open pop up at the bottom of the screen. “Alert me” also sends a browser notification while the board is open in a tab: reliable on a computer, but phones pause background tabs after a few minutes, and iPhone browsers don’t offer these alerts. ESPN doesn’t post starting lineups before tip, so the board can’t confirm starters. Once a game tips, its tags freeze with its lines.
Not modeled in the Odds projection: injuries, minutes changes, rest and blowouts (the NBA over score covers some of these). Check news before betting, and treat edges over 10 points with suspicion: the market usually knows something the averages don’t.