Glossary

Plain-English definitions of every betting term used in this dashboard. Terms in the UI link here.

Spread: The number of points the favorite must win by for a bet on it to pay.; The point spread is the book’s handicap. Bills -3.5 means the Bills must win by 4 or more for a Bills bet to win; a Dolphins +3.5 bet wins if the Dolphins win outright or lose by 3 or fewer. We write every line as the home team’s expected margin, so +3.5 means the home team is favored by 3.5.
Cover: A team covers when it beats the spread, not just the opponent.; If the Bills are -3.5 and win by 7, they covered. If they win by 3, they did not cover and a Bills spread bet loses even though the Bills won the game.
Total (over/under): A bet on the combined score of both teams being over or under a number.; If the total is 44.5, an Over bet wins when the teams combine for 45 or more points, an Under bet wins at 44 or fewer. Wind and cold push totals down; domes do not.
Vig (juice): The book’s built-in fee. At -110 you risk 110 to win 100.; Both sides of a spread are usually priced at -110. The two implied probabilities add up to about 104.8%, and the extra 4.8% is the book’s margin. It is why a 50% bettor loses money and why you need to win 52.4% of -110 bets to break even.
De-vig: Removing the book’s margin to find the market’s fair probability.; Scale both sides’ implied probabilities so they sum to 100%. A -110/-110 market de-vigs to 50/50. A -120/+100 market de-vigs to about 52.4% / 47.6%. The de-vigged number is what we compare our probability against.
Break-even: The win rate you need at a price to make zero profit. 52.4% at -110.; Our probability must clear break-even, not 50%, before we bet. A pick we rate at 51% at -110 is a losing bet even though we think it is more likely than not.
Edge: How far our fair line is from the book’s line, in points, on the side we like.; If our fair line makes a game Bills -5 and the book has Bills -3, we have 2 points of edge on the Bills. In the NFL, 2 to 3+ points against an efficient line is significant, and most games are a pass. A 5-point-plus disagreement almost always means something is wrong with the inputs, which is why anything over the suspect cap is flagged instead of celebrated.
Fair line (market blend): Our model line blended with the market line: 65% model, 35% market.; The model prices every game from scratch, then the fair line pulls that number 35% of the way toward the current market line. The closing NFL line is the sharpest public number there is, so a model that ignores it drifts 2 to 3 points away on ordinary games and mistakes noise for edge. Blending is the single biggest lever for landing closer to the close; the edges that survive it are the real ones. The pure model line is kept alongside so the inputs can still be benchmarked against the market on their own.
Monte Carlo simulation: We play each game 10,000 times and read probabilities off the distribution.; A single number hides the spread of outcomes. Each simulated game draws a number of possessions for both teams, a good-day / bad-day shock for each, and a touchdown, field goal or nothing on every possession, with probabilities set so the expected scores match the fair line and total. The margins that come out pile up on 3, 7 and 10 the way real NFL games do, so a line of -2.5 versus -3.5 is priced as the big difference it is. The same 10,000 games price any spread, total, alt line, team total or moneyline.
Star tiers: Bigger edge, bigger tier: 1-Star at 1.5 points (2.0 on totals), 2-Star a point higher, 3-Star two points higher.; The tier sets the target stake in units (1, 2 or 3) before quarter-Kelly caps it. A tier table that runs backwards, where 3-Star bets win less than 1-Star bets, is the red flag the previous model raised; it is tracked on the home page and in the backtest.
Units: One unit is 1% of bankroll. We bet in units, not dollars.; Betting a fixed share of bankroll is how you survive variance: at a 55% hit rate you still lose 45% of bets and losing weeks are normal. A 1-Star is one unit, 2-Star two, 3-Star three, each capped by quarter Kelly and by hard limits per bet (3%), per game (4%, because a spread and a total on the same game move together) and per day (12% of bankroll across a slate).
Price ladder (play-to number): The worst line at which each tier still applies. Past the 1-Star number, it’s a pass.; If the fair line is Bills -4.6 the ladder reads “3-Star to Bills -1.0, 2-Star to Bills -2.0, 1-Star to Bills -3.0, pass past that”. Line shopping means taking the best available number inside the ladder; a line that has moved past the play-to number is no longer a bet, however much we liked it an hour ago. Grading records whether the pick-time line was still inside the ladder.
Matchups: OL vs pass rush, passing game vs coverage, run game vs run defense. A tiebreaker capped at ±0.5.; Ratings are additive; matchups are the interaction. A top pass rush against a leaky offensive line is worth a little more than the two ratings alone say. Each component is the product of the two teams’ standardised rates over their last 8 games, scaled so one-sigma against one-sigma is a quarter point, shrunk for small samples, and capped at half a point in total. It breaks ties; it never makes a bet on its own.
Situational angles: Trends such as “home dogs of 3+” or “teams off a bye”. A confidence tag, never more than ±0.5 points.; Every angle has a written reason and is tracked walk-forward against the spread. It earns credit only with at least 100 prior games, and its quoted hit rate is shrunk toward 50% and capped at about 54%, because a trend that “hits 73%” on 40 games is noise. The market already prices the obvious; angles are allowed to nudge the number, not drive it.
CLV beat rate: The share of bets whose line moved in our favour by close. Target 55%+.; Among bets where the line moved at all, how often did it move toward our side? 50% means the market is indifferent to what we see; 55%+ over hundreds of bets means we are consistently ahead of the close, which is the best-known predictor of long-run profit. This is the first thing the roadmap asks to track, because it shows up in weeks rather than seasons.
Closing line value (CLV): Whether the line moved toward your side after you bet. The best predictor of long-run profit.; If you bet Bills -2.5 and the game closes Bills -3.5, you beat the closing line by a point. Over hundreds of bets, consistently positive CLV means you are seeing something the market later agrees with. Win rate bounces around for a whole season from luck alone; CLV shows up immediately. It is the first number on the home page for that reason.
Kelly stake: Bet size that maximizes long-run growth if your probabilities are right. We use a quarter of it as a cap.; Full Kelly stakes (edge divided by odds) are aggressive and assume perfect probabilities. Nobody’s are, so the star tier sets the target in units and quarter Kelly caps it, never more than 3% of bankroll on one bet. Stakes are stored as a fraction of bankroll and shown in units or dollars.
Key numbers: Margins that happen far more than a bell curve predicts: 3, 7, 10, 14, 6.; Because of how football scores, games end by exactly 3 about 15% of the time and by 7 about 9%. Our probability model puts extra mass on those numbers, so a line of -2.5 versus -3.5 is treated as the big difference it really is.
Sigma: How wide our uncertainty is around a predicted margin. About 13 points for NFL spreads.; Even a perfect prediction of the “true” margin misses the actual score by about 13 points on average because football is noisy. Sigma is fit from our own past errors and is never tuned to make bets look stronger; the simulator’s dispersion is matched to it. A smaller sigma would make every probability look more confident than it deserves.
Calibration: When we say 56%, do we win about 56% of the time?; A calibrated model’s stated probabilities match real outcomes. The calibration chart buckets our probabilities and plots how often the pick actually won. Points above the diagonal mean we were under-confident, below mean over-confident. If the model is over-confident we shrink it until it is honest.
Suspect: An edge over 5 points on the fair line. Assume bad data before assuming a great bet.; The previous model treated huge disagreements with the book as its best bets. In practice they are almost always a wrong starting QB, a stale line, or a data bug. We flag them, bet nothing, and ask a human to check.
Lean: A side we like a little, under the betting threshold. No stake.; Leans are shown so you can see which way the model points on games it will not bet. They are not recommendations.
Team rating: Points per game better or worse than an average team, split into offense and defense.; Ratings come from one regression over every play, fitting all 32 offenses and 32 defenses at once so each team is judged against the opponents it actually faced. The play data is cleaned first: garbage time is stripped out, turnover plays are regressed almost fully toward what the play was worth before the turnover, and red-zone finishing is regressed heavily. EPA per play is the spine, with success rate and yards per play blended in. Defense is reported as points prevented, so positive is good on both sides.
Market-implied rating: The power ratings you can solve for from the book’s own spreads.; Given every closing spread, least squares recovers the ratings the market is implicitly using. Our ratings correlate about 0.89 with them late in the season. The rating model itself never sees a betting line; the market enters once, at the end, through the fair-line blend.