Betting Glossary
168 terms, plain words first. Open "For experienced bettors" for the precise version.
The Basics
A bet on which team wins the game, whatever the score. Bears −150 means you risk $150 to win $100 if the Bears win.
A bet on the winning margin. Bears −3.5 must win by 4 or more; Lions +3.5 win the bet by losing by 3 or fewer, or winning.
For experienced bettors
Both sides are usually priced near −110, so the price matters as much as the number. A better number at a worse price can still be the worse bet.
A bet on the combined score of both teams. Over 44.5 wins if they score 45 or more together; Under 44.5 wins at 44 or fewer.
Baseball's spread, almost always 1.5 runs. Yankees −1.5 must win by 2 or more; the other side at +1.5 wins by losing by 1 or winning.
For experienced bettors
Because so many baseball games are decided by one run, the favorite's run line is usually priced at plus money and the underdog's at minus money.
Hockey's spread, almost always 1.5 goals. Rangers −1.5 must win by 2 or more; the other side at +1.5 wins by losing by 1 or winning.
For experienced bettors
Late empty-net goals turn many one-goal games into two-goal wins, which is why the puck line is priced the way it is. Price it from the full margin distribution, not the moneyline alone.
A company that takes bets and sets the prices, like DraftKings or FanDuel. Consensus Edge compares the prices of 15+ sportsbooks for every game.
The price a sportsbook offers on a bet. It tells you how much you win, and it also hides how likely the book thinks the outcome is.
The number a bet is graded against, like −3.5 on a spread or 44.5 on an over/under. People also say "line" when they mean the odds.
The US way of showing odds. −150: risk $150 to win $100. +130: risk $100 to win $130. Minus marks the favorite, plus the underdog.
For experienced bettors
−110 on both sides is the standard price on spreads and over/unders. The gap between the two sides' prices is where the vig lives.
Odds shown as your full return per $1, stake included. 2.30 returns $2.30 on $1, a $1.30 profit. The board can show odds this way.
For experienced bettors
Minus prices: decimal = 1 + 100/|A|. Plus prices: decimal = 1 + A/100. −110 → 1.91; −150 → 1.67; +130 → 2.30. Break-even chance = 1/decimal: 1/1.91 = 52.4%.
The fee a sportsbook builds into its odds. At −110 on both sides you risk $110 to win $100, so the book keeps a cut whichever side wins.
For experienced bettors
−110/−110 is a 4.55% hold. Lower vig makes a book cheaper to bet into; it does not make the book sharper or its prices more accurate.
−110 implies 110/210 = 52.38%. Both sides: 104.76%. Hold = 1 − 1/1.0476 = 4.55%.
To win a spread bet. A 7-point favorite covers by winning by 8 or more; the underdog covers by losing by 6 or fewer, or winning.
A tie with the line, so your stake comes back. Bears −3 winning by exactly 3 is a push: no win, no loss.
For experienced bettors
Half-point lines (−3.5) can never push. In a parlay, a pushed leg usually drops out and the payout shrinks to the remaining legs.
The team expected to win. It has minus odds, like −150, and gives points on the spread, like −3.5.
The team expected to lose. It has plus odds, like +130, and gets points on the spread, like +3.5. Underdogs pay more when they win.
A game with no favorite on the spread. The spread is 0, so your team just has to win.
Betting the favorite. Favorites win more often, but their short prices mean they still have to be priced right to be worth betting.
For experienced bettors
Likely and good value are different questions. The game card's Chalk Analysis shows the favorite's EV at the price, so you can see what backing the likelier side actually costs.
One bet built from several bets, where every leg must win. Two −110 legs pay about +264, but you have to hit both.
For experienced bettors
Books hold more on parlays than on single bets, because the vig compounds across legs. Legs from the same game are rarely independent; see Same-Game Parlay.
Decimal odds multiply: 1.909 × 1.909 = 3.645, about +264. For independent legs the chances multiply too: 52.4% × 52.4% = 27.4% to break even.
Your standard bet size. On Consensus Edge 1 unit = 1% of your bankroll, so 2u on a $1,000 bankroll is $20.
The money you set aside only for betting. Consensus Edge sizes every suggested stake as a share of it, so stakes grow and shrink with it.
Our estimate of how often a team wins this game, like 62%. It is a chance, not a promise: a 62% team still loses about 4 times in 10.
Our name for what the sportsbooks think: 15+ books' prices with the vig removed, blended into one chance. It is the market price, not "square money."
For experienced bettors
Each book's prices are de-vigged with the Shin method, then weighted by the inverse of its vig, so low-vig books (often offshore) count more. Unlike "square money," Square says nothing about who is betting; it is the price itself.
Our name for Consensus Edge's own models: team ratings and stats turned into a win chance. It is not "sharp money," which means bets from professional bettors.
For experienced bettors
What runs differs by sport. The NFL rating starts from the market spread; machine-learning (LightGBM) game models run for the NBA only; other sports use rating and signal models. Weights are re-tuned daily against results.
Our name for prediction markets like Polymarket and Kalshi, where people trade on outcomes with real money. A 62¢ contract reads as about a 62% chance.
For experienced bettors
On moneylines, Crowd counts more when the money behind a price is deep and less when it is thin; on spreads, weights follow which sources have a price. It is a third, independent price, not a read on who is betting: neither "square money" nor "sharp money."
Our final chance for an outcome, blending Square, Sharp and Crowd into one number. Every BET, LEAN and NO EDGE starts from it.
Our strongest recommendation: the price beats our fair number by enough to justify the full suggested stake at your risk setting.
For experienced bettors
It needs our chance to clear a floor (54% on moneylines; break-even plus 2 percentage points on spreads and over/unders, never below 40%), positive EV above the sport's floor, and enough sources behind it. The edge is price versus probability, not how likely the side is to win.
A smaller recommendation: the price is in your favor, but by less than a BET, so the suggested stake is smaller.
For experienced bettors
LEAN is also an edge grade, below A and B: a moderate edge, worth a smaller position.
No bet worth making here at today's price. It doesn't mean this team loses: either the price is fair or it's a market we don't grade.
For experienced bettors
Most NO EDGE cells mean "not graded," not "fairly priced": switched-off markets, edges too large to believe, and edges with too few sources all show NO EDGE. If you bet one anyway, see "Betting It Anyway?"
A button that opens your sportsbook's bet slip with this bet already filled in. It works with FanDuel, DraftKings, BetMGM and Caesars.
For experienced bettors
When you come back we ask whether you placed it, so the bet can be tracked on your page. Same-game parlays open at DraftKings only.
The sportsbook you use, picked above the board next to American/Decimal. We then show your book's price beside the best price, and Place on buttons open your book.
The game has started, so we stop pricing it. The card shows our last pregame read and the time it was frozen.
For experienced bettors
If no frozen copy exists, the numbers follow the live market instead and the card says so. Once the game ends the label reads Final.
No sportsbook offered this bet, so there is no price to compare and nothing to grade.
Finding Value
The win chance a price implies, which is also the win rate you need to break even. −200 implies 66.7%: win more often than that and you profit.
For experienced bettors
Each side's implied figure is exactly its break-even rate. Add both sides and you get more than 100%; the excess is the vig.
Minus prices: |A| / (|A| + 100). Plus prices: 100 / (A + 100). −200 → 200/300 = 66.7%. +130 → 100/230 = 43.5%.
The win rate you need at a price just to not lose money. At −110 it's 52.4%; at +150 it's 40%.
For experienced bettors
Best Bets marks it on each meter, and the Your price card prints it for the price you type. A side is only worth betting when our chance sits above it.
−110: 110/210 = 52.4%. +150: 100/250 = 40.0%.
The chance of an outcome with the vig taken out. If both sides are −110, each side's fair chance is 50%, not 52.4%.
The price a bet would have with no vig, matching its fair chance exactly. A true 50% bet's fair odds are +100, even money.
For experienced bettors
Example: −150/+130 implies 60.0% + 43.5% = 103.5%, a 3.4% hold. Shin fair for the favorite is 58.3% (in a two-way market Shin equals splitting the excess evenly); proportional scaling gives 58.0%. The 60% is the raw implied price, not a fair one.
Fair American from chance p: if p ≥ 50%, −100·p/(1−p); else +100·(1−p)/p. 58.3% → −140. 40% → +150.
Our chance turned into odds. If our number says 60%, our fair price is −150, and any better price, like −130, is in your favor.
For experienced bettors
60%: −100 × 0.60/0.40 = −150. At −130 you'd need only 130/230 = 56.5%, so a 60% chance gives about 3.5 points of edge.
What a bet earns on average per $100 if you could make it many times. +$5 EV means about $5 profit per $100 over the long run.
For experienced bettors
EV is the only honest test of a single bet, but it depends entirely on the probability you plug in. We show EV at the best price and, separately, at your book's price.
EV per $100 = p × profit − (1 − p) × 100. 55% at −110: 0.55 × 90.91 − 0.45 × 100 = +$5.00.
What a bet is worth on average at your own sportsbook's price, which can be worse than the best price. Type your price into Your price to see it.
How much better your price is than the true chance warrants. If we say 55% and the price implies 52.4%, the edge is about 2.6 points.
For experienced bettors
On Consensus Edge an edge is always price versus probability. A 75% favorite can have no edge, and a 40% underdog can have a big one.
When the odds pay more than the true chance deserves. A 50% bet offered at +110 is an overlay: you're being paid extra.
When the odds pay less than the true chance deserves. A 50% bet offered at −120 is an underlay: you're paying extra.
Checking several sportsbooks for the best price on the same bet. Getting −105 instead of −110 means more profit every time you win.
For experienced bettors
Betting −108 instead of −110 is worth about 0.9% of everything you bet, more than many handicapping edges, and it applies to every bet you make.
−110 pays 0.909 per $1; −108 pays 0.926. The extra 0.017 per win, at a ~52% win rate, is about 0.9% of money bet.
The best price for a bet across all the sportsbooks we track right now. We show it next to your book's price so you can see what you'd give up.
A label on a call saying how strong and trustworthy its advantage is, from A (strongest) down to LEAN, plus warning labels when we won't act.
For experienced bettors
Grades rank the edge after our adjustments, with stricter bars on moneylines than on over/unders and extra sport-specific floors. A grade sits beside the BET/LEAN recommendation; it does not replace it.
Our strongest tier: a real, priced edge the system trusts.
A graded recommendation: solid, but thinner than an A.
The edge is not usable after our checks, so wait. It fails for reasons ranging from a suppressed market to an effective edge of zero after AI adjustments.
For experienced bettors
DEFER fires when: (1) the market is suppressed (switched off), (2) there is no book price to act on, (3) the effective edge is zero or negative after the AI adjustment cap, or (4) the edge is below the market's floor. On the Exchanges board, Defer means the same: no call yet.
The number is large and the system does not believe it. It is almost always a stale line or a bad price, not free money.
For experienced bettors
It fires when EV is above our believability limit: 30% in most markets, 10% on NFL spreads. Real edges in liquid markets are a few percent.
For each market, the sides most likely to hit, with our chance next to what the price implies. Likely is not the same as good value.
For experienced bettors
A Best Bet is a recommendation only when it carries a graded call (BET or LEAN). Eligible when our chance is above 50%, EV at the best price is no worse than −10%, and moneylines are no worse than −300. Your risk setting changes the ranking.
Ranked by a certainty-equivalent return, EV − (λ/2)·Var, where Var = p(1−p)·d² per unit staked and d is the decimal price. λ: Conservative 2.0, Moderate 1.5, Standard 1.0, Aggressive 0.25.
Shown on a NO EDGE moneyline: the most we'd suggest risking if you bet it anyway, as a small range in units. Damage control, not a recommendation.
For experienced bettors
The range runs across the four risk settings, with your own setting marked.
The chance Consensus Edge itself puts on an outcome, from our blend. When a list says "books' number" instead, the chance comes from sportsbooks alone.
Shown when our own number isn't used in a market: the chance is the sportsbooks' blended price with the vig removed, not our model's.
A side that carries a BET or LEAN recommendation. Only graded calls get the green chip; other sides on a list are shown for context.
When your sportsbook's line differs from the one we graded, like −3.5 at your book versus our −3. The button warns "Not our bet" and names ours.
For experienced bettors
Half a point can change the bet completely, especially across a key number. Our EV applies to our number only.
One setting for your stakes and Best Bets ranking. Aggressive stakes 10% of full Kelly, our ceiling; other stops stake less. Top-bar gear on desktop, Settings in the phone menu.
For experienced bettors
Each stop stakes a share of full Kelly: Conservative 2.5%, Moderate 5%, Standard 7.5%, Aggressive 10%. The setting also sets how heavily Best Bets penalizes variance. Stored in this browser, not on your account.
Best Bets variance penalty λ in EV − (λ/2)·Var: Conservative 2.0, Moderate 1.5, Standard 1.0, Aggressive 0.25. A smaller stake and a heavier variance penalty move together by design.
A formula for how much of your bankroll each bet deserves, based on your edge and the payout; no edge, no bet.
For experienced bettors
No model's win chance is exact, so we never stake full Kelly: at most one-tenth of it, 10% at Aggressive, less at the other risk stops. Full Kelly for 55% at −110 is 5.5% of bankroll; one-tenth is $5.50 per $1,000 (0.6u). Full Kelly is capped at 10%, so the largest suggested stake is 1% of bankroll.
f* = (b·p − q) / b, where b = profit per $1, p = win, q = 1−p. Our stake = f* × share × bankroll, where share is your risk setting's share of full Kelly, 2.5% to 10%.
How we turn a price and our chance into BET, LEAN or NO EDGE, and into a suggested stake in units.
For experienced bettors
Floors first: our chance must reach 54% on moneylines, and break-even plus 2 percentage points on spreads and over/unders (never below 40%). Then the edge must clear the sport's EV floor with enough sources agreeing. Stakes run 2.5% to 10% of full Kelly by risk setting; full Kelly is capped at 10%, so the largest is 1% of bankroll.
Betting the same amount every time, however good the bet looks. Simple and steady. Our record and ROI figures assume a flat $100 per bet.
The first price a sportsbook posts on a game. It moves as news and money arrive.
For experienced bettors
Openers are the least informed prices a market will see, which is where early bettors find the most mispricing, usually at low limits.
The last price before the game starts. With all the news in, it is the most accurate price there is, which is why beating it matters.
How a price changes from open to close. Bears −3.5 moving to −3 means the market now thinks the game is a bit closer.
For experienced bettors
Moves come from news (injuries, weather, lineups) and from respected money. Who moved first matters more than how far: a move that starts at sharp books and spreads is information; a lone retail book drifting is usually not.
The share of bets that won. Alone it misleads: 55% at −110 makes money, but 55% at −200 loses, because −200 needs 66.7% to break even.
Profit divided by the money you bet. A 5% ROI means $5 profit for every $100 wagered.
For experienced bettors
Yield is the same thing: profit ÷ turnover. Our record's ROI assumes a flat $100 per bet.
How much you're up or down overall, in dollars or units.
A searchable log of past recommendations with our chance, the price, EV and the result, so you can check the record yourself.
Sharp
Whether you got a better price than the final one before the game. Bet at −110, it closes at −130: you beat the close.
For experienced bettors
Measured in percentage points of fair probability between your price and the close. Your page shows mean CLV and % beat close. Win-loss needs hundreds of bets to mean anything; consistent positive CLV shows skill much sooner.
A chart of closing line value across all picks. A line that keeps rising means the picks keep beating the closing price.
Winning margins that happen far more often than others. In the NFL, 3 and 7 are the biggest key numbers, so −2.5 and −3.5 are very different bets.
For experienced bettors
Because so many games land on them, a half-point across a key number is worth far more than one elsewhere. Basketball margins are spread out, so key numbers matter much less there.
The half-point on a line, like the .5 in −3.5. It means the bet can't push.
For experienced bettors
A hook is worth the chance that the game lands exactly on the whole number it crosses. Across a key number like the NFL's 3 that is a lot; across most numbers it is very little.
When enough of the most respected, fastest-moving sportsbooks move a line the same way within minutes, usually because pros bet it or news broke.
For experienced bettors
Our NFL steam alerts fire when two sharp operators (Pinnacle, BetOnline, Circa, Bookmaker) move the same game the same way by at least 2 percentage points within 5 minutes. On a spread, a move is sized by what the points are worth, not by how many there are.
Bets from professional bettors, which sportsbooks respect and move their prices for. Not the same as our Sharp source, which is our own model.
For experienced bettors
We don't see who bets or how much. We infer pro action from fast, same-direction moves across several sharp books (see Steam Move).
Bets from casual bettors, who tend to like favorites, overs and popular teams. Not the same as our Square source, which is the sportsbooks' blended price.
When a line moves against the side most people are betting: a classic hint that bigger, smarter bets are on the other side.
For experienced bettors
Spotting it needs ticket and money percentages, which Consensus Edge does not collect. We infer pro action from fast multi-book moves instead.
A price a sportsbook hasn't updated after news or a market move. It can be a real chance, and it's why a price that looks too good usually is.
For experienced bettors
Large EVs on stale lines are why SUSPECT exists. We also drop a book's price from the blend when it lags the freshest book by more than 6 hours.
How much a sportsbook's prices lead the market. Leading books like Pinnacle and Circa move first and take big bets; most others copy them.
For experienced bettors
Sharpness is price discovery, not low vig. DraftKings and other US retail books are not sharp books. Our Analytics ranking scores each book's closing moneyline prices with the Brier score.
The share of money a sportsbook keeps on a market, from its vig. −110 on both sides is about a 4.5% hold; lower is cheaper for you.
For experienced bettors
Analytics ranks books by average hold. Low hold makes a book cheaper to bet into; it doesn't make its prices more accurate.
Hold = 1 − 1/overround. −110/−110: 1 − 1/1.0476 = 4.55%.
The sum of both sides' chances as the odds show them, always above 100%. −110/−110 adds up to 104.8%; the extra is the book's cushion.
For experienced bettors
−110 → 52.38% per side; 2 × 52.38% = 104.76%. Hold = 1 − 1/1.0476 = 4.55%.
All the money bet on a game or market. A big handle means lots of interest; it doesn't tell you which side is right.
How far apart the sportsbooks' fair prices are on a game. Tight means they agree; Wide means they don't, which can mean news or a mispriced book.
For experienced bettors
Measured as the standard deviation of books' vig-free chances: under 2 percentage points is Tight, 2 to 4 is Normal, 4 or more is Wide.
When two bets tend to win or lose together, like a favorite winning and the over hitting. Treating them as unrelated misprices a parlay.
A parlay with several bets from one game. The legs often move together, so the price should reflect that; books usually charge extra for it.
For experienced bettors
Our parlay pricing applies a per-sport correlation adjustment to legs from the same game instead of treating them as independent. Same-game parlays open at DraftKings only.
The natural swings in results. A bettor with a real edge can still lose for weeks; a short run proves little either way.
For experienced bettors
A 55% bettor at −110 will routinely see long losing runs. Size so a bad stretch can't knock you out; that is what fractional Kelly is for. Judge a process on hundreds of bets and on CLV, not on a month.
Baseball bets settled after five innings, so mostly about the starting pitchers. MLB cards show F5 moneyline, spread and over/under when books offer them.
A sportsbook licensed outside the US, like Pinnacle or BetOnline. Their prices feed our Square blend; they are not licensed by US states.
A sportsbook licensed by a US state, like DraftKings, FanDuel, BetMGM or Caesars. The My book list groups books as Regulated or Offshore.
A market where people buy and sell yes/no contracts on outcomes. Prices run from 0 to 100¢ and read as chances: 62¢ is about 62%.
For experienced bettors
Prices come from traders, not a bookmaker, so there is no built-in vig; the venue charges a fee instead. Our Crowd source reads these prices.
A prediction market for trading on outcomes. It runs an international, crypto-funded exchange and a separate US exchange regulated by the federal government (the CFTC).
A US prediction market regulated by the federal government (the CFTC), traded in dollars. It offers game winner, spread and over/under contracts.
The US regulator over Kalshi and Polymarket US. It makes them legal, regulated exchanges rather than sportsbooks.
A yes/no share that pays $1 if the outcome happens and nothing if not. Buying "Yes" at 65¢ means paying for a 65% chance.
Money waiting to trade near the current price. More liquidity means it takes more money to push the price around, so the price is more reliable.
All the money already traded on a market. High volume means many traders have tested the price.
Several markets on one game at different lines, like over 215.5, 218.5 and 221.5. Prices should step down smoothly from one line to the next.
For experienced bettors
Our Exchanges board checks each ladder for consistency and holds back a call when neighboring lines disagree.
Venues like Kalshi and Polymarket where you trade yes/no shares with other people instead of betting against a sportsbook. Our Exchanges board compares their prices with a fair value.
For experienced bettors
Fair value comes from sharp sportsbooks with the vig removed; cost is the venue's price plus its fee at a $100 order. Calls are Buy, Post, No side or Defer. Alternate lines estimated from the main line show Not rated and are never a buy.
What a trade on an exchange really costs: the price plus the venue's fee. A 52¢ share with a 1¢ fee costs 53¢ all-in.
On an exchange, placing an order at your own price and waiting for someone to take it, instead of buying right now at the asking price.
For experienced bettors
The Exchanges board shows "Post at X¢" when waiting for a better fill is worth it. Depending on the venue, posting may cost a lower fee or earn a rebate.
How much money is available to trade at or near a price. A great price with $20 behind it can't take a real-sized bet.
For experienced bettors
The Exchanges card charts the dollars available up to our maximum entry price.
An exchange market's rules for paying out, like whether overtime counts or what happens if the game is postponed. They can differ from a sportsbook's rules.
For experienced bettors
We read each contract's rules before grading it; a contract whose rules we can't confirm gets no grade.
Under the Hood
A way to remove the vig that allows for books overcharging big underdogs more than favorites. We use it to turn each book's prices into fair chances.
For experienced bettors
In a two-way market Shin gives the same answer as splitting the excess evenly between the sides; with more outcomes it shifts more of the margin onto longshots.
In the Square blend, each sportsbook counts in proportion to 1 divided by its vig, so a 2% book counts twice as much as a 4% book.
For experienced bettors
weight = 1 / max(vig %, 0.5); weights are normalized to sum to 1, then Square = Σ weight × that book's Shin fair chance, renormalized so both sides sum to 100%.
Your starting estimate before new information. The market's price is the natural starting point; evidence has to be strong to move far from it.
How Square, Sharp and Crowd weights are set game by game. On moneylines a deeper prediction market counts more; on spreads weights follow which sources have a price.
The share each of Square, Sharp and Crowd gets in the consensus. Weights vary by sport and game and are re-tuned daily, so we don't print fixed percentages.
One input to our Sharp model, like rest, injuries or a team rating. A game card lists the signals behind its number.
For experienced bettors
Signal weights are re-tuned every 24 hours, so we describe signals in words rather than fixed percentages. The methodology page has the full math.
A small model that learns from past results how best to combine our sources' predictions.
For experienced bettors
A logistic regression trained on past source predictions. It needs at least 100 samples and goes live only if it improves the Brier score. It is switched off for the NBA.
Whether our percentages come true at the rate we say. Of all the games we call 70%, about 70 in 100 should win.
A chart of predicted chances against how often those outcomes really happened. A model whose chances come true at the stated rate follows the diagonal.
A group of predictions in the same range, like 55 to 60%, used to check, range by range, whether our chances come true.
A method that fits an S-shaped curve to raw model outputs to fix steady over- or under-confidence.
For experienced bettors
We use it as the fallback when there is too little data for isotonic regression.
A method that corrects raw predictions while keeping their order, so a higher raw number never becomes a lower chance.
For experienced bettors
Used once a sport and market has at least 30 samples; below that, Platt scaling is the fallback.
A score for probability forecasts: the average squared gap between the forecast and what happened. Lower is better; always guessing 50% scores 0.25.
For experienced bettors
There is no universal "good" number. The fair test is our Brier against the closing line's Brier on the same games.
Forecast 70%: if it wins, (0.70 − 1)² = 0.09; if it loses, (0.70 − 0)² = 0.49. Average over all games.
Pulling extreme predictions back toward 50% so the model doesn't overstate its confidence.
For experienced bettors
The amount is set per sport and market and adjusted automatically as results come in, so we don't quote fixed values.
A machine-learning tool that builds many small decision trees and adds them up into one prediction.
For experienced bettors
Our LightGBM game models run for the NBA only today: a calibrated moneyline classifier plus margin and combined-score regressors, trained on several seasons of games.
One input variable fed to a machine-learning model, like rest days or a team's efficiency.
For experienced bettors
Our NBA moneyline model uses 30 features.
A model that predicts a number, like the winning margin or the combined score, rather than a yes/no outcome.
A model that predicts a category, like win or lose. Our NBA moneyline model is a classifier that outputs a win chance.
An average that counts recent games more than older ones, so it reacts to form without forgetting the season.
How many points a team wins or loses by.
For experienced bettors
How widely results scatter around the spread turns a predicted margin into a cover chance. Our standard deviations, in points: NBA 13.5, NFL 12.73, college football 15.682, college basketball 11.5.
The bell curve: most results land near the average and fewer far away. We use it to turn a predicted margin into chances.
For experienced bettors
Not everywhere: NFL and college football alternate lines use measured margin tables, which capture how often games land on key numbers like 3 and 7.
One number for a team's strength, used to predict margins. If A rates 5 and B rates 2, A is about a 3-point favorite on a neutral field.
A rating where teams gain points for wins and lose them for losses, more for upsets. A higher Elo means a stronger team.
For experienced bettors
In the NFL our "Elo" is the market spread converted into a rating, not an independent model.
A saved copy of our prediction at a moment in time. We keep an opening and a closing snapshot for every game.
Our prediction when we first see a game's odds. It is never changed afterwards, so it is a fair record of what we said early.
Our latest prediction before the game starts, frozen at the start. Our accuracy is measured on it.
The date from which a sport's analytics are counted. Older predictions, made before a fix, are left out so they don't distort the record.
Every 24 hours our system re-tunes its signal weights and blend weights based on how accurate recent predictions were.
A comparison of our model's predictions with sportsbook closing prices: where we disagreed, and which side was right.
How each signal has done historically, to show which inputs add value and which are noise.
A football stat for how much a play changes a team's expected score. Positive EPA helped; negative hurt. Teams are compared by EPA per play.
Points scored minus points allowed per 100 possessions. A +5 team outscores opponents by 5 per 100 possessions.
Possessions per game. Faster teams create more scoring chances, which pushes the combined score up.
A per-minute basketball rating of a player's overall stat production. We use it to estimate how much a missing player hurts.
The edge a team gets from playing at home. It differs by sport and is built into every prediction.
Playing on two days in a row. Tired teams tend to play worse, so it's built in as a fatigue penalty.
Days off between games. More rest generally helps, and a rested team facing a tired one has a measurable edge.
How much a missing player changes a team's expected margin, based on the player's role and quality.
An NFL signal for how much the starting quarterback matters, including when a backup starts.
For experienced bettors
Our margin model prices a quarterback change in points. A cap of 15 percentage points applies only on a fallback path, not to the main model.
How tough a team's opponents have been. Key in college sports, where schedules vary a lot.
How a team performs in close games late on. It is one of the signals shown on some cards.
A college football efficiency rating that adjusts for opponent quality. It is one of the signals behind our college football numbers.
Hockey's luck gauge: shooting percentage plus save percentage. Teams far above or below normal tend to drift back toward it.
A tennis Elo rating kept separately for each surface: hard, clay and grass. Players can be much stronger on one than another.
The share of a team's possessions a player uses while on court, through shots, free throws or turnovers.
For experienced bettors
When a high-usage teammate is out, others absorb those possessions, and books can be slow to adjust player prop lines.
The share of swings that miss completely. It is one of the strongest predictors of a pitcher's strikeouts.
The share of batted balls hit at the ideal speed and angle for extra-base hits. A strong predictor of home runs.
Expected batting average based on how hard and at what angle the ball was hit, regardless of defense. It predicts future hits better than batting average.
An overall hitting measure based on quality of contact, weighting each outcome by its run value.
How a ballpark changes scoring. Some parks boost home runs, others suppress them, and projections adjust for it.
How a hitter does against left-handed versus right-handed pitching. Most hit better against opposite-handed pitchers.
Hockey minutes a player spends on the power play. Top-unit players get more shot and goal chances.
Expected goals per 60 minutes, based on shot location and type. It predicts future goals better than actual goals do.
The share of shots a goalie stops. Extreme values tend to drift back toward the league average.
Player props are bets on one player's stats, like points or strikeouts. Our model-based props are paused right now.
For experienced bettors
When running, the scanner projected each stat from season averages, recent form and matchup, then blended the result with the market's own price before computing EV.
The math our paused props scanner used to turn a projected stat into a chance of going over or under the line.
For experienced bettors
MLB and NHL count stats such as strikeouts, hits, shots and goals use Poisson; higher-volume stats such as NBA points and NHL saves use a normal (Gaussian) curve.
How the paused props scanner measured value: our blended chance against the best price across sportsbooks, as profit per $100.
The bars a prop had to clear for BET or LEAN. They differ by sport and are stricter than game lines.
For experienced bettors
BET / LEAN need both EV and probability edge: NBA 6% & 4pp / 2.5% & 2.5pp; MLB 5% & 4pp / 2% & 2pp; NHL 8% & 5pp / 4% & 3pp. Both conditions must hold.
The paused scanner trusted the market more than its own model: each prop's chance blended a minority share of the model with the market's own price.
For experienced bettors
Model weight in the blend: NBA 0.40, MLB 0.40, NHL 0.15 to 0.30 by stat (0.30 by default), NFL 0.20. No machine-learning layer runs on props.
How a player has done lately, weighted toward the most recent games. The paused props scanner mixed it with season averages.
For experienced bettors
NBA projections blend 60% season average with 40% recent-form average.
Shifts to a player projection for the opponent: positional defense in the NBA, pitcher quality and park in MLB, opponent shots and goalie in the NHL.
How an opponent defends a position (guard, forward, center) rather than overall. A weak overall defense can still be strong against centers.
When a key teammate is out, the paused NBA props model passed part of their usage to the remaining players.
For experienced bettors
About 60% of the missing player's usage was redistributed, in proportion to each remaining player's share.
Inputs to the paused NBA props model: season averages, recent form, usage rate, positional defense and teammate absences.
Inputs to the paused MLB props model: season stats plus whiff rate, barrel rate and xBA, park factors and platoon splits.
Inputs to the paused NHL props model: season stats plus power-play time, expected goals, opponent defense and goalie save percentage.
A model for counts like goals or strikeouts. From an expected rate, like 0.8 goals a game, it gives the chance of each count.
Another name for the bell curve, used for high-count stats like goalie saves, where the bell shape fits well.
Books have real-time lineup news and minutes projections we don't. Treat any props number as a baseline for your own research.