A clear reference for the betting terms used on Gridiron & Wine play cards — from beginner basics to sharp-bettor concepts.
If a term shows up in a daily NHL, NBA, or MLB pick — moneyline, run line, edge, CLV, unit, walk-forward, platoon split — you'll find it defined here. Entries are short and concrete. Where the math matters, we show it.
A bet on whether a team covers the point spread, not whether they win outright. The dominant bet structure in NBA, NFL, and college basketball — and the bet most exposed to point-spread inefficiencies that edge-based models target.
How it works. The favorite is listed at a negative number (-6.5 means they need to win by 7+ to cover). The underdog is listed at a positive number (+6.5 means they can lose by 6 or fewer, or win outright). Most spreads are priced at -110/-110 — you risk $110 to win $100, regardless of which side you take.
Pushes and half-points. A “push” happens when the final margin lands exactly on the spread (e.g., a -3 favorite wins by exactly 3). The bet is refunded. Books often use half-point spreads (-3.5 instead of -3) specifically to eliminate pushes — the half-point is meaningful enough that hooks at common key numbers (3 in NFL, 4 in CBB, 7 in NBA) move the line by 5-15 cents of juice depending on the sport.
Why ATS is structurally interesting. Unlike moneylines, where the price reflects win probability, spread bets are designed to be roughly 50/50 propositions at -110 juice. That means the entire opportunity lives in identifying spreads where one side’s true cover probability exceeds 52.4% — the breakeven needed to overcome standard vig. Small percentage gaps (53-54% true probability) are tradeable edges on volume.
Key numbers and line shopping. NFL spreads cluster heavily at 3 and 7 (the most common margins). NBA spreads cluster less but still skew toward 3, 5, 7. A half-point at a key number is worth meaningful expected value — the difference between -2.5 and -3 in NFL is roughly 10 cents of juice because so many games land on exactly 3. Bettors who consistently shop multiple books for the best half-point on key numbers pick up structural edge that compounds across a season.
The average margin by which a team wins or loses across a sample. Used in baseball and hockey models as a richer alternative to win-loss record — a team that wins 55% of games by an average of 3 runs is structurally different from a team that wins 55% by 1 run, and ARM captures that difference where binary W/L cannot.
Why it matters. Run differential is more predictive of future wins than past wins themselves — a phenomenon known as Pythagorean expectation. A team underperforming their run differential is due to regress upward; one overperforming is due to regress downward. ARM is the simplest expression of this concept. Combining ARM with rolling-window analysis (last 10 games, last 30 days) catches teams in transition that aggregate season stats miss.
Running a betting model on historical game data to measure how it would have performed had it been live during that period. Done correctly, backtesting validates a strategy before risking real money; done incorrectly, it’s the single most common path to overconfidence in a strategy that won’t survive contact with live markets.
Walk-forward validation. The standard for honest backtesting. The model is trained only on data prior to each test window, then tested forward in time. Period 1: train on 2018-2021, test on 2022. Period 2: train on 2018-2022, test on 2023. And so on. Each test period is genuine out-of-sample data the model never saw during training. Strategies that survive walk-forward across multiple periods are far more likely to survive live.
Common backtesting failures.
Why backtesting alone isn’t enough. Even a well-designed walk-forward backtest can be unconsciously curve-fit (the modeler picks features and thresholds based on hindsight). The real validation only comes from live performance — specifically CLV across a meaningful live sample. Read why CLV is the only stat that can’t be backtest-gamed →
The dedicated capital a bettor sets aside specifically for sports betting, kept fully separate from living expenses, savings, and discretionary spending. The size of the bankroll determines how much each unit is worth in dollar terms.
Why separation matters. Sports betting is a high-variance activity. A profitable +3% ROI strategy can drawdown 15-25% in any given 100-bet window. If your bankroll is “whatever’s in checking,” a normal drawdown can wipe out rent money. If it’s a separate, walled-off pool you decided in advance you could lose entirely without consequence, drawdowns are merely uncomfortable — not catastrophic.
Right-sizing the bankroll. The rule of thumb: never start with more than you can afford to lose. Common sizing benchmarks:
Refilling vs growing. Two schools: refill (top up bankroll to original size when it drops, treat losses as cost of doing business) or grow (let bankroll compound up and down, adjust unit size as bankroll grows). Compound growth is mathematically optimal for the long-run bettor; refilling is appropriate for recreational bettors who want a fixed exposure ceiling.
The discipline test. A bettor who can’t stop themselves from dipping into living expenses after a cold streak doesn’t have a bankroll — they have a problem. Bankroll separation is the single highest-leverage discipline in long-run betting; everything else is downstream of it. Read more on bankroll management and fractional Kelly →
A late, often meaningless, score that flips a spread bet from a loser to a winner. A team trailing by 10 with 30 seconds left who scores a garbage-time touchdown to lose by 3 instead of 10 just gave their +6.5 backers a backdoor cover. The favorite "won" the game decisively but failed to cover the spread.
Why it's a recurring betting phenomenon. Late-game strategy diverges from game-winning strategy. Teams up by a margin will stop running their full offense; teams down by a margin will press aggressively for points that don't change the outcome. Both behaviors compress the final margin systematically — and the compression direction usually favors the underdog spread bettor.
The data. Backdoor covers are real and measurable. Across NBA and NFL, roughly 8-12% of games involve a meaningful spread swing in the final two minutes that wouldn't have happened in a tied-or-close game. Sharp models for spread bets that account for game-state-conditional scoring patterns extract meaningful edge from the asymmetry.
A bet that looked won until a late, unlikely event flipped the result. Classic examples: a missed extra point pushing a covering favorite into a push, a meaningless garbage-time touchdown by the trailing team that backdoors the cover, a goalie pulled with 0.4 seconds left leading to an empty-net goal that swings a puck line.
Why it doesn't matter for long-run results. Bad beats and lucky covers cancel out over volume. A bettor who fixates on bad beats is selectively remembering — the same player has had equally lucky beats they conveniently forget. The math of betting only cares about decisions made at edge prices, not the variance of any single outcome.
The practice of paying additional juice to move a spread or total in the bettor's favor — e.g., paying -130 instead of -110 to bet a -3 favorite at -2.5. Buying points around key numbers (3 and 7 in NFL, 5 and 7 in NBA) is sometimes +EV; buying points elsewhere is usually a vig leak.
The math. Each half-point a sportsbook charges anywhere from 10 to 25 cents of juice depending on the underlying line and key-number proximity. Buying off -3 to -2.5 in NFL routinely costs 20-25 cents because so many NFL games land on exactly 3 — the half-point is genuinely valuable. Buying off -8 to -7.5 typically costs 10 cents because the half-point is less likely to matter.
When it's worth it. Only at key numbers, and only when the model's probability that the result lands on that exact number is high enough to justify the extra juice. Recreational bettors buy points indiscriminately because it "feels safer"; sharp bettors buy points only when the math says the half-point covers the marginal cost.
The difference between the odds you took on a bet and the price that same bet closed at right before the game started. If you bet a moneyline at +130 and the closing line was +115, you got a better number than the market eventually settled on — that’s positive CLV. If you bet at +130 and the line drifted to +145, the market moved against you and your CLV was negative.
Why it matters more than win rate. Single-bet outcomes are dominated by variance. Across 100 bets, even a +3% ROI strategy can finish above or below breakeven by luck alone. CLV doesn’t have that problem: it’s fixed the moment the line closes, regardless of who wins. A bettor averaging +2 cents of CLV across 200 bets is almost certainly a winning bettor — the closing line is the most efficient price the market produces, and consistently beating it is the strongest available signal of long-term edge.
Why sharp books care. Every major sportsbook tracks CLV at the account level. A customer who consistently beats the closing line is taking value the book didn’t intend to give — and accounts get limited based on CLV, not win rate. A losing customer with positive CLV is more dangerous to a book than a winning customer with negative CLV; the math will eventually correct.
The +EV mindset is the upstream cause; CLV is the downstream proof it’s working. Read the full guide on why CLV predicts long-run ROI →
A bet with roughly 50/50 probability of winning. Standard -110 spread and total markets are designed to be approximate coin flips by construction. Calling a bet a "coin flip" in betting slang usually means "I don't have any real edge here, I'm just guessing."
Why coin flips are losing bets. At -110/-110, a true 50/50 bet pays $0.91 on $1 risked when you win and costs $1.00 when you lose. Over time you lose 4.5% on every dollar wagered to vig. Coin flip bets aren't break-even — they're systematic vig leaks. Only bets where your edge clears vig are worth taking.
To win a spread bet by enough to beat the handicap. A -6.5 favorite "covers" if they win by 7 or more. A +6.5 underdog "covers" if they lose by 6 or fewer (or win outright). The bet pays whether the result was decisive or barely covered — the bettor either covers or doesn't.
Cover rate vs win rate. A favorite that wins 80% of its games might only cover the spread 50% of the time, because the spread is designed to make the cover decision close. ATS records measure cover rate (against the spread), not raw win-loss. A 53% cover rate at standard -110 juice produces a small profit; a 50% cover rate breaks even before vig.
The peak-to-trough decline in bankroll between two points in time. Drawdowns are an inevitable feature of any high-variance activity — the question isn’t whether they happen but how deep and how long, and whether your bankroll and bet sizing are designed to survive them.
Two ways to measure. Max drawdown is the largest peak-to-trough decline ever observed. Average drawdown is the typical decline over a defined window. Both matter. A strategy with low max drawdown and frequent shallow drawdowns is fundamentally different from one with rare but severe drawdowns — even if both have identical long-run ROI.
Worked example. A truly profitable +3% ROI strategy across 1,000 bets will routinely have stretches where the running bankroll is down 15-25% from its peak. Those drawdowns can last 100-300 bets. A bettor who panics and stops at 20% down has effectively converted a winning strategy into a loser by quitting at the worst moment.
Drawdown and bet sizing. Your maximum tolerable drawdown determines your maximum responsible bet size. A bettor who can’t emotionally survive a 30% drawdown should be using smaller unit sizes (and likely smaller Kelly fractions) than the math says they could. The optimal Kelly bet on paper assumes infinite emotional tolerance; real-world Kelly fractions exist precisely to bring drawdowns into a survivable range.
The two failure modes. Undersizing exits a strategy too early during a normal drawdown. Oversizing turns a normal drawdown into bankruptcy. Right-sized betting threads the needle: drawdowns hurt but don’t end the game, and the long-run math has time to play out. Read more on sizing bets to survive expected drawdowns →
The team a sportsbook believes is less likely to win — represented by positive American odds (+150, +200, +320) on the moneyline or by getting points on the spread (+3.5, +7). Backing dogs is structurally favorable for sharp bettors because public money disproportionately backs favorites, which leaves dog prices systematically a few cents above fair value.
Why dog moneylines are a recurring source of edge. If 70% of public bets on a game land on the favorite, the sportsbook has to shade the favorite's price higher (and the dog's price lower) to balance their book. The shading pushes the dog's number above what model-derived fair value would be. A bettor whose model identifies the right dogs at the right prices is harvesting that shading consistently.
Home dog vs road dog. Home underdogs are statistically the most profitable bet type in NHL and a recurring source of edge in NBA spread markets. The public has an emotional bias toward backing favorites who are also marquee teams, which pushes those favorite prices higher and rewards home-dog backers across long samples.
The gap between your estimate of true probability and the market’s implied probability on the same outcome. Edge is what every +EV bet is built on; it’s the structural difference between paying fair value and paying less than fair value for the same payoff.
The math. If a model projects a team to win 48.3% of the time and the market price implies 43.9% (the +128 American odds), the raw edge is 4.4 percentage points (48.3 − 43.9). That’s the percentage of expected value before vig — the cushion you have over the price you’re paying.
Raw edge vs net edge. Sportsbook vig eats into your edge. On a -110 spread, the book takes roughly 2.4% per side. So a 5pp raw edge becomes a ~2.5pp net edge after juice. The minimum bar for a profitable strategy isn’t “positive raw edge” — it’s “positive edge after vig and variance.”
Sample-size reality. A 3-4pp edge bet is profitable in expectation but can lose 8 of 10 sessions through variance. The math only sorts itself out over hundreds of bets at the same edge profile. Bettors who quit at the first cold stretch are quitting on math that hadn’t had time to land.
Edge is the engine; unit sizing is the throttle; CLV is the gauge that tells you the engine is firing. Together they describe the entire mechanical structure of profitable betting. Read the full pillar on finding mispriced odds →
A predictive metric used in hockey and soccer that estimates the probability a given shot becomes a goal, based on shot location, shot type, defensive pressure, rebound or rush context, and angle. xG measures underlying scoring rate beyond raw shot counts — a critical distinction in low-event sports where a single bounce can swing a result.
Why raw shots lie. A team can be outshot 35-22 and still generate more dangerous chances. Twenty shots from the perimeter against a set defense are worth roughly the same xG as five high-danger chances from the slot. xG-for and xG-against capture which team is actually generating quality opportunities, regardless of the volume that shows up in a box score.
The math. Each shot gets an xG value between 0 and 1 based on its location and context. A wrap-around from a bad angle might be 0.03 xG. A point-blank rebound at the side of the net might be 0.55 xG. Team xG is the sum across all shots in a game. Models are trained on years of historical NHL/MLS data with labeled outcomes — for any shot at coordinates X, in context Y, what fraction historically went in?
How betting models use xG. A team outscoring their xG by a wide margin over a 10-game stretch is regressing toward their xG, not extending the gap. A team being outscored despite winning the xG battle is the inverse — due for positive regression. Sharp models weight xG-differential more heavily than goal-differential because xG predicts future goals better than past goals do.
Where xG is most useful. Hockey totals (over/under goal lines), full-game moneylines, and goalie evaluation. A goalie facing 30 xG worth of shots and surrendering 2 goals is performing well above expected; facing 1.5 xG and surrendering 3 goals is the inverse. The NHL goalie workload model uses xG as a primary input to evaluate save-percentage context.
The average dollar outcome of a bet over an infinite number of trials. Positive EV (+EV) means the bet returns more than it costs in expectation; negative EV (-EV) means it loses money over the long run, no matter how often individual bets win.
The formula. EV = (Pwin × payout) − (Plose × stake). For a $1 bet on a team at +128 (decimal odds 2.28) with a 48.3% true win probability: EV = (0.483 × 1.28) − (0.517 × 1.00) = +$0.10 per dollar risked. That bet has +10 cents of expected value — over a few hundred trials, you’d gain about a dime for every dollar you put down.
Why win rate alone doesn’t tell you EV. A 60% winner at -150 has EV = (0.60 × 0.667) − (0.40 × 1.00) = $0.00 — pure break-even, no profit, despite winning 60% of bets. A 48% winner at +120 has EV = (0.48 × 1.20) − (0.52 × 1.00) = +$0.056 — profitable, despite losing more than half the bets. The seat is the price, not the win rate.
Variance vs EV. A bet’s EV is its long-run truth. Its variance is how dramatically individual outcomes will swing around that average. You can be a +EV bettor and lose money for three months straight; you can be a -EV bettor and win for two months. The math converges only with enough trials.
Every official Gridiron & Wine pick is a +EV bet at the moment of publication — the model’s probability beats the market’s implied probability by enough to clear our minimum edge threshold. Read the full +EV pillar →
Any bet structure that isn't a straight moneyline, spread, or total. Examples: parlays, teasers, pleasers, futures, props, same-game parlays. Exotics typically carry significantly higher vig than core markets, which makes them poor expected-value bets even when individual angles are correct.
The general rule. Exotics exist because they have high recreational appeal. Books offer them because the hold is higher and the math is less obvious to the average bettor. Sharp bettors stay in core markets for volume and only touch exotics when they've identified a specific structural edge (e.g., uncorrelated +EV legs in a parlay).
To bet against a specific side, source, or trend. Fading a touted public play (going opposite the popular side) is sometimes a winning strategy because public bets are disproportionately on the wrong side, shading lines past fair value.
Fading the public. Common wisdom that "fading the public is +EV" is only partially true. It works on certain bet types — primarily NBA spreads and NFL spreads — where the public's bias creates measurable line shading. It doesn't work universally. Sharp bettors don't fade for ideological reasons; they fade because their model says the unpopular side has positive expected value at the current price.
Fading a specific source. Some bettors track a touted handicapper's record and bet opposite. If the handicapper is genuinely losing money long-term, fading them is +EV by definition. If they're break-even or slightly winning, fading them is vig-leaking. The math matters more than the narrative.
Read more on why public-bias mispricing creates fade opportunities →
The team a sportsbook believes is more likely to win — represented by negative American odds (-150, -200, -350) on the moneyline or by giving points on the spread (-3.5, -7). Favorites are usually overpriced because the betting public disproportionately backs them, which pushes the favorite's moneyline higher and the underdog's lower.
Why heavy favorites bleed money long-term. A -300 moneyline favorite needs to win 75% of the time just to break even. Even teams that "should" be -300 against an opponent will often have true win probability closer to 70% because of variance, injury risk, and the limitations of pre-game projection. The math is brutal on heavy favorites; the small percentage shortfall vs implied probability compounds into negative expected value over volume.
When favorites are worth backing. When your model's probability comfortably exceeds the implied probability — e.g., a -150 favorite (60% implied) that your model has at 67%. The 7pp edge is real and bettable. Backing favorites blindly is a losing strategy; backing favorites at edge-justified prices is just normal +EV betting.
A bet that settles based on the score at halftime, not the final result. First-half markets exist in NFL, NBA, college football, and college basketball. Common bet types: 1H moneyline, 1H spread, 1H total.
Why first-half markets are useful. They isolate the starting personnel and the team's primary game plan, removing late-game variance from substitutions, garbage-time scoring, fouls, and pace adjustments. For bettors who specifically identify edges in starting lineups or first-half tempo, 1H markets capture those edges cleanly without the noise of full-game outcomes.
The MLB analogue. First-five-innings (F5) markets serve the same function in baseball — they isolate the starting pitchers and the lineups they face, removing bullpen variance. The MLB F5 article goes deep on why bullpen-free betting is structurally cleaner for pitcher-edge plays.
Hold and limits. First-half markets typically run slightly higher hold than full-game markets (5-7% vs 4.5%) and lower limits, because the books recognize the markets are more easily modeled than full games and want sharp action capped.
A bet on a long-term outcome — usually a championship winner, division winner, season-long award (MVP, Cy Young), or season win total. Futures pay out only when the underlying event resolves, which can be months in the future.
Why futures hold is high. Books typically run 20%+ hold on futures markets — far higher than the 2-5% on core single-game lines. The reason: futures stay open for months, the books face material adjustment risk if a team gets hot or cold, and recreational bettors love the long shot payouts. Books bake significant cushion into futures prices to protect themselves.
When futures can be +EV. Early-season pricing on teams whose offseason changes haven't been fully priced. Mid-season pricing on teams the market has overreacted to (down on a slump, up on a hot streak). Late-season pricing on teams positioned for a playoff run the market has missed. Sharp futures bettors track team-level fair value across the entire season, not just at the season-opening prices.
Hedging futures. A futures ticket that gets close to cashing (your Super Bowl pick is in the playoffs) can be hedged by betting the other side of subsequent games to lock in some profit. The math only works if the hedge captures more than the second-bet juice costs — see Hedge for the full breakdown. Read more on identifying mispriced futures →
Placing a bet on the opposite side of an existing position to lock in a guaranteed profit or limit a loss. Hedging trades expected value for certainty — you’re paying juice on the second bet to reduce variance on the combined position.
The classic hedge scenario. You bet a team to win the Stanley Cup at +500 in October for $100. By the Final, your team is one win away. Their opponent is now priced at -200 on the deciding game. You can “hedge” by betting the opponent for an amount that guarantees a positive return regardless of outcome. A $200 hedge on the opponent at -200 cashes a flat $100 if your original bet loses (you risked $200 to win $100), and cashes $500 − $100 = +$400 if your team wins. The hedge locks in profit either way; without it, you have a 50/50 between winning the full +$500 and losing your $100.
The math. A hedge is only worth taking if the guaranteed locked-in profit beats the expected value of holding the position. If your futures ticket is priced at fair value relative to the live odds, hedging just pays double juice. If the futures price moved meaningfully against the market (which is the normal case for a ticket that survived to a deep round), hedging captures the market’s revaluation as locked-in P&L.
Live betting hedges. When a line moves significantly in-game (your team leads by 21 at halftime), you can hedge by betting the live spread or moneyline on the other side. Same logic: lock in some of the unrealized gain by paying juice on the second bet.
When not to hedge. If your edge on the original bet hasn’t changed, hedging is just paying juice to reduce variance you should be willing to absorb. Recreational bettors over-hedge; pros hedge only when the expected dollar gain from certainty beats the dollar cost of the second-bet juice.
The percentage a sportsbook expects to keep on a market over time, after paying winners. Hold is the cumulative effect of vig across a balanced book — if the book gets even action on both sides of a market, the hold is what they retain.
The math at -110/-110. Both sides imply ~52.4% probability. The combined over-round is 4.8%. If the book takes $100 on each side, they pay out $191 to whichever side wins and keep the remaining $9. That’s a 4.55% theoretical hold on the $200 in total wagering. Even-money sounds fair until you do the math.
Hold by market type.
Why hold matters to bettors. Hold is the headwind. Your edge has to clear hold before any profit shows up. A 5pp raw edge on a 5% hold market produces 0pp of expected profit. The same edge on a 2% hold market produces +3pp. Sticking to low-hold markets isn’t glamorous — no one brags about taking core NBA spreads at sharp books — but it’s where the math actually works.
Sharp books vs square books. Pinnacle and Circa run low hold on core markets to attract sharp action. DraftKings and FanDuel run higher hold on the same markets because they target recreational bettors who don’t price-shop. The same bet at two books can have meaningfully different EV based on hold alone.
Betting half the bet size that full Kelly would recommend. Half-Kelly cuts the bankroll variance dramatically while preserving most of the long-run growth — the standard real-world sizing rule for bettors who care about both growing the bankroll and surviving it.
The math. Full Kelly maximizes the expected logarithm of bankroll growth assuming the bettor's probability estimates are perfectly accurate. In practice, probability estimates are noisy. A small overestimate of edge tells full Kelly to bet large amounts — sometimes disastrously large. Half-Kelly cuts that overestimate risk in half while only sacrificing about 25% of the theoretical growth rate. The trade-off is heavily skewed in favor of cutting back.
Why half-Kelly is the default. If full Kelly recommends 10%, half-Kelly recommends 5%. The reduction in drawdown variance is large (peak-to-trough swings shrink by roughly half), while the long-run growth rate only drops marginally. For most bettors, the survival math of smaller bets matters more than the theoretical-optimal growth of larger ones.
Quarter-Kelly and lower. Bettors with high uncertainty about their probability estimates (most quantitative models qualify) typically use quarter-Kelly or even smaller fractions. The smaller the fraction, the smaller the bet size, the safer the bankroll, and the slower the growth. Pick the fraction that lets you bet through normal drawdowns without panicking. Read the full Kelly guide →
The total dollar amount wagered on a market, sportsbook, or sport over a given period. Handle is the gross wagering volume — what bettors put down, not what the book kept. Handle minus payouts equals book revenue (also called hold dollars or gross gaming revenue).
Why handle matters. High-handle markets are more efficient because more money corrects mispriced lines faster. NFL Sunday core moneylines have the highest handle of any U.S. sports market, which is also why those markets carry the lowest hold and are hardest to beat on a per-bet basis. Low-handle markets (small-conference basketball, niche prop markets) are less efficient but capped at low limits, which makes them hard to scale.
Handle as edge signal. A specific game seeing handle disproportionate to its expected market size (a Tuesday night MLB game getting NFL-Sunday-level volume) usually signals sharp money has identified an angle. Books pay attention to handle distribution as a proxy for where sophisticated bettors are positioning.
The half-point on a spread or total — the ".5" in -6.5, +3.5, or 220.5. Hooks exist to eliminate pushes. A bet of -6.5 either wins or loses; -6 can push. The half-point gives the book a deterministic settlement on every bet and the bettor a deterministic profit or loss.
The cost of a hook. Each half-point is worth some amount of juice, varying by line. Hooks at key numbers cost the most. Buying off -3 to -2.5 in NFL might cost 20-25 cents of juice because so many NFL games land on exactly 3. Buying off -8 to -7.5 might cost only 10 cents because the half-point is less likely to be the deciding margin.
Why hook math matters. Over a season of betting at -110 standard prices, a bettor who consistently buys hooks at non-key numbers leaks meaningful EV to the extra juice. A bettor who only buys hooks at key numbers (where the half-point is most valuable) is capturing structural edge. Read more on how small price differences compound across volume →
The win rate at which a given moneyline bet breaks even — the percentage embedded in the price the sportsbook is offering. Converting odds to implied probability is the first step in deciding whether any bet has positive expected value.
The formulas. For positive American odds: implied = 100 / (odds + 100). For negative American odds: implied = |odds| / (|odds| + 100). Worked examples:
How edge is measured. Compare your fair-value estimate to the implied probability. If you think a team wins 48.3% and the price implies 43.9%, you have 4.4pp of raw edge. If you think they win 50% and the price implies 52.4%, you have negative edge and should pass. Every bet decision starts with this comparison.
The vig hidden in the two sides. Both sides of a -110/-110 market sum to 52.4% + 52.4% = 104.8%. That extra 4.8% over 100% is the sportsbook’s margin baked into the prices — the vig. A truly fair market would sum to exactly 100%; the over-round is how the book gets paid regardless of outcome.
Read the full guide on how implied probability drives the +EV decision →
Betting on a game that's already in progress. Lines update in real time as the score changes, possessions flip, and momentum shifts. In-game has exploded since 2018 as books built faster pricing infrastructure and bettors adopted mobile-first habits.
Why in-game markets carry higher hold. 5-10% is typical, well above pre-game's 2-4%. The reason: books are pricing under time pressure with incomplete information, and they need cushion against the bettor seeing something the algorithm hasn't priced yet. Live moneylines, live spreads, and live totals all carry that hold premium.
Where in-game edge lives. Bettors with sport-specific knowledge that beats the live algorithm. Examples: NBA bettors who know rotation patterns and can predict when a star is about to re-enter; NFL bettors who recognize game-script changes (a team that goes pass-heavy after falling behind); MLB bettors who track bullpen usage in real time. The edge isn't free — it requires watching games actively rather than placing bets and walking away.
Hedging in-game. Live betting is the natural venue for hedging pre-game positions. A bettor up significant unrealized profit on a halftime lead can lock in some of that gain by betting the live spread or live moneyline against their pre-game side. Whether the hedge is worth doing depends on the price you're paying vs the variance you're trading away. See Hedge for the math.
The sportsbook’s commission, baked into the odds rather than charged as a separate fee. Identical to vig. The reason -110 is not a true coin flip is that the book offers $0.91 for every $1 you risk — if you and an identical bettor on the other side both wagered $110, the book pays out $210 on the winning side but collected $220 total. The $10 difference is the juice.
The hold math. A standard two-sided market at -110/-110 carries roughly 4.55% theoretical hold on the combined wagering. Reduced-juice markets at -105/-105 cut that to about 2.4%. Over thousands of bets, that difference compounds: a bettor who consistently shops for reduced juice keeps roughly twice as much expected profit as one who always takes -110.
Where juice gets worse. Core moneyline and spread markets at sharp books carry low juice (2-3%) because they need to attract sharp action. Futures markets often carry 20%+ hold — the book builds in massive cushion because lines stay open for months and adjustment risk is high. Prop markets typically run 8-15% hold. Live in-game markets run 5-10%. The closer you stick to core markets at competitive prices, the smaller the juice headwind on every bet you place.
Beating the juice. The minimum bar for a winning strategy isn’t “positive expected value” in the abstract — it’s “positive expected value after the juice you’re paying on each bet.” A 5pp raw edge bet at -110 becomes a ~2.5pp net edge after vig. Strategies that look profitable on paper often die quietly to juice cost over volume.
Read how juice + raw edge combine to determine your true EV →
The mathematically optimal bet size for maximizing long-run bankroll growth, given an estimated edge and known payoff odds. Derived from log-utility maximization — Kelly is the unique bet size that maximizes the expected logarithm of your bankroll over infinite trials.
The formula. f* = (bp − q) / b, where b is decimal odds minus 1, p is your estimated win probability, and q is 1 − p. Worked example: 5pp edge on a +120 dog. p = 0.55, q = 0.45, b = 1.2. f* = (1.2 × 0.55 − 0.45) / 1.2 = 17.5% of bankroll. That’s full Kelly.
Why nobody bets full Kelly. The formula assumes your probability estimate is exact. If you say 55% but the true probability is 52%, full Kelly tells you to bet 17.5% — the math actually wants 0%. Probability estimates are always noisy, so full Kelly produces ruinous overbets when the model is slightly off. Real-world Kelly bettors use fractional Kelly: half-Kelly (8.75% in the example) or quarter-Kelly (4.4%) cut the variance dramatically while preserving most of the long-run growth.
Tiered sizing as discrete Kelly. Most published pick services use a tier system (1u, 1.5u, 2u, 3u) instead of computing a fresh Kelly fraction per play. The tier ladder is essentially a discrete approximation of fractional Kelly — bigger edge maps to bigger tier, with explicit caps that prevent any single bet from blowing up the bankroll on a model miss.
Read the full guide on Kelly for real bankrolls → · See how the upstream edge gets computed →
A directional model opinion that did not clear the threshold to become an official pick. Leans are useful context — they signal a side the model is interested in — but they're not sized aggressively because the edge isn't large enough or the signal isn't stable enough to risk full units on.
What separates a lean from an official pick. Official picks must clear multiple gates: minimum edge (typically 3-5 percentage points after vig), multi-signal agreement, and historical edge stability across the specific tier. A lean clears one or two of those gates but not all. Maybe the edge is real but small. Maybe one model component disagrees with the rest. Maybe the matchup type has a thinner historical sample than the model's confidence intervals require.
How to use leans. Free-tier subscribers see leans only. Paid subscribers see both leans and official picks. Think of leans as the model's "watchlist" — useful for confirming a sportsbook line shopping decision, for learning the model's reasoning style before committing, or for taking a small recreational position when the lean aligns with your own read. Never size a lean the way you'd size an official pick.
The change in odds, spread, or total between when a market opens and when it closes. Line movement is the betting market's running estimate of fair value being updated in real time as money flows in and as new information lands.
What drives it. Three forces. Sharp money (large, sophisticated bets) moves lines fast because books respect that action. Public money (high volume of small bets) moves lines slower and often in the wrong direction. Information (lineup changes, weather, injuries, late scratches) moves lines almost instantly when books detect the news.
Reading line movement. A line that moves against the public's betting percentage is "reverse line movement" — typically a sign of sharp action on the unpopular side. A line that moves in the direction of heavy public money is just absorbing volume. Sharp bettors pay attention to which side the LINE is moving toward, not which side the PUBLIC is betting on.
Why closing matters most. The closing line has absorbed every bet, every news update, and every adjustment the books made. That's why CLV measures performance against the close — it's the market's final, most-informed estimate. Beating the closing line on volume is the strongest available evidence of edge.
To bet the favorite — either giving points on a spread ("laying the points") or paying negative juice on a moneyline ("laying the juice"). Synonymous with backing the chalk. The bettor takes on the obligation of the favorite covering or winning outright in exchange for the negative-odds price.
The math. Laying -150 means you risk $150 to win $100. Your bet needs to hit 60% of the time just to break even. Laying -3 in NFL means you need your team to win by 4+ to cover; they can win by 3 and you push, win by 2 and you lose. Laying juice or points is structurally harder than getting them — which is why dogs and underdogs are statistically a more profitable bet shape for the average bettor.
When laying is correct. When your model's probability comfortably exceeds the breakeven that the negative price implies. Laying -150 (60% breakeven) with a model probability of 67% is +7pp edge — a clean +EV bet. Laying because "the favorite is better" without specifying a probability above breakeven is just paying juice with extra steps.
Comparing prices for the same bet across multiple sportsbooks and taking the best available number. Line shopping is the highest-ROI workflow habit in sports betting — a few minutes per bet captures expected value that compounds across thousands of bets per year.
Why it works. Sportsbooks don't post identical prices. The same game can have a moneyline of -148 at one book, -142 at another, and -135 at a third. The price difference reflects each book's own risk position, their internal model's projection, and the action they've already taken. A bettor who only uses one book is, by definition, paying random pricing — sometimes the best number, sometimes the worst, averaging out to mediocre.
The math. Across an MLB season of ~250 bets at $100 risked each, the difference between always taking the best of 3 books vs always taking the first line offered is roughly $1,000 of expected profit. That's a 4% ROI swing on identical bets, just from price discipline. Stretch the math across multiple seasons or multiple sports and line shopping becomes the single biggest controllable input to long-run profitability.
How to do it. Maintain accounts at 3-5 books with different pricing philosophies. DraftKings and FanDuel tend to lead on certain markets; Pinnacle and Circa (where legal) tend to lead on others. Use an odds-comparison tool or a manual rotation through your tabs. The few minutes per bet pay back dramatically. Read more on how line shopping amplifies edge → · Read how line shopping converts to positive CLV →
Synonymous with in-game betting. Wagering on a game that's currently being played, with lines that update in real time as the score, time remaining, and possession shift. See In-Game Betting for the full breakdown.
An underdog that the bettor believes is materially better than the moneyline price implies. "Live" in the colloquial sense doesn't refer to in-game betting — it means the dog has a real chance to win outright, not just a chance to cover the spread.
How a dog becomes "live." Several common patterns: a key injury or scratch on the favorite that the line hasn't fully absorbed; a structural matchup that favors the dog (slow-pace team facing a fast-pace favorite who can't run their offense); a road favorite playing in a hostile environment after travel; a rested goalie or starter facing a fatigued counterpart. The market often under-prices these spots because the headline numbers (team rankings, recent records) don't capture the matchup-specific dynamics.
A bet a tout or bettor claims is guaranteed to win. There is no such thing in sports betting. Every bet carries risk; "lock" is marketing language used to oversell confidence on plays that don't deserve it.
The reality. Even a 65% probability bet (which would be a massive edge over standard spread prices) loses 35% of the time. Across a season, that's 35 losses out of 100 such bets. A real sharp bettor talks about probability and edge, not certainty. Anyone selling "lock of the day" content is selling certainty they cannot deliver — and the bettors who buy it pay for the marketing in the form of bad EV.
A bet with a low implied probability of winning but a high potential payout. Futures markets (championship odds at +5000, season MVP at +2500) are the canonical long shots; high-leg parlays and exotic props also fit.
Why long shots are structurally bad bets. The hold on long shots is almost always 20%+ — books pad the price to protect themselves against unlikely outcomes that would pay out massively. Even when a long shot has positive expected value in the abstract, the variance is so high that a typical bankroll cannot survive enough trials for the math to converge. A +5000 bet that's a true 1-in-30 (+3.3% true probability) is +EV but takes hundreds of attempts to land profitably — and a single hit doesn't validate the strategy.
A bet on which team wins outright, with no point spread. Moneyline prices are expressed in American odds — positive numbers for underdogs, negative numbers for favorites — and reflect the implied probability of each side winning.
How the math works. A +150 moneyline says you'd win $150 on a $100 risk if your team wins. A -150 moneyline says you risk $150 to win $100. Both prices imply specific win probabilities: +150 = 40% breakeven, -150 = 60% breakeven (see Implied Probability for the formulas).
Why moneyline matters in low-scoring sports. In NHL and MLB, where margins of victory are typically 1-3 goals or runs, moneyline is the dominant betting structure — point spreads (puck lines, run lines) exist but trade away too much edge for the marginal bettor. In NBA and NFL, moneylines are still used but spreads carry more volume because the higher scoring rates make the spread the more natural risk-adjusted bet.
Dog moneylines are where edge typically lives. Public bettors disproportionately back favorites, which pushes favorite prices artificially high (and dog prices artificially low). A model that finds 48% probabilities at +130 (43.5% implied) is finding edge precisely because the public's bias against dogs lets those prices drift to +130 instead of fair value at +108. Read more on why dog moneylines are structurally mispriced →
The discipline of sizing bets relative to bankroll in a way that survives normal variance and maximizes long-run growth. Money management is the difference between a bettor with edge who profits over time and a bettor with the same edge who busts during an unlucky stretch.
The core principle. Bet size should scale with edge, not with confidence or excitement. A 5pp edge bet deserves a larger stake than a 2pp edge bet; both deserve a smaller stake than a 10pp edge bet. The mathematical framework for this is Kelly Criterion, applied at a fractional discount (typically half-Kelly or quarter-Kelly) to absorb model uncertainty.
The two failure modes. Over-sizing: betting too large relative to bankroll, which produces ruinous drawdowns during normal variance and ends the bettor's career before the math converges. Under-sizing: betting too small relative to bankroll, which means you never capture meaningful dollar profit even when your edge is real. The optimal sizing threads the needle: large enough to grow the bankroll meaningfully, small enough to survive expected drawdowns.
Bankroll separation. Money management starts with a dedicated bankroll separated from living expenses. A bettor who can't keep betting capital walled off from rent money has no money management — they have a tilt risk that will eventually empty both pools.
Slang for a low-probability bet with a high potential payout. Synonymous with long shot. Often used in the context of futures (preseason championship futures at +10000) or large parlays. Same structural math problems as any long shot: high hold, high variance, low probability of converging within a bettor's lifetime sample.
A bettor whose presence (or whose touted picks) seems to jinx outcomes. "Don't be a mush" is a friendly accusation between bettors after a brutal beat. There's no statistical reality to mushes — variance is variance and outcomes don't care about superstition — but the term captures the emotional reality of betting through cold streaks.
A game or market that a sportsbook has temporarily stopped offering. The book pulls the line — takes it off the board — because of breaking news (a star scratch, weather concerns, injury rumors that move the projection meaningfully) and won't accept bets until they've updated their model.
Why books take games off the board. Information arrives faster than the book's algorithm can re-price. Rather than offer a stale price that sharp bettors can pick off, the book pulls the market and reposts when they've digested the news. The temporary OTB period is the book's defense against being arbitraged.
What it tells the bettor. When a game suddenly goes OTB, something material is happening. Bettors who can identify the news driving the OTB before the line reopens can position for the reopen. This is hard — books pull games faster than most news services publish — but it's a recurring source of edge for bettors who track confirmed-pitcher feeds, lineup announcements, or injury reports in real time.
Also referred to as: "off the market," "no action," "frozen line."
A single bet that combines two or more individual bets ("legs"). All legs must win for the parlay to cash; any single losing leg loses the entire bet. Payout multiplies the individual leg prices together, which is why parlays look attractive (big payouts on small risk) and why they're structurally a bad bet for the average bettor.
The math against you. Each leg of a parlay carries its own vig. A two-leg parlay of -110 legs has roughly 9% hold (vs. 4.5% on either leg standalone). A four-leg parlay has 15-20% hold. A ten-leg parlay can have 30%+ hold. The book takes a bigger margin on parlays precisely because the math is hostile to the bettor.
When parlays are actually +EV. Only when the underlying legs are individually +EV AND uncorrelated. If you have a 5pp edge on Leg A and a 5pp edge on Leg B, parlaying them captures roughly compound edge — assuming the outcomes are independent. The catch: in practice most parlay candidates are correlated (e.g., a team's ML and their first-half ML), which destroys the compound-edge math and gives the book another structural advantage.
Same-game parlays. Same-game parlays (SGPs) are the worst structure for bettors. The book prices them assuming correlations they understand far better than the public — common SGP hold is 25-40%. Recreational bettors love them because the payouts look enormous; sharp bettors avoid them entirely.
The recreational case. If parlays are pure entertainment ("I want $5 to potentially win $400 on this longshot card"), the math is irrelevant — you're paying a small known cost for a thrill. Just don't confuse them with a serious profit strategy.
The performance difference for a baseball hitter or pitcher against same-handed vs opposite-handed opponents. Most hitters perform meaningfully better against opposite-handed pitchers (a left-handed hitter facing a right-handed pitcher, for example). Most pitchers perform better against same-handed hitters.
Why it matters in betting. Public bettors and sportsbook lines typically use season-aggregate stats (a team's overall wRC+, a pitcher's overall FIP). Platoon-aware models look at the specific matchup: how does this lineup project against this pitcher's handedness profile? A team that's a 105 wRC+ overall might be 125 vs. left-handed pitching and 92 vs. right-handed pitching. Facing a lefty starter, they're a meaningfully better offensive bet than their season stat suggests.
The asymmetry sharp models exploit. Late-game pinch-hitter substitutions are one of MLB's biggest hidden price-moving events. Managers will pull a same-handed hitter for an opposite-handed pinch-hitter in high-leverage spots. The market is slow to price these substitutions because they happen late and lineups aren't always confirmed in advance. Models with handedness-aware projections capture edge here that aggregate-stat models miss entirely.
A bet on a specific event within a game that doesn't depend on the final score — things like "will Player X score over 18.5 points?", "first team to score?", "anytime touchdown scorer?". Props proliferated with the rise of legalized betting in the U.S. and now make up a meaningful share of recreational betting volume.
Why prop markets are inefficient. Player props are graded individually by each book based on internal models — and those models vary wildly in sophistication. A starting NBA player's points prop might be set at 24.5 at one book and 22.5 at another. That's not a 1-point line difference; that's the books disagreeing on the projection by enough to create exploitable gaps. Line shopping props is more valuable than line shopping spreads.
Why books accept the inefficiency. Prop limits are small. Most books cap player props at $500-$2,000 per side, which means even if you find a 10pp edge, you can't bet enough volume to hurt them. The combination of small limits, high hold (often 8-15%), and high recreational appeal makes props a profitable market for sportsbooks despite being beatable in the abstract.
Sharp prop betting. Specialists who focus on a single sport's props (NBA player points, NFL receiving yards, MLB strikeouts) and build dedicated projection models for those markets find edge that generalist bettors miss. The work is high — every individual game requires lineup confirmation, recent-trend updates, matchup adjustment — but the per-bet edge can be 5-15pp, far above what's available in core moneyline markets.
The NHL equivalent of a point spread. Almost universally set at ±1.5 goals — the favorite must win by 2+ goals to cover, or the underdog must lose by 1 goal (or win outright) to cover. Standard juice is meaningfully off -110 because of the asymmetric scoring structure of hockey.
Why -1.5 is a real obstacle. Roughly one-quarter of NHL games are decided by exactly 1 goal (the "loser point" rule guarantees a meaningful share of OT/SO finishes). A favorite winning by exactly 1 covers neither -1.5 (loses the spread) nor pushes (no 1.5-goal pushes). That's why favorites priced at -200 on the moneyline often run +160 or +180 on the puck line — the math of "must win by 2" is genuinely hard, and the price compensates.
Where the value lives. Public bettors overpay for favorite puck-line tickets ("I want better than -200 odds, so I'll take -1.5 +180"). Sharp models find value in the opposite direction: home underdog +1.5 at -180 cashes whenever the underdog wins outright OR loses by exactly 1. In a sport where one-goal losses are common, that's a much higher cover rate than +1.5 implies for the underdog.
How models size puck-line edges. A puck line is fundamentally a derived market: the model needs both a moneyline win probability AND a margin-of-victory distribution. Models that calculate puck line fair value from raw goal projections (not just moneylines) tend to find structural mispricings the public misses. Read more on derivative markets and edge →
A bet that finishes exactly on the line, resulting in the stake being refunded with no profit and no loss. The most common scenario is a -3 NFL favorite winning by exactly 3, or a total set at 220.5 in NBA finishing exactly at 220.
Why books use half-points (hooks). Pushes are revenue-neutral for the book — they refund the stake. Half-point lines (-3.5 instead of -3) eliminate pushes entirely, which lets the book guarantee a winner/loser on every bet and collect the vig deterministically. The half-point itself is worth meaningful expected value to one side or the other.
Push frequency. NFL spreads push on key numbers (3, 7) more often than the average game would suggest because game results cluster at those margins. Buying a half-point off -3 to -2.5 in NFL might cost 15-20 cents of juice precisely because of how often games land on 3. Across a season, the cost of buying hooks adds up; for that reason, most sharp bettors only buy half-points off the most important key numbers (3 and 7 in NFL, 5 and 7 in NBA).
Parlay impact. In a parlay, a single-leg push reduces the parlay by one leg — the remaining legs still need to win, but the parlay payout drops to reflect the smaller number of active legs. A two-leg parlay with one push and one win pays as a single straight bet. A three-leg parlay with one push pays as a two-leg parlay.
The historical results record of a bettor, model, or system. Past performance is a useful diagnostic only when accompanied by sample size, time horizon, and methodology disclosure — and even then it doesn't predict future results.
The two ways past performance lies. Cherry-picked windows (showing only the months when the strategy worked) inflate apparent edge. Insufficient sample (50-100 bets dressed up as proof) can show any result by variance alone. A 60% NFL ATS record across 30 bets is statistically indistinguishable from random; a 53% record across 1,000 bets is meaningful evidence of edge.
Why CLV beats past performance for evaluation. Past performance is a lagging indicator dominated by variance over short windows. CLV is a leading indicator that converges to a real answer faster. A strategy with positive CLV across 200 bets is almost certainly profitable long-term, even if the 200-bet ROI is negative due to variance. A strategy with negative CLV across the same sample is almost certainly losing, even if the ROI happens to be positive. Read why CLV evaluates a process while past performance evaluates an outcome →
A game with no point spread — both teams are projected as equal, and the bet settles on the moneyline alone. Pick'em lines appear on tightly-matched games where the book's model can't justify making either side a favorite by even half a point.
Why pick'em is interesting. Pick'em moneylines are usually priced -110/-110 or -115/-115 — meaningfully lower juice than the typical -125 to -150 favorite/-105 to +130 dog structure. That makes pick'em games one of the lowest-vig core markets available. For bettors with a real edge on which team is going to win, pick'em markets convert that edge to expected dollars more efficiently than higher-juice favored markets.
The inverse of a teaser. A pleaser modifies spread or total lines in the sportsbook's favor (e.g., moving a +3 underdog to -3) in exchange for dramatically higher payout odds. All legs must still win for the pleaser to cash.
Why pleasers exist. They're the book's equivalent of a parlay with the math weighted against the bettor even more aggressively. A two-team six-point pleaser typically pays around +700 (7-to-1). The math required to make a pleaser +EV is brutal — the bettor needs to be right not just on direction but on margin by a significant amount on every leg.
When pleasers make sense. Almost never. The implied probability of hitting a typical two-team pleaser is around 12-15%; the payout implies the same. Books offer pleasers because recreational bettors are attracted to the long-shot payout. Sharp bettors don't touch them.
Any bet with positive American odds (+105, +130, +250, etc.) — meaning the bet pays more than 1:1. Plus-money bets are typically underdogs (winning is less likely than 50%). The terminology contrasts with "minus money" or "juice" on favorites.
Why plus-money bets are the dog-bettor's home base. A +130 bet returns $130 on $100 risked — meaningfully better than the $90.91 return on a -110 bet. Across a season, plus-money bets at edge-justified prices produce higher dollar profit than minus-money bets at the same edge percentage. That's why dog moneylines and home-dog plus-money plays are the volume backbone of most sharp model output.
The collective betting behavior of recreational bettors — high volume of small bets, biased toward favorites, marquee teams, and overs in totals markets. The public's predictable patterns are a recurring source of edge for sharp bettors who position opposite the consensus.
Why the public bets predictably. Recreational bettors back teams they like watching (Cowboys, Lakers, Yankees), teams that just won (recency bias), and outcomes that are more fun to root for (overs, favorites). They don't price-shop and they don't track CLV. The aggregate effect: sportsbooks see disproportionate action on certain sides, which forces them to shade lines against that public action to balance their book. The shading pushes prices past fair value — and that's where sharp edge lives.
Reading public betting percentages. Most sportsbooks publish (or third parties scrape) the percentage of bets on each side of a market. A line where the public is 75% on one side but the line moves toward the other side ("reverse line movement") is one of the cleanest sharp signals — it means large bets are coming in on the unpopular side, which is almost always sophisticated money.
Fading the public isn't automatic. Public-side bets aren't always wrong. Public bettors sometimes back legitimate favorites who deserve the price. The edge comes from identifying the spots where public-driven shading has moved the price past fair value — not from blindly fading every popular side. Read more on public bias as a source of mispriced odds →
The MLB equivalent of a point spread, almost always set at ±1.5 runs. The favorite must win by 2+ runs to cover; the underdog gets +1.5 runs added to their final score. As with NHL's puck line, the ±1.5 structure creates dramatically different juice from the moneyline price.
Why -1.5 is harder than it looks. Roughly one-third of MLB games are decided by exactly 1 run. A team winning 4-3 covers the moneyline but loses the run line. That's why a -160 moneyline favorite typically runs +120 or +130 on the run line — the structural difficulty of winning by 2 in a low-scoring sport is genuinely worth the price differential.
Where the value lives. Two angles. First, plus-money favorite run lines (Team X at +130 RL) where the model projects a multi-run win driven by a starter mismatch or bullpen edge — that's a +EV bet because the price overcompensates for the difficulty. Second, home underdog +1.5 RL at a juicy minus number where the dog loses by exactly 1 frequently enough to make the cover rate beat the price.
Why the run line is a derivative market. Like puck lines, run lines are derived from a moneyline + margin-of-victory distribution. Models that project run differential directly (not just W/L probability) find structural mispricings that simple moneyline models can't see. Read more on price-vs-fair-value across markets →
Net profit divided by total amount risked, expressed as a percentage. The single most useful metric for comparing bettor performance across different sample sizes, bet structures, and bankroll sizes.
The formula. ROI = (Net Profit / Total Risked) × 100. Worked: across 100 bets at $100 stake each, $10,000 was risked. If the bettor netted $300 in profit, ROI = ($300 / $10,000) × 100 = 3%.
Why ROI beats win rate. Two bettors can post identical win rates with radically different profitability. A 53% NFL ATS bettor (-110 standard) nets about 1.3% ROI — barely above breakeven. A 49% MLB underdog bettor averaging +140 on their plays nets about 12% ROI on the same sample size — profitable despite a sub-50% win rate. Win rate without payout context is meaningless.
Sample size and ROI noise. Even at a true +5% ROI strategy, individual 100-bet windows will range from -5% to +15%. ROI converges to its true value over hundreds to thousands of bets, not dozens. Anyone publishing a flashy ROI on 30 bets is selling noise, not signal.
ROI vs CLV. ROI tells you what happened. CLV tells you whether your process is sound. Over short windows ROI lies; CLV doesn’t. Both matter, but if forced to pick one for evaluating a bettor’s long-run prospects, CLV is the more honest indicator. Read why CLV predicts ROI before ROI does →
A betting structure that combines multiple parlays from the same set of selections. A bettor who picks 4 teams in a "round robin by 2s" gets every possible 2-team parlay combination from those 4 selections — six parlays in total, each independently graded.
Why bettors use round robins. They reduce parlay variance. Instead of needing all 4 legs to win for a single payout, the bettor can have some parlays win even when one or two legs lose. The trade-off: total stake is much higher (6 separate bets) and the per-parlay edge math still applies — every leg's vig compounds in every parlay.
The math reality. Round robins are still subject to the same parlay math problems. Each underlying parlay carries 9%+ hold. The structural feature of "covering some parlays even when one leg loses" is just spreading the bet across more individual parlays — it doesn't make the underlying math friendlier. Round robins are slightly less variance than a single big parlay, but they don't fix the EV problem.
A bettor who consistently finds edge, beats the closing line, and gets respected (or limited) by sportsbooks because of it. "Sharp" describes both individual bettors and the books that cater to them (Pinnacle, Circa, Bookmaker) where lines are set to attract sophisticated action rather than recreational dollars.
What makes someone sharp. Three things, in order. First, a process that estimates true probability better than the market on a defined set of bets — usually a quantitative model, sometimes deep specialist knowledge of a single sport or market. Second, the discipline to only bet when the edge is real and to size bets to the edge (not to gut feel). Third, consistent positive CLV across hundreds of bets, which is the math-honest measurement of whether the first two are actually working.
How books detect sharps. Account-level CLV tracking is the primary mechanism. Books also watch for bet timing (sharps often bet right when lines open or right when news breaks), bet structure (sharps don't parlay heavily), and price-shopping behavior (sharps move accounts to whichever book has the best line). Most retail books limit sharp accounts to $50-$200 per bet within months of identifying them.
Sharp vs square markets. A "sharp book" like Pinnacle takes high limits, runs low hold, and adjusts lines based on the largest bets. A "square book" like DraftKings caps individual bets low, runs higher hold, and adjusts lines based on volume of public action. The same game can have meaningfully different prices at the two — and that gap is where line shopping pays off. Read more on the +EV mindset that defines sharp bettors →
The handicap a sportsbook adds to one team to balance betting action between two sides. The favorite "gives" points; the underdog "gets" points. Standard juice is -110/-110, meaning the book is positioned to take 4.55% hold if action lands evenly on both sides.
How spreads work. If the spread is -6.5 for the favorite: the favorite must win by 7+ to cover the spread; the underdog covers if they lose by 6 or fewer, or win outright. Half-point spreads (-6.5, +6.5) eliminate pushes; whole-number spreads (-7, +7) allow the result to land exactly on the line, refunding both sides.
Why spread markets are designed to be 50/50. The spread is the line at which the book believes equal action will land on both sides. If 80% of bets are pouring in on the favorite, the book will move the spread (e.g., from -6 to -7) to attract action on the underdog. The end result of all this market-making is a closing spread that reflects the consensus of sharp money + adjusted-for-public-action — i.e., the most efficient estimate of margin-of-victory the market produces all week.
Where edge lives in spreads. Three sources. First, key-number arbitrage: NFL games cluster at margins of 3 and 7 more than the spread suggests, so half-points at those numbers are worth meaningful EV. Second, late-breaking news that hasn't fully moved the line. Third, models that capture matchup-specific drivers (pace adjustment in NBA, pitcher leverage in MLB) the consensus market is slow to price. Read more on finding mispriced spreads →
A rapid, coordinated line movement across multiple sportsbooks within a short window — typically minutes. Steam moves usually indicate large sharp action hitting the market simultaneously. The line at one book moves; other books copy the move to avoid being left with stale prices that sharps could arb against.
Why steam matters. Steam is a leading indicator of where the closing line will land. If a line moves from -3 to -4 on a steam move two hours before game time, the closing line will almost certainly be at -4 or worse. Bettors who get on the move before the steam are taking the favorable price; bettors who wait until after lose the opportunity.
Chasing steam vs anticipating it. "Steam-chasing" is betting on a steam move after it's already happened. Most recreational bettors do this — they see the line moved and assume the sharps know something. The problem is that by the time the move is visible, the favorable price is already gone and you're often betting the same side at a worse number than the sharps got. Sharp bettors don't chase steam; they anticipate it from their own models or from independent reads on news/lineups.
What steam reveals about the market. Consistent steam in one direction across a season often signals a structural inefficiency — a category of bets the public reliably misreads. Sharp bettors who identify those patterns can position ahead of future steam rather than reacting to it. Read more on why closing-line value is the math behind steam →
Slang for a sharp bettor. Used interchangeably with "sharp" (see Sharp for the full definition). Generally describes a bettor with a documented track record of positive CLV, edge-driven sizing, and account limitations from books that have identified them as a long-term winner.
A recreational bettor whose process consistently produces negative expected value. Squares lose money over time because they bet favorites, parlays, marquee teams, and overs without regard to price — and they don't track CLV or shop lines. The opposite of a sharp.
What makes a bettor a square. Three habits, in order. First, betting on which team they think will win rather than on whether the price is right. Second, taking the first line they see without comparing to other books. Third, sizing bets emotionally — bigger after a win, smaller after a loss, biggest on the games they care about. The combination produces a steady drip of negative EV that accumulates into a season-ending loss.
Square sportsbooks vs sharp sportsbooks. The terminology applies to books too. DraftKings, FanDuel, and BetMGM are "square books" because their pricing model targets recreational bettors — they cap sharp bettors quickly and balance lines against public action. Pinnacle and Circa are "sharp books" because they take high limits, run low hold, and adjust lines based on the largest bets. Same game, materially different prices.
Why books love squares. Square bettors are the books' core profit center. The math of -110/-110 spread markets only works for the book if a large share of the bettors are systematically wrong. Squares provide that share. Sharp bettors profit because the books need square bettors to fund the long-run economics of the market — and the sharp side captures the EV the books leave on the table after squaring up their books against public action. Read more on the structural divide between sharp and square →
A pre-set limit on the amount a bettor will lose in a given day, week, or session. When the loss threshold hits, the bettor stops betting until the next defined period. Stop losses exist to prevent tilt — the emotional escalation that destroys bankrolls during cold streaks.
Why stop losses help discipline more than math. Mathematically, stop losses don't improve long-run EV. If your bets are +EV, walking away from a losing session means missing additional +EV opportunities. But mathematics assumes infinite emotional control, which no human has. The real value of a stop loss is preventing the bettor from chasing losses with progressively larger bets — the behavior that turns a normal drawdown into bankruptcy.
How to set one. A common rule: max daily loss = 3-5% of bankroll. With $10K bankroll, that's $300-$500 per day. Hit the limit, log out, come back tomorrow. The rigidity is the point — anything that requires in-session judgment defeats the purpose.
Stop losses vs unit discipline. A bettor with rigorous unit sizing rarely needs a stop loss because their bet sizes are already calibrated to bankroll. Stop losses are most useful for bettors who occasionally deviate from unit sizing (placing extra bets, increasing sizes after a win, etc.) and need a hard backstop against the worst version of themselves. Read more on bankroll discipline →
A bet on the combined final score of both teams, irrespective of who wins. The bettor takes either the over (combined score exceeds the line) or the under (combined score is less than the line). Standard juice is -110/-110.
How totals are set. The sportsbook projects expected combined points/goals/runs from team offensive ratings, defensive ratings, pace (in NBA/NHL), starting pitcher quality (in MLB), weather (where relevant in outdoor sports), and a vig adjustment. The opening total reflects the book's pre-action estimate; the closing total reflects how money flowed during the betting window.
NBA totals. Sensitive to pace — the same two teams playing at 105-pace produce a total in the 230s, at 95-pace produce a total in the 210s. Pace-adjusted efficiency is what makes NBA totals tractable for models that go beyond surface stats. (The NBA pace article covers the model's approach in depth.)
MLB totals. Driven by starting pitcher quality, park factors, weather (wind in particular), and umpire zone tendencies. The F5 total — first-five-innings only — is one of the cleanest totals markets in baseball because it isolates the starters and removes bullpen variance. The F5 article covers why this matters.
NHL totals. Lower-event sport (typical totals are 5.5 or 6 goals). A single goal swings the result decisively, and goaltender form is the largest driver. Read more on how goalie workload moves NHL totals →
The verb form of teaser. "I teased the Cowboys from -7 to -1" means the bettor moved the spread by 6 points in their favor by accepting a teaser structure. See Teaser for the full mechanics.
The emotional state in which a bettor abandons discipline and starts making decisions driven by frustration, anger, or desperation rather than process. Tilt typically follows a string of losses — the bettor chases by increasing bet sizes, betting on games they wouldn't normally touch, or grasping at long-shot parlays to "get back to even."
Why tilt destroys bankrolls. The math of betting only works at calibrated bet sizes on edge-justified bets. Tilt produces oversized bets on poorly-evaluated games — the exact recipe for ruinous drawdowns. A bettor who tilts once a month and bets 3-5x their normal unit size during the tilt session can blow up an otherwise winning year in a single bad weekend.
How to prevent tilt. Pre-commitment mechanisms. Stop losses are the most common — a hard daily limit on losses that forces the bettor to walk away before tilt escalates. Other techniques: pre-defined unit sizing rules (no exceptions), automatic deposit caps at the sportsbook level, separating the betting bankroll from any other money source (so chasing losses is constrained by the dedicated pool).
Tilt isn't a character flaw. It's a predictable human response to variance. Every bettor experiences it. The successful long-term bettors aren't the ones who never tilt — they're the ones who set up structural defenses so that tilting can't blow up the bankroll. Read more on bankroll discipline that survives tilt →
A market where both sides are bettable (moneyline, spread, total). Contrasted with one-way markets like futures (only "to win the championship" — no opposite side). Two-way markets are inherently more efficient because arbitrage and balancing action constrain the prices on both sides.
Why two-way matters. When a book takes heavy action on one side of a two-way market, they can adjust the line and rely on the other side to absorb action. The market self-corrects. One-way markets don't self-correct because there's no opposing side to balance — that's why futures hold is so much higher than core moneyline hold.
The implied probability tell. A perfectly fair two-way market would have both sides' implied probabilities sum to exactly 100%. Real markets sum to 102-110% — the over-round is the book's vig. The smaller the over-round, the sharper the market and the lower the hold. Comparing over-rounds across books is one way to identify which book is sharpest on which markets.
A standardized bet-sizing convention, almost always defined as 1% of bankroll. With a $10,000 bankroll: 1u = $100, 1.5u = $150, 2u = $200, 3u = $300. The convention exists for one reason — it lets pick publishers and bettors compare performance across wildly different bankrolls.
Why 1% is the standard. 1% per bet survives normal variance comfortably. A 20-bet losing streak at 1u costs 20% of bankroll — painful but recoverable. The same losses at 5% per bet costs 100% and ends the game. The 1u = 1% convention emerged because it sits at the sweet spot between “big enough to matter” and “small enough to survive variance over thousands of bets.”
Tier sizing. Most pick services tier bet sizes by conviction: 1u (standard play), 1.5u (slightly stronger edge), 2u (high-edge play), 3u (rare high-conviction). The tier ladder is essentially a discrete approximation of fractional Kelly — bigger edge gets a bigger bet, capped at the highest tier to prevent any single play from blowing up the bankroll.
Honest unit reporting. The integrity test for a pick service is whether they publish their unit-sized record honestly — including the losing days. A unit-denominated record is the only way to compare track records across services with different bankrolls, betting styles, and dollar amounts. Anyone reporting only dollar profit or only win rate is hiding either how big they bet on winners (size variance) or how often they bet at all (selection bias).
Read the full guide on tier-based unit sizing → · See how edge size determines unit size →
Short for vigorish. The commission built into sportsbook odds — functionally identical to juice. The reason both sides of a market priced at -110/-110 sum to roughly 104.8% implied probability instead of 100% is that the sportsbook needs that 4.8% over-round to cover their cost of operating and produce a profit margin.
How vig affects every bet. Standard NFL and NBA spreads at -110 require you to win 52.4% of bets just to break even on a season’s volume. Hit 50% and you lose 2.4% of every dollar risked. Hit 55% and you net 2.6% — the 5pp edge above breakeven is your real ROI, not the raw 5pp above 50%.
Beating the vig. Every betting strategy’s first job is to clear the vig. Strategies that look profitable in raw win-rate terms but don’t survive vig math go broke quietly. This is also why CLV is the standard sharp metric: a bettor with positive CLV is, by definition, getting prices the market would later regard as too generous — meaning they’re overcoming the vig structurally, not occasionally.
See also: Juice · Hold · The +EV framework that survives vig →
A backtesting methodology where the model is trained only on data prior to each test period, then tested forward in time. Walk-forward is the methodological gold standard for honest model evaluation because it eliminates look-ahead bias — the model cannot see future data during training.
How it works. Suppose you have 5 years of game data (2020-2024) and want to validate a model. Period 1: train on 2020-2022, test on 2023. Period 2: train on 2020-2023, test on 2024. Each test period is genuinely out-of-sample. Performance averaged across all walk-forward windows reflects what the model would have produced live, not what it can curve-fit on the full dataset.
What walk-forward catches. Overfitting (model tuned to historical noise rather than real signal), look-ahead bias (accidentally using future information at decision time), and time-period dependence (a model that worked great in 2021 but failed in 2024 will show that decay across walk-forward windows instead of being masked by aggregate stats).
What walk-forward doesn't catch. Selection of which features to use is still potentially curve-fit by the modeler (the human picks features based on hindsight about what worked). Live deployment is the only true validation; walk-forward is the strongest pre-deployment check. CLV after deployment is the post-deployment check that can't be backtest-gamed →
A bet. Used interchangeably. "I placed a wager on the Cowboys" and "I placed a bet on the Cowboys" mean the same thing. "Wager" tends to sound more formal and appears more in industry/legal contexts; "bet" is the conversational default.
A bettor who places extremely large wagers — typically $50,000+ per bet, sometimes seven figures. Whales are a tiny share of the betting population but move meaningful action and can shift lines on their own when they bet at sharp books.
Whale economics. Sportsbooks have a complicated relationship with whales. A large losing whale is enormously profitable; a large winning whale can wipe out a book's month in a single weekend. Books carefully evaluate large-account behavior to decide who they'll take action from. Recreational whales (who happen to bet big but bet square) are welcomed; sharp whales are limited fast.