This page is loading and it needs JavaScript. The written page below is served without it.
A research tool for player and team props in NBA, NFL, MLB and WNBA. For every prop it shows how often that player has cleared that exact line in their own game log, how many games stand behind that rate, and the price posted at 7 sportsbooks, so the record and the price sit side by side before you decide anything. It does not take bets, sell picks or predict results.
Four surfaces, one board behind them, so a number never changes when you click through to it. The prop board is the home page: every prop on the slate, ranked by hit rate, weighted by sample, filterable by odds and book. The AI research assistant answers questions in plain English, grounded in the box scores, the posted odds and the hit rates the board reads. Player and team research draws game by game hit rate charts with splits by home and away, opponent and timeframe across every league we cover. The parlay builder takes your leagues, legs, wager and risk, assembles a ticket from tonight’s scored pool under a correlation cap, and shows its working on every leg. Around them: odds compared across 7 sportsbooks, and bet tracking graded on the box score.
Yes. Parlay Builder publishes an MCP connector, so Claude or ChatGPT can read the research directly instead of you reading it out to them. The board in your chat: tonight’s ranked props, a player’s hit rate across five windows, two props compared and the graded record, asked for in plain English and answered from the same data the site reads. Read only, and scoped to you: every tool is a read, it cannot place a bet, move money or change anything in your account, and the key reaches no other subscriber’s data. Our research, not a price feed: it carries measured history and the board’s own ordering, and sportsbook prices stay on the site, where our licence permits showing them. How to connect it is on the AI connector page.
A general chatbot generates the numbers along with the sentence, so a hit rate it quotes may be a plausible-looking invention. Parlay Builder’s chat is mostly not a language model: it routes your question to an intent and then runs ordinary code against the same database the board reads, so the figures come back as cards with the sample and the source attached. A model is used for exactly two jobs: working out which question you asked, and reading a bet-slip screenshot you upload. It is never asked what a player’s hit rate is, because it does not know and we do.
No. It is research tooling. There is no wallet, no deposit, no wager and no settlement of anything with money in it. When you have decided on a bet you take it to your own sportsbook. That also means nothing here is advice: it describes what has already happened at a price you are being offered, and the decision stays yours. If betting has stopped being entertainment, help is at 1-800-GAMBLER (1-800-426-2537).
Six kinds of thing: find what is worth looking at on tonight’s board, compare players or a whole market on one measure, explain what a number means, build a ticket from the scored board, diagnose a leg you are already considering, and track how the slips you saved settled. What it will not do is answer from data we do not have: weather, pitch mix, park factors, injuries and a correlation figure are all outside the feed, and it says so plainly rather than guessing.
NBA, NFL, MLB and WNBA, for both player props and team markets, priced across DraftKings, FanDuel, BetMGM, Caesars, BetRivers, theScore Bet and Hard Rock Bet. NHL is deliberately not offered: there is no scoring model behind it, so listing it would advertise something the product does not run. Coverage of a league is only as live as its season. An off-season league keeps its full history and has no prices. Counted at 15 Nov 2026, 09:17 UTC, when this page was built: NBA, 61,041 player games of history and 893 priced props on the board; NFL, 14,044 player games of history and 483 priced props on the board; MLB, 395,459 player games of history and no player props posted at the time; WNBA, 17,790 player games of history and no player props posted at the time.
Counted from the box score: Hit rates are measured over a player’s actual game log, with the sample printed next to every rate. Priced against the live line: Every comparison runs against the odds actually posted at DraftKings, FanDuel, BetMGM, Caesars, BetRivers, theScore Bet and Hard Rock Bet, not a house number. Graded when the game ends: Tracked bets settle against final official stats. Wins and losses both stay on your record. Empty when the slate is empty: Off night? The board says so. We would rather show you nothing than show you something made up.
It does not sell picks or forecast outcomes: every number on the board is a measurement or a model estimate, and both can be wrong. It does not cover every league, because a league is offered only once the model can score it. And it does not hide the losers: props that miss the bar stay visible under Market Watch with the reason on the card, and the slips you track settle with losses beside wins.
Two feeds, kept separate. Results come from official box scores, the same per-player game logs the leagues publish, and every hit rate on the site is counted from those, with the number of games printed beside it. Prices come from DraftKings, FanDuel, BetMGM, Caesars, BetRivers, theScore Bet and Hard Rock Bet: 7 books, and the ingest drops any other book before the data lands, so no price on the site is from somewhere we cannot name. The row counts per league, and when they were counted, are in the answer on which leagues are covered.
How often a player has cleared one exact line in the games behind it, counted from the official game log. It is counted at the whole number the bet settles on, so Over 24.5 needs 25 or more, and only over games the player actually played. A record of past games, not a forecast of the next one. It is not a win rate: a win rate is how often your bets win, and a hit rate is a count of past games at one number. The windows are the last 3, 5 and 10 appearances plus this season and last, never games his team played without him, and each one prints the games behind it. Rates also move with context, home or away, opponent and rest, and every split carries its own sample.
Five games the player actually appeared in: never five games their team played. For a starting pitcher those are very different windows: five starts can span five weeks, while his club plays most nights. A game the player missed leaves the denominator rather than counting as a miss, and a game he played and recorded nothing in stays in it as a real zero. Postseason and spring games are in no window; every rate on the board is regular season only, counted at the exact line posted.
Not one window, but five. Last 3, last 5, last 10, this season and last season are shown together, each with the games behind it, so a hot fortnight is visible as a hot fortnight rather than as a season. A short window on its own is the most misleading number in this whole category: three games admit only four possible percentages, and one of them is 100%. Shown next to the ten-game and season rates it stops being a claim and becomes what it is, a recent streak with its length attached. Split them by home and away, by opponent and by days of rest, and chart the game-by-game log against the line underneath. Every split carries its own sample size, and a split too thin to mean anything says so rather than printing a confident percentage over four games.
Yes. Look a line up on its own, without building anything. Pick a player, a stat and a number, and the page answers the only question that matters first: how often has this actually happened, and over how many games. The research page and the player pages both do this.
Two readings, not a rating of the player. The letter grade starts from the measured hit rate: how often that player has cleared that exact line in their own game log (A 80%+ · B 70 to 79% · C 55 to 69% · D 50 to 54% · F under 50%). The sample caps the letter: under 10 games it can be no higher than C, and under 15 games no higher than B. A line that has not cleared more often than its own price needs to break even is held at B or lower. The number of games behind the rate is printed beside the letter. Separately, hit rate vs price compares that rate, pulled toward the market in proportion to how little history stands behind it, with the probability the posted price implies, with the sportsbook’s margin left in. A high grade says the line has been cleared often; a positive hit rate vs price says the history has run ahead of the price. Neither says anything about tonight.
A measured hit rate of 80% or better on that exact line, and more than the band. An A also needs at least 15 games and recent games that back the rate up; where recent games are not on hand, only 90% or better is an A. These checks can only hold a letter down, never raise it. Below 22% at +150 or longer it reads Longshot instead of F. The thresholds are absolute rather than a curve: a weak slate is supposed to grade badly, where a percentile scale would mint the same number of A grades every night whatever was on offer. An A describes the record, not tonight.
No. A high grade is history, not a forecast: it means the line has been cleared often in the games on file. A letter is a band of that past hit rate, not a forecast win rate and not a ranking of bet quality. When the ordering was tested against results, no band of it came out ahead. And even a probability estimate that is exactly right loses whenever the odds say it should: a true 60% loses about 4 times in 10. An A can lose and an F can win, and neither outcome means the grade was wrong, because the grade is a statement about a record rather than a prediction about a result. Any tool that tells you otherwise is selling you something. Every individual bet can lose, and a parlay multiplies both the payout and the ways it can fail.
By hit rate, weighted by sample: 55% over 200 games ranks above 58% over 12, because a long record is harder to fluke than a short one. Prices from -300 to +250, the range most readers can actually take, sort above the extremes either side of it. The order describes history; it is not a Parlay Builder prediction about which props will win. Each card also carries a letter grade from the player’s own game log and, where at least two sportsbooks quote both sides of a line, the market’s own fair price: their prices with the built-in margin taken back out, and the median across those books. A prop only reaches the ranked board at all if its hit rate clears the price’s break-even by a minimum margin, it holds a minimum grade, and it has a ten-game sample and a price the book has confirmed recently; the rest of the slate stays on the page under Market Watch, with the reason on every card.
It is the gap, in percentage points, between the measured hit rate and the break-even implied by the price you are being offered, after a short record is pulled toward that price and with the sportsbook’s margin left in. A reading of +2 means the history has been slightly ahead of that price. It is a description of what has already happened, not a prediction. Most of a well-priced board sits near zero, and a large number is more often a stale price than an opportunity.
The slate is every posted player line for the day, laid out game by game, so you can look up a particular player or work through one fixture. Top Props is the model’s ranked shortlist across every game, filtered to the props that cleared its own eligibility bar: most of the slate does not. The slate shows you what exists; Top Props shows you where the model thinks the price is wrong. Neither is a tip, and the same grade on a prop means the same thing on both.
Because props are scored from the lines sportsbooks have posted, and a prop only exists once a book has priced the game. Between a slate finishing and the next one being posted there is a real window with nothing to show. MLB and WNBA player lines are generally posted about a day before first pitch or tip-off, so later days of the look-ahead often have games with no props yet. A prop whose game has started comes off the board too, because it can no longer be bet. And the board refuses to pad itself: if nothing on tonight’s slate clears the bar, it says so rather than filling the space with the least-bad options.
Prices are re-read every couple of minutes for games starting soon, and each prop carries the age of the price it was scored against; anything that has not been re-quoted for six hours drops off the board rather than sitting there looking live. Box scores load four times a day, so a game that finished last night is in the hit rates by morning. Nothing here marks a market as suspended, though. A book can pull a line at any moment, so check the news before you act on a number.
It searches tonight’s already-scored board for the combination that best fits the leg count, risk level and target payout you set. Each candidate leg carries its measured hit rate, the sample behind it and how its price compares to that history; the builder prefers legs across different games, because two props in one game do not settle independently and pricing them as though they do overstates the ticket.
A ticket can show three different percentages. Hit rate: How often a player has cleared one exact line in the games behind it, counted from the official game log. Model estimate: The model’s chance for one leg: its hit rate, pulled toward the price when the record is short. An estimate, not a record of past games. Combined chance: Every leg’s model estimate multiplied together, with the legs treated as independent. Legs from one game move together, so on a same-game ticket the estimate allows for those legs instead. The builder lists it under Slip checks, not on the ticket face. Neither is a forecast of this ticket. A chance is not a return.
No parlay is certain, and this one is not either. A parlay multiplies both the payout and the ways it can fail, and the sportsbook’s margin compounds with every leg you add. The builder will tell you when the ticket it assembled is priced against you. It shows the negative number and says so on the card rather than hiding it, because reaching a payout target you asked for sometimes means stringing together prices that add up to a losing bet.
The structured parts cannot: hit rates, samples, lines, prices and grades are read out of the database by ordinary code and rendered as cards, so there is no step where a model could invent one. The written explanation around them is a different matter, and we would rather say so than pretend otherwise. Phrasing can be clumsy, an emphasis can be wrong, and a question can be misread as a different question. That is why every card shows the sample and the source underneath: if a sentence and a number disagree, the number is the one that came from the data.
The AI answers from real box scores, current prices and measured hit rates, with a short sample pulled toward the market so a three game streak is not read as a trend. It describes what has happened and what the price implies. It does not predict results, so treat every answer as research, not a pick.
No, and we do not claim to. We measure the market's fair price and show the history behind every number, and we measured the claim itself rather than assuming it either way. Across roughly 96,000 graded props in an offline replay, the board tracked the de-vigged consensus of five books to within the margin those books charge, and no configuration tested cleared that margin. Historical testing showed that calibration alone is not enough to overcome sportsbook pricing, which is exactly why this product is built on price, context, sample quality and transparency rather than a claimed advantage over the market. What the work did buy is calibration: shrinkage and eligibility filtering materially improved calibration in the historical replay. Expected calibration error fell from 0.2319 to 0.0552, though meaningful calibration error remains. So the accurate description is a fast research tool that shows its working. The full replay, figures and limits together, is on the methodology page.
No. We do not beat the closing line and do not claim to. We measure the market's fair price and show the history behind every number; if a tool tells you it beats the book, ask to see the measurement. Ours: across roughly 96,000 graded props the model tracked the de-vigged consensus across five books to within the margin those books charge, and no configuration tested cleared it. Historical testing showed that calibration alone is not enough to overcome sportsbook pricing, so the board is built to do the things that hold up: line-shop seven books, score every prop the same way against real history, and show the sample behind each number, in seconds rather than an evening.
One, for Quick Picks. The Quick Picks page shows how its published legs have settled since October 3, 2026, counted leg by leg rather than ticket by ticket, against final official box scores. The wider board's record, every prop it surfaces graded wins and losses together, is kept internally to check how the scoring holds up. Both describe past results and forecast nothing. Also public: the offline replay on the methodology page, with its figures and limits, and the record of your own tracked slips on your account.
Yes. Free access is capped, with no plan to pick. Without a subscription you can open the research pages, the top 3 ranked props, past Quick Picks (3 picks at a time), 2 AI parlay builds of 3 legs and 3 AI chat messages a day. The one paid plan is Parlay Builder ($7.99/month, or $79.99/year): checkout takes a card, there is no trial, and you can cancel any time.
Yes, within a daily allowance. Without an account you can browse the top 3 ranked props and get 2 AI parlay builds of 3 legs a day. More builds, saving and tracking come with the subscription, $7.99 a month or $79.99 a year.
Free access is capped, with no plan to pick. Without a subscription you can open the research pages, the top 3 ranked props, past Quick Picks (3 picks at a time), 2 AI parlay builds of 3 legs and 3 AI chat messages a day. An account on its own adds nothing to that. Everything else comes with the subscription, $7.99 a month or $79.99 a year.
No. Browsing the board and reading hit rates and posted odds needs no signup. The top 3 ranked props are open to browse; the rest of the board comes with the subscription, $7.99 a month or $79.99 a year. Without a subscription, and without an account, you also get 2 AI parlay builds of 3 legs and 3 AI chat messages a day.
It is the board a prop builder starts from. Pick a league and a market and every prop the books have posted is here with the player’s measured hit rate, the number of games behind that rate and the posted price beside it; open any player to build the selection out with a different line, a different side, the splits and the game log. There is no charge for that and no account is needed to browse it.
Partly: without a subscription you get 2 AI parlay builds of 3 legs a day, with no account needed. The subscription, $7.99 a month or $79.99 a year, removes the daily cap and adds the suggested leg swaps and best-book pricing on the ticket it hands back.
Access stops at the end of the period you paid for. Nothing you saved is deleted. Your slips and follows stay on your account and open again when you subscribe.
Yes. Plan changes and cancellation run through the billing portal from your account page, not through checkout, because a second checkout would start a second subscription rather than move you between plans. A cancellation takes effect at the end of the period you have already paid for, so you keep the plan until then rather than losing it the moment you click. There is no cancellation fee and no minimum term. Your saved slips and their settled results stay on your account afterwards.
A proposition, or prop, is a bet on one statistic inside a game rather than on who wins it: a player’s points, rebounds, assists, threes, strikeouts or passing yards, or a combination such as points plus rebounds plus assists. Most props are offered as Over or Under a number, and the price on each side reflects the book’s view of the chance plus its margin.
One bet made of several legs, every one of which has to win for the ticket to pay. The payout is the legs’ prices multiplied together, which is why it is bigger, and why a parlay misses far more often than any one of its legs. The builder shows the legs’ combined history beside the combined price, flags legs that share a game because they do not settle independently, and labels a ticket low, medium or high risk.
They are quoted against $100. A negative price such as -150 means you stake $150 to win $100; a positive price such as +130 means a $100 stake wins $130. Every price implies a break-even rate. For a negative price, divide its size by its size plus 100: -150 is 150 out of 250, or 60%. For a positive price, divide 100 by the price plus 100: +200 is 100 out of 300, or 33.3%.
-110 means you stake $110 to win $100, or $11 to win $10. It is the standard margin on most spreads and Over/Under props, and it implies a break-even of about 52.4%: you have to win more than 52.4% of -110 bets just to break even.
The average a bet would return per stake over many repeats, if the probability behind it were exactly right. It is arithmetic on an estimate: the chance of winning times the winnings, minus the chance of losing times the stake. At a 60% chance and +110 that is 0.60 × $110 = $66, minus 0.40 × $100 = $40, or $26 per $100 staked. The answer is only as good as the 60%. Parlay Builder does not claim to find bets with a positive EV: its estimate is measured history pulled toward the posted price, not a read on tonight, and the offline replay found it tracking the market rather than beating it.
No. Hit rate vs price is measured in percentage points: how far the hit rate, pulled toward the price when the sample is short, sits from the break-even the posted price implies. EV is measured in money per stake. They share a sign but never a magnitude, because the price converts one into the other: a reading of +2 points is worth about +4% at +100 and about +40% at +1900. Props are compared in points on purpose, because ranking by raw return would put a long shot with a tiny chance at the top of every slate. Neither number is a promise.
A parlay whose legs all come from one game, such as a player’s points with his team to win. Legs in one game tend to move together, so books price them differently from legs in separate games. The builder flags legs that share a game so you can see the link rather than pricing them as though they were independent.
Fewer legs mean a better chance the ticket lands. Each added leg multiplies the payout and the ways to lose, and it multiplies the book’s margin too, so a long ticket gives up more to the margin than a short one.
Comparing the same bet across sportsbooks and taking the best price. A few cents on the odds or a half point on the line adds up over many bets. Parlay Builder reads prices from DraftKings, FanDuel, BetMGM, Caesars, BetRivers, theScore Bet and Hard Rock Bet; the subscription shows which book has the best one.
For most people, no. Books build a margin into every price, so a bettor has to win more often than the price implies just to break even (52.4% at -110), and variance means even a sound process has long losing runs. Our own record of past picks has landed below that break even. Research like hit rates and price comparison helps you see what you are paying for; it does not make betting a reliable way to make money.
Decide the amount you are comfortable losing before you look at a board, and treat it as spent: most bets lose a little to the book’s margin over time, and nothing on this site changes that. Parlays lose more often, because every leg has to land and the margin is charged on each one, so a bigger payout is the price of a smaller chance rather than a reward for picking well. Losing stretches are normal even on lines with a long record, and raising stakes to win back a loss is how a budget stops being one.
Help is free and confidential: call or text 1-800-GAMBLER (1-800-426-2537) or visit ncpgambling.org. Only bet what you can afford to lose, set deposit, time and loss limits and keep to them, never chase losses with bigger bets, and do not bet under the influence or to escape stress. You must be 21 or older, or of legal age where you live, to bet.
Odds and lines come from licensed data feeds and refresh on a schedule; hit rates and grades are Parlay Builder’s own calculations.