Methodology: how SlateProof prices bets and checks facts

How does SlateProof estimate a fair price?

For each exact outcome it takes the prices of sharp sportsbooks and exchanges, removes each book's margin, and blends what is left. A venue's own price is never used to judge itself.

Removing the margin (de-vigging). The implied probabilities of every outcome in a market add up to more than 100%, and the excess is the book's margin. A -110 / -110 market implies 52.38% on each side, 104.76% in total, so the margin is 4.76% and the fair split is 50% each. SlateProof uses the power method by default. For -150 / +130, the raw 60.00% and 43.48% (103.48% in total) become 58.40% and 41.60%, a method that puts a larger share of the margin on the longer price than simple scaling does. A one-sided market, such as a “3+ shots” line with no opposite side, is never de-vigged and gets no fair price from that book.

Which books count.

• Only reference books: sharp sportsbooks and exchanges. Pinnacle counts most on game lines but less on player props, where an audit found it no sharper than the exchanges. Daily fantasy apps never count.
• Quotes older than 10 minutes, suspended or withdrawn quotes, and markets with a margin above 10% are ignored. An exchange quote counts only with at least $25 resting on every side, and carries full weight at four times that.
• Books that copy one another share one vote (for example Pinnacle and its sister book, or Kalshi and Polymarket).
• A venue never sets its own fair price. To judge a quote at a venue, SlateProof builds the fair price without that venue and without the books that copy it.
• For pricing promotions only, DraftKings' and FanDuel's own two-way prices are blended in as well. That is an input to the estimate, not a preference for where to bet.

Exact lines. A 20+ line and a 25+ line are different bets, and standard and alternate labels for the same exact outcome are one bet. A line that no reference prices directly may be estimated from the same player's other lines with a fitted distribution, only for statistics with small averages such as hockey shots or strikeouts; it is labelled extrapolated and carries extra uncertainty. Yards, basketball points and basketball and football totals are never extrapolated. Ambiguous periods and rows without a player ID are skipped.

What are expected value and conservative EV?

Expected value is the average profit per dollar staked if a bet could be repeated many times: EV = fair probability × decimal price − 1. At a fair probability of 52% and a decimal price of 2.00 (+100), EV is 0.52 × 2.00 − 1 = +0.04, or 4 cents per dollar. If the fair probability were really 50%, the same bet would be worth nothing.

Because the fair probability is an estimate, SlateProof also reports a conservative EV: the same calculation with the probability shaded down by its own uncertainty. Uncertainty grows when books disagree, when markets are wide, when little money rests on an exchange quote and when a line is extrapolated. An edge is a model claim, not a guarantee.

How is an arbitrage found and checked?

An arbitrage exists when the same exact proposition can be bought on every outcome for less than it pays: the prices, after each venue's taker fee, satisfy sum of 1 / decimal price < 1. Stakes are split in proportion to 1 / price, so every outcome returns the same amount. With illustrative prices of 2.06 on the Over and 2.04 on the Under and $500 in total, $248.78 on the Over and $251.22 on the Under return $512.49 either way: a profit of $12.49 (2.5%). Real stakes are rounded to whole dollars (to cents for long shots), and the profit shown is the worst case after rounding.

Every candidate then has to pass these checks. Anything that fails is blocked, or shown with a warning that names the risk.

• Same proposition, whole outcome set. The same event, period, participant, line and outcomes at every leg. Two legs at one venue never count.
• Fresh quotes. Each price must have been in its venue's latest delivery, be at most 10 minutes old when captured, and be captured at least 2 minutes before the start. The dashboard treats an arbitrage as actionable only while its snapshot is at most 30 minutes old, and re-ages it in your browser.
• Net of fees. Prices are measured after the venue's taker fee (the table below).
• A minimum return. 0.2%, or 0.5% when a leg is on an exchange, because whole contracts, fee rounding and rounded odds eat a thin margin.
• Exchange depth. At least $50 must rest at the quoted price. Depth reported within 10 minutes caps the total stake; older or unknown depth is a warning.
• A stale-line guard. Each leg is compared with the median fair price of at least 3 other books. A leg that is better than that consensus by about 5 points at even money is flagged (place that leg first), and by about 15 points it is blocked as a likely stale or erroneous line that a book may void. With fewer than 3 other books, nothing confirms the prices, and an arbitrage of 5% or more is blocked as too good to verify. A sportsbook price unchanged for 6 hours is flagged as a possible slow line.
• A buyable price. An exchange whose two sides add up to 100% is showing a midpoint, not a price you can buy, and is blocked.
• Settlement differences are disclosed. A player who does not play (sportsbooks void; an exchange may settle differently), a postponement, period or 90-minute scope, NFL ties, and whole-number lines that can push. A period moneyline that can tie is blocked unless every leg is at a book known to refund the tie.

Exchange fees, as researched and used in these calculations (checked October 6, 2026):

How are promotions valued?

Promotions are valued with the same fair probabilities as ordinary bets. For a stake S, a book's decimal price d, a fair probability p, and a promotion's own caps and minimum odds:

• Bonus bet (the stake is not returned): EV = p × (d − 1) × B. Hedged with cash on the other side at decimal price d2, the hedge is B(d − 1) / d2 and the locked profit is B(d − 1)(d2 − 1) / d2. A $25 bonus bet at +400 (5.00) hedged at 1.25 needs a $80.00 hedge and locks $20.00, 80% of its face value.
• Profit boost: the boosted price is d + min((d − 1) × boost × S, cap) / S and EV = S(p × boosted price − 1). A 50% boost on +150 (2.50) with a $10 maximum stake and a $5 cap pays extra winnings of $5, for a boosted price of 3.00.
• Parlay boost: the same, applied to the product of the legs' prices and fair probabilities, with the token's minimum odds enforced. Legs on one event or sharing a participant are not combined automatically.
• No-sweat bet: EV = S(p × d − 1) + (1 − p) × c × S, where c depends on how the refund is paid: a bonus bet 65%, site credit 95%, cash 100% of its face value.
• Early-win token: EV = S((p + L) × d − 1), where L is the chance that the team leads by the token's margin at some point and still loses. L was fitted offline to historical games: NFL play-by-play from 2011 to 2025 (nflverse, CC BY 4.0), and MLB and NBA game records.
• Sign-up offers are valued from their own terms (bet-and-get, win-and-get, first-bet refund, exchange credit) using each book's current bonus-bet conversion rate, and carry their state variants and end dates.

Promotions are priced from a built-in catalog of what each sportsbook commonly gives existing customers, plus each venue's new-customer offer. A figure that no source gave, such as a typical maximum stake, is marked as assumed. Many tokens are targeted, so your account may not see every promotion, and the terms in your app are the ones that count. A visitor may enter a federal tax bracket, kept on their own device, to see after-tax values; only 90% of gambling losses are deductible from 2026, and only if you itemize. State tax is not included, and this is not tax advice.

How are decisions frozen, graded and measured?

Every flagged single, arbitrage and promo pick is frozen the first time the engine sees it as actionable: the exact venue, price, fair value, uncertainty, sources and timestamps. Later sightings are observations of that decision, not new bets, and a frozen record is never edited. A correction is a new revision beside the original.

Grading. After the game, decisions are graded from final scores and box-score statistics (PropLine's free endpoints). Grading fails closed: a player missing from the box score stays pending for 24 hours and then becomes unresolved, never a silent void; a score that disagrees with the event is rejected; a tie on a two-way moneyline is a push; period markets stay unresolved until their exact period facts are available.

Closing-line value (CLV). CLV asks whether the price taken beat the market's last word: the bet's EV measured against the fair price at the close. A close counts only if it was observed after the decision and within 30 minutes of the start. Otherwise the decision has no CLV, rather than an invented one.

The paper staking convention. Paper results are what the moves would have made, not money anyone was paid.

• Watch-list singles: one flat unit per first flagged sighting.
• Arbitrage: one decision per proposition and lane (every leg at an exchange, or any other mix of venues), at the first moment it was actionable, worked out at the bet size on the track-record slider ($500 by default), capped at the depth the venues showed, as if every leg was accepted. A second figure counts only the arbitrages still present at the next price check.
• Promo picks: each promotion's own amount, frozen once a day when the first leg starts in 30 to 120 minutes, and graded as a whole ticket. Early-win tokens that did not pay in the final score are counted as losses, which is a lower bound because a final score cannot show an in-game lead.
• Days run midnight to midnight UTC. Where ranges are reported, they come from resampling whole days.

Why is the watch list unproven?

Because no test so far shows that its flags make money. In an offline test on 5 October 2026 of about 214,000 resolved MLB player-prop bets from March to June 2026 (the SmartStake dataset, CC BY 4.0), a higher predicted EV did not bring a higher return. The groups that a screen would have flagged lost roughly 2% to 3% of their stakes on average, and most of their 95% ranges include zero, so the test shows no edge rather than proving there is none. One cohort showed positive closing value against the reference and still lost money, so closing value alone is not proof of profit. On MLB game lines, closing value was about +0.4% at EV of 1% or more, which is not demonstrated either.

So the watch list stays paper-only, and the dashboard says so, until its own frozen forward record shows positive closing value and positive returns with ranges that support it. SlateProof does not describe any result as outperforming the market without enough observations and their uncertainty. Arbitrage and promo hedges rest on arithmetic instead: their open question is whether they can be placed, which the track record measures with the share still open at the next price check. In an internal replay of 44 captured snapshots on 5 and 6 October 2026, 12 distinct arbitrages appeared, 0.24% to 2.75% after fees, about $1 to $14 on a $500 total. Read that as a sense of scale, not a result.

What does SlateProof not know?

• Books that PropLine does not carry: Caesars, bet365, ESPN Bet, BetParx, Bally Bet, Circa and Borgata. Offshore books, sweepstakes apps and non-US exchanges are never execution venues. Pick'em apps (PrizePicks, Underdog and Sleeper) are used only for pick'em entries whose every pick is hedged at a legal sportsbook or exchange, and they never set a fair price.
• Your accounts. Promotions tied to an account, your limits, and whether a venue has restricted you are invisible to it.
• Current prices. PropLine's free plan allows 1,000 requests a day, so snapshots are minutes apart and some events are checked less often. A price older than about ten minutes must be re-checked in the app.
• Rules and fees that changed after the dates below. They are researched by hand and refreshed on request; nothing updates them automatically.
• How a venue will settle an edge case. Voids, postponements, player did-not-play and tie rules differ, and a venue's own rules decide.
• Whether a market is legal where you are. The venue's own app and your state regulator have the final say, and this is not legal advice.

Where does the data come from, and when was it checked?

Live odds, final scores and statistics come from PropLine. The facts that change slowly are dated:

These come from the operators' own pages where they could be read and otherwise from at least two trackers; sources are listed with each venue in the data. A refresh is a manual re-check followed by an update of the page and its date. If a date looks old, trust the venue's own app. See the state hub, sportsbooks and exchanges for the current tables.