Methodology

How we measure sportsbook margins

By the OverroundHQ Research Desk · Updated 14 July 2026

A margin index is only worth publishing if the numbers can survive hostile questions. Here is the entire procedure — including what we do when a book fights being measured.

The formula

For any market we compute the overround: convert every outcome’s decimal odds to an implied probability (1 ÷ odds), sum them, subtract 100%. A two-way market at 1.926 / 2.020 gives 51.92% + 49.51% = 101.43% — a 1.43% margin (the worked example Pinnacle uses in its own margin explainer). Three-way markets add the third term. Where a comparison needs hold (the book’s keep per unit of handle) we label it explicitly: overround ÷ book total.

Sampling rules

Same fixture, tight window. A cross-book comparison only counts if every book was sampled on the same event within a ≤15-minute window. Prices from different moments are never compared side by side.
Logged out wherever the book allows it. Cloudbet, Thunderpick and Duelbits render odds to a logged-out browser (verified 8 Jul 2026); Duel’s odds come from its sportsbook platform’s public odds API, which answered a fully anonymous client. Books that wall odds behind a CAPTCHA (Stake, Sportsbet.io at our last check) stay out of the basket rather than in it with asterisks.
Proof or it didn’t happen. Every row keeps a screenshot or raw API dump with a timestamp in our archive, and shows its capture date on the site. Nothing on OverroundHQ is quoted from another affiliate’s “we tested”.
Pre-match and live are different animals. Live prices carry structurally higher margins; every live sample wears a live badge and is never averaged with pre-match rows.
Outcome order as displayed. Where a book’s listing order is ambiguous (some esports layouts), we record odds exactly as displayed and say so, rather than guessing which side is which.

When numbers become claims

No medians before n≥6. Until a book has six samples on a market type, the site shows individual dated rows labelled “sampling, n=X” — never an average. As of the first round (8 Jul 2026) every book is still below that bar, and the site says so wherever numbers appear.
Two-week shelf life. A sample older than 14 days gets a stale flag; stale rows never headline a comparison. Odds move much faster than withdrawal policies — we treat freshness as part of accuracy.
Operator claims stay attributed. Cloudbet’s “margins typically of 2%–3%” is real — from their blog, dated June 2020. Our 8 July 2026 sample of their World Cup 1X2 ran 4.90–5.48%. Both facts appear together, each with its date; readers draw the conclusion.
No invented ratings. No star scores, no “trust ratings”, no aggregateRating markup. The measured number is the review.

What the basket is (and isn’t)

The launch basket is Duel, Cloudbet, Thunderpick and Duelbits — the four crypto books where anonymous odds access was verified first-hand. It is not a “best sportsbooks” list: inclusion means we can measure it honestly, nothing more. Books get added when they pass the same access test, and the history page records every sampling round, including ones that make a basket book look expensive.

Money and independence

Some “Visit” buttons pay us a commission (each page discloses this above the first such link). Commissions never touch the numbers: rows are ordered by measured overround, cheapest first, whoever that is. If a partner book samples expensive, the row says so — that is the product.

Corrections

If you can show a row is wrong — mispriced capture, wrong outcome order, broken math — write to the address on the about page. Verified corrections are applied to the row itself with a visible note, not quietly. The calculator is open for anyone to re-run our arithmetic.