How we measure the price of a bet

1 · The fee is inside the odds

A bookmaker prices a soccer match three ways: home, draw, away. Divide 1 by each price to get the chance the book is implying. One of the three outcomes must happen, so a fair market's chances add to exactly 100%. A real market adds to more, and the excess is the bookmaker's fee on that match.

This book charges 4.81% on this match. Divide each implied chance by 1.0481 and you also recover what the book itself thinks is fair: the home price should have been 2.20, not 2.10. One book's three prices carry both numbers, what it believes and what it charges on top.

2 · Only like is compared with like

This is the rule that does the most work, and the easiest one to get wrong.

• Same market type. A two-way market carries a structurally different margin from a three-way one. Blending them produces a confident number that means nothing, so we never do it.
• Same fixture, where we claim it. Book A's cup slate cannot be compared with Book B's league slate. This matters enough that we publish two different numbers and label them differently, rather than one number that quietly mixes the two:
  • The weekly index is each book's own average across its own three-way board. It answers "what did this bookmaker charge last week", and the market count beside each book tells you how wide a board that average came from. It is not a head-to-head.
  • The case study is the head-to-head: only fixtures at least two books priced, compared outcome by outcome, with the shared-fixture count for every pair of books printed underneath so you can see exactly how far the comparison stretches.
• Match result only. We read the standard 1X2 market. Handicaps, half-time markets and promotional early-payout variants are excluded, because they look like a 1X2 and are priced differently.

3 · How the sampling works

1. Every Thursday at 08:30 SAST, when the weekend is fully priced, the collector reads each book's public pages once.
2. robots.txt is fetched and obeyed for every origin on every run, including separate API hosts, which carry their own rules.
3. Margins are computed with a shared library, so every book gets identical arithmetic, and that library is covered by an automated test suite.
4. A plausibility band rejects any market outside 0.5% to 20%. Anything outside that is an extraction fault, not a discovery, and is dropped with a note rather than published.
5. A book that cannot be read is listed as unread with the reason. Never silently dropped, never estimated.
6. Every weekly table is archived permanently. Corrections are printed beside the original, never instead of it.

4 · What the collector will never do

These are not preferences. They are enforced in the code and they are why some bookmakers are absent from our tables.

• Never log in. Public pages only. Authentication turns terms of service into a binding contract in a way that browsing does not.
• Never evade a challenge. A CAPTCHA, a Turnstile or a 403 means that book is recorded as blocked and we move on.
• Never carry identity. No cookies, no credentials, no affiliate tokens are sent with a collection request. The collector reads prices and has no business carrying anything else.
• Never hammer a site. One page or a handful of small documents per book, once a week, spaced out. That is a smaller footprint than a single human visitor.
• Never hide. The collector identifies itself by name in its user agent and points at a page explaining what it is.

Currently absent because of their own published rules: BetBus, Sportingbet. Those are their decisions and we respect them. The full list, with reasons →

5 · The rules we never break in what we publish

No tips, no predictions, no bankers, no locks. No lottery-number content ever, because nobody can predict a random draw and we will not pretend otherwise. No number invented, no gap papered over, no margin published against a real bookmaker's name until it has been measured on identical fixtures. When we are not certain, the page says so in plain sight.

6 · Known limitations, stated up front

• A snapshot is not a live feed. Prices move constantly. Every figure is timestamped and correct only for that moment.
• Coverage is partial and named. We currently measure 4 bookmakers. That is not the whole market and we never imply otherwise.
• Slate differences distort averages. A book's overall average reflects the fixtures it happens to price. That is exactly why the case study reports the shared-fixture number too.
• One week is not a trend. Until the archive has depth, treat any single week as a measurement rather than a verdict on a bookmaker's pricing policy.

7 · Check us

Take any price on this site, do the sum, compare. The last collection ran on 24 July 2026 and covered 204 markets. If something does not match, tell the desk at [email protected]: the correction gets published beside the original, dated, and the error stays visible in the archive.