How every number on this site is made

Exact fractions, not decimals

Every engine works in exact rational arithmetic (Fraction, never floating point). The Pass Line edge is not "about 1.41%"; it is exactly 7/495, derived from a win probability of exactly 244/495 across the 36 outcomes of two dice. Exactness is not pedantry; it is what makes claims checkable. It is how we can say that free odds cost exactly zero, and how we caught the industry-wide misquote below.

The verification chain

• Reference checks: every engine carries a table of published reference values and fails the entire build if a computed figure deviates from them. When the two disagree and re-derivation proves the computation right, the reference is corrected and the discrepancy documented in the code.
• Structural invariants: facts that must hold regardless of any reference: every European roulette bet must share one edge; la partage must be exactly half the European baseline; a better payout can never cost more; the probabilities of a full game tree must sum to exactly 1. Invariants survive even when the references themselves are wrong.
• Transcription checks: after a build, pages are re-read and the figure in each page is compared to the engine's own output, so the right number cannot silently land on the wrong page.
• A quality ratchet: sixteen structural checks (headings, links, metadata, orphans) whose recorded ceilings only ever go down. A regression fails the build.

Derived, attributed, or refused

Each figure is labelled by where it comes from, and the honest answer is not always "we computed it":

• Derived: craps (all bets, from 36 dice outcomes), roulette (three wheels), baccarat (a complete walk of the 416-card, eight-deck game tree without replacement), sic bo (216 outcomes), three card poker (all 22,100 hands), video poker hold decisions (all 2,598,960 deals), hold'em equities (all 1,712,304 boards enumerated, not simulated).
• Attributed: figures whose derivation is beyond a static build, the 99.54% full-game return of 9/6 Jacks or Better, and the main-game edges of Ultimate Texas Hold'em and Pai Gow Poker, are quoted from published analyses (Pai Gow Poker's for the house way of setting hands) and say so on the page.
• Refused: where a figure could not be derived or reliably attributed, the page says so rather than guessing.

Model choices, stated plainly

• Blackjack uses an infinite-deck model: card draws are treated as independent (a ten is always 4/13). This sits about a tenth of a point above a six-deck shoe analysis: our base S17/DAS/3:2 figure is 0.512%, and a finite shoe under the same rules comes out slightly cheaper for the player, because the model ignores how the shoe depletes. Rule differences are computed by re-running the whole calculation with the rule changed, and match published values.
• Buy bets divide by the wager alone: we publish 5.00% where many references say 4.76%. Both describe the same money; the 4.76% divides by wager plus commission (20/21 of 5.00%). We state our denominator; most sources do not.
• En prison is not la partage: the number almost everyone quotes as "1.35% for both" is wrong. La partage is exactly 1/74 = 1.3514%; single-imprison en prison is exactly 19/1369 = 1.3879%. The derivation is here.

Who, then, is the author?

No invented names, no stock-photo "editorial team". The pages are drafted and edited with the engines' output as the source of truth, and the site is operated by a private individual (contact: admin@allinblind.com; more on what this site is for). Where a claim is not mathematical, venue details, tournament dates, gambling law, pages say what they were checked against and when, and cite their sources: primary ones (regulators, statutes, operators' own material, official schedules) wherever they could be read, otherwise a named legal summary. Trust the derivations, not a byline: every figure links to the page that derives it, the complete dataset is published at /data/edges.json (CC-BY), and the whole reference is readable at /llms-full.txt. If you find a number that fails your own check, write to us; the build will not ship until the disagreement is resolved.

How the words are made. The prose is drafted with AI writing tools from the engines' output and from primary sources, and checked against both before it ships; the illustrations are AI-generated. No figure is an AI tool's estimate: every number is computed by the engines described above or quoted from a cited source.