What we measure. What we refuse to.

Every number on this site is read from an operator's own terms or help pages or from a checkable public record, carries the date it was read, and is asserted against the dataset every time the site builds: 32 casinos, 170 source documents, 18 published refusals, 32 of 32 payout records fully verified, newest read 2026-09-02. The overall score is capped at 9, because two inputs an honest 10 needs, a complaint ledger and self-measured payout times, do not exist here yet.

What we measure today

Payout terms, clause by clause. For 32 casinos we record the stated processing window, the fee policy, the KYC gate and any published no-KYC ceiling, each read from the operator's own pages with its clause reference and read date. A stated window is always rendered as the operator's claim, never as a measured time.

Computable cashback. 10 programmes carry a published rate ladder, base and cadence; 9 are computable. We record wagering multipliers per tier, and record the absence when an operator publishes none.

Rakeback, kept separate. 12 operators pay a share of house edge per bet. It is never mixed with cashback because the bases differ by two orders of magnitude.

Community records, cited. Delay flags come from checkable records: court reporting, regulator actions, corroborated complaint threads. The flag is independent of the operator's claims, which is why a casino can state "instant" and still carry one.

The refusals. 18 programmes we refuse to compute value for, each published with its reason on the refusals ledger.

How verification works

A record is verified only when all three hold: named sources, high or medium confidence, and a dated read. Anything else is marked unverified, and unverified claims are never stated as fact, anywhere. When we could not confirm an operator's claim on its own cited pages, the review says so in a "What we refuse to say" section instead of repeating the claim.

Counts in prose are asserted in code at build time. If this page says 32 casinos and the dataset holds a different number, the site fails to build. Records older than 30 days are listed in a freshness report and re-verified before any deploy.

How the score works: RAW Score v0, verbatim

RAW Score v0 = 0.25 x disclosure + 0.2 x commitment + 0.4 x record + 0.15 x freshness, each axis 0 to 10, rounded to one decimal, then two caps. Disclosure counts the five facts a withdrawal depends on that the operator actually publishes: fee policy, minimum withdrawal, withdrawal ceiling, no-KYC ceiling, operating entity and licence. What an operator will not tell you is scored, never imputed. Commitment grades the stated payout window and fee class from the operator's own terms; an unverified record is credited nothing, because we do not state unverified claims as fact and we do not award points for them either. Record starts from the cited community payout record (a delay flag scores 0, no flag on record 7, a cited fast record 10, since absence of a flag is not an endorsement) and subtracts for a confiscation clause, a wagering gate, and a refusals-ledger entry. Freshness grades the age of our own read of the record, full marks inside 30 days, zero past 150.

The caps: a casino with a cited delay record cannot score above 4.9, whatever it publishes, because a published promise does not outweigh a documented failure to pay. And no casino can score above 9 until we can measure payouts ourselves and hold a complaint ledger: the final point belongs to the inputs that do not exist yet, so v0 scores what is verified today and says what it cannot see. Payment is not an input, no field a payment can reach is an input, and there is no hand-set per-casino number anywhere: the formula is versioned in code, its monotonicity is enforced by tests (a delay flag can never raise a score, publishing a fact can never lower one), and changes get a dated version bump on this page.

What we do not do yet, on purpose

No self-measured payout times. Real payout speed is not measurable without funded test withdrawals. We have not run them yet, so no minutes figure appears on this site. When the timing programme exists, every measurement will publish its timestamps.

No complaint ledger. Mediating and documenting complaints properly is infrastructure, not a page. Until it exists we do not claim complaint statistics.

No uncapped score. The full RAW score is defined as majority-weighted on measured payouts and verified complaints, and neither input exists yet. RAW Score v0 therefore scores only what is verified today, prints the axis table that produced it, and stops at 9: pretending to see those inputs would make it exactly the marketing artifact this site exists to replace. The cap lifts when the inputs exist, with a dated version change.

No affiliate links today. When they arrive they will be disclosed on every page, and they will decide nothing about rank or copy. That separation is the product.

What operators can pay for, all of it disclosed on the page it touches. Expedited verification moves an operator up our reading queue, never changes what the read finds, and the verified page carries a small badge saying so. Commercial partnerships (affiliate deals and launch packages) buy a tracked link and labeled placement, never rank, score or copy; affiliate links are marked as affiliate links. The full terms are on the operator page.

Re-evaluation and right of reply

Any operator may request a re-evaluation of their review every 3 months. A re-evaluation is a fresh dated read of the operator's live documents against the criteria on this page: the clause-by-clause payout read, the fee policy, the KYC gate, the licence record, the cited community records. The criteria are public precisely so an operator can fix what a review found without asking us; the next read records the fix. Operators who do not want to wait can use the disclosed expedited queue described above, which changes the date of the read and nothing else. Operators also have a right of reply: a response from a verifiable operator address is published on the page it concerns, next to the evidence we cited. Details on the operator page.

The question bank

We track 816 real questions players ask in this niche, harvested from search behavior, community threads and help pages. 180 are answered on this site today, every one computed from verified data. 617 are in the verification backlog: we would rather show you a gap than an invented answer. And 19 we refuse to answer at all:

What we log during the beta

We are running a beta, and during it we record two things when someone searches this site: the question typed, and whether we had an answer for it. That is the whole record. We do not store an IP address, a cookie, a session identifier, a device fingerprint or a referrer, and there is no account to attach a question to, so a logged question cannot be traced back to the person who asked it. Nobody but us can read the log, including other visitors.

We do it because a question that returns nothing is the most honest content gap there is. Keyword tools tell us what people type into Google, which we have found repeatedly is not the same as what people actually want to know. A question asked here, in someone's own words, is better evidence than either. The log exists for the beta only and is deleted when the beta ends. If we ever want to keep it, we will say so here first.

Straight answers