How this list is ordered

Each check is a yes-or-no question about a single field in an operator's record. There is no partial credit and no judgement inside a check: either the field is filled from a document that was read, or it is not. The weights are unequal because the fields are — a verification clause that binds the operator is worth more than a coin list that can change on a Tuesday.

The right-hand column is how many of the forty-one actually pass, and it is the most useful part of the table. The verification-figure check is passed by five brands out of forty-one, which is the clearest measure on this site of how little this market commits to in writing.

The obvious way to build a score like this is to average over the fields you happen to hold. It is also wrong, and wrong in a direction that flatters exactly the operators a reader should be most careful with. A brand publishing two facts, both of them good, scores full marks on a denominator of two and lands at the top of the table above a brand that published nine facts and got eight of them right.

So the denominator here is not the number of fields read. It is ten, always, for every brand. A field with nothing in it contributes zero points to a total that still has to reach ten. The practical effect is that a thin record cannot rise, and the brands at the bottom of this site's list are there because they publish little, which is precisely the finding the list is supposed to report.

The same rule governs empty lists. Where an operator's coin, studio or restricted list is empty here, the document has not been read. It does not mean the operator takes no coins, carries no studios or restricts nobody. An empty list earns no points and is never rendered as a zero.

The register-link check asks whether the record carries a licence register URL. It does not ask where that URL goes, and thirty-nine of the forty-one pass it — but only thirty-three of those URLs lead to something a third party runs. The other six point back at the operator's own site or terms page. Six brands are therefore collecting a full point for a link that confirms nothing, and the licence page names them rather than leaving that buried in the code.

The restricted-list check is narrower still, deliberately. It scores only whether an operator publishes a restricted list at all — a disclosure signal, the same kind as publishing a coin list. The contents of those lists are never read, never compared and never scored. An earlier version of this code carried a per-brand comparison of territory lists against the subject's; it was removed rather than left sitting unused, because a scoring field nobody renders is exactly how that sort of material creeps back into a site later.

One further check has nothing to act on at present. The code tests each status word for the strings revoked, withdrawn and cancelled, and any brand carrying one is sorted below every brand that does not, whatever else its record holds. No operator in the current forty-two records trips that rule. It is described here so the mechanism is visible before it is ever needed.

Sorting runs in three passes. A revoked licence goes below every live one first. Then brands are ordered by score, highest first. Then ties are broken alphabetically by brand name, which is arbitrary on purpose — an alphabetical tie-break cannot be gamed and cannot quietly encode a preference. Scores are rounded to one decimal place for display.

No brand among the sites like Roobet scored here reaches ten. The highest score in the set is 9.0, held by Wild.io, which passes nine of the ten checks and fails only the verification-figure check. The median is 7.3. The bottom two, at 3.5 and 3.3, are Nitrobetting and BetPlays, the two records with no licence authority, number or register link on file at all.

The subject is scored by the same code as everything it is compared with, and it does not do well. Roobet passes six of the ten checks and scores 6.0. Thirty-one of the forty-one alternatives score higher, three tie with it and seven score lower.

Two of the four failures are the register checks, and both trace back to one thing. Roobet publishes no certificate link, and the Curacao certificate register returned no result for any domain queried on 28 August 2026 — including brands that certainly hold Curacao licences — so the number cannot be resolved in either direction. That leaves no register URL and no status word on file, and both checks score zero. Unresolved is not a finding that the licence is invalid; it is a finding that the check could not be run.

The score does not measure whether a casino pays out, how quickly it pays out, whether its bonuses are worth taking or whether its games are fair. None of those could be established from a document, and none was tested. A brand scoring 8.5 has published more than a brand scoring 5.5. That is the entire claim.

It also does not measure how good a licence is. Curacao and Anjouan licences count identically here, and the licence-live check asks only whether a status word is on file and does not read as revoked. Weighing one regulator against another would be a judgement, and this site's method is reading documents rather than judging the bodies that issue them.

The commercial arrangement is kept outside the calculation entirely. This site carries one paid placement, labelled as one wherever it appears, and it sits outside the ranked table. No weight in the code refers to it, and the ordering would be identical if it were removed tomorrow. Everything above is checkable against src/lib/score.ts rather than taken on trust; if a figure on this page ever disagrees with that file, the file is right. Corrections go to [email protected].