How we score casinos

Two scores, deliberately kept apart

Most review sites publish one number. A single figure, from 1 to 10, that is supposed to compress licensing, terms, payouts, complaints, game selection and how the site feels to use into one digit — and, quietly, in a great many cases, how much the operator pays.

We publish two, we keep them apart, and we never let one absorb the other.

• The TrustAudit Index is our editorial safety audit. It is what we found, and every deduction is published with its reason.
• The JuryScore is the community verdict, computed from Juror testimony. It is what players found.

These measure different things and they can legitimately disagree. An operator can be structurally sound and unpleasant to deal with. An operator can be liked by its players while carrying a clause that will one day cost one of them everything. Averaging those two facts into a single number destroys the information in both. We would rather show you the disagreement, because the disagreement is often the most interesting thing on the page.

The one average we do publish, and why it is not the same thing

A review page also shows an Overall: the mean of the readings we hold on that operator — the TrustAudit Index, an Experience mark for how good the product actually is, and the JuryScore. It appears only where at least two of the three exist, and it is published underneath all three rather than in place of them.

That is a summary, not a replacement, and the difference is the reason the parts stay on the page. An operator can be structurally sound and unpleasant to deal with. An operator can be liked by its players while carrying a clause that will one day cost one of them everything. One figure cannot hold both of those facts at once, so the three readings are shown separately and the Overall sits below them. If the Overall is the only number you read, you have read the least informative thing on the page.

What we do not do is average our audit with anybody else's. We read the public complaint databases and review platforms closely, and a pattern enters an operator's record only where at least two independent ones report the same thing — every one of those reports is linked under the finding it supports, so you can read the original and disagree with us. But their scores are theirs. Printing another reviewer's rating in the place a reader came for ours hands them the comparison and the credit at once, and this site is being built to do that job rather than to relay it.

Ratings on other scales are quoted inside a source, never converted: 1.2 out of 5 on a platform where almost nobody posts after a good experience is not 2.4 out of 10 of anything.

The TrustAudit Index: an open ledger

The TrustAudit Index starts at 10.0. Every operator, on the day it enters the register, is at 10.0.

From there we subtract, and every subtraction is shown on the page with its reason and its size. Not a category rating. Not a summary. The actual arithmetic:

• Licensing. The best verified licence the operator holds, weighted by the regime that issued it. No verified licence at all is the heaviest deduction available in this category.
• Fair terms. Every unfair clause we have catalogued in the operator's terms costs points, scaled by severity — a dormancy fee is minor, a vague bonus-abuse clause is major, a maximum-win cap on real-money play is predatory. A clause that merely exists in the terms costs a fraction of the points. A clause we have recorded the operator enforcing against a player costs the full amount. See unfair terms to look for.
• Withdrawal limits. A low monthly cap is a deduction, because it is a cap on your access to your own money. See withdrawal limits: the hidden cap.
• Complaints. Black points from Tribunal cases over the last 24 months, judged relative to the operator's size. A dozen unresolved cases means something different at a very large operator than at a small one, and a raw complaint count would let the biggest operators look worst simply for being biggest. So we divide by a size band. It is the only fair way to compare.
• The reputational record. A licence and a set of terms are documents an operator writes about itself, and a company can hold both in good order while its players spend a fortnight being asked for a fifth proof of address. So we also read the published complaint databases and review platforms, and record a pattern only where at least two independent platforms report the same thing. Each one is graded by what is actually reported — service friction, a payment that is late, or a balance that is kept — because those are different events rather than degrees of one. The charge is then divided by the operator's size band, for the same reason the complaint penalty is: documented complaints rise with customer numbers, and counting them unadjusted would rank the largest operators worst and call it evidence. These are other people's accounts, sourced and linked on the page, never restated as findings of fact.
• Blacklists. Accepted external blacklist evidence is a deduction. Our own blacklist is not a deduction — it is a floor, capping the score outright and forcing a Guilty verdict.
• Related casinos. Operators under common ownership pool a weighted share of each other's complaint history. Reputation is a group asset, and so is a group's record. This can cut both ways and it is capped in both directions.
• Responsiveness. The only line that exists to add points. An operator with an approved representative who answers Tribunal cases on time, reliably, earns half a point. It is small, it is not enough to rescue a bad operator, and it is not meant to be — the related-casino line above can move up to 1.5 points in an operator's favour, three times as far, but that is a correction rather than a reward.

Then the score is clamped to the 0–10 range, and the verdict band follows from it.

That is the whole model. There is no adjustment step at the end, no editorial override, no field in the admin panel where a number can be typed. The score is a function of the recorded facts, and if you disagree with the score, you can point at the specific line in the ledger that you think is wrong — which is precisely the argument we want to have.

The verdicts

The bands exist because a score of 8.4 and a score of 8.2 are not meaningfully different, and pretending otherwise gives false precision. The band is the honest resolution of the measurement. The number underneath it is there for anyone who wants to check our working.

The JuryScore: testimony, weighted

The JuryScore is computed from Juror testimony — reviews left by players. It is not an average of star ratings, because a plain average is trivially easy to distort and tells you almost nothing when the sample is small.

Three things shape it:

Verification weight. Testimony from a Juror who has proved they held an account and played there counts substantially more than an anonymous rating from a fresh account. This is the main defence against astroturfing, and it works because it is expensive to fake in the way that matters.

Recency decay. Testimony ages out on a half-life. An operator that was excellent three years ago and is poor now should not be able to coast on its old reviews, and one that has genuinely improved should be able to escape its past. A review from last month carries much more weight than one from three years ago.

A Bayesian prior. A new operator with two glowing reviews is not a 10.0 — it is an operator with two reviews. The score is pulled toward a neutral centre until enough real testimony exists to move it. This is why an operator with a handful of perfect scores will show a JuryScore well below 10: the score is telling you the truth about how much is actually known.

Senior Jurors — people with a track record of substantive, corroborated testimony — carry a modest additional weight. Not because their opinion is worth more, but because their record makes them expensive to counterfeit.

The cold-start rule: "Not yet rated"

This is the rule we are proudest of, and it is the one that costs us the most traffic.

When we do not have enough evidence to rate an operator, we say so. We do not publish a number.

"Enough" is a defined quantity rather than a judgement call. An operator's TrustAudit is not published until at least three of its six evidence channels have been examined: licensing checked against the regulator's own register, the terms read for unfair clauses, the monthly withdrawal cap, the typical payout time, the KYC turnaround, and the reputational record searched. Below three, the file is too thin for a published score.

Above it, the score publishes and is labelled provisional, with the outstanding checks listed by name. That labelling is not a hedge: because every unopened channel can only ever subtract, a score built on a partial file is an upper bound. Saying "this operator scores no higher than X, and here is exactly what we have not finished" is a claim we can stand behind.

Not a provisional number. Not a default. Not the industry average, not a placeholder 7.0, not a "3.0 until proven otherwise". The operator's page says Not yet rated, and it explains what is missing.

The temptation to do otherwise is real, and it is commercial. A page with a number on it converts better than a page that says we do not know. Every ranking table looks better fully populated. Sorting is easier when every row has a value.

But a made-up number is not a lower-quality fact. It is not a fact at all, and printing one would corrupt the only thing that makes any of the rest of this worth reading. A score is a claim. A claim we cannot support is a claim we do not make. If that means an operator sits unrated on our site for six months while we accumulate evidence, so be it. The rule is recorded as a decision in our public record and we do not intend to revisit it.

The same principle runs through everything else here: where we do not know a payout time, the field reads unknown rather than an estimate. Where a licence cannot be matched on the regulator's register, the operator's page says we could not match it, and no licence is claimed on its behalf. We would rather have a page with gaps in it than a page with inventions in it.

The firewall

The plainest section on the site.

Commercial terms have no path into any score. Not a small weighting. Not an adjustment "to reflect partnership quality". No path.

• Whether an operator pays us has no code path into the scoring functions. The scoring code cannot read a commercial field, because the commercial fields are not passed to it. That is enforced structurally, in the code, not by a policy that a future employee could quietly relax.
• Paid placements exist. They are labelled Sponsored, they are visually distinct, and they are never mixed into a score-ranked list as if they had earned their position.
• Every link to an operator carries an affiliate disclosure at the point of the click. Not in the footer. At the link.
• We link to operators we score badly, because a register with the bad entries taken out of it is not a register. Blacklisting is the one exception, it follows published criteria applied the same way to everyone, and it caps the score in the open rather than quietly removing the page. Whether an operator pays us is not one of those criteria and cannot become one while they are published.

The full technical detail — the exact weights, the formulas, the code — is on our methodology page, and the commercial arrangements are itemised on our transparency page. We would rather you checked than believed us.

Why the ledger is public

A score whose reasoning is hidden asks for trust. A score whose reasoning is published asks to be checked — and being checked is the only mechanism by which a score can actually earn trust.

So every deduction on every operator page shows its reason and its size. If we are wrong, the error has an address, and you can put your finger on it. That is the whole argument, and it is made at greater length in why we publish every penalty.