Eight layers, from source to correction.
Research runs through the same eight layers for every jurisdiction, operator and product. The last layer feeds back into the first.
- 01DiscoverFind regulators, operators, datasets, experts and testable products.
- 02ObserveCapture declarations, measurements, tests and events, separating what we saw from what we were told.
- 03EvaluateJudge each claim on authority, directness, independence, corroboration, recency and relevance.
- 04ModelDerive comparisons, trends and estimates, keeping inputs, assumptions and uncertainty.
- 05PersistStore entities, atomic facts, evidence and validity dates once, independent of any page.
- 06InterpretWrite jurisdiction profiles, operator assessments and rankings, keeping confidence apart from quality.
- 07PublishRelease research with status, scope, sources, dates, limits and corrections.
- 08ImproveDetect gaps, changes, contradictions and stale findings, then recheck and update.
Every fact has a state and a type.
The state says whether a claim meets the evidence threshold for its field. The type says how we know it. A missing value is never shown as zero or as a settled fact.
Coverage counts what applies.
Coverage measures how complete a market record is against the fields that apply to it. A negative finding, such as no regulator, is a finding. An unknown is not. Fields that cannot exist under a regime are left out: a prohibited market has no bonus rules to research, and a licensed market has no blocking orders against it.
Legal status of online casino, sports betting, online poker, lottery and land-based casino, the regulator and the governing law.
When the current regime started, minimum age, self-exclusion register, bonus rules and tax on winnings.
Player liability, site blocking, payment blocking and advertising rules.
Coverage is kept separate from confidence. A complete record can still rest on weak sources, and the two are never combined into one score.
How evidence is weighed.
Evidence strength is judged for each claim, not for each source. A credible source can be wrong about a specific fact, and ten copies of one report are one source. Our own network repeating a fact never counts as confirmation.
| Dimension | The question we ask |
|---|---|
| Authority and competence | Is this source positioned to know this specific claim? |
| Directness | Primary record, firsthand observation, interpretation or repetition? |
| Independence | Do apparently separate sources share an origin, owner, dataset or incentive? |
| Corroboration | Do independent observations agree under the same definitions? |
| Method quality | Are sampling, instruments and calculations adequate for the claim? |
| Temporal validity | Was it true at the relevant time, and is it still? |
| Contextual relevance | Does it apply to this jurisdiction, product, user and metric? |
| Consistency and uncertainty | What contradictions, errors or open assumptions remain? |
When we estimate, and when we refuse to.
An estimate must pass seven gates. Legal status, licence validity and operator authorisation are never estimated from market proxies. Every estimate is labelled, shows a range, and names its model version.
- 1EligibilityIs this attribute estimable, and is estimation allowed for this use?
- 2InputsAre the observations relevant, independent and current enough?
- 3AssumptionsWhich comparability, sampling and stability assumptions are needed?
- 4ValidationCan it be checked against known markets or held-out observations?
- 5UncertaintyCan we report a defensible range or sensitivity?
- 6DisclosureIs it labelled estimated, with model version, period and limits?
- 7ReassessmentWhat new data or drift would change or withdraw it?
How assessments and rankings are built.
Each assessment names what was compared, for which user, where, with which evidence and which method version. Eligibility is a gate, not a deduction. Results that move with small weight changes are shown as a band.
- Confidence is not quality. A well-supported negative finding lowers a score.
- Popularity is not quality. Market share describes; it does not score.
- An operator's promise is not a tested result.
- A rank only holds for its comparison set, criteria, place and date.
- Appearing on a commercial page is not a research conclusion.
Online casino, live casino, slots, mobile, new casinos, crypto, sports betting, esports and lottery each have their own dimensions. All share the same evidence rules. Every weighting profile is published with its rationale and version.
The world model underneath.
Findings are stored once, as atomic facts linked to stable entities, with the dates they held. Pages, reports, the AI analyst and the API all read the same facts. Changes flow out through reviewed jobs, never silent rewrites.
{
"subject": "brand:4127",
"predicate": "licensed_by",
"object": "regulator:dk-spm",
"context": { "jurisdiction": "DK", "product": "online_casino" },
"valid_from": "2019-01-01",
"valid_to": null,
"state": "known",
"info_type": "declared",
"obs_class": "official_record",
"evidence": ["src:dk-register-2026-10-11"],
"verified_at": "2026-10-12",
"recheck_by": "2026-11-11"
} From draft to correction.
Gates scale with consequence. Licensing claims, safety statements and market estimates need more review than a game list.