01Received
A submission arrives through the report form and is stored privately. It is not a public record, not a source for publication yet, and not visible outside the review console.
Source standards
A source is not a link. It is a record in its own right — with a publisher, a type, a reliability rating, a retrieval date and an archive snapshot — and every claim GambleLABS publishes has to reach one.
TypedRatedArchived at captureRe-read, not assumed
01Taxonomy
Source types
Every source is filed under exactly one of twelve types. The type is a constrained value in the database, not free text, so a source cannot be quietly described as something stronger than it is.
- Official register
- A register maintained by a named authority. Authoritative for the status it shows on the date we retrieved it, and for nothing after that date. Registers are re-read rather than assumed to be stable.
- Regulatory notice
- A decision, sanction or public warning issued by an authority, read from that authority’s own domain. Recorded with its procedural status, because a notice under appeal is not a final finding.
- Court record
- A judgment, order or filed document from the court itself. First-instance decisions are recorded as first-instance, and where an appeal is known to be pending the record says so.
- Corporate filing
- A registry filing, set of accounts or ownership disclosure. Authoritative evidence of what was filed and by whom — not, on its own, evidence that the contents are accurate.
- Company statement
- A company’s own terms, policies, published position or statement to us. Primary evidence of the company’s position; secondary evidence, at best, of the events it describes.
- Press release
- The subject’s account of an event. Used for the subject’s position, for dates and for quotations. Never used as the evidence for a regulatory outcome when the regulator has published its own notice.
- News report
- Secondary throughout. Used to locate primary material and, where an outlet has done first-hand reporting, cited as that outlet’s reporting rather than absorbed as fact.
- Academic
- Peer-reviewed or institutional research, cited for method and aggregate findings, with the study’s own stated limits carried across rather than dropped.
- Technical record
- DNS, WHOIS, certificate-transparency entries, captured HTTP responses, payment descriptors. Recorded with the method and the exact timestamp so that a third party can repeat the observation.
- User-submitted
- Material sent to us by a player, a former employee or an insider. Never publishable on its own, always redacted before any public use, and governed by the consent the sender gave.
- GambleLABS analysis
- GambleLABS’ own analysis, including anything produced with the Evidence Engine. It is filed as a source so that reasoning is auditable, and it never counts towards corroborating its own conclusion.
- Other
- Anything that fits none of the categories above. Requires an explicit note describing what the material is and how it was obtained, or it does not go on file.
Where a single document could sit under two types — a regulator’s press release announcing its own enforcement notice, say — it is filed as the weaker of the two, and the stronger document is sought separately.
02Weighting
Reliability ratings
Type describes what a source is. Reliability describes how much weight it can carry. They are separate fields because a news report can be verified and a corporate filing can still be self-serving.
- 01Primary — official
Issued by the body with legal responsibility for the fact: the regulator that granted the licence, the court that ruled, the registry that holds the filing.
WeightSufficient on its own to support a documented record, provided the party is correctly identified.
- 02Primary — corporate
Issued by the subject itself: filed accounts, published terms, a signed statement, a licence declaration on its own site.
WeightSufficient for what the company says about itself. Not sufficient for whether what it says is true.
- 03Secondary — verified
A second-hand account that we have checked back against a primary document, or that we have confirmed with the body concerned.
WeightCounts towards corroboration. Does not replace the primary document it points at, which is filed alongside it.
- 04Secondary — unverified
A second-hand account we have not yet been able to check — typically a report of a decision we cannot find published, or a claim about a company we cannot tie to a filing.
WeightCannot support a claim about a named party. It can justify opening a file and looking for the document.
- 05Unassessed
On file, not yet rated. The default state of every newly captured source, including everything an ingestion adapter has fetched but no reviewer has read.
WeightCarries no evidential weight at all. A record cannot be published on unassessed sources.
Ratings are applied by a person, not inferred from the domain a document came from. A ministry website can host a stale copy of a register; an independent outlet can hold a document nobody else has. The rating records what we did to check, and it can be downgraded as well as raised.
03Durability
Why we archive everything
Every source record has a field for an archive URL alongside the live one, and capturing it is part of filing the source rather than a later tidy-up.
Regulatory material disappears routinely and without malice. Authorities restructure their sites, retire old notices after a statutory period, move registers behind session-bound search forms and replace PDFs with database views. Operators redraft terms overnight. A record whose only support is a URL that no longer resolves is, in practice, an unsupported record — and the reader has no way to tell the difference between a claim we could once prove and one we never could.
So a source is archived at the moment it is used, not when the link breaks. The source row keeps both addresses: the live URL, because readers should be sent to the authority’s current page wherever possible, and the snapshot, because that is what the claim actually rests on.
What the snapshot is for
- Proving what a page said on a date. The retrieval date is stored with the source; the snapshot is the evidence for it.
- Surviving a takedown. A notice withdrawn after an appeal is still a fact about what was published, and we record both the notice and its withdrawal rather than deleting the history.
- Letting a reader check us. Independent verification is the whole point. It only works if the reader can reach the same document we read.
Its limits, stated
Some material cannot be snapshotted: paywalled filings, registers that will not render without an interactive session, documents supplied to us privately. Where that is the case the source record says so, describes precisely how the document was obtained and, for anything supplied to us, states whether we may republish it. We do not claim an archive we do not hold.
What a source record holds
- Title and publisher
- Live URL and archive snapshot
- Source type and reliability rating
- Publication date, where stated
- Retrieval date
- Jurisdiction it relates to
- Reviewer notes on how it was obtained
Link rot is the default
Treat any gambling-regulatory link older than a few years as likely dead. That assumption shapes the schema: the archive field is not an optional extra bolted on for tidiness, it is how a claim stays checkable.
04Material sent to us
User-submitted evidence
Player reports, bank statements, chat transcripts and screenshots are among the most useful material this platform receives, and the most dangerous to handle carelessly.
02Checked
A reviewer examines the material for internal consistency, checks dates and amounts against anything independently verifiable, and records what could not be checked as well as what could.
03Corroborated
A single submission never supports a public claim about a named company. It has to meet a register entry, a filing, an authority’s notice, or independent submissions that could not have been coordinated.
04Redacted
Anything published is redacted first — names, account numbers, addresses, identity documents, transaction identifiers — and the redaction note is shown to readers so they know what was removed.
Consent governs the whole path. A report is private by default and stays private unless the person who sent it consented, specifically, to publication — and consent to publish an anonymised case file is not consent to be named, quoted or contacted. Withdrawing consent is always available and takes effect on the public record, not merely on our intentions.
Uploaded files are held outside the web root under opaque names, checked against their declared type before they are accepted, and opened only by the two roles permitted to see them. They are never republished as supplied.
If you have sent us something
Your report is private, your contact details are encrypted at rest, and nothing about your case is published without your explicit consent and a human review.
05Protected sources
Anonymous sources
People inside this industry who talk to journalists lose their jobs, and sometimes more. Anonymity is a protection we take seriously — and a cost we are honest about.
What we promise a source
- We agree the terms before, not after: on the record, unattributable, or background only. Once agreed, the terms hold, including if the story becomes more valuable without them.
- Identifying material is separated from the working file. Contact details are encrypted at rest and reachable by a deliberately small number of people.
- We do not publish detail that would identify a person by elimination — a role, a location, a date and a department are often enough on their own, and we treat them as identifying.
- We will say when we cannot protect someone. If a claim could only have come from one person, we say so to that person before publication and let them decide.
What anonymity does not buy
- It does not lower the evidence standard. An anonymous account is a single source, and the single-source conditions apply to it in full.
- It does not remove the reader’s right to weigh the source. The record states what kind of person the source is and what interest they have, in terms precise enough to be useful and general enough to be safe.
- It does not remove the subject’s right of reply. The claim is put to the named party in specific terms — as specific as we can manage without identifying the source.
- It is not available for making an allegation we cannot otherwise support. Where corroboration fails, the claim is not published, and anonymity does not change that.
06The invariant
Every claim must reach a source
This is the rule the database itself is built around: claim tables carry source links, and a published record shows its source count whether that count is healthy or zero.
The chain
- 01A reader sees a claim on a record.
- 02The record shows how many sources are attached to it.
- 03Each source names its publisher, its type and its retrieval date.
- 04Each source carries a live URL, an archive snapshot, or a stated reason it has neither.
- 05The reader reaches the document and checks the claim without asking us.
Where a chain is incomplete, the platform shows the gap rather than hiding it. A record with no sources attached is marked as having none — the provenance bar prints no sources attached rather than omitting the line — and a record that has never been reviewed says that too instead of displaying a comforting date.
The same rule governs everything the ingestion adapters bring in. Fetched material is not a published record; it is unassessed source material waiting for a person. Which upstream registers are actually connected, and what is outstanding on the rest, is set out in full on the data sources page.