Mapping the gambling stack: a method note

This note describes how GambleLABS represents the gambling industry as data: what an entity is, what a relationship asserts, how evidence is graded, and what conclusions the resulting graph does and does not support. It is published so that any claim drawn from the model can be tested against the reasoning behind it.

Why a graph

The gambling industry is not a list of companies. It is a set of layered relationships — ownership, operation, software supply, content distribution, payment processing, marketing, hosting — in which the same commercial function is routinely split across several legal persons in several jurisdictions. The layers are described in Who actually owns the site you are betting on and Operator, platform, white label.

A flat directory cannot represent that. It answers "which companies exist?" when the questions that matter are "who controls this?", "who supplies it?", "who is answerable?" and "what else is connected to the same parties?" Those are questions about edges, not nodes.

Entities

An entity is a legal person or an identifiable business layer, recorded once and classified by kind: operator, company, platform, white-label provider, game provider, aggregator, payment provider, affiliate, hosting provider or licence holder. One organisation may legitimately appear under more than one kind if it performs more than one function; the classification describes a role in the stack, not a taxonomy of firms.

Every entity record states its own completeness — stub, partial or substantive — because a thin record is an accurate representation of thin evidence, and dressing it up as a dossier would misrepresent what is known. Editorial status is separate again: a record exists internally long before it is published, and presence in the database is never an implicit claim of verification.

Relationships

A relationship is a directed, typed edge between two entities: owns, subsidiary of, ultimate parent of, operates, powered by, supplies games to, aggregates for, processes payments for, licensed by, hosts, markets for, shares infrastructure with, former parent of.

Two properties of the model matter more than the vocabulary.

Edges are time-bounded. Each carries validity dates where known. Corporate structures change; a relationship that was true is not therefore true now, and an undated assertion is an assertion about an unspecified moment.

Edges carry their own evidence. Sources attach to the relationship, not to the entities at either end. This is deliberate: the fact that two companies are well documented says nothing about whether the link between them is.

Evidence grading

Every relationship is graded on two independent axes.

Evidence status describes the documentary position:

• verified — supported by a primary source, typically an official register, a corporate filing or a direct statement by the party itself.
• corroborated — supported by two or more independent sources of good standing, none of them primary.
• single-source — one source only, sufficient to record but not to rely on.
• unverified — asserted somewhere, not yet substantiated. Recorded at this grade precisely so the gap is visible rather than hidden.
• disputed — a party has contested it, and the contest is recorded alongside the claim.

Confidence — high, medium or low — is a separate judgement about the strength of the inference given that evidence. The two are not the same. A single primary document can support a high-confidence conclusion; three secondary reports of each other can support only a low-confidence one.

One part of this is enforced by the software rather than left to editorial discipline: an edge cannot be graded verified or corroborated unless a source record is attached to it. The editorial console refuses that grading when it is set on a new edge with no source, and refuses it again on every later edit of an edge that still has none. The lower grades are deliberately recordable without a source, because unverified exists to hold an assertion that has not been substantiated — the grade is what tells the reader so.

The equivalent control on entity dossiers is stricter. A dossier cannot move to published while it has no source attached, and the last source cannot be detached from one that is already live.

Everything else here is editorial policy, applied by people. It is stated so that a reader can hold GambleLABS to it, not because a database is holding us to it.

What the graph can demonstrate

• That a specific relationship was documented, by a specific source, at a specific date.
• That several brands share a documented operator, platform or supply route.
• That an entity's structure changed, and when the change was recorded.
• That a claim made in public conflicts with a primary record.
• Where evidence is absent — which is itself a finding, and one the model is designed to make visible rather than hide.

What the graph cannot demonstrate

• Liability. Ownership is not responsibility for debts. Establishing that one company controls another says nothing about who owes a player money.
• Intent. A structure that separates functions across jurisdictions may be tax-driven, licensing-driven, historical, or evasive. The edges look identical in every case.
• Causation. Two entities sharing infrastructure, an address or a director may be connected in a way that matters, or may be customers of the same service. Shared hosting in particular proves nothing — see Domain, host and brand.
• Completeness. The absence of an edge means no evidence has been recorded. It is not evidence of absence, and the interface never renders it as such.
• Current truth. Every record carries a last-reviewed date because the graph describes the state of the evidence when it was last examined, not the state of the world today.

The most frequent failure in network analysis is treating a densely connected diagram as a finding. Density is a property of the data collection, not of the industry.

Sourcing and review

Claims resolve to sources, each classified by type and reliability, with retrieval dates and archived copies where available, so that a claim survives link rot. Records are reviewed on a stated cycle and the review date is published. Corrections are permanent, attached to the record they correct, and never silently applied.

Where machine assistance is used in processing source material, the analysis is recorded as a distinct artefact with its provider, model, pipeline stage and human review status, and no such output reaches a public record without human approval. That process is described at the Evidence Engine.

The full standards — source classification, evidence thresholds, corrections policy and right of reply — are published in the methodology.