Social network analysis has acquired a reputation as either an arcane discipline or a presentational flourish. It is neither. It resolves three specific questions that an enumeration of directors cannot, and each resolution holds only under conditions worth understanding before any weight is placed upon it.
A diligence file conventionally arrives as a set of attributes belonging to a single entity: incorporation particulars, directors, shareholders, filings, adverse coverage. Each attribute is singular. The entity is active. The director is named. The address is registered. Every proposition concerns one node in isolation.
The propositions that alter a commercial decision are seldom singular. They are relational. That a company is trading is information. That it shares a registered address with fourteen other entities, three of which share a director with a party previously identified as acting on instruction, is a decision.
Social network analysis is the discipline of constructing the graph in which relational propositions become visible, then interrogating that graph in a manner that is reproducible and open to audit. Nodes represent entities: natural persons, companies, addresses, telephone numbers, vessels, domains. Edges represent asserted relationships: directorship, shareholding, registration, litigation, co-mention. Provenance is preserved at the level of the individual edge, so that any downstream insight remains traceable to the finding that created it.
First question: who else occupies this address?
Address clustering identifies registered addresses shared by an improbable number of distinct entities. It remains the most durable shell indicator in the discipline, principally because incorporation agents are economical with their arrangements.
Where it fails. Serviced offices and free zones. In Dubai, Mauritius or the City of London, several thousand entirely legitimate companies may share a single registered address. The cluster constitutes a line of enquiry rather than a finding. It establishes where to direct attention; a human being must then establish whether co-location signifies anything at all.
Second question: what is the distance to a designated party?
Proximity analysis computes the shortest path between the subject and any designated node within the graph. A single hop denotes a direct relationship with a sanctioned party and is treated as critical. Two or three hops merit examination. Beyond that threshold, in any sufficiently dense commercial graph, near-universal connectivity renders the measure meaningless.
Where it fails. Path composition. A two-hop connection running through a shared directorship in a trading company differs materially from a two-hop connection running through joint mention in a single news article. Any proximity result presented without the intervening edge types is not interpretable. Composition governs significance more decisively than length.
Third question: is this individual substantively present in the role?
Nominee indicators identify persons recorded as director, shareholder or agent across an implausible number of entities in the absence of corroborating evidence of substantive involvement: no correspondence, no operational function, no professional footprint consistent with the direction of that many businesses.
Where it fails. The measure is heuristic, and published without verification it is defamatory. Professional company secretaries and corporate service providers lawfully hold dozens of appointments. The output constitutes a question for human resolution and never a conclusion fit for a client report.
The graph determines where a person should be sent. It does not substitute for the person.
Reproducibility as a precondition
An insight incapable of recomputation is incapable of defence. Where a report asserts that a subject stands two hops from a designated entity, a reviewer must be able to trace that assertion to a deterministic computation over a known state of the graph: identical algorithm, identical version, identical parameters, identical underlying findings.
Thresholds therefore belong in configuration rather than in code. Every insight records the graph version against which it was computed. The graph itself is append-only. An analysis performed today and repeated in six months will diverge, legitimately, and the analyst will be able to state precisely what changed and why.
The limit of the method
A graph constructed solely from open sources inherits every deficiency of those sources. In jurisdictions where ownership is declared rather than verified, the graph will reproduce the declared fiction with complete fidelity. Network analysis conducted over a partial record yields a confident cartography of the wrong territory.
For that reason the analysis is run over the combined corpus: open-source records and documented human sourcing in a single graph. The human layer supplies the edges that filings omit, which is to say the relationship every participant in the market understands and no registry has ever recorded. Constructed on data alone, the graph delivers structure without truth. Constructed on both, it delivers a map on which a decision can properly rest.