Most public records are dull on their own. A company register says that a person became a director on a given day. A filing says that one company holds shares in another. A gazette notice says that a firm has moved its registered office. A procurement register says that a contract was awarded. Each is a small, true, dated statement, and hardly any of them, read alone, tells you anything new.
Put them side by side and something else appears. The director is also a director of eleven other companies. The shareholder is a holding company whose own shareholder sits in a low-tax jurisdiction. The new registered office is shared by four hundred firms. The contract went to a company that did not exist two months earlier. None of these facts is in any single record. Each is a relation between records. That is why, in our work, we say the link is the finding.
What a single record cannot say
Every record is written for one purpose. A register of directors tells the public who is responsible for running a company. A shareholding filing tells the state who owns it. A procurement notice tells losing bidders who won. Nobody writes a record to explain how it relates to all the others.
So the questions that matter are almost never answered in one place. Who controls this company? is answered by following a chain of shareholders until it reaches a person. Do these two firms act together? by finding the people, addresses and advisers they share. Was this company set up for one deal? by putting its creation date next to the contract date and the date it was wound up. Each answer is a path across several records, often from different registers and countries.
The scale makes this hard to do by hand. The UK's company register alone held 5,485,994 companies at the end of September 2026, each with its officers, owners and filings. Disclosure rules help, but only so far. Companies House requires every company to identify its "people with significant control", usually anyone holding more than 25% of the shares or voting rights. Holders acting together are meant to be counted as one. But a stake split among holders who look independent, each under the line, appears as control only if someone declares the arrangement. The register shows what is declared, not necessarily what is true; what it means depends on what you join it with.
Patterns that repay attention
Some shapes recur often enough that we look for them on purpose. All of them have ordinary explanations as well as interesting ones.
Chains of holding companies. Company A is owned by B, which is owned by C in another country, which is owned by a trust. Each step is visible somewhere; the whole chain nowhere. Long chains are normal in large groups. One that ends in a jurisdiction that publishes little is a reason to keep reading.
The same nominee on many boards. Person X sits on the boards of hundreds of unrelated companies. Usually X is a professional director supplied by a service firm. The useful question is who appointed X, and for whom.
One address, hundreds of firms. A single office can be the registered address of hundreds or thousands of companies, usually a formation agent or an accountant. It says who did the paperwork, not who owns the business. A small address shared by a handful of firms that also share a director is a different matter.
Officers who move together. A group of people leave Company A within weeks and reappear, in the same roles, at Company B. Sometimes it is a team changing employer. Sometimes B is A under a new name, with the old debts left behind.
A company created just in time. Company D is incorporated six weeks before it wins a public contract, files no accounts and is dissolved a year later. The dates sit in three registers. Together they describe a vehicle built for one transaction.
Ownership that moves just before it must be shown. A stake changes hands days before a reporting date and changes back shortly after. Each filing is correct on its own day. The pattern appears only when the dates are laid on one line.
A link is a question
It would be easy to read every one of these shapes as wrongdoing. That would be a mistake, and an unfair one. Many shared addresses belong to formation agents. Long ownership chains are the normal structure of international groups. Family businesses share directors because the directors are related.
So our rule is that a link is a question, not an accusation. We state the connection in the plainest words the records allow, then ask what would explain it, and we try the innocent explanations first: the agent, the group, the family. Only what remains after that deserves a reader's attention.
Behind that rule sit three habits that the technical side of our work is built around, and that a companion note, Why most links are noise, describes. Every link carries its dates, because two relations that never overlapped cannot connect two people. Every link carries its source, so that a reader, a lawyer or the person named can check each step. And every claim that two records describe the same person or company is a recorded decision that can be reviewed and undone.
Why this matters
Public records are one of the few sources anyone can consult on equal terms. They are filed under legal obligation by the people and companies they describe, and kept by public bodies. That makes them unusually trustworthy as raw material, and easy to misuse, because a list of true facts can be arranged to suggest something false.
None of this is new. Archivists, auditors and investigative reporters have long worked this way, with paper and patience. What has changed is that registers once visited in person are now published as data, and that the questions asked of them have grown: who stands behind a supplier, a counterparty, a donor, a new shareholder. The answer is rarely written down anywhere. It has to be assembled, one dated and sourced link at a time, and then tested against every ordinary explanation before it is allowed to mean something.