Entities
Registers of the facilities, companies and countries on record, with the ownership chains between them.
Of 130,394 organisation records, 2,019 have a ticker, 2,637 an SEC filer number and 1,075 a legal-entity identifier. Everything else resolves by name
Give it a name, an alias, a ticker, a filer number, a legal-entity identifier, a tracked site, a designated entity or a sovereign limited partner. It tries twelve stages in a fixed order and stops at the first that produces a candidate, then climbs the documented ownership chain above whatever it finds.
Resolve an entity
Five queries, three different rungsEach of these is a real query against this record, and the rung on each chip is the one that caught it when this page was built. Following one lands where typing it would land. Two of these are caught earlier than the rung they were chosen to show. An exact organisation name is rung 7 and it stops the stages below it, so a designated entity or a limited partner that also exists as an organisation name is caught as a name. Those later rungs fire only for handles the organisation table does not have. The label on each chip names the rung that actually caught it.
76,399 of those records have a page of their own; the rest are largely variant spellings and duplicate keys the canonical map folds onto one that does, and they are searched precisely so a query with one of them lands on the right record. 64% of the ownership edges (66,625 of them) were read directly from SEC filings and use the document’s own address, so a hop opens at its primary source instead of at a citation of one. The rest fill in private companies and foreign parents no securities regulator sees, and each is marked as that.
An organisation resolves onto its own record, where its ownership chain is a section of that page; a sovereign limited partner has no company record and resolves here. Each hop of the walk above the entity names the source it came from, the date that source speaks to, and a confidence stamp on the line itself. Conflicting parents both appear, and a chain that runs out of public record says so, with the gap labelled as a gap. Where a query matches more than one entity you are asked to choose, and where it matches nothing the page says what was tried.
What it can be handed
The stages run in the order below and the first one to produce a candidate stops the rest, which matters because they are not equally likely to fire. The exact-key stages run first and reach the least; nearly every organisation in the record has no structured handle, so in practice this is a name matcher with a handful of fast paths, and knowing which rung caught your query tells you how much to trust the hit.
The first rung is the only exact key that reaches the whole record, and it is the one this platform mints. Every identifier issued by somebody else (a securities regulator, a legal-entity registrar, an exchange) reaches between 0.8 and 2.0 per cent of the organisations here, because those identifiers exist only for entities with a registration obligation, and most of the corporate structure that compute runs through has none. That asymmetry is what the public identifier record looks like, and no amount of collection here would close it.
Trigram similarity, the last rung, runs only when every stage above it has come back empty, and what it returns lands with low confidence stamped on it. It keeps a query moving when the record spells a name differently. The rung above it usually saves the query first: the alias table holds 15,364 recorded spellings (former names, legal-entity forms, local-language renderings and acquired entities) against 130,394 organisation names, which is why an obsolete or translated name often resolves exactly.
Working detail: what a hit means
1 · Precedence, and why it is fixed
The stages are tried in one order for every query, and the first to produce a candidate stops the rest. A resolver that reordered its stages by what looked promising would return different entities for the same string on different days, which is the one property a citable identifier resolution cannot have.
Where a stage produces more than one distinct entity you are asked to choose. The tool never picks a winner between two records that match equally well.
2 · What each rung is counting
Each rung counts handles. A stage matches a string or a key, so the figure counts the strings and keys available to match: names, alias rows, designation entity names, site names, identifiers. One organisation has many aliases and at most one of each identifier, so no rung is a count of companies.
Each count is taken as the role that stage reads as, which is not the same for all of them: the identifier, name, alias and similarity stages search every organisation record, while the site and designation stages search only what an anonymous reader can browse. Counting them uniformly understated five rungs by more than an order of magnitude before it was caught.
3 · The walk above the entity
From the resolved entity the ownership graph is climbed through documented edges, to a ceiling of eight hops. Each hop names the class of source that established it and the date that source speaks to. A walk that reaches no documented parent is recorded as terminating, not as a one-entity chain.
4 · What is not published
One resolution stage matches a company key from a subscription corporate database. That stage works, and the size of the population it matches against is not published, so the figure shows no bar for it. An unmeasured rung and an empty one are different facts.
The same boundary governs what a resolved record can show: where a relationship was established through a subscription source, what renders is its class and the public primary that establishes it. The subscription record itself does not render.
The Investigator is entity-centric: you give it an entity and it climbs that entity’s ancestors. The impact pre-scorer reads the same records from the other end: you give it a hypothetical designation and it returns every facility the cascaded designation reaches.