Which registers can decide your file, and which cannot
Source inventory read off the product repository on · endpoints probed live and dated
Every quality figure on this page comes from an internal, repository-authored fixture. There is no external holdout and no third-party validation. The method is on /accuracy.
On this page
Built for these files. Wrong for these.
The coverage question is answerable in two lists. If your file is in the first, every row can be decided against a register and cited to a record a third party can open.
A file of UK companies, US SEC filers, French or Norwegian companies, UK and EEA regulated-market issuers, or any row carrying an LEI or a register-qualified company number. Names, towns, postcodes and domains all work as input, and the file does not have to be clean.
A file of German private companies or Delaware LLCs. Those registers publish nothing a decision can be cited to, at any price, and no amount of engineering changes it. Both are set out in full at the foot of this page.
Entity Bind resolves companies and organisations only. It does not resolve people, vessels, bank-account ownership, sanctions exposure, emissions or geolocation, and its results may not be used for eligibility decisions about individuals, including employment, credit and tenancy.
The nine sources a request reaches today
The pill is the licence tier the running system enforces; the line under it is the instrument your procurement lawyer asks for. Open the row to see what was measured on it.
Live
9 sourcesWired into the running deployment, on by default, and exercised by the measurements published on this site. A request can reach these today.
Open Government Licence v3.0
Company profile, previous company names, registered office, register status, UK SIC 2007 codes, accounts and confirmation dates, and persons with significant control at company level.
what was measured
On a UK company's legal name the register of incorporation outranks every other source, and that ordering lives in the code rather than in a style guide. Five personal fields the register volunteers on a PSC — date of birth, nationality, country of residence, name elements and address — are dropped at the ingestion boundary and never reach storage, pinned by a named offline test. Corporate PSCs are kept: over the recorded log, 252 companies carried PSC facts and 311 owners, 26 of them corporate, and all 26 carried the full identification block that makes the next ownership hop resolvable.
- Companies House register search
The register itself. The adapter reads the same records through the API rather than through this search front end, and the five personal fields the register volunteers on a person with significant control are dropped before storage.
- Open Government Licence v3.0
The instrument itself. It is what makes these facts storable and resellable rather than return-only, and it is the document counsel asks for.
Open Government Licence v3.0
The filed accounts themselves, with the figures machine-tagged inside the document in the format known as inline XBRL: average employees during the period, and turnover for the small minority of companies that must file a profit and loss account.
what was measured
A separate switch from the company profile, because it is a different host, a different rate-limit bucket and a payload two orders of magnitude larger. Measured over 18,894 real filings from one Accounts Data Product day: average employees present on 90.3%, net assets on 85.9%, assets less current liabilities on 80.3% — and turnover on 0.9%, because only 2.13% of the register has to file a profit and loss account at all. Plan on the headcount and the balance sheet; treat turnover as a bonus. That floor is a limit of UK company law rather than of the adapter, and it is why the listed-issuer source below exists.
- Companies House developer hub
The API documentation. The accounts documents are a different host and a different rate-limit bucket from the company profile, which is why they are a separate row and a separate switch.
- Open Government Licence v3.0
The instrument itself. It is what makes these facts storable and resellable rather than return-only, and it is the document counsel asks for.
CC0 1.0
LEI, legal name, other and previous names, registered and headquarters address, BIC, ISO 20275 legal form code, the issuing register's authority code with the number held in it, and direct and ultimate parent LEIs where both parties hold one.
what was measured
The only source in the inventory that grades its own facts: every record carries a corroboration level, so an uncorroborated self-reported name can be demoted below a filed one rather than trusted equally. It is also where an OpenCorporates identifier can be obtained without touching OpenCorporates' data or terms, because GLEIF publishes it under CC0.
- GLEIF
The body that publishes the record is the body that assigns the code, which is what makes an LEI authoritative by definition rather than corroborated.
- CC0 1.0 Universal
The dedication in full. Nothing here is conditional on attribution, which is why this material can be resold on the plan that carries resale rights.
CC0 1.0
The register of registers: 1,071 business registers, used to qualify a company number by the authority that issued it rather than by the country it sits in.
what was measured
Reference data rather than an evidence source: it is never asked about a customer's row. It exists because a company number is only an identifier inside the register that issued it, and 220 of those 232 jurisdictions run more than one register. Nothing yet consumes the qualified pair as a merge rule. Turning it into one would move the published bound on how often an accepted match is right, so it waits on a measurement rather than arriving with the adapter.
- GLEIF
The Registration Authorities list is published by GLEIF as a code list alongside the LEI records. It is reference data: it is never asked about a customer's row.
- CC0 1.0 Universal
The dedication in full. Nothing here is conditional on attribution, which is why this material can be resold on the plan that carries resale rights.
United States federal public domain, 17 U.S.C. §105
CIK, EIN, tickers and exchanges, state of incorporation, business and mailing addresses, switchboard phone, and former names with the dates they applied between.
what was measured
Field fill, measured over 40 large filers: EIN, business address, mailing address, state of incorporation and phone at 100%, former names with date brackets at 57%, and LEI at 2%. Nothing is built on the EDGAR LEI. The mailing address is where the SEC posts, frequently a registered agent or a post box, so it is kept in its own field and never written as a registered office.
- SEC EDGAR company search
The filer universe this source reads. It is a register of filers rather than a register of incorporation, so a company absent from it is not a company that does not exist.
United States federal public domain, 17 U.S.C. §105
As-filed revenue, total assets, employee counts and fiscal year end, each citable to the accession it was filed in and each carrying its own period in the value.
what was measured
About 6,300 filers carry structured financials against roughly 10,387 on the submissions endpoint, so a third of filers have an identity here and no figures. Concept coverage for CY2024 is uneven in the same way: assets on 6,253 entities and net income on 6,043, but the two revenue tags on 2,961 and 2,490 — neither covers half the population, and at least one large filer reports both.
- SEC EDGAR company search
The same filers. The structured financials come from a separate API host, and about a third of the filers listed here have an identity there and no figures.
Licence Ouverte 2.0 (Etalab)
SIREN and SIRET, legal name, NAF activity code, establishments, employee band and registered address, reached through the keyless government search route.
what was measured
The largest single tranche of resale-safe entities available anywhere, and the register with the hardest name problem in the inventory. See the parity section: a French file that arrives with names alone has a structural ceiling nothing in the resolver can lift.
- Annuaire des Entreprises
The keyless government search route this adapter uses. It is the same SIRENE universe INSEE publishes; what it cannot fix is that a third of French companies share their exact name.
Norwegian Licence for Open Government Data (NLOD) / CC BY 4.0
Organisation number, historic names, legal form, website, telephone, structured postal and business addresses, NACE codes, incorporation date, headcount, share capital, VAT-register flag, and three independent insolvency and dissolution booleans.
what was measured
The richest schema tested, and the reason the record model distinguishes register status from status detail: a single status word cannot carry what Norway publishes in three separate fields. With Finland it is one of only two registers anywhere in the inventory that supplies a website, and it is the only one that supplies a phone number — facts that are otherwise scraped and therefore uncitable.
- Enhetsregisteret, Brønnøysund Register Centre
The open data service itself. It is the richest schema tested and the only source in the inventory that supplies a telephone number.
- CC BY 4.0
One of the two instruments named above; the Norwegian Licence for Open Government Data is the other. Both require attribution and neither restricts resale.
A revocable permission — "no restrictions on the ways that the data can be used" — not a licence instrument
IFRS revenue, assets, equity and profit for listed issuers, tagged in inline XBRL under the ESEF regime.
what was measured
It exists because the UK companies whose turnover anyone actually asks about file on paper at Companies House: several household-name retailers filed 20 of 20 recent accounts as scanned PDFs. The same companies file again under the disclosure rules, where turnover is tagged on about 83% of reports rather than about 1%. The FCA's own National Storage Mechanism is deliberately not used: its terms forbid including any part of it in a private electronic retrieval system, which is precisely what this product is. Because the permission is revocable, only the extracted numbers are republished, each carries the terms and the retrieval date, and the terms are re-read on every refresh.
- filings.xbrl.org
The index this adapter reads. What it carries is a revocable permission rather than a licence instrument, so only the extracted numbers are republished and the terms are re-read on every refresh.
Four things get called coverage. One of them answers a request.
The distance between the first state and the last is the distance between a slide and a working integration. We use the word for state three, and quote a figure only for state four. The two groups below sit at states one and two.
- 01
Somebody read the source
The endpoint was called once on a dated day and its status recorded. Enough to cost a build. Not enough to answer a request.
- 02
An adapter exists
Code that maps that source's payload onto the record model, with tests. Most vendor coverage maps describe this state, two short of an answer.
- 03
It is switched on here
Wired into the running deployment, holding its credential, reachable by a resolution. Two of ours have sat here for months on a licence question.
- 04
Its behaviour has been measured
A fixture built from that register's own bulk snapshot, run through the resolver, results published. Five registers have reached it.
The rest of the inventory is complete, not curated: the blocked ones, the credential-less ones and the ones we rejected on their own terms are all here, so you can see where a register you need would land.
Built, not measured
7 sourcesThe adapter ships complete and nothing about its behaviour has been measured, so nothing about it is claimed. Two shapes sit here and they are not the same. EUIPO, USPTO and the two metered vendors are silent: switched off or holding no credential, so no request leaves. Finland, Czechia and Poland are the opposite — keyless, on by default, and reached today by any record that states that country or states none at all — and what they are missing is the fixture, not the wiring. Read the note on each row for which it is.
each adapter, and what it is missing
Open Government Licence v3.0
Latest charity income with both ends of the financial period it covers, joined to the company by the registered company number the charity publishes.
File-backed and offline: the daily tab-delimited extract of the whole register is read once at start-up and served from memory, so no network call happens during a resolution and it is the one source declared in the sub-processor test with no hosts at all. It has no configured default and stays switched off until an operator points it at that extract, because a silently empty register would look like a company that is not a charity. What was measured when the index was built: of 171,778 main registered charities, 162,685 carry a latest income and 31,509 publish a company number, 96.07% of which join the live register. Only main charities are read, so a linked sub-charity's income is never double-counted, and the 53 company numbers claimed by two main charities are published for neither. Scotland's OSCR is deliberately not read: its licence is the Open Government Licence plus two accepted conditions, which is a different instrument.
- Charity Commission for England and Wales
The regulator that publishes the daily register extract this source reads offline. Scotland's OSCR is a separate regulator under a different instrument and is deliberately not read.
- Open Government Licence v3.0
The instrument itself. It is what makes these facts storable and resellable rather than return-only, and it is the document counsel asks for.
CC BY 4.0
Business ID, EUID, full name history with validity dates, registered address, TOL classification, website and VAT registration entries.
Keyless, on by default and genuinely reached: a record stating Finland, or stating no country at all, is routed to it today, because a national register is skipped only when the record states a different country. It is also the only source anywhere that supplies the EUID — the sole pan-European registry-issued key in existence, assigned under the Business Registers Interconnection System. What it has no evaluation fixture for is quality, so there is no measured figure for Finland.
- Finnish Patent and Registration Office
The register that runs the Business Information System jointly with the Tax Administration. It is the only source anywhere that supplies the EUID.
- CC BY 4.0
Attribution is a condition of the licence rather than a courtesy. It is carried in the citation you get back anyway.
None. The bulk files are catalogued nationally as CC0 1.0; the REST endpoint this adapter uses is catalogued nowhere and carries no licence statement at all
IČO, VAT number with its registration status, roughly 40 NACE codes, a structured address, a per-record last-update date, and which constituent register holds each assertion.
Keyless, on by default and genuinely reached — a record stating Czechia, or stating no country at all, gets a call today. What it has no fixture for is measured quality, so no figure appears against it. Configured pass-through rather than open, so it retrieves a candidate and stores nothing. Two things would resolve that, in preference order: move ingestion to the catalogued bulk files, whose CC0 position is verified, or get a written statement from the ministry about the API. Raising the tier is a deliberate act by somebody who has read the terms, not an edit to a constant. It is also the only register that reports which of its own constituent registers holds each fact, which maps onto per-field provenance almost natively.
A CC0 label on the national data portal, contradicted by the Ministry's own article 15 re-use conditions
KRS number, REGON, NIP, legal form, the register's own as-of date, and the prior registration in the legacy RHB register.
Keyless, on by default and genuinely reached, on the same routing rule as Czechia. Pass-through for the reason above: two published statements about the same data disagree, so the narrower reading stands until one of them is settled. No evaluation fixture, so no quality figure. It is also the most people-dense payload in the inventory — the register extract carries the management board, the supervisory board and, for some company forms, shareholders by name — so it is read by a fixed allow-list of paths rather than by iterating the response. Poland runs three parallel number spaces, which is one of the reasons a company number is never treated as unique within a country.
A copyright grant that expressly excludes database rights
EU trade marks and Madrid registrations designating the EU, with the applicant name and address they were filed under.
On by default and disabled in practice. Access is free but not self-service: it needs completed sandbox integration tests, an officer's passport and a company-register excerpt, one subscription per organisation. This deployment holds no credential, so the adapter opens no socket and reports a missing credential on the sources endpoint. The open tier is refused at construction rather than merely unset, because EUIPO's terms forbid substantial extraction “however performed”.
- EUIPO
Access is free and not self-service: sandbox integration tests, an officer's passport and a company-register excerpt, one subscription per organisation. This deployment holds no credential, so the adapter opens no socket.
Public domain within the United States only, with international copyright expressly reserved
US trade marks and Madrid designations of the US, with the owner name and address on file.
The only source switched off by default. It needs no credential, so a default of on would be a default that calls, and the only free route that can be searched by owner name is undocumented and unversioned. The documented USPTO APIs cannot answer the question at all: the trademark status API is lookup-only by serial or registration number, and a customer's row carries neither. The public-domain statement being scoped to the United States is a live legal question for a UK-domiciled seller, so the tier is pass-through until somebody with advice says otherwise.
- USPTO
The documented APIs are lookup-only by serial or registration number, which a customer's row does not carry. The public-domain statement is scoped to the United States, which is a live question for a UK-domiciled seller.
Commercial terms that prohibit warehousing
Company-shaped fields taken out of a people database: a company name, a description, sometimes a domain.
Both adapters ship, neither holds a key here, and neither makes a request without one. They propose no candidates, only describe an entity something else has already identified, and the restriction comes from a live probe: asked about HSBC — 8 Canada Square, London — this class of vendor placed it in Londonderry, New Hampshire, and pointed its LinkedIn field at a recruitment sub-page, while the free UK register returned the correct address for the same company. Only the company-shaped keys are ever read out of a response, so anything person-shaped nested in a company record is dropped by never being looked at.
Planned
27 sourcesA costed build with no code behind it, or a build stopped on a specific unresolved licence question. Not available and not promised for a date. They are listed so you can see where a register you need would land, and what would have to be settled first.
each planned build, and what blocks it
Open Government Licence v3.0
The full register as a monthly snapshot, plus a real-time change stream.
The snapshot behind the published UK fixture was taken this way. Wiring it as a standing source is a separate build from the per-company API already live.
- Companies House developer hub
The bulk product and the streaming API are documented here. The snapshot behind the published UK fixture was taken this way; wiring it as a standing source is a separate build.
- Open Government Licence v3.0
The instrument itself. It is what makes these facts storable and resellable rather than return-only, and it is the document counsel asks for.
Open Government Licence v3.0, personal data excluded, corporate insolvency notices only
Corporate insolvency notices from 1998 onward, each carrying a structured company number and a previous company name on a permanent record URL.
The best citation in the inventory: it hands over the join key rather than forcing probabilistic name matching, and gives a previous company name for free. 3,317,918 notices. It supplies the event; the register supplies the state, and the two are kept apart.
- The Gazette
Each notice sits on a permanent record URL and carries a structured company number, which is why this is the best citation in the inventory: it hands over the join key instead of forcing a name match.
- Open Government Licence v3.0
The instrument itself. The scope taken here is narrower than the licence allows: corporate insolvency notices only, personal data excluded.
No dataset licence attaches — a protocol response, not a database
The tenant identifier behind a domain, from one unauthenticated OpenID discovery call, and therefore the cluster of domains a single organisation controls.
Verified clusters: microsoft.com, linkedin.com and github.com return one tenant; meta.com, instagram.com and whatsapp.com return another. Microsoft binds a domain to a tenant only after a DNS TXT proof, so this is verified domain control by the same organisation rather than an inference from a web page. It is an unordered cluster with no evidence of direction, it is not a parent-child relationship, and it lags divestitures by months.
Public signed artefacts; the log monitor is Apache-2.0
Domain portfolios from the subject-alternative-name sets of every browser-trusted certificate issued since 2018.
Planned as a self-hosted monitor against the public logs. The obvious hosted shortcut is explicitly rejected as a dependency: it returned HTTP 502 on three of three attempts, publishes no service level and states no licence. Shared hosting and CDN certificates bundle unrelated tenants into one name set, so this is corroboration and never a merge signal on its own.
Commission Decision 2011/833/EU, unverified on the host
Designations, of which 1,772 are enterprises, each citing the EUR-Lex instrument that created it.
The only sanctions source that meets the record-level citation bar without qualification, because the citation is to the legal instrument and the legal instrument is the fact. 916 of the 1,772 enterprise records carry at least one structured identification. A designation becomes a storable fact only where the list record and the entity share a strong identifier; otherwise it is returned as a screening result, which is an assertion about the matcher and never a fact about the company.
- EUR-Lex
The register of the legal instruments each designation cites. That is what makes this the only sanctions source meeting the record-level citation bar without qualification: the citation is to the instrument, and the instrument is the fact.
United States federal government work
Eleven US export-control and restricted-party lists behind one search, with the Federal Register notice per record.
- International Trade Administration
The publisher of the consolidated list. It is a search across eleven separate US restricted-party lists rather than a list of its own, and each record cites its Federal Register notice.
United States federal government work
267,253 federal debarment and exclusion records.
The query API is capped at ten requests a day for a non-federal user without a role, which makes per-lookup enrichment impossible. The bulk asynchronous extract is the only viable path, and that is a different and larger build.
- SAM.gov
The exclusions data itself. The query API is capped at ten requests a day for a non-federal user without a role, so per-lookup enrichment is impossible and the bulk extract is the only viable path.
Unverified; only the validity boolean is treated as open
A live validity answer for a VAT number.
Reproducible but not linkable: a caller can re-issue the identical request and get the identical answer, but there is no URL to open, which is materially weaker than a record-level citation. Storing the name and address it returns is blocked on an unfinished legal read, and several member states return dashes for both as national policy.
Open Government Licence v3.0
Public contract awards in the Open Contracting Data Standard, naming the supplier as the buyer recorded it.
Free reuse with attribution required
All Belgian registered entities, with establishments and VAT numbers.
CC BY 4.0 (Danish basic data)
All Danish registered entities.
Access is by request rather than by key, which is registration friction rather than a licence problem.
CC BY 4.0, high-value dataset slice only
All Swedish organisations, in the slice published under the EU high-value datasets regime.
Sweden dropped its fees in February 2025 because of the EU high-value datasets regulation, which has required member states since June 2024 to publish basic company data free, machine-readable, by API and in bulk. That regulation is the single largest change in open company data in a decade, and most of the planned European rows below exist because of it.
- Bolagsverket
Fees were dropped in February 2025 under the EU high-value datasets regulation. Only the slice published under that regime is in scope here, not the whole register.
- CC BY 4.0
Attribution is a condition of the licence rather than a courtesy. It is carried in the citation you get back anyway.
CC BY 4.0
Roughly six million Italian companies.
Blocked. Italy stood up a new portal under the high-value datasets regime, but the API is unverified and the distribution looks bulk-only, so there is no stable per-record URL to cite yet.
Japanese government standard terms of use, unverified
Every Japanese corporate number, with the name and address held against it.
CC BY, unverified
Every Australian Business Number holder, with trading names and GST registration.
- Australian Business Register, ABN Lookup
The publisher of the bulk extract. The licence is recorded as CC BY and unverified, so nothing is claimed about what may be stored from it.
Bilateral signed agreement, terms unverified
All Estonian entities across eight published datasets.
- Centre of Registers and Information Systems
The Estonian registrar. Access to the eight datasets is by bilateral signed agreement rather than by open licence, and those terms are unverified.
An acceptable-use grant behind signed terms, not an open licence
All Irish companies.
- Companies Registration Office, Ireland
An acceptable-use grant behind signed terms, not an open licence. That is the distinction that decides whether Irish facts could be stored, and it has not been settled.
Unverified
All Swiss entities, keyed by UID.
The documented route returns 401 and access is by email request. There may be a keyless distribution on the national open-data portal that bypasses it entirely, which is worth ten minutes before anybody budgets the slow path.
- Zefix, the Swiss central business name index
The documented API route returns 401 and access is by email request. There may be a keyless distribution on the national open-data portal that bypasses it, which is worth ten minutes before anybody budgets the slow path.
Paid commercial terms, €6.40 a month plus €0.02 a query
All Dutch entities.
Blocked. Per-query pricing on a resolution path is a different commercial shape from everything else here, and the terms may make it return-only rather than storable.
- KVK, the Netherlands Chamber of Commerce
Paid per query. That is a different commercial shape from everything else in this inventory, and the terms may make it return-only rather than storable.
Unverified; bulk extraction likely barred
Regulated firm reference numbers and permissions for UK financial services firms.
- Financial Conduct Authority
The FCA publishes the Financial Services Register. Its terms for bulk extraction are unverified, which is why this row is wired at pass-through in principle and has no adapter in practice.
Terms published in a separate document and unverified
2,875 market codes, 2,293 of them active, used to normalise the free-text exchange names filings carry.
CC0 1.0
53,291 LEIs and 21,335 LEI-and-website pairs.
Open, free and weak authority. Useful for corroborating a fact a register already asserts, never for settling one.
- Wikidata
Open, free and weak authority. Useful for corroborating a fact a register already asserts, never for settling one.
- CC0 1.0 Universal
The dedication in full. Nothing here is conditional on attribution, which is why this material can be resold on the plan that carries resale rights.
Open for the index metadata; archived page content is return-only
Dated snapshots of a company's own site, for previous names and prior addresses with a citable retrieval date.
- Internet Archive Wayback Machine
The index metadata is open; the archived page content is return-only. A dated snapshot of a company's own site is the company's assertion about itself, ranked accordingly.
The facts are not copyrightable; the page they sit on is
The registered name, register number, register court and VAT number a company is required by law to publish on its own website.
The one route to a German register court number that does not inherit somebody else's terms, because the company itself is legally obliged to publish it. It is still a company's own assertion about itself rather than a register's, and it is ranked accordingly.
Etalab Open Licence for the .fr registry
Domain registration dates and the registrant organisation where the registry publishes it.
A domain registration date is not an incorporation date, and this source must never be allowed to write one.
- Afnic, the .fr registry
The one registry whose RDAP service has been verified. A per-registry survey is outstanding, and a domain registration date is never allowed to write an incorporation date.
A protocol response, not a dataset
Mail routing and domain verification records, as weak evidence that two domains are operated by one organisation.
Live with a short time to live and no as-of date, so it corroborates and cannot be cited as a dated fact.
CC BY-NC 4.0, so a commercial licence is required
472 datasets and 4.09 million entities, refreshed every two to six hours.
Only usable under a reseller or OEM licence. The cheaper screening licence is internal-use with redistribution prohibited, which makes it pass-through, which this system's write path refuses to store — so buying the cheaper one would produce a source that can be queried and never retained.
- OpenSanctions
Only usable under a reseller or OEM licence. The cheaper screening licence is internal-use with redistribution prohibited, which this system's write path refuses to store.
- CC BY-NC 4.0
The non-commercial clause is the whole issue: it is why a commercial licence is required rather than optional.
Counts in this inventory are approximate and are recounted before they are quoted into a contract. The five registers with properly counted universes are further down. 29 of these 43 rows open at their publisher, and where the licence has a canonical deed that is linked too. The other 14 carry no link because their register address could not be established with certainty from here; read that as not yet checkable on this page, never as no register exists.
If the row you need is built but not live, name it and we will give you the real blocker, not a roadmap date.
Talk to usA register mirror is not a resolution
Your row is a name somebody typed, a town that may be the billing address, and a domain that may belong to the parent. A copy of Companies House answers “what is company 00617987”. Useful, cheap, and several people do it well.
It is not the question your file arrives with. That question is which of 5,464,811 live UK companies the row in front of you is, answered with evidence a third party can open.
We hold no proprietary corpus and sell no database. This page is an account of which registers a decision can be grounded in, and where one cannot.
In a product whose entire pitch is per-field citations, a fact we cannot cite to something a third party can open is worse than no fact: it reads as an answer.
Every row in the inventory is judged against that bar: a resolvable URL or a documented endpoint for the specific record. Sources that fail it are recorded with the reason.
Five registers, counted and compared
A bare name decides a UK row most of the time and a French row far less often, and that is a property of the register. Each one's live legal-person universe was counted from its own bulk snapshot, and the share carrying a unique normalised name measured over it. The census counts name collisions, not legal identity.
| Region | Register | Universe counted | Unique normalised name |
|---|---|---|---|
| United Kingdom | Companies House | 5,464,811 live companies | 99.53% The baseline every other region is measured against. |
| United States | Colorado Secretary of State | 2,191,009 extant legal persons | 97.76% State-level. A first review of the universe rule wrongly read 'live' as 'compliant' and excluded 1.13 million extant delinquent entities; the rule was corrected. |
| United States (filers) | SEC EDGAR | 6,471 tickered filers with complete rows | 99.91% Reported beside the combined figure and never blended into it: closed-end fund families at one address measure lower on the digital tier, which is a register-shape finding rather than an engine one. |
| France | SIRENE | 5,841,117 legal persons | 67.96% A third of French companies share their exact name — 1,003 live companies sit on one name. A French file needs a town or postcode column, and with one it measured 2,998 of 3,000 on the postcode tier. |
| Norway | Brønnøysund | 450,754 corporate forms | 97.78% |
To cite or reuse these figures, the census has a page of its own, with the method, each register’s snapshot and licence, and a CSV of exactly the numbers above.
This is a property of the register, not of the resolver. Nothing resolves a genuinely ambiguous bare name; we answer one with a shortlist and an abstention. The consequence is a column. A French file carrying a town or a postcode resolves very differently from one carrying only a name.
Against a bar of 0.9565 declared before the run, not chosen after it. Scope: United Kingdom, United States (Colorado) and EMEA (France and Norway).
Measured with no model spend at all, on identifier and digital tiers plus retrieval and fragmentation, using the same seeded, sibling-complete builder in every register. It does not carry the live-adjudicated name tiers across, which rest on jurisdiction-free calibration keys and have been recorded in the UK only.
what the figure is, and what it is not
What it is: the posterior probability of the measured contract holding — safety, decided tiers, retrieval and fragmentation — across the registers above.
What it is not: live adjudicated accuracy outside the United Kingdom. The name-tier expectations for Colorado, France and Norway are predictions written down before any spend. Finland, Czechia and Poland have adapters and no fixtures, so no quality figure exists for them, and “EMEA” here means France, Norway and the United Kingdom.
Where these universes were counted
Each register's own front door. The counts and the uniqueness shares are ours, measured over that register's bulk snapshot; the universe is theirs.
- Companies House register search
The register the baseline was measured on. The name-uniqueness figure is the register's own property, not the engine's.
- Colorado Secretary of State
Chosen because it publishes the whole register as public-domain bulk data, not for the size of its economy. One state register of fifty; Delaware publishes nothing at any price.
- SEC EDGAR company search
A register of filers, not of incorporation. It is reported beside the Colorado figure and never added to it.
- Annuaire des Entreprises
The SIRENE universe, through the keyless government search route. A third of these companies share their exact name, which is a fact about France rather than about the resolver.
- Enhetsregisteret, Brønnøysund Register Centre
The open data service the Norwegian fixture was built from, by the same seeded, sibling-complete builder that produced the UK's.
Internal fixtures only. The historical 2026-08-29 per-tier recording covers 317 GB companies with identifier-rich arrivals seeding the catalogue first; the context tier includes country and registered-town prose. Its legacy replay has no full-prompt-verified calls and cannot certify current calibration. The September 5 CI reports cover 1,176 labelled pairs and 705 records over 250 companies using offline stubs. These are not independent customer holdouts or live production validation; the 99% minimum-input product target remains unmet. Scope is company and organisation identity only — not people, vessels, bank-account ownership, sanctions or emissions. Results may not be used for eligibility decisions about individuals, including employment, credit and tenancy.
A page each, for the four measured jurisdictions
If your file spans more than one of these registers, a sample beats a call.
Start freeWhat you may keep, and where that is enforced
Every fact carries a licence tier, and the tier decides whether it may be stored or only returned. The rule is a refusal on the write path, not a clause in an agreement.
Open
Public domain, open-licensed, or self-derived from our own cited research. Companies House under the OGL, GLEIF under CC0, SEC EDGAR as US federal public domain, SIRENE under Licence Ouverte 2.0, Brønnøysund under the NLOD.
you may · enforced by
- You may
- Store it, redistribute it, and resell it on the plan that carries resale rights. Attribution is required by most of the instruments and is carried in the citation you get back anyway.
- Enforced by
- Nothing has to stop you. This is the only tier the persistence layer will accept without qualification, so the resale-safe subset is resale-safe by construction rather than by promise.
Needs check
The terms have been read and are genuinely ambiguous, or the licence statement lives somewhere the data does not. A provisional marker, not a legal position.
you may · enforced by
- You may
- Read the instrument beside the source, not the tier name, because this row is not uniform. Most sources in this state have no adapter or are wired at pass-through. One live source is not: the ESEF annual reports, whose permission is explicitly revocable rather than a licence instrument. Its extracted figures are stored and exported like an open fact, because a number is not copyrightable and only the numbers are republished. Ask us for the specific reading if a source in this state is on your critical path.
- Enforced by
- The configured tier, never the label. Czechia and Poland are wired at pass-through; Japan, Australia, Estonia, Switzerland and the market-identifier list have no adapter at all. ESEF is the exception: every fact it produces carries its terms and its retrieval date, and those terms are re-read on each refresh, so a withdrawn permission surfaces at the next refresh.
Licensed
Storable while a data licence is active, and only for its term. Nothing here is currently licensed by us.
you may · enforced by
- You may
- Nothing today, because we hold no such licence. If one were taken, facts from it would be storable for the term and withdrawn at the end of it.
- Enforced by
- Facts are withdrawn when the licence lapses rather than served past its term. In the product that undertaking is called the degradation contract.
Pass-through
Contractually prohibited from warehousing. It may be returned to the caller who asked for it and may never be retained here.
you may · enforced by
- You may
- Read it in the response. It is not in the export, it is not in the golden record — the single merged view assembled per entity — and it is not resellable.
- Enforced by
- Three layers with one rule. Saving an entity that carries a pass-through fact raises an error before any write happens; the serialised copy of the decision has that material stripped while the in-memory object keeps it, because it may be returned and must not be stored; and the step that picks a winning value for each field — survivorship — drops pass-through facts before a golden record is assembled at all.
Resale is decided field by field
Every field in the export carries its own resellable flag. A record assembled from four sources does not have one licence, so one label on the whole record would be a fiction.
what the flag is made of
It is the conjunction of two facts, published separately so you can tell which one is the limit: the data’s own licence, and your contract as reported on the limits endpoint. So a plan without resale rights never makes an open-licensed fact look restricted — it shows an open fact you are not contracted to resell.
Resale rights sit on the Scale plan at $799. What that plan includes
Storage and resale are separate rights. Check the licence and caller resale flags on each field.
where the boundary stands today
Real, tested and currently unexercised by traffic: every source callable without a credential declares the open tier, and the two metered vendors that would trip the guard hold no key here.
What we can key on, and what we can hand back
A resolved entity is a hub with cited edges to the identifier systems your stack already speaks: a company number qualified by the register that issued it, an LEI from the body that issues them.
GLEIF, which issues them
the qualification
Authoritative by definition: the body that publishes the record is the body that assigns the code.
The register of incorporation, with the authority code from the GLEIF register-of-registers list
the qualification
A company number plus a country is not a unique key and never was: 1,071 registers serve 232 jurisdictions, 220 of those jurisdictions run more than one, German numbers are qualified by the court that holds them, and Poland runs three parallel number spaces. Treating the pair as unique produces false positives, which is why the qualifier is carried on the record.
SEC EDGAR
SEC EDGAR
the qualification
EDGAR gives the exchange as free text, so a ticker is only unambiguous once the exchange is normalised to a market identifier code — which is itself a planned source, not a live one.
The home register (Czechia, Poland, Norway, Finland, Belgium, Denmark), checkable against the EU validation service
Finland's PRH feed
the qualification
The only pan-European registry-issued key that exists. Supply is thin because most member states do not publish it yet.
SIREN and SIRET in France, the organisation number in Norway, REGON and NIP in Poland, IČO in Czechia, the business ID in Finland
SEC EDGAR, at 100% fill on the sample measured
the qualification
Carried as a typed tax identifier rather than as a bare number, because a tax identifier without its scheme is a string that looks like several different things.
GLEIF, and national registers that publish it
GLEIF
Your own file, or your incumbent
the qualification
Accepted on input, matched on, and carried through. We cannot supply one: it is a proprietary identifier, and four sweeps of the open web found 35 records anywhere that publish one. If a D-U-N-S is the key your stack speaks, it has to come from the party that issues it.
GLEIF's own CC0 payload
the qualification
Obtainable without touching OpenCorporates' data or terms, because GLEIF publishes the mapping itself.
A planned source, not a live one
the qualification
Listed because it is the strongest available evidence that two domains are controlled by one organisation, and because it is not yet wired. It clusters; it does not order, so it can never establish which of two companies is the parent.
Not found is not the same as not configured
which one you were told
An empty field means one of two things and the response says which: the register publishes none, or the source that would answer is not switched on in your deployment. Collapsing those into a null turns a crosswalk into a claim that no such identifier exists.
A number without its register is a string
why the pair is carried
Company number plus country is not a unique key, and treating it as one produces false positives, not misses. The qualified pair is carried on the record before any rule is allowed to depend on it.
What we cannot reach
Limits of open data and of a deliberate boundary, not backlog items with a date on them. No amount of engineering closes most of them.
The largest economy in the EU publishes no API, no bulk product and no stable per-record URL, on terms that exclude reuse. Every commercial workaround is a scraper inheriting those terms, and the one open German dataset anybody cites stopped being maintained in 2019. No build is planned until the high-value datasets regime changes the position.
There is no federal company registry. Of twenty-seven bulk-capable states, four are free with no account — Alaska, Colorado, Connecticut and Ohio. Florida is a fixed-width flat file over FTP. California and Texas charge.
Delaware, the most important formation state, offers no bulk product and no API at any price: selling that data is part of its own business model. What exists here is SEC filers plus named states, and Colorado is the one measured. Nothing about a Delaware LLC is answerable from an open source, ours included.
Nothing wakes up nightly to refresh your records, and we sell no monitoring product.
what you get instead
Temporal evidence. Every fact carries when it was first seen, when it was last checked and how many times it has been confirmed, and a stale fact is served with its staleness visible. Nothing is silently refreshed and nothing is silently withheld. Survivorship ranks on last confirmed, so the most recently confirmed value wins. If a fact has to be current on a date, ask again on that date and read the timestamps you get back.
A private German GmbH, a French SARL or a US LLC yields no revenue, no assets and no profit from any open source. There are no proprietary risk scores — no failure score, no credit limit, no payment-behaviour index.
the whole reading
As-filed figures exist for roughly 6,300 US filers and for UK companies that had to file accounts. That is the whole of it. Where figures do exist they are as-filed and not standardised across jurisdictions: a US GAAP revenue tag and a UK FRS 102 turnover figure are not comparable line items and are not presented as though they were.
We supply the inputs a buyer can reason from, each citing a primary record, and we supply no score. There is no open equivalent of a trade-payment exchange and there cannot be one, because it is a data cooperative, not a public record.
Ownership is visible only where somebody had to disclose it: direct and ultimate parents where both parties hold an LEI, and UK persons with significant control above the filing threshold at company level. Below the threshold, and for most private companies worldwide, the chain is invisible. We can say we found no parent. We cannot say there is no parent, and the API keeps those two answers apart.
If you already run a licensed company-data feed or a master data management platform, run Entity Bind alongside it. It is the independent decision and evidence layer: it tests, repairs and governs the matches those systems propose, and it will contradict them with a citation when they are wrong.
The large licensed data providers hold hundreds of millions of records, an identifier the rest of your stack already speaks, and the deepest ownership and financials data in the category. We hold a fraction of their records and always will, and no global company graph — we are not building one. What we do is check the matches they propose and name the ones we would not have made. Every comparison, including where we lose
Officers, directors, beneficial owners as individuals, politically exposed persons and every contact-at-company record sit outside the product by design. They are not a gap being closed. The single exception is UK persons with significant control at company level, with all five personal fields discarded at the ingestion boundary. A buyer who needs a named human contact needs a different vendor. The scope limit itself is stated in full at the top of this page.
Sanctions and screening are not a capability here. A screening result is an assertion about a matcher, never a fact about a company, and false-positive rates above ninety per cent are normal: writing one into a record as a fact would manufacture a defamatory statement with a citation on it.
Once you know what the registers hold
Three questions follow a coverage answer, including the one that follows a no, and each has a family of pages behind it rather than a paragraph.
- Which column would settle a rowWhich of the identifier columns already in your file settle a match on their own, who issues each one, the exact shape a value takes, and what goes wrong with it.Open
- What to do about a register we cannot reachEntity Bind beside the alternatives, with the rows where the alternative wins counted rather than omitted, and the one category this product refuses to be sold into.Open
- How any of this is measuredLong-form method: blocking and candidate generation, Fellegi–Sunter, calibration, survivorship, and how to measure any of it on your own file.Open
Start with your own company records.
You get back which rows landed in which register, which columns would settle the ones that did not, and what we refused to decide. We write nothing back to your systems.
300 credits. No card.
For a guided review of your file, explore the Merge Safety Sprint →
Coverage and sources: your questions
Which countries do you actually cover?
Five registers carry measurements: the United Kingdom (Companies House, the baseline), the United States via Colorado and via SEC filers, France (SIRENE) and Norway (Brønnøysund). Combined three-region parity against the UK baseline measured 0.999955 against a pre-declared 0.9565 bar, at zero model spend, with no false binds in any register. Adapters for Finland, Czechia and Poland are wired but carry no evaluation fixtures, so they have no measured quality and we do not sell them as covered. Everything else in the source inventory is a planned build with a cost in engineer-days rather than a present capability. The claim is per register, never per continent.
Do you cover Germany?
No, and it is a hole rather than limited coverage. Germany has no lawful bulk source: the commercial register offers no API, no bulk product and no stable per-record URL, and its terms exclude reuse. Every commercial German workaround we found is a scraper inheriting those terms, which fails the bar this product sets for itself — a fact we cannot cite to something a third party can open is worse than no fact, because it reads as an answer. The EU High-Value Datasets regulation has required member states since June 2024 to publish basic company data openly, and that may change the position. Until enforcement does, the disposition is not to build, and the answer to a German file is no.
Do you have US coverage?
Scoped, and the scoping matters. There is no federal company register. SEC filers are wired through EDGAR, and at state level only Alaska, Colorado, Connecticut and Ohio publish bulk data free with no account; Colorado is the one with a measured fixture, at 2,191,009 extant legal persons and 97.76% name uniqueness. Delaware, the most important formation state in the country, sells no bulk product and no API at any price. There is also no quality claim for SEC filers yet: the wide digital tier measured lower there because closed-end fund families share near-identical names at one address. That is a finding about the register's shape rather than about the engine, which is why it is reported beside the combined figure instead of blended into it.
Why does France need an extra column?
Because a third of French companies share their exact name with another live company. The name-uniqueness census puts France at 67.96% against the United Kingdom's 99.53%: 1,003 live companies sit on one name. That is a property of the register rather than of the engine — no resolver can resolve a genuinely ambiguous bare name, and the answer there is the shortlist. Add a town or a postcode and it resolves: on the postcode tier, a French fixture measured 2,998 of 3,000. So a French file wants a postcode or commune column mapped, and if you do not have one, that is the finding rather than the failure.
Where does the data come from?
Public registers and cited web research, not a purchased corpus. Live and exercised today: UK Companies House including persons with significant control, SEC EDGAR submissions and XBRL, GLEIF's LEI records, Norway's Brønnøysund and France's SIRENE. The bar every source has to clear is verifiability at record level: a stable resolvable URL or documented API endpoint for the specific record, not a homepage, not a session-bound search form and not a bulk file the reader has to download and grep. Sources that fail it are recorded with reasons rather than quietly dropped. Where two sources disagree, the register of incorporation wins — and a live probe found a paid vendor placing HSBC in New Hampshire where Companies House had it in Canary Wharf.
How often is the data refreshed?
There is no cadence. Nothing revalidates on a schedule: a fact's `last_checked_at` advances only when a resolution happens to re-observe the same value, and a background revalidation job does not exist. What you get instead is the age, stated. Every fact carries when it was first seen, when it was last confirmed and how many times, the API exposes an age in days per field, and the entity reports the age of its stalest field rather than its freshest. A stale fact is served with its staleness visible, never silently refreshed and never silently withheld. If you need scheduled re-checking, that is a roadmap conversation rather than a feature.
Do you have ownership or UBO data?
Some, cited, and bounded. Ownership rides along from GLEIF Level 2 relationship records and the UK persons-with-significant-control register, with every edge carrying its basis and its certainty — because GLEIF's "parent" means accounting consolidation while a credit bureau's means majority legal ownership, and rendering both as one parent field is wrong in a way nobody can see. Traversal defaults to one hop and may be raised to ten, is cycle-safe, fails closed on ambiguous or name-only owners, and always populates what it did not walk, so you cannot infer depth from an empty field. It is bounded by LEI adoption, so an incomplete path is an absence of coverage rather than an absence of owners.
Do you have company financials?
For UK companies, headcount and balance sheet yes, revenue essentially no — and the reason is law rather than implementation. Only about 2% of the live UK register has to file a profit and loss account at all, and under 1% of filings in a sampled day carried a machine-readable turnover figure. Over 18,894 real filings, average employees reached 90.3% coverage and net assets 85.9%. Turnover for large issuers lives in ESEF instead, tagged on roughly 83% of regulated-market issuers. For a German GmbH, a French SARL or a US LLC there is no revenue, assets or profit available from any open source at any price. Quote headcount and balance sheet; treat revenue as a bonus.
Do you screen for sanctions?
No. Sanctions, adverse media and risk scoring are out of scope, and there is no risk boolean anywhere in the schema by design. A screening result is an assertion about a matcher rather than a fact about a company: false-positive rates of 90–99% are normal in production know-your-business screening, and writing one into a record as "sanctioned" would manufacture a defamatory fact with a citation attached. What a screening tool needs from us is the identity underneath it — which company this row actually is, with an evidence trail — so that its list matching runs against a resolved counterparty rather than a spelling. That is the layer we sell, beside a screening vendor rather than instead of one.