Forty-five days before the LEI becomes the US entity standard, the SEC's register of 981,355 entities lists 773

What we measured

Four register surfaces, all keyless, all fetched on the same day. The EDGAR bulk submissions file is the SEC's entity register: 981,355 entity records, refreshed daily, one JSON document per CIK. Every record carries a top-level field named lei. The GLEIF golden copy is the global LEI register: 3,403,856 records in the 17 August publish, of which 358,275 are US entities. The SEC's structured form datasets for 2026 Q2 record what regulated funds told the SEC on Forms N-PORT and N-CEN, including their own LEIs, their series' LEIs, and the LEIs and CUSIPs of 5.3 million portfolio holdings. The GLEIF ISIN-LEI mapping file links 9,135,428 securities to their issuers' LEIs, openly and daily.

Finding one: the field is empty

The lei field is populated in 773 of 981,355 EDGAR entity records. That is 0.079 per cent. Among the 7,992 listed operating companies in the register, 25 carry an LEI. Apple's LEI has been ISSUED at GLEIF since 2012; its EDGAR record says null. The same is true of essentially every household-name filer.

The operational consequence is direct. Any institution that needs to join SEC filings to sanctions lists, to counterparty risk systems, to Basel reporting, or to any of the datasets keyed on the LEI cannot do it from the SEC's register. Every consumer of EDGAR data rebuilds the same crosswalk privately, badly, or not at all. From 1 October the identifier the register ignores is the standard the register's operator has adopted.

Finding two: what the field contains when it is not empty

Of the 773 populated values, 667 are valid LEIs. The other 106 are a museum of what happens when a register publishes a field it never validates. There are telephone numbers, including "(646) 508-0022" and "424-231-9100". There are IRS employer identification numbers. There are entity names typed into the identifier field, including "DANGEROUS 7 SPV 2 LP" and "WUWALLACEFAMILYTRUST". There is a Connecticut state registry string, US-CT.BER:3091043. There are fourteen 20-character strings that look like LEIs and fail the ISO 7064 check digits, and there are thirty-five records whose LEI is the literal text "N/A".

The LEI carries two check digits precisely so that software can reject a corrupted value. The same government already relies on this: a HMDA loan identifier must begin with a valid LEI and end with a computable check digit, by regulation. EDGAR runs no such check on its own LEI field and republishes whatever was filed.

Finding three: the SEC holds the crosswalk it does not publish

In a single quarter, 1,973 fund registrants stated their LEI to the SEC on Forms N-PORT and N-CEN. The entity register surfaces 12 of them. For 1,954 CIKs we can show the operator holds a valid LEI on one publication surface while the entity register's field for the same CIK sits empty. Silence here is not missing data; it is a published position of the register, and it is wrong.

Where both surfaces are populated they can disagree. The entity register says CIK 892538, SunAmerica Series Trust, has LEI 549300E40BQMHI2LOX26. GLEIF says that LEI belongs to SunAmerica Asset Management, LLC, the trust's investment adviser. The trust's own filings on Form N-PORT report 549300YDIAXUNCUXFM44, which GLEIF registers as SunAmerica Series Trust against exactly this CIK. The register carries the manager's identity on the fund. We found the same defect class last week in a different US register: the FDIC's BankFind record for Associated Bank, N.A. carries the LEI of its holding company. Two regulators, one failure mode: identity assigned to the wrong side of a control relationship at the register boundary.

Finding four: the reverse crosswalk exists, at GLEIF

27,704 LEI records in the golden copy name EDGAR as their registration authority and carry an EDGAR identifier in their registeredAs field: 22,672 series IDs and 5,019 CIKs. GLEIF, a Swiss foundation, publishes a mapping from LEIs to SEC identifiers every day under CC0. The SEC publishes no mapping in either direction, and its own investment company series and class register, 19,340 series, has no LEI column at all. The authoritative crosswalk between the US securities register and the US-adopted entity standard is maintained outside the United States, by the counterparty register.

Finding five: the two halves disagree 100 times

Because GLEIF records which EDGAR series each LEI belongs to, and because funds report their series LEIs to the SEC on N-PORT, the two halves of the same mapping can be reconciled for 12,604 series. They agree 12,504 times. One hundred fund series carry two different LEIs on the two sides of the register boundary.

The specimens tell you how identity actually decays. Voya Ultra Short Income ETF is 254900FKI2RDASVD0175 at GLEIF, ISSUED and valid. Its N-PORT filings say 254900FK12RDASVD0175, with the letter I replaced by the digit 1. The reported variant fails the check digits, so the checksum the SEC does not run would have caught the SEC's own data. Three VictoryShares ETF series hold each other's LEIs in a cycle, the unmistakable signature of a column shift in someone's reporting pipeline, now frozen into regulatory filings. Eleven checksum-valid LEIs reported to the SEC do not exist in the GLEIF golden copy at all: plausible identifiers that resolve to nothing.

The operational consequence: any system that keys fund exposure, fee analysis, or counterparty aggregation on the reported LEI will silently split one fund into two entities or merge two into one, and no schema validator on either register can see it, because each register is internally consistent. The defect exists only at the boundary.

Finding six: most US LEIs are not current anyway

Of 358,275 US LEI records, 202,698 are LAPSED: 56.6 per cent, materially worse than the global lapse rate. Among the LEIs asserted on SEC surfaces the status distribution is better but not clean, and the full breakdown is in the repository's governance summary. Adoption without renewal is not identification; it is an inventory of expired claims.

The findings, graded

All figures are computed from register surfaces fetched on 17 August 2026: the EDGAR bulk submissions file, the GLEIF golden copy published at 08:00 that day, the GLEIF ISIN-LEI mapping, and the SEC structured form datasets for 2026 Q2. All are living systems, so a later run produces different totals while the method reproduces exactly.

The method, transferable

None of this required privileged access. The method is the same one we have now run against fund registers, insurance registers, scholarly records, bank registers, learning standards, and enterprise knowledge bases.

First, model identity honestly. An identifier in a register is not a property of an entity; it is a dated claim by a named register. Our OWL model reifies every published value as an IdentifierAssertion carrying the value exactly as published, the scheme the field claims, whether the value conforms to the rules that scheme declares for itself, and, for LEIs, whether the value resolves at GLEIF and with what status.

Second, make the schemes self-describing. A SKOS registry declares each scheme's length, character set, and check-digit algorithm as data, so the validator has no hard-coded rules and adding a register means adding data, not code.

Third, make defects first-class. Each defect class is one SHACL shape, so the validation report is the findings table. Cross-register comparisons are ReconciliationObservation nodes; a register holding an identifier on one surface while another sits empty is a CoverageObservation.

Fourth, never trust one computation. Every headline is computed set-based in Python and independently via SPARQL and SHACL over the graph, and the build fails if they disagree. This gate has caught a real bug in every register we have audited, including, twice, our own pipelines.

Prior art, credited

FIBO, maintained by the EDM Association, models legal entities and US registry identifiers conceptually, and its maintainers are actively improving LEI constraints; our findings sit downstream of FIBO's scope, at the layer where published values meet declared rules. The OFR's 2018 staff discussion paper by Liju Fan and Mark Flood prototyped OWL over bank registry data and remains the closest federal prior art. The Data Foundation's work on FDTA implementation, including its argument that the Act is about meaning sharing rather than file formats, frames the policy need this measurement serves. GLEIF's own data quality programme assures what issuers submit; nothing in it constrains what a downstream register republishes, which is exactly the gap measured here.

What a regulator or a fund complex should do with this

If you operate a register: validate identifier fields against their schemes at ingestion, publish the crosswalks you already hold, and treat an empty field as an assertion you are making. If you run regulatory reporting at a fund complex: reconcile your reported LEIs against the golden copy quarterly; the 100 disagreements above are all detectable from your side with the checksums and the open files.

Common questions

Open, reproducible, and free to use

The repository, with the ontology, the SKOS scheme registry, the three SHACL layers, the harvest and reconciliation pipeline, the query library and the build report, is public: code under MIT, ontology and documentation under CC BY 4.0. The build report lists what could not be obtained as carefully as what could, and the graph of 525,387 triples regenerates from about 2.5 GB of open register data with the scripts as shipped.

Related work: the same register-boundary discipline found the mirror-image defect at the FDIC in the US bank register, was proved on the US fund universe, and the category it belongs to is set out in Register assurance: why every public register fails at its boundary.