The largest index fund is missing from the open identifier map

Scope is the load-bearing fact

The model is deliberately small: a product hierarchy of four levels, and identifiers as first-class nodes rather than string properties. Every identifier assertion carries three things a bare literal cannot: the scheme it belongs to, the source system that asserted it, and its computed check-digit state. The SKOS registry behind it declares each of twenty schemes across six markets (US, UK, Ireland, Australia, Canada, Mexico) as entity-scoped, issue-scoped or venue-scoped, and SHACL enforces the attachment: an LEI names a legal entity, an ISIN names an issue, a ticker names an admission to one venue. Most of what goes wrong in fund data is an identifier sitting at the wrong level of this diagram.

The same class listed on two venues has two listings, two tickers, possibly two SEDOLs, and one ISIN. Put the ticker on the share class, as most spreadsheets do, and the model cannot even represent the dual listing, let alone govern it.

Three sources that must agree, measured where they do not

Each source lands in its own named graph, so provenance survives the merge and cross-system disagreement is a query rather than an audit project. The build takes 82 seconds including full-file validation of every GLEIF pair; the SHACL gate covers the whole 1.29 million triple graph in 27. This matters because the industry default is to treat identifier reconciliation as a quarterly project. At these speeds it is a continuous property of the data.

Nine findings, all reproducible

Figures throughout this page are dated to the 14 August 2026 build, run against the GLEIF file dated 8 August 2026 and four pinned quarters of Form N-CEN. The SEC series/class register and the GLEIF file are fetched as latest, not pinned, so a rerun today will not reproduce these exact totals; see "Open, aligned, and reproducible" below.

The strongest result: where the registration authority owns the pipeline, the arithmetic holds

The GLEIF file, all 9,119,948 ISIN-to-LEI pairs of it, checksum-validated in full, not sampled, returns zero ISO 6166 (ISIN) and ISO 17442/ISO 7064 (LEI) check-digit failures. The zero is not a vacuous pass: the validators embed corrupted test vectors (a broken Apple ISIN, a broken GLEIF LEI, a broken HSBC SEDOL) and refuse them on every run. The scope is narrow by design: check-digit arithmetic only. It says nothing about deduplication, about whether any given LEI registration is still current, or about whether a given ISIN-to-LEI pair is actually the correct pairing; none of that was tested. Within that narrow, honestly-stated scope, 9.1 million machine-maintained records are clean.

The control group is where the value sits. An LEI carries two check digits under ISO 7064 MOD 97-10, so a corrupted one is mechanically detectable by anyone who cares to divide by 97. Of the 14,960 LEI values self-reported across four quarters of N-CEN filings, 19 distinct values fail that division, a rate of 0.13%. Two of those 19 are additionally malformed, not even 20 characters of the right alphabet, a subset of the 19 rather than an addition to it. One filer submitted 00000000000000238096, a zero-padded integer wearing an LEI's length. The Guardian Insurance separate account and a Teucrium ETF file LEIs that fail the checksum outright. None of this is exotic fraud; it is what happens when a regulated form field carries no validation and downstream nobody divides by 97. The check is pure arithmetic, so even a 0.13% failure rate is a real defect, not a rounding artefact: it should be zero, and machine-maintained data shows that zero is achievable. Where the value is retyped into a filing, it rots at a measurable rate. That difference, not any exotic technology, is the case for validating at the point of entry with executable constraints.

A different kind of signal, not necessarily a defect: 214 funds file their registrant's LEI as their own, and 13 LEIs are shared across sibling funds of the same umbrella, the largest spanning 27 distinct SEC series under one LEI. This is very likely a legitimate series-trust structure, one legal entity hosting many SEC series that are not separately incorporated, which is exactly why the ontology's own SHACL shapes grade both patterns Warning, not Violation. It is a genuine question about identifier granularity, worth asking the filer about, not evidence of a broken ID.

The hole in the open fabric sits where the money is

Every US ETF trades on an exchange, so every US ETF has an ISIN. Yet only 497 of the 4,053 self-reported ETF funds in the graph, 12.3 percent, have any ISIN against their LEI in GLEIF's open mapping. The Vanguard 500 Index Fund is among those missing the link: LEI issued and current, ISIN present in every commercial terminal on earth, absent from the open map that is supposed to link the two. VOO is not missing an identifier; GLEIF's open file simply does not carry the LEI-to-ISIN pair for it, a known gap in voluntary ISIN reporting by the legal entity. The open identifier fabric is thinnest precisely for the instruments the public trades most.

The enclosure goes deeper. GLEIF maps ISINs to the fund's legal entity, but which share class an ISIN names is knowledge that lives in the CUSIP registry, which is licensed. From public data alone, 259 of the 19,803 funds in the graph, 1.3 percent, resolve their ISIN to a class, and only because they have one class and one ISIN so the pairing is forced. For everything else, the last hop of the join between the world's two great open identifier systems runs through a paywall. A separate, structural gap sits beside it: the SEC's own series/class register carries no trading-venue field at all, for any listing; the only venue source is Form N-CEN's ETF-only exchange table (Item E.1), which non-ETF funds never complete because it does not apply to them. Only 980 of 30,238 listings, 3.2 percent, resolve a venue; the other 29,258 (96.8 percent) are mostly ordinary open-end mutual fund share classes that transact at NAV and were never going to have a venue to report. The ontology's own SHACL shape grades this Warning and says so explicitly: not an error in the record, a scope limit of the public schema.

We say this precisely because the model refuses to fudge it: fund-level ISIN assertions attach through a dedicated property that says "this entity issued this security, class unresolved", and promotion to class level happens only when the data earns it. An ontology that silently guessed the attachment would demo better and govern worse.

Policy as shapes, arithmetic as code

The governance layer is 15 SHACL shapes plus six SPARQL-based business rules, and the division of labour is deliberate. Check-digit arithmetic runs in the pipeline and is asserted into the graph; shapes then require the recorded result to be true. Encoding MOD 97 in SPARQL is possible and unwise. The rules that pay for the machinery are the conditional ones: an exchange-traded fund whose share class has no listing is flagged at the moment of assertion, which is the executable form of "an ETF you cannot trade is a launch defect". 73 of 4,053 ETF funds (1.8 percent) hit this rule, the single Violation-severity result in the whole governance layer; every other finding on this page is graded Warning. The same rule family catches identifiers attached at the wrong scope level, conflicting LEIs from different sources, and funds whose registered classes exceed the count on their own annual report (2,148 of them, Warning-severity, part timing artefact, part genuine drift, and the shape severity says which interpretation it earns).

Warnings are findings, not noise. A fund with no LEI from any source (38.3 percent of the graph, mostly funds outside the N-CEN window) is not a broken record; it is a governance decision waiting for an owner. Encoding that distinction in shape severities is what turns a validation report into a work queue.

Open, aligned, and reproducible

The ontology, the twenty-scheme SKOS registry, the shapes, the pipeline, the tested SPARQL library and a committed example subgraph (the Vanguard Index Funds family, 1,371 triples) are public under CC BY 4.0 and MIT. The FIBO alignment is graded and every target IRI was verified live against the EDM Council's published Turtle before assertion; where FIBO has no counterpart, the alignment says so instead of inventing one. Every figure on this page traces to one of four pipeline commands run against about seventy megabytes of public downloads. Run them again today and the method reproduces exactly, but the SEC register and the GLEIF file will have moved on, so the exact totals will not.

Related work: our FinanceBench verification study applies the same evidence-before-assertion discipline to financial question answering, and the machine-validated ontologies study measures when an ontology can actually reject a wrong statement. Working on fund data, identifier governance or semantic data foundations in financial services? Write to fabio@thetesseractacademy.com.