Benchmarking UBO chain resolution across 80+ jurisdictions

The problem nobody admits

Every compliance team relies on corporate registries to resolve UBO chains. Almost every team also quietly knows the output is wrong more often than anyone wants to say out loud. Published figures from commercial providers hover around "95% coverage" — which is true if you ignore trusts, SPVs, nominees and any structure that routes through Jersey, Guernsey, BVI, Cayman or Delaware.

Our benchmark

We built a 2,400-entity test set from real onboarding files, evenly distributed across:

• UK Companies House, DE Handelsregister, LU RCS, FR RNE and 22 other EEA registries
• US Delaware Division, SEC EDGAR, and state-by-state secretary filings
• Offshore: BVI, Cayman, Jersey, Guernsey, Isle of Man — where we fall back to leaked datasets and filed beneficial-ownership registers
• MENA: UAE freezones, Saudi commercial register, Egyptian investment register
• India (MCA), Singapore (ACRA), Hong Kong (CR), Japan, and 20 more APAC jurisdictions

Headline numbers (entity chain fully resolved)

• Direct registries only: 48% — worse than the marketing decks claim.
• Commercial data vendors (top 3 combined): 71%.
• Mercurium disambiguation pipeline: 91%.
• Mercurium + human analyst loop on the remaining 9%: 98%.

The gap between commercial vendors and us comes entirely from how we handle the tricky parts — nominee chains, trusts with named settlors but unnamed beneficiaries, offshore SPVs with only partial beneficial-ownership registers. We don't stop at the registry. We go to the filed document, read it, and chain it forward.

What we're publishing

The test set schema (not the test data — for obvious reasons) and the evaluation harness are going open-source. We'd like to see every vendor in this space benchmarked against the same hard cases rather than the same two easy ones.