How Keizu determines a UBO

Walking the graph, not just reading the top of it

Keizu resolves ownership by walking the shareholding graph beneath a company and multiplying effective percentage down every verified path. Someone holding 30% of a company directly, and separately 55% of a holding company that itself owns 70%, controls 68.5% overall - computed as one merged figure, not reported as two separate stakes. That merge happens at the person level: every path that reaches the same natural person gets summed before the threshold test is applied, not evaluated path by path.

Two kinds of edge, and only one of them is exact

Not every ownership signal a register publishes carries a percentage. Keizu's resolution engine treats two edge types differently:

• Filed percentage. A register has published an exact figure for a holding - e.g. "42% of ordinary shares." This is the only edge type that contributes a real, summable number toward a threshold test.
• Consolidation / sole-shareholder. A real ownership or control signal exists - for example a GLEIF accounting-parent relationship, or a register stating a sole shareholder without a percentage - but no exact figure is attached. Once a chain passes through one of these edges, everything beneath it, including the edge itself, is reported only as an upper-bound ceiling. It is never merged with an exact figure, because doing so would overstate precision the source data doesn't actually have.

This distinction matters in practice: a report that says "confirmed 42% owner" and one that says "up to 100%, unconfirmed beyond a consolidation link" are answering the same question with very different confidence, and Keizu never collapses that difference into a single misleadingly precise number.

The threshold: AMLR's 25%, not a fixed constant

The EU's AMLR (Regulation (EU) 2024/1624) sets the general beneficial-ownership test at 25% or more of shares, voting rights, or equivalent control - notably "or more," not "more than," so a holding of exactly 25% qualifies, which the older AMLD5 wording did not. A lower threshold, down to 15%, is available only where the European Commission identifies a specific sector as higher-risk by delegated act. As of today no such act sets 15% for gambling, payments, or crypto - so those sectors sit at the general 25% test like everyone else. Keizu treats 15% as an explicit manual override a firm can request for its own risk appetite, never as a silent default, because defaulting to it would assert a stricter legal position than currently exists in EU law.

Where a chain stops, and why that's reported honestly

A resolution chain terminates for exactly one reason, drawn from a closed, named set - never a silent empty result:

• Resolved to a natural person - the terminating success case.
• Fiduciary / nominee - the register names a fiduciary, not the person who actually controls the holding.
• Register closed or not published - the jurisdiction's disclosure regime doesn't make this data available at all.
• Below threshold - a real holding exists, but it doesn't clear the applicable percentage test.
• No natural owner - a genuinely ownerless entity, such as a foundation or trust with no natural beneficiary.
• Circular holding or depth exceeded - structural stops that prevent an infinite or unreasonably long walk.
• No data available or fetch failed - the data simply hasn't been retrieved yet.

These split into two kinds that matter for how you should treat the answer. Permanent stops are facts about a jurisdiction's disclosure regime - fiduciary/nominee, register closed, not published, below threshold, no natural owner, circular holding, depth exceeded - and are never silently retried, because retrying them wouldn't change the answer. Transient stops - no data available, fetch failed - are queued for a retry, because they're a gap in what's been fetched so far, not a fact about the register. A foundation confirmed to have no natural owner is a complete answer. A fiduciary/nominee stop or a closed register is a gap Keizu cannot close on its own, and it is disclosed as exactly that - never presented the same way as a clean resolution.