Canonical identifier

A canonical identifier is a preferred handle for an entity after identity decisions have been made. Its value should remain stable when display names or attributes change. The identifier's namespace and issuer matter: the same string in two namespaces may refer to different entities, while two identifiers can legitimately refer to the same entity under different schemes. The identifier labels a decision about identity; it does not prove it.

In practice, canonical identifiers link source records, claims and revisions to a persistent subject. External identifiers can be retained as cross-references with their issuing schemes and validity periods. If a false merge is later corrected, the mapping history should show which records moved and which claims were affected. Reusing an identifier for a different entity would make older references ambiguous.

For a defensible record, a stable identifier lets a reviewer follow the same subject across time while inspecting changes to the evidence. Entity resolution establishes the referent; the canonical identifier names it; a golden record presents its selected attributes. An identifier is therefore a continuity device, not an independent statement of factual correctness.

The hard problems arise when identity decisions change. If two entities were wrongly merged, one identifier must be split into two; if two identifiers were found to denote the same entity, they must be merged. In both cases the old identifier must continue to resolve, with a record of what happened, because other systems and documents refer to it. The usual approach keeps the old identifier as a redirect or a deprecated alias that points to the successor or successors, with a date and a reason, and never reassigns an identifier to a different entity. The article on identifier versioning after corrections examines these cases for graphs in detail.

Treat an identifier as a claim of identity, not as the entity. A record that cites an identifier on a date is making a statement that depends on the identity system's state on that date, which links the topic to point-in-time reproducibility: to reproduce an earlier result one needs to know which entities the identifier denoted then.

Ask whether any identifier has ever been reused for a different entity; the answer should be never. Ask what happens to a citation of a merged identifier; it should resolve with a visible redirect. Ask whether any identifier embeds a name or address; it should not. Ask whether external identifiers carry their issuing authority and the date of confirmation. Ask whether the identity system itself is versioned, so that the mapping on a past date can be recovered.