Entity Attribution: From Address to Identity
An address is not an entity. Attribution is a scored mapping, published with the evidence that produced the score.
Context
Explorers label addresses as a display convenience. Research and risk need a mapping that can be wrong, and a way to say how wrong.
Method
Seeds come from protocol deployments, attested registries and clustered funding. Each edge carries a method and a confidence. The published table is the current best label per address, not the full graph.
Findings
High-confidence labels are sparse. Most addresses remain unlabeled, which is the correct default. Forcing a name produces the kind of silent error the rest of the pipeline is built to avoid.
Implications for the data model
entity.attributions.confidence is required. Downstream models that need a name must filter, not coalesce to “unknown entity” as if it were a finding.