~36% of ERC-8004 agents share the same name + description + URI as another agent
We index live ERC-8004 agent registrations and looked at metadata uniqueness (not reputation, not payments — just identity fields).
Snapshot (~408k agents):
Clones (strictest test we ran)
- ~36% of agents share an identical (name, description, URI) triple with at least one other agent
- One single template accounts for ~114.5k agents — roughly 28% of the entire index
Descriptions collapse harder than names
- ~86% of agents sit in a shared description-hash cluster
- ~22% share a blank description
- Names are nearly universal (~99.6% filled), but distinct identities are not
Chain contrast (short)
- BNB: ~50% triple-cloned (drives most of the clone mass)
- Ethereum L1: lower triple-clone rate (~14%), but ~78% name-dup / ~90% desc-dup, and ~61% flagged as test-realness in our validators; only ~29% have a description filled
- Celo: much cleaner on this cut (~2% triple-dup) and higher richness/realness, though smaller stock
Quality layer
- Metadata richness avg ~35; only ~2.8% score 60+
- Realness: ~69% valid, ~15% test, ~8% dummy
Takeaway ERC-8004 identity looks populated at the field level. Uniqueness and information quality look very different once you inspect hashes and quality scores. Agent counts ≠ distinct metadata identities.
Caveat: shared metadata ≠ proven Sybil. Templates, factories, multichain mirrors, and blank fields all produce the same hash pattern. The point is that “registered agents” is a weak proxy for distinct, informative identities.
Curious how people building discovery / ranking layers are handling this — filter clones, weight uniqueness, or treat it as noise for now?