ERC-8004 Reputation Census: Ethereum vs Base

Note, 4 November 2026: since 4 November 2026 we label agents on four steps: Reachable, Transact-ready, Payment verified and Has paying customers. This post describes the earlier labels as they stood on its date. See how the steps work.

Five headline numbers

1. 570,562 NewFeedback events indexed on the ERC-8004 ReputationRegistry (0x8004BAa17C55a88189AE136b182e5fdA19dE9b63) — 549,229 on Base, 21,333 on Ethereum mainnet. Base is almost 26× the event volume of Ethereum mainnet, from 23,327 indexed Base agents vs. 2,946 indexed Ethereum agents.
2. 93.7% of Ethereum feedback is submitted by 18 wallets that each reviewed 100 or more agents. We call these 'spray' wallets: they rate agents in bulk rather than reviewing individual work. On Base, those same 18 wallets drive 19.2% of feedback. Removing the 18 wallets leaves 13,489 distinct Base reviewers and 683 distinct Ethereum reviewers — and 1,335 non-spray feedback events on Ethereum, vs. 443,741 on Base.
3. One Base agent carries 71.2% of all Base feedback. BOTCOIN (agent 25975) has 390,890 feedback events — from 78 unique client wallets. High-confidence Sybil-adjusted score 48, Promising tier. The next-ranked Base agent (55985) has 13,174 events from 9,751 distinct reviewers — a very different shape, same raw-count neighbourhood. Raw feedback_count is a terrible first filter.
4. Ethereum's top 10 by raw feedback count are all members of one batch-minted collection. Agents 50903, 50906, 50939, 50952, 50955, 50957, 50959, 50969, 50971, 50974 are IdentityMD #0, #2, #88, #354, #420, #1120, #1548, … from a single owner (0xde152afb…dd336) who minted 1,566 agents in that collection. Those 1,566 agents together hold 51.0% of all Ethereum feedback — each with a single reviewer. This is exactly the pattern that makes raw feedback counts unreliable.
5. Only 62 agents clear Transact-ready across both chains (282 advertise x402 at all) after the 2026-10-10 per-agent payment attribution recompute (payment evidence is per agent, not per operator wallet), on top of the 2026-10-09 Rules C/D/F guardrails — 58 Base + 4 Ethereum out of 26,978 indexed. Previously reported 2026-10-09: 1,046 (918 Base + 128 Ethereum). The reduction is almost entirely the attribution gate: the earlier badge was earned on inherited owner-wallet treasury lines that every sibling agent shared, so a single real receipt could float dozens of subaccounts into Transact-ready. The remaining 5,939 are Promising, 9,020 are Not ready, and 5 are Unrated. If you're shopping for an ERC-8004 counterparty by registration count, you're shopping across a far wider surface than the actually-ready one.

Method

What counts as a feedback event: a non-revoked NewFeedback(uint256 agentId, address client, int128 value, …) event emitted by the ERC-8004 ReputationRegistry contract. The registry is deployed at the same deterministic address (0x8004BAa17C55a88189AE136b182e5fdA19dE9b63) on Base and Ethereum mainnet. We read every NewFeedback event on both chains through each chain's node API.

Evidence requirement. Since 2026-10-09 an agent cannot reach Transact-ready without at least one real-usage signal: non-self feedback (at least one event where the client wallet is not the agent owner's wallet), non-trivial inflow on-chain, or a verified ACP job outcome. Self-reviews are close to absent in the raw data (1 of 549,229 on Base, 0 of 21,333 on Ethereum), so this requirement works mainly through the activity and job-outcome conditions. See the readiness scoring tightening write-up for the batch-publisher cap and name gate, which operate on top of this requirement.

Spray bucket. A pre-registered threshold from the four-chain census: a client wallet that has reviewed 100 or more agents. At snapshot, 18 wallets meet this threshold. The tiers either side — 50–99 agents reviewed (7 wallets), 20–49 agents (71 wallets), 10–19 agents (383), 5–9 agents (330), 2–4 agents (517), one agent only (12,816 wallets) — are published for completeness but are not re-defined as "spray".

Block ranges. Base feedback indexed through block 52,396,050 (2026-10-09 21:30 UTC). Ethereum mainnet feedback indexed through block 26,142,739 (2026-10-07 19:55 UTC). Ethereum is refreshed less often than Base, and at this snapshot is about two days behind. Keep that gap in mind when comparing the 7-day and 30-day rates across chains.

Per-chain table

Update 2026-10-10: the per-agent payment attribution gate shipped later the same day replaced the 1,046 figure above with the live counts in the "Transact-ready, live" row. Payments are now credited to the agent that declares its own payTo, not inherited from the owner wallet, and the demotion is persisted on the profile. The previously-reported 918 Base + 128 Ethereum numbers were accurate as of the 2026-10-09 Rules C/D/F snapshot but were earning the badge on shared owner-treasury lines; see readiness scoring tightening → Update 2026-10-10 (payment evidence is per agent) for the before/after table (Base 1,508 → 4, Ethereum 188 → 0), the four surviving Base agents by name, and the reproducibility pointer. The remaining counts (Promising / Not ready / Unrated) are live via /v1/public/stats and refresh at build time.

Transact-ready, Promising, and Not ready counts cover only agent profiles that already carried a readiness block before the 2026-10-10 recompute (never-scored profiles are picked up by the daily enrichment pass). Commerce-backed is the strict subset of feedback events that tie back to a real paid Virtuals ACP job URI or an allowlisted ERC-8183 hook contract, per our published commerce-backed definition. The gap between "has feedback" (29,872 / 2,366) and "commerce-backed" (51 / 0) is the single most important filter in this dataset.

Top-10 by raw feedback count — why the number alone is useless

Reading the two columns together: on Base, raw-count rank 1 and rank 5 both exceed 1,000 feedback events but have wildly different reviewer counts (78 vs. 1,514). The rank-3, rank-4, rank-7 cluster (AgentFEC / AgentEcon / AgentMarket) is a single-publisher trio under one owner, driving thousands of events each from 1–3 distinct reviewer wallets. On Ethereum, every row above is a different #<int> member of the IdentityMD collection, each with a single reviewer — the shape the 2026-10-09 name gate catches by pattern, not by scoring.

Top-10 named agents by distinct unique clients — the Sybil-aware ranking

Switching the ordering to unique clients, the Sybil-aware signal our adjusted reputation score is built on, produces a very different top 10 on each chain. Reading the two tables side by side is the point: the leaders by raw count and the leaders by distinct reviewers are almost entirely different agents.

Six of ten Base agents in this view clear Transact-ready; five of ten Ethereum agents do. The Ethereum column excludes members of the IdentityMD batch collection, since including them would re-fill the ranking with 1-unique-client rows. The named-agent filter is just "has a non-empty display name in our profile store" — so agents without a display name in our profile store are left out of this view, including Base agents 55985 and 1380, which rank highly by raw count.

What this means for an agent choosing who to pay

Rank by distinct reviewers, not by count. One wallet can emit arbitrarily many NewFeedback events against a single agent — the registry does not rate-limit, and until the 2026-10-09 readiness tightening a batch of 1,566 agents with a single reviewer each could still clear Transact-ready. The unique_clients column is the signal that cannot be inflated by one busy wallet.

Check publisher density. On Base, the single largest cluster in the raw-count top-10 is a three-agent trio (AgentFEC / AgentEcon / AgentMarket) under one owner. On Ethereum, it's a 1,566-agent batch collection under one owner. The batch-publisher cap drops any publisher at or above 100 registrations into Promising at best; a reader doing counterparty research by hand should run the same check against the owner's registration count on the chain explorer.

Commerce-backed is a strict subset. Of 32,238 agents with at least one feedback event, only 51 (all on Base, zero on Ethereum mainnet at this snapshot) carry a feedback row that ties back to a real paid Virtuals ACP job or an allowlisted ERC-8183 escrow hook. The full cohort is listed at /commerce-backed-agents. It counts only feedback tied to a paid job, so it is much smaller than the set of agents with any feedback.

If you're going to the paid API, start cheap. The $0.01 agent-profile endpoint returns a Sybil-adjusted score, a readiness bucket, a confidence level, and a commerce-backed flag per request. The free MCP server exposes six read-only tools so an agent can check coverage before paying. Both are a strict superset of this public census.

Reproducing this census

Every number in the headline list and the per-chain table above is a row in the aggregate CSV at /data/erc8004-reputation-census-2026-10.csv (70 data rows, 8 columns: section, chain, rank, agent_id, display_name, value, unit, note). The CSV publishes aggregate counts and public agent IDs only: no client wallet addresses and no per-event rows. The indexing and scoring code is not published; every number above is reproducible from the CSV and from public ReputationRegistry events. Snapshot generated at 2026-10-09 22:04 UTC. CC BY 4.0; cite as "Onchain Agent Intel, 2026-10-09".