State of ERC-8004 · Report 14: The x402 Seller Cohort That Persists, Joined to On-Chain Identity

An independent audit (Bitquery, fetched 2026-09-09) reports that the median Base x402 seller earned about $2.50 across August and that fewer than 6 in 100 July sellers survived to August, with a recurring seller base of roughly 1,300. That is the first outside corroboration of what our own snapshots already show: the per-call x402 wrapper is not a persistent business for most of the sellers in it. The question outside measurements cannot answer — and the one this piece adds — is who the sellers that DO persist are, at the address level, and how many of them carry on-chain identity or commerce-backing.

Published 9 September 2026 · Every count below is a SAMPLE of the CDP Bazaar catalog (one Base-weighted x402 discovery source), not a census of x402 activity. The persistence cut is a two-snapshot intersection between 2026-08-15T23:13:50Z and 2026-09-09T02:37:30Z (~25 days apart, 30-day rolling windows overlapping by ~5 days — a persistence PROXY, not a strict calendar-month cohort). The identity join reads the erc8004_agent_profiles.sqlite owner table as of the same 2026-09-09 build; the rating cut reads reputation_agent_summary in audit_trail.db as of the same date. Bitquery figures are cited as an independent adjacent measurement, not adjudicated by this piece. CC BY 4.0 — license.

The Bitquery numbers land where our earlier reports already were, from a different data source. Their August measurement of the Base x402 seller population reports a $2.50 median seller earning across roughly 54,383 measured sellers, fewer than 6 in 100 July sellers surviving to August, and a recurring seller base of about 1,300. Treat that as adjacent measurement, not competitor framing — they are indexing wallet-level payment flow; we are indexing endpoint catalog + on-chain identity. Neither of us alone answers the useful question, which is: of the sellers who actually persist, how many walk back to a registered identity, and how many of those identities are backed by anything smarter than a plain EOA?

What we add.

Our dataset is a join between the CDP Bazaar catalog (endpoints marked quality.l30DaysUniquePayers ≥ 1, i.e. paid by at least one distinct wallet in the trailing 30 days) and the on-chain ERC-8004 identity registry on Base and Ethereum. That gives us three things Bitquery does not index: (i) which endpoints appear in a paid state at more than one point in time, (ii) which of their pay_to addresses walk back to a registry-owned agent identity, and (iii) which of those identities are commerce-backed via a smart-account payTo rather than a bare EOA. The persistence layer here is a two-snapshot intersection, not a calendar-month cohort; the boundary is stated in every section that uses it.

Method + denominators.

Two Bazaar snapshots are compared. Early snapshot: 2026-08-15T23:13:50Z. Late snapshot: 2026-09-09T02:37:30Z. Both are ~24–25 days apart on rolling-30d denominators, so the windows OVERLAP by roughly 5 days — every persistence figure here is a two-snapshot intersection at that window boundary, not a strict July-vs-August calendar cut. An endpoint is “earning” if the CDP-published quality.l30DaysUniquePayers field is ≥ 1 at that snapshot moment; that predicate does not distinguish a legitimate sole customer from an operator funding wallet, which is disclosed here and elsewhere on the site. A publisher is a distinct lowercase pay_to address occurring in the earning set. The identity join uses the owner field of the ERC-8004 agent profile store (Base + Ethereum). Every ratio below carries its own explicit denominator.

The persistence shape.

Endpoint-level persistence over the ~25-day span. Earlier earning set (2026-08-15 snap, ~30d window Jul 16–Aug 15): 14,979 endpoints with payers ≥ 1. Later earning set (2026-09-09 snap, ~30d window Aug 10–Sep 9): 14,494 endpoints. Endpoints present in both (persistent): 7,423 (49.56% of the earlier set). About half the endpoints that were earning by the CDP-published payer field ~25 days ago are still earning by the same field today. That is a broader shape than a strict month-boundary cohort would give — the ~5-day overlap between the two rolling windows makes 49.56% a soft upper bound on what a strict calendar cut would yield.

Publisher-level persistence is a tighter cohort. Earlier earning publishers: 1,404 distinct pay_to addresses. Later earning publishers: 1,453. In both: 935 (66.60% of the earlier set). Two thirds of publishers that were earning ~25 days ago are still earning today — a much healthier persistence rate than the seller-level survival figure Bitquery reports. That gap between what we and Bitquery see is exactly the overlap caveat above: neither number is wrong; they are different denominators on a different window.

The identity join over the persistent set.

Concretely, joining the persistent-publisher cohort to on-chain identity: of the 935 publishers earning in both the 2026-08-15 and 2026-09-09 Bazaar snapshots, 22 (2.35%) resolve to an ERC-8004 agent owner, with the top persistent publisher accounting for 5.02% of persistent-endpoint pay_to occurrences (top 10 at 33.61%), 4 rated (any on-chain feedback in reputation_agent_summary), and 7 commerce-backed (aa_type set to a real smart-account label, not a bare EOA). Endpoint-level the same cohort reads 856 of 7,423 persistent endpoints (11.53%) owned by an ERC-8004 identity; the endpoint number is bigger than the 2.35% publisher number because a small number of identity-carrying publishers own many endpoints. That is the co-location point: 22 pay_tos next to 856 endpoints is the shape you would miss if the 11.53% or the 856 was quoted alone.

Publisher concentration inside the persistent set.

The persistent-publisher cohort inherits the same top-heavy concentration our other reports have measured for the earning set. The top persistent publisher (0x50ab2018c06c6e4eaa9ba52057eb55ed284912fc) accounts for 607 of 12,094 persistent-endpoint pay_to occurrences (5.02%). The top 10 persistent publishers together account for 4,065 occurrences (33.61%). This is why the publisher-level cut matters: a small handful of persistent publishers own a third of the persistent-endpoint occurrences, and identity coverage clusters with them — the top persistent publisher is identity-carrying, which is what drags the endpoint-level identity share (11.53%) so far above the publisher-level share (2.35%).

What this does NOT show.

The two Bazaar snapshots are ~25 days apart on rolling-30d denominators; they overlap by ~5 days. This is not a July-vs-August calendar cut, and no figure here can be re-quoted as a permanent month-to-month number. Compare to Bitquery’s stricter month cohort with that boundary in mind.
“Earning” here is payers ≥ 1 in a 30-day rolling window — a legitimate sole customer and an operator funding wallet look identical from on-chain data alone. That caveat cuts against every persistence and identity figure on this page.
SAMPLE (CDP Bazaar catalog rows), not POPULATION. The CDP Bazaar is one Base-weighted x402 discovery source; endpoints not indexed there are invisible to us, so every share here is conditional on inclusion.
The identity join uses the owner address of an ERC-8004 agent as the walk-back key. A commercial operator that publishes an x402 endpoint from a different treasury than their registered agent owner will read as “no identity” here even if a manual look would find the connection.
“Rated” means feedback_count > 0 in reputation_agent_summary — a single unweighted feedback event. “Commerce-backed” means the ERC-8004 owner has an agent with a smart-account aa_type, which is a payment-affordance proxy, not an actual settled-with-a-real-buyer proof.
A 402 is a payment challenge, never a buyer.
External Bitquery figures ($2.50 median, ~54,383 sellers, <6/100 survival, ~1,300 recurring) are cited as adjacent measurement, not adjudicated by this piece. Bitquery indexes wallet-level payment flow; we index endpoint catalog + on-chain identity — different lenses on a shared shape.

So what.

Bitquery’s finding and ours describe the same market from opposite sides. Their measurement kills the “huge active seller base” framing at the wallet-flow layer. Ours adds the identity layer on the survivors: about two-thirds of publishers that were earning ~25 days ago are still earning; of those, only 22 walk back to a registered identity, only 7 are commerce-backed, and only 4 have any on-chain rating. That is the shape a buyer is actually facing when they try to trust an x402 endpoint on Base today — a persistent seller cohort measured in the hundreds, not the tens of thousands, and a persistent identity-carrying subset small enough to itemise.

The x402 Seller Cohort That Persists, Joined to On-Chain Identity.