The agent-payments category has a scoring problem. Look at registration counts and it looks large — the ERC-8004 identity registries hold 25,618 agents across Base and Ethereum mainnet as of this page load, and Coinbase's public x402 Bazaar catalog exposes tens of thousands of paid endpoints. Look at settled economic activity and it looks tiny. Both are measurable. Both have been measured this month — independently, with different methodologies, on different sides of the market. The shape is the same either way.
This post reconciles the two whole-catalog measurements. The job is not to argue the finding — we have been publishing our own version of it since Report 01 — but to state both predicate definitions explicitly, note the caveats each team's dataset carries, and close the loop with the third independent source (an Imperial College academic paper) that finds the same conclusion from the identity-registry side.
Supply side (ours): a small single-digit percent of the CDP Bazaar has real paying demand
Our supply-side measurement takes the entire Coinbase CDP x402
Bazaar discovery catalog and cuts it against CDP's own
quality.l30DaysUniquePayers
field. The finding, as of the most recent daily snapshot on
Report 07,
is that a small single-digit percent of the whole indexed
catalog crosses the strong-filter cohort at
≥5 unique payers in 30 days. That report
renders the exact live number and the full exclusive-bucket
payer histogram on each build, so the citable claim never goes
stale — read
Report 07 for
the current snapshot's headline and the
market-map dataset for the
underlying cohort ladder.
The predicate is important. "Listed in the CDP Bazaar" is a supply-side signal; "≥5 unique payers in 30 days" is a demand-side threshold. The catalog itself is the denominator; the tiny high-payer cohort is the numerator. The finding is that listing is not demand — the two are almost decoupled at the catalog scale.
Demand side (x402-list.com): the top ten services take almost all of the volume
On 2026-08-15 we re-verified an independent operator's
whole-catalog measurement from the demand side.
x402-list.com operates a
public /api/v1/stats endpoint
exposing settled volume, settlement count, unique buyers, and
concentration ratios across the services it measures. Every value
below is a frozen citation — quoted verbatim from
data.traction in the JSON body,
computed_at=2026-08-15T19:43:47Z,
as_of=2026-08-15T18:09:23Z — and
pinned by test so a later edit here cannot drift the numbers
silently.
| Metric (x402-list.com, trailing 30 days) | Value, as measured 2026-08-15 |
|---|---|
| USDC settled (30d) | $206,753 |
| Settlement count (30d) | 5,473,120 |
| Unique buyers (30d) | 5,745 |
| Measured services | 423 |
| Top-10 services share of settled volume | 97.9% |
| Largest single service share of settled volume | 83.3% |
| Top buyer inside the largest service | 98.7% |
| Coverage of measured flow (their own field) | 17.5% |
| Largest service's volume imported (not natively observed) | true |
Source: https://x402-list.com/api/v1/stats
as measured 2026-08-15. Frozen citation — do not extrapolate the
exact percentages to today.
The caveats — both teams', stated up front
The convergence between our number and theirs is the whole story of this post, so both teams' honest caveats have to sit beside the number, not below it.
x402-list.com measures 423 services, not the whole x402
universe. Their own
coverage_of_measured_flow_pct
field self-reports as 17.5%. The
"top-10 services = 97.9% of volume" claim is 97.9% of the
volume they measured across their 423 services, not
97.9% of all x402 volume in the wild. The correct reading is
"inside a large-enough demand-side sample, ten services
dominate."
Their top-1 service's history is imported, not
natively observed.
top_1_service_imported: true in
the same JSON body means x402-list attributes the largest
service's 30-day volume via an imported signal rather than a
directly-measured x402 handshake. The 83.3%
top-1 share is therefore a directional upper bound on that one
service's dominance, not a raw observation.
Our predicate and theirs are different measures.
Our whole-catalog scan filters on
≥5 unique payers in the last 30 days per endpoint,
exposed through CDP's
quality.l30DaysUniquePayers
field. Theirs filters on
share of measured settled USDC volume at the
service level. They are not the same predicate; a service that
crosses our "5-payer" bar can still be small in absolute
settled volume, and a service that captures a large share of
volume can be dominated by one payer (which theirs also
reports for the largest service — 98.7% of its 30-day volume
from a single buyer). The two predicates converge in shape —
both report that a tiny slice of the catalog owns essentially
all of the paid activity — but the exact numbers are not
directly comparable.
Both predicates, side by side
Any comparison of two whole-catalog numbers has to begin with a table of what each team actually measured.
| Predicate | Definition |
|---|---|
| Ours (supply side) On-Chain Agent Intel, daily snapshot |
An endpoint in the Coinbase CDP x402 Bazaar catalog
qualifies for the strong-filter cohort iff CDP's own
quality.l30DaysUniquePayers
field is ≥5 at snapshot time. Denominator
is the whole catalog as returned by
api.cdp.coinbase.com/platform/v2/x402/discovery/resources
(a paginated whole-catalog fetch). No probing, no
per-endpoint call; the finding is a straight cut of CDP's
own data. Fresh cut published daily on
Report 07
and on the
market-map dataset.
|
| Theirs (demand side) x402-list.com, frozen 2026-08-15 |
A service qualifies for the "top-N" ranking by its share of
settled USDC volume observed by x402-list.com over the
trailing 30 days, across the 423 services it has
instrumented. Denominator is x402-list.com's measured flow
(self-reported at 17.5% of the wider x402 universe), not
the whole x402 economy. Values exposed at
https://x402-list.com/api/v1/stats
under data.traction.
|
Both predicates try to filter out demand-free listings; both report the same shape.
Update — August 2026: what changed since publication
Three post-publication data points sharpen the finding rather than soften it. All three were re-verified against their public primary sources on 2026-08-16 before this section shipped — a note is called out below where our task-brief framing did not fully reproduce.
Mainstream financial press now cites a ~93% year-to-date decline in x402 settlement volume. A 2026-08-13 Yahoo Finance markets piece ("x402 Settlement Volume Plunges 93% YTD, but Cloudflare Could Revive AI Agent Payments") reports the headline figure and frames Cloudflare's Monetization Gateway (GA 2026-07-01) as the most credible near-term revival lever — "Cloudflare could change that equation by placing x402 payments in front of a large base of websites and developers." The underlying month-30 activity was reported four weeks earlier by CoinDesk (2026-07-15): "about 75 million transactions over the past 30 days" moving "about $24 million" in aggregate.
Read those two numbers together and the analytic point is cleaner than any single percentage. Payment count stayed high while payment value collapsed. On a rough divide that puts the average x402 payment on the order of a third of a US cent — roughly $0.32 per settlement — which is what you would expect from bot pings, retries, and automated probes, not from a paying human buyer or an AI agent transacting for real work. In a market whose average payment is trivially small, the scarce good is not the rail; the scarce good is a reason to trust the counterparty. That sharpens the trust-filter thesis this post opened with rather than merely repeating it.
We are ETH-and-Base-aligned; the headline volume figures span the wider x402 economy and much of it is Solana-hosted. The reader should read the 93% number as market-wide, not as a comment on the ERC-8004 identity-registry cohort our own supply-side scan measures.
Chainalysis has now published its own x402 measurement (Inside x402: 100M Agentic Payments on Base, 2026-06-03) — a fourth independent read after ours, x402-list.com's, and the Imperial College paper. Honest scoping note: Chainalysis is a crypto-native blockchain-analytics firm, not a public regulator or a general-industry outlet, so count it as an additional industry-analytics corroboration rather than a cross-industry one. What Chainalysis measures is value-size concentration rather than service-count concentration — a different axis than our three earlier sources, but the same underlying story. Verbatim: "Take transactions of $1+, which represented 49% of volume in early 2025; by early 2026 they had 95% share. Meanwhile, transactions of between 10¢ and $1 collapsed from 46% to just 4% over the same time period." The tiny-value tail of the payment distribution has evaporated on Base specifically.
Fireblocks joined the x402 Foundation and shipped an Agentic Payments Suite. Per the primary source (Fireblocks press release, 2026-05-20) they joined alongside Google, Microsoft, Visa, and Coinbase in the foundation membership; their specific contribution is the security-and-governance extension — "delivering the security extension that adds request integrity and spend governance to the protocol." Sit that beside Cloudflare's edge Monetization Gateway (GA 2026-07-01) and the shape of the industry response is clear: the rail is being commoditised by the largest possible vendors, which leaves the trust layer — the "should I let this counterparty spend my USDC" question — as the still-unbuilt part.
Honest correction against our task brief: the Fireblocks announcement went out on 2026-05-20, not July 2026 as some downstream aggregators and our own drafting notes had it. Verified against the PR Newswire primary source before this update shipped.
Why the convergence matters
Two operators, two different methodologies, two different sides of the market, three months apart — and the finding is the same: an x402 directory at the catalog scale is almost entirely long-tail listings with no paying demand behind them. The paying demand — such as it is — lives in a handful of services at the head, and inside the largest of those a single buyer dominates.
That shape has a direct consequence for how a buyer should shop
the category. In a market this concentrated, a
bigger index is a vanity metric. A
trust filter on top of the index is the actual product
— the small cohort of services that a real buyer has paid, more
than once, over a real window. That is exactly what our
commerce-backed cohort
surfaces on the identity-registry side. Today the cohort stands
at
44
ERC-8004 agents whose
ReputationRegistry rows tie back
to a real, paid on-chain job — Report 03
walks the predicate in long form and
Report 05
sits it alongside the raw registration and live-endpoint counts.
The third independent source: Imperial College's ERC-8004 study
The x402 side of this is the payments/discovery layer. The
identity-registry side has been measured independently too. An
Imperial College London study
(arXiv:2606.26028,
Xihan Xiong, Zelin Li, Wei Wei, Qin Wang, William Knottenbelt,
Zhipeng Wang; CC BY 4.0; data frozen 2026-05-13) reports that
only 3% (Ethereum), 4% (BSC),
and 15% (Base) of ERC-8004 identity
registrations satisfy their trust bar — a resolvable
registration file and at least one declared live
service endpoint. In their own words, "most registrations
are placeholders rather than active agents," and the
ReputationRegistry
"cannot function as a trust signal." Our own reconciliation
against that paper, with both predicates side by side, is at
Can Trustless Agents Be Trusted? Reconciling Imperial College's Frozen May Snapshot Against Our Live August ERC-8004 Data.
Three independent measurements now agree that the ERC-8004 / x402 registries are mostly noise. Their headline percentages are not directly comparable — different predicates, different denominators, different chains — but the direction is the same on all three sources and the practical implication is the same on all three: whatever the true "real agents" percentage is, it is nowhere near the raw registration count. Counting registrations measures nothing; a trust filter is the product.
What we're doing about it
We publish the trust-filter continuously.
/commerce-backed-agents is
the free, indexable hub of every ERC-8004 agent whose
ReputationRegistry row ties back
to a real, paid on-chain job (ERC-8183 / Virtuals ACP outcome or
an allowlisted hook contract). The canonical predicate lives at
smartcontractauditpro/commerce_backed.py,
with a divergence-regression test keeping the per-agent
classifier and the SQL predicate in agreement. Underlying the
cohort is a live continuous index —
25,618
ERC-8004 agents on Base + Ethereum mainnet, of which
3,323
answer at their advertised endpoints (Base
7%,
Ethereum mainnet
58%)
and
277,710
ReputationRegistry feedback
events indexed. Liveness is one code path
(smartcontractauditpro/liveness.py::is_live)
continuously tracked since
2026-07-17.
On the x402 supply side, we publish the whole-catalog cut daily. The /x402-bazaar-market-map dataset — with matching JSON and CSV exports under CC BY 4.0 — carries the cohort ladder (≥1 / ≥5 / ≥10 / ≥100 payer thresholds), the exclusive-bucket payer histogram, and the add/remove/price-change delta panel against the previous snapshot. Report 07 is the long-form treatment; the dataset is the daily companion.
Attribution and reuse
The x402-list.com figures on this page are frozen citations,
quoted verbatim from that operator's public
/api/v1/stats endpoint as
measured 2026-08-15. Reuse governed by
x402-list.com's own terms.
The Imperial College paper figures are quoted under CC BY 4.0. Full attribution: Xihan Xiong, Zelin Li, Wei Wei, Qin Wang, William Knottenbelt, Zhipeng Wang (Imperial College London), Can Trustless Agents Be Trusted? An Empirical Study of the ERC-8004 Decentralized AI Agent Ecosystem, arXiv:2606.26028 (v1 2026-06-24, v2 2026-07-08).
The Chainalysis, CoinDesk, Yahoo Finance, and Fireblocks quotations in the August-2026 update above are short excerpts included under fair-use citation practice; each is dated, sourced with a direct link to the primary URL, and re-verified on 2026-08-16. Rights in those excerpts remain with the original publishers.
Our own supply-side and identity-side numbers are served live
from /v1/public/stats,
/x402-bazaar-market-map,
and Report 07
and refresh daily. Redistribution of our figures is welcome
under CC BY 4.0 with a link back to this page.