The State of ERC-8004 series has been walking the same argument down one level at a time. Report 01 introduced the registered-vs-live gap. Report 02 narrowed that gap to the readiness distribution — the small share buyers could actually integrate with. Report 03 narrowed it again to the commerce-backed cohort: the agents whose ReputationRegistry rows tie back to a real, paid on-chain job. Every step took the pile of numbers and made it smaller and more honest. This post takes the next natural step — applying the same lens to the chains themselves and asking which of them are actually generating that signal.
The answer, measured across our own index, is unambiguous. Ethereum mainnet and Base are where real ERC-8004 agents ship. BNB Chain, once you filter for anything but a raw registration count, is a placeholder farm dressed as an index. So we cut it.
The verdict from our own index
Our 2026-07-16 chain-quality audit — read the full doc at
docs/chain-quality-audit-2026-07.md
— measured every registered ERC-8004 agent on Base, Ethereum, and
BNB against the same live-endpoint predicate we use everywhere else
on the site (endpoint returns 2xx AND declares at least one
machine-readable surface: MCP, OpenAPI, or
/.well-known). The result of that
snapshot:
| Chain (2026-07-16 audit) | Registered | Live | Live rate |
|---|---|---|---|
| Ethereum mainnet | 1836 | 1268 | 69.1% |
| Base | 19164 | 2013 | 10.5% |
| BNB Chain | 73469 | 108 | 0.15% |
Snapshot from the 2026-07-16 chain-quality audit. Current-day figures for the two chains we kept are live-sourced in the next table.
BNB is where the "big index" story starts and where the honest read of that story ends. It carried nearly four times as many registered agents as Base and Ethereum combined, and roughly one in every 680 of them was actually reachable. Base's live rate was two orders of magnitude higher; Ethereum's was three.
The tail was worse than the aggregate. In the same audit, a single
BNB owner
(0xEBf9…C11, mint contract
PurrfectClawNFT) held over
eighteen percent of the entire BNB registration set — north of
thirteen thousand agents by itself, every single one of which
failed the live-endpoint check, had zero declared capability, and
had never received a USDC payment. The second-largest BNB owner
cluster (698 agents) showed the same profile: zero on every
quality signal. When we tightened the filter to bulk minters
(owners with 100+ agents each), BNB's bulk cohort of just over
fourteen thousand agents scored
zero on any endpoint capability, zero on 2xx responses,
and zero on USDC ≥ $10. Base's bulk cohort of nearly thirteen
thousand agents was around ten percent real by the same tests.
Not high, but not zero.
Cost of scanning a chain that's 99% placeholders
The chain-quality audit also priced the decision honestly. At the time of the audit, BNB was consuming roughly ninety percent of our Alchemy CU spend — most of it activity-tracker scans across wallets that would never move a dollar — for approximately one live agent per two hundred registered. Base was running about 440 times more efficient per live agent; Ethereum about 410 times more efficient. Scoping the BNB scan to quality-signal wallets (task 0155) cut the bill by around 71%, but the underlying trade did not get better. BNB continued minting roughly 1500 new placeholder registrations per day, with occasional bursts of over 9000 in a single day, and none of that new supply ever produced a commerce-backed feedback row.
On the flip side, keeping BNB indexed required carrying two
registries in every discovery cycle (the cross-chain
0x8004A169… standard registry and
the BNB-team-forked BRC8004 registry
0xfA09B33…, which held twenty-five
total agents at the time of the audit), maintaining a chain-specific
health check and payment-monitor branch, and threading the "BNB
Chain" narrative through every piece of copy on the site. Small
individually, real in aggregate.
Commerce-backed reputation is the real trust signal — and BNB never carried any
The single measurement that made the call unarguable is the one
Report 03
was built around. Reputation on ERC-8004 is permissionless — anyone
with a handful of wallets can manufacture endorsements at zero
cost. The 2606.26028 arXiv audit found that most feedback across
every chain evaporates once you apply a Sybil-adjusted uniqueness
filter. What does not evaporate is
commerce-backed reputation: a feedback row whose author
and subject demonstrably transacted on-chain via an ERC-8183
AgentCommerce job outcome or the
equivalent hosted Virtuals ACP job pattern, atomically settling a
USDC payment. A Sybil farm can't produce one without first paying
the agent — at which point it's a paying customer.
Applying that filter to our own index today:
| Measure (live, all-Ethereum index) | Value |
|---|---|
| ERC-8004 agents indexed (Base + Ethereum) | 25,618 |
| Live agents (endpoint 2xx AND capability declared) | 3,323 |
| Live rate — Base | 7% |
| Live rate — Ethereum mainnet | 58% |
| ReputationRegistry feedback events indexed | 277,710 |
| Agents that have received any feedback at all | 29,128 |
| Commerce-backed agents (got-paid cohort) | 44 |
| Job-outcome feedback rows (T1 / T3 evidence) | 979 |
Live-source: /v1/public/stats.
Liveness itself has been continuously tracked since
2026-07-17.
The commerce-backed cohort we keep coming back to lives on Base — see the /commerce-backed-agents hub for the ranked list. On the 2026-07-27 snapshot referenced in the Ethereum-alignment brief (docs/bnb-vs-arbitrum-alignment-2026-07.md), the commerce-backed count on Base was 44. On BNB it was zero, on every historical snapshot ever taken. Zero the day we set up the indexer; zero at Report 01; zero when we shipped the commerce-backed cohort hub for Report 03; zero when we ran the decision brief the week before this post. Six months of scanning, zero commerce-backed feedback rows produced. The signal simply wasn't there.
Framed the other way: on Base, our 979 commerce-backed feedback rows sit next to 277,710 indexed ReputationRegistry events. Most of that larger number is low-signal batch-bot writes — the exact failure mode the commerce-backed predicate was designed to filter around. The remaining, narrow, honest cohort is what we sell. Dropping BNB removed exactly zero rows from that cohort.
What "all-Ethereum" means going forward
We now index Base and Ethereum mainnet. Both are Ethereum-aligned
in the strict sense: Ethereum is the L1, Base is a canonical
Ethereum L2, and both share the same
0x8004… cross-chain registry
deployments the ERC-8004 spec uses. The narrative is clean, the
pipeline is one fewer chain-specific branch, and the marketing
headline of "the ERC-8004 index" now reads as the
Ethereum-ecosystem coverage number instead of a placeholder count
inflated by a chain we didn't respect.
The next chain we plan to add is Arbitrum One. The 2026-07-16
audit confirmed via a bounded public-RPC probe that the standard
ERC-8004 IdentityRegistry
(0x8004A169…) is deployed on
Arbitrum, along with the ReputationRegistry
(0x8004BAa1…). Arbitrum's
developer culture is Ethereum-native — DeFi, perp, and vault
teams that publish agent-style automation — and early probes
suggest a small, high-quality cohort in the low hundreds. That
composition is the exact profile our paid surface is designed
around. Adding it is a follow-on task; nothing on Arbitrum is
indexed yet and the ecosystem-alignment brief
(docs/bnb-vs-arbitrum-alignment-2026-07.md)
walks the sequencing. Base + Ethereum + Arbitrum is the target
composition; today, the two chains live on the site are the two
that were already generating paid signal.
What a buyer actually gets from an all-Ethereum surface
The paid API answers the same
questions with a smaller, tighter index behind it. On any
/v1/intel/agent call for Base or
Ethereum mainnet, the response carries the Sybil-adjusted
reputation score, the commerce-backed status and job-outcome
count, the hook-type classification (Virtuals ACP vs
self-deployed
AgentCommerce.sol), the
by-service revenue breakdown, and the 7d / 30d velocity delta
over the commerce-backed subset. Every one of those fields is
derived from a per-chain pipeline that now only has to worry
about signal that actually exists. No BNB placeholder tail to
classify around, no dead-owner cluster inflating the reputation
histogram, no zero-endpoint agents to filter out before scoring.
Three commerce-backed agents you can inspect right now — all on Base, all with real job-outcome trails:
- Lunara (Base) — one of the deepest job-outcome rows in the current commerce-backed cohort.
- Ethy AI (Base) — hosted ACP integration with an actively growing job trail.
- Capminal (Base) — another top-of-cohort commerce-backed profile.
Every one of those pages is free. Every one of them draws its signal from the same all-Ethereum pipeline that now powers the paid surface.
The trust filter, not the registry dump
The line the series has been walking down since Report 01 lands
here. "Biggest index" is a marketing claim any script that reads
Transfer(from=0x0) can make. Live
agents plus commerce-backed reputation is a claim only a data
source willing to actually probe endpoints and cross-reference
payment events can make. We chose the second surface, deleted the
chain that couldn't support it, and are optimising the pipeline
for the composition of the ecosystem that actually ships. Base
plus Ethereum today; Arbitrum next. All-Ethereum.