Settlement across the ERC-8004 agent economy

Measured first-party from each registry’s first block, across the 26 networks the standard has reached and the four where settlement exists. Pinned to Base block 51,416,524, BNB Chain 122,357,356, Ethereum 25,996,329, Polygon 93,953,274. Every figure below comes from one run and resolves to a sealed snapshot.

---

The money ladder, all three tiers

A single number for “the agent economy” is the easiest thing to get wrong here, so this states three and never sums them.

Base:

• Gross USDC touching a registered agent wallet: $399,441,244 across 2,007,091 transfers.
• Both legs of the transfer are registered agents: $10,591,460 across 32,254.
• Both legs registered and the two parties distinct: $222,043 across 20,098, mean $11.05.

Ethereum: $75,784,604 gross, $170,955 both-legs, $5,644 between distinct registered parties.

Polygon: $13,633 gross, fifty cents both-legs, nothing between distinct parties.

BNB Chain: $0.00011 total, across 2,526,999 transfers. That is not a rounding error and not a typo. The indexed instrument is an 18-decimal Binance-Peg USDC, and the largest single transfer among those two and a half million is worth $0.0000004.

The distance between the first and third tier is the whole reading. Gross volume touching an agent wallet is 1,800 times the value those agents moved between each other.

What the gross is actually made of

Most of the difference is not agents trading. We called name(), symbol() and asset() on the largest destinations of agent-originated USDC on Base and they identify themselves:

• The largest, $108,891,170 in and nothing back, holding no USDC at all: a pass-through router.
• The second, $33,530,074 in, nothing back, zero balance: the same shape.
• The third, $8,638,876: Spark USDC Vault, an ERC-4626 yield vault.
• The fourth, $7,786,774 in and $8,053,188 out: Spicy USDC Vault.
• Also FARM_USDC (Harvest), EVK Vault eUSDC-49 (Euler), Aave Base USDC, Moonwell USDC.

Each contract’s address is in the sealed snapshot beside its figures, which is what the snapshot is for: this page carries the reading, and the bytes carry the identifiers.

On Ethereum the same picture, with Spark Savings USDC, the Uniswap V4 PoolManager and CoW Protocol’s settlement contract.

These are yield positions and swaps taken by wallets that happen to hold an ERC-8004 identity. An agent depositing into Aave is not an agent economy, and counting it as one inflates the sector by a factor of thirty-two.

We built a rule to separate commerce from DeFi automatically, by flagging venues whose flow with the agent set is two-sided, and discarded it: a vault returns USDC on redemption, so Spicy and Harvest passed. When the value concentrates in a dozen addresses, the honest method is to identify each one.

The one market

Removing DeFi leaves a single identifiable marketplace with material volume in the whole ecosystem: TermiX’s settlement contract on Base.

$6,259,351 in across 171,149 transfers, $6,133,958 out across 171,141, from 6,335 agent wallets, at a mean of $36.57 per transfer.

High frequency, low value, two-sided and near-balanced. That is what escrow with delivery guarantee looks like, and nothing else in the ecosystem looks like it.

Is that 6,335 wallets, or one operator holding 6,335 addresses?

We asked, because a population count that cannot tell those apart is worth nothing, and this dataset contains a registry where exactly that is true.

It is not one operator, on the evidence available. The largest wallet is 0.1% of flow through the contract. The top ten are 0.9%. The Gini coefficient over per-wallet flow is 0.19, the median wallet moved $1,850, and only twelve of the 6,335 moved under $100.

That rules out a whale. It does not, on its own, distinguish a broad market from a programmatically provisioned cohort, because a distribution that flat is also what you get when addresses are funded alike and scripted alike. Real marketplaces are usually power-law shaped, with a few large users and a long tail; this one is not. We can measure the flatness and we cannot measure intent, so we state both readings and prefer neither.

The capital underneath it

Throughput is not capital. We traced how much money entered those 6,335 wallets from outside the system they form with the contract.

$139,518 gross in. $4,984 out. $134,533 net.

Ninety-seven per cent of it arrives through a single contract, which we followed on chain to $6.16 billion inbound from 202,879 distinct senders. That is general Base infrastructure, not a TermiX contract, and it is what makes this externally-sourced capital rather than an artefact of how we classify an address. Its address is in the snapshot.

Block timestamps were absent for most settlement rows, so rather than fetch millions of them we binary-searched the chain for the first block of each week. Every boundary below is a real block whose timestamp was read; nothing is interpolated.

Week	Deposits	Net external capital, cumulative	Throughput, cumulative	Turnover
6 Aug	$303,172	$33,955	$303,172	8.9x
13 Aug	$964,473	$33,956	$1,267,645	37.3x
20 Aug	$1,775,233	$85,528	$3,042,878	35.6x
27 Aug	$1,454,449	$135,413	$4,497,326	33.2x
3 Sep	$892,084	$134,539	$5,389,410	40.1x
10 Sep	$477,262	$134,533	$5,866,673	43.6x

External capital arriving in the last two weeks of that window: $329, then $11. Newly funded wallets in those weeks: zero, then zero. Throughput over the same two weeks: $892,084 and $477,262.

The capital base stopped growing on 27 August while volume kept compounding. Turnover rises not because money moves faster by design but because the numerator kept going and the denominator stopped.

What this does and does not say

It says the same $134,533 has been recycled roughly forty-four times. That is an observation about capital intensity, not about honesty. A marketplace that holds funds in escrow and releases them on delivery should show high turnover; that is what escrow is for.

It does say that a headline of the form “$X of funded agent order volume” can be literally true and describe very little underlying capital, depending entirely on what “volume” counts. TermiX publicly cites $11.5M, a figure of theirs that we have not verified and carry only as theirs. If it counts gross throughput, it is consistent with everything measured here and there is nothing to explain.

We asked TermiX which of the two readings the protocol means, publicly and before writing this, in TermiX-official/aacp-whitepaper#2. It is unanswered. Two caveats we owe that fact: the account we asked from had no organisation filled in on its profile, and the repository is close to dormant, with seven stars and no prior open issues. The question may never have been seen, and we are not treating silence as an answer.

Declared limitations

• USDC only. Wallets funded in ETH and swapped, or funded on another chain, are invisible here. $139,518 is a floor, not a total.
• The settlement scan is anchored on the agent set. A transfer between two addresses where neither is a registered agent is not indexed, so money reaching a platform contract directly from a non-agent is not something we can see.
• Registrations are not verifications. Every identity count in this dataset counts transactions anybody can send. We do not claim the agents behind them are real, and counting them is not verifying them.
• The flatness of the wallet distribution has two readings and we cannot choose between them, as stated above.
• No reply from TermiX, with the two caveats above about why that may mean nothing.
• This is a snapshot, not a history, everywhere except the six-week series.

Method

The exclusion rule that decides which counterparties count is the engine’s own, not a second implementation written for this report. Instrument decimals are read from each token contract rather than assumed, which matters: assuming six on BNB Chain overstates that chain by a factor of one trillion. Freshness is measured as scan coverage rather than the age of the last event, because a quiet chain is not a stale one. Before writing any adverse claim we ran the four benign-hypothesis tests over the settlement data and wrote against the residual; on Base, 179 of 197 standalone self-payers also pay distinct counterparties, and the unexplained residual is eighteen addresses and $6.91.

Verifeth indexes the ERC-8004 identity and reputation registries continuously, from each registry’s first block, on every network the standard has reached.