We measured $121k across 1,171 agents in 30 days. We can't yet show that any of it is independent.

Agenstry has published observed on-chain inflow per agent for months. The number is honest about what it counts — gross inbound stablecoin transfers to wallets agents declare on their own cards — and our methodology page has always carried the caveat that we cannot tell an arm's-length customer from an operator moving its own money.

That caveat was a disclosure. As of this week it is a measurement.

We built a payer-independence check and ran it across every agent with observed inflow. This post is the first output: the real distribution, what it found, and — more importantly — what it could not find and why.

The headline

Over the 30 days ending August 6, 2026:

That first pair of money rows is itself a finding. 1,171 agents settle into only 477 wallets, so adding up per-agent revenue counts a lot of the same money twice — 17.41% too much. The de-duplicated figure is the one we consider the market size; the attributed figure is what you get if you trust a per-agent leaderboard and sum the column.

Every scored agent sits on the bottom rung, insufficient_evidence_of_independent_payers.

That label is deliberately about us, not about them. It means: money arrived, we watched it arrive, and we cannot see who sent it. Payer-identity coverage across the index is 0% — we hold transaction-level sender records for zero of the 1,171. Without a sender, "independent" is not a question we can answer, and a check that cannot answer its question should say so rather than guess.

We are publishing a result of zero because the alternative — waiting until the number flatters us — is how measurement infrastructure loses the right to be believed.

What the ladder actually is

Five rungs, ascending, each describing the strength of our evidence:

1. insufficient_evidence_of_independent_payers — inflow observed, payers invisible to us
2. single_payer_observed — one payer accounts for effectively all volume
3. concentrated_payers — volume concentrated in a handful of payers
4. recurring_independent_payers — several distinct payers, repeat demand, none dominant
5. independently_paid — many distinct payers, low concentration, repeat demand, no self-referential inflow

None of these rungs says "fake" and none says "fraud". Those words do not appear anywhere in the labels, the flags, or the code, and a test asserts they never will. The reason is not squeamishness. A low rung is overwhelmingly a statement about Agenstry's coverage, and dressing a coverage gap as an accusation about an operator would be both wrong and, if we ever got it wrong about a real business, indefensible.

What we could measure

Payer identity is missing, but settlement structure is not. Those checks run against the daily ledger and wallet linkage, and they found real, differentiated signal:

The shared-wallet number is the one that changed how we read our own leaderboards.

842 of 1,171 agents — 71.9% — settle into a wallet that at least one other indexed agent also settles into. The largest single cluster is 149 separately-listed agents sharing one address. Each of those 149 is credited with the same $124.73 over the same 348 transactions, because that is what arrived at the wallet they share. Sum the per-agent figures across that cluster and you get roughly $18,585 of apparent activity generated by $124.73 of actual settlement.

Nothing about that is misconduct. Running a fleet of agents out of one treasury is a perfectly ordinary architecture. But it means per-agent inflow in a shared cluster is the same money seen from several sides, and anyone — us included — who adds those figures up is double-counting. The check now says so on every affected agent page, and the paid wallet_intel skill returns the attribution warning at the wallet level.

Concentration, and the limits of a seven-day ledger

Two more things the structure checks surfaced, both about history rather than intent:

• $119,309 of the $142,194 attributed — 83.9% — sits in ten agents. A single agent accounts for $102,809, or 72.3% of everything we observed. (Both figures are on the per-agent attributed basis, so they are directly comparable to each other.) The agent economy's measured revenue is not a broad base; it is a very short head with a long, thin tail.
• Median active days: 6. Our daily ledger currently spans June 29 to August 6, and most agents have a week of it. 261 agents have fewer than three active days — too little to characterise any pattern, which is why they carry short_observation_window rather than a verdict.

single_day_concentration fires on 190 agents where 90%+ of window volume landed on one day. A launch, one large customer, and a scripted burst all produce that shape. We cannot separate them from a week of ledger, so we flag and say why.

Three shapes, anonymised

We do not name agents beside a low-confidence label. The pattern is the finding; the operator is not.

A — one wallet, 149 agents. Exclusive-wallet checks fail by construction. $124.73 credited identically 149 times. Independence cannot be established from a shared address, no matter how clean the rest looks.

B — 99.6% on one day. Exclusive wallet, $208.33 across 23 transactions, but only 2 active days out of 7 observed, and essentially all of it on one. Two flags, both about not having enough history.

C — the cleanest shape we hold, still rung one. Exclusive wallet. Active every observed day. No day above 18% of volume. No repeating daily amounts. The largest observed inflow in the entire index. It passes every settlement-structure check that exists — and it stays on the bottom rung, because we hold no sender records for its wallet.

Example C is the honest summary of this whole exercise. 169 agents carry no blocking caveat other than the missing payer identity. They are not stuck because they look suspicious. They are stuck because the one thing that would promote them is the one thing we cannot yet see.

Independent context

We are not the only people asking whether agentic settlement volume means what the charts imply. In When HTTP 402 Meets the Blockchain: Risks on Emerging x402 Payments (arXiv:2607.19545, July 21, 2026), Qinying Wang, Yong Yang, Yuan Chen, Shouling Ji and Mathias Payer measure roughly 119 million recent Base and Solana transactions across 15 major x402 facilitators serving 60,000+ sellers and 360,000+ buyers. Their focus is facilitator security and centralisation rather than payer independence — they report, for instance, over $202,000 burned in gas and fees on settlement attempts, $5,800 of it on reverting submissions alone.

A note on sourcing, because it matters for a post about evidence standards: we went looking for a widely-repeated claim that a specific double-digit percentage of Base x402 settlements is fictitious and a further large share internal. We could not substantiate those figures in the preprint they are usually attributed to, so they do not appear here. If you have the primary source, tell us and we will cite it.

Limitations

Stated plainly, because a method post without them is marketing:

• Zero payer coverage, as of this measurement. Every payer metric — distinct payers, concentration, recurrence, self-referential share — is stored null, not 0. Null means unmeasured. A zero would read as a measured finding against an operator, and we will not ship that ambiguity. This is a state of our ledger on a given date, not a permanent property of the index.
• Short window. Roughly five weeks of daily ledger, most agents with seven days in it. The 90d trailing window is currently bounded by how long we have been recording, not by the window length.
• Inflow ≠ revenue. These are gross inbound stablecoin transfers to declared wallets. Not netted, not audited, not proof a sale happened.
• Wallet linkage is our own. Shared-wallet detection sees agents we index. Two agents settling to one address that we only know one of will not be caught.
• A high rung will never mean "verified legitimate". It will mean we observed many unrelated wallets paying, over enough time, without self-referential inflow. That is a statement about observation, not endorsement.

What moves the number — and why this post has a shelf life

One thing: transaction-level payer records. The scoring functions that produce concentration, recurrence and payer≈payee overlap are already written and already run — they return null today purely because the sender ledger behind them is empty. Fill it and the same code produces the same metrics with values in them, on the same five-rung ladder, with no change to the labels.

That ledger is not waiting on new code. Our on-chain scanner already retains payer-side transfer evidence; it simply has not completed a full pass since the table shipped. We ran the whole pipeline against a populated sender ledger in a development environment to confirm the ladder works end to end, and it does: agents distribute across all five rungs, including the top one.

So treat the distribution above as a dated snapshot, not a standing fact. It is what we could evidence on August 6, 2026. The moment the scanner backfills, payer coverage stops being 0% and this table changes — which is the point of building it. The live figures are always at /authenticity and /authenticity.json; if you are citing us, cite those.

Until then, the honest answer to "how much of the agent economy's revenue is independent?" is: we don't know, and now we can tell you precisely how much we don't know.

Per-agent labels are live on every agent page with observed inflow, free, with the checks and measurement date attached. The method, the live distribution and the worked examples are at /authenticity. The full numeric breakdown and the 7/30/90-day trailing windows are available through the paid get_agent_full and payment_intelligence skills on REST, A2A and MCP.