HWHow we measure Agent GDP

The question the number answers

Agent GDP answers a narrow question: how much value settled, in public logs this index knows how to attribute, on agent payment protocols, over a stated window? It does not answer how much commerce agents influenced, how much of that commerce was honest, or how large the card-based agent checkout market is. Those are different questions and they need different evidence.

The figures on this page are read from the same server-side path as the Agent GDP page. They are cached for about five minutes. The timestamp and the dataset name are part of the figure. Sample rows are fixtures whose hashes start with sample-. They are not chain history, and a page that is showing them says so.

What a counted payment is

An x402 row is a USDC transfer whose transaction sender, on EVM chains, or fee payer, on Solana, is a registered facilitator settlement wallet. The buyer is the token sender. The seller is the token recipient, the onchain counterpart of payTo. The chains are Base, Polygon, Solana, and Arc. Arc history starts on 16 September 2026, the public mainnet date. The x402 backfill starts on 6 May 2025, the date of Coinbase's launch post.

An MPP row is a Settled event on the Tempo escrow precompile, mainnet chain id 4217. Only deltaPaid, and only when it is greater than zero, is summed. The token is taken from the channel open. pathUSD, OUSD, USDC.e, and USDT0 are accepted at 6 decimals. ChannelClosed is ignored because it repeats lifetime settled value. Testnet is ignored.

A 402 response, a signed mandate, an open checkout session, a Bazaar listing, and an unpaid invoice are not payments. They can explain a payment. They do not add one.

Raw volume and the organic screen

Raw volume is the sum of those settlements before any seller filter. Facilitator fee legs stay in it. Organic volume applies version 1 of the seller screen to the window you are looking at. All four tests have to pass: at least $100 of revenue, at least 3 distinct buyers, an average payment of at least $0.001, and a seller address that is not a facilitator wallet. Average payment is revenue divided by the number of payments.

The screen is recomputed per window, so a seller can be organic over 30 days and fail over 7 days. Failing the screen does not delete the raw row. The two series are published together so the adjustment is visible. Passing is not a finding that the seller has human customers. Related wallets can clear a three-buyer test. That limit is the wash trading problem, and it is not papered over with a private filter.

Windows, payers, and rollups

The Agent GDP page headlines use 24 hours, 7 days, and 30 days. The 7 and 30 day payer and seller counts from the headline tables are distinct across the window. The 24 hour payer count is built from hourly buckets and is not distinct across the day. Chart buyers are distinct inside each bucket. Adding the chart will not reproduce the headline buyer count. Transaction counts do add, because a payment is in one bucket.

Dashboards read hourly and daily rollups, which keep history. Raw transfers in the database are kept for 180 days and then archived. A chart that goes back further than 180 days is using rollups, not a rescan of receipts. Rankings rebuild on an hourly job. A learn page can be a few minutes behind the chain, which is what the timestamp is for.

Use-case payers are a third definition: payers summed per seller, so one buyer of two sellers counts twice. They are labeled on the use-case pages and should not be quoted as the buyer total.

Protocols in the total, and protocols left out

A protocol is indexable only when a public chain log, or a published settler list plus a public transfer, identifies the payment without a custodial API. x402 and Tempo MPP sessions meet that test. AP2, ACP, L402, and the card-network agent products do not. They remain on the protocols list at no added volume. The shares below are only the indexable set. A note under the table appears when the split is temporarily taken from facilitator labels rather than protocol rollups. That note is part of the figure.

Settled volume for the last 7 days, by payment standard. Updated Oct 25, 2026, 12:16 AM UTC. Share is that row divided by the standards in the total.

Protocol	Volume	Payments	Share
x402	$40,133	186,912	100.0%
MPMPP	$0.072956	80	0.0%

How a token amount becomes a dollar

There is no price oracle. Named 6-decimal dollar tokens are priced at one dollar. For x402 that is USDC on the scanned chains, using the Circle contracts the methodology lists. For MPP it is the four Tempo tokens named above. Arc's 18-decimal gas token is excluded so network fees are not booked as commerce. Any other token is skipped. Skipping undercounts new assets. Pricing them at a guessed rate would put a fictional dollar figure in a total that is otherwise a sum of logs.

Catalog prices in the Bazaar are not used as volume. A resource listed at one price and settled at another contributes the settled amount. Call counts in the catalog are a quality signal in CDP's product. They are not transaction counts in Agent GDP.

What would make a citation wrong

Quoting a figure without the window, the dataset, and the timestamp. Adding chain totals to protocol totals. Adding organic to raw. Treating a metadata protocol's absence as a measured zero. Treating organic volume as proof against wash trading. Using the sample dataset as if it were mainnet. Adding an AP2 mandate or an ACP checkout on top of an x402 transfer for the same purchase.

The versioned rules live on the methodology page and at /api/v1/methodology. If this article and that page ever disagree, the methodology page is the one that was computed against. This article is the explanation of those rules, with the live totals beside them.