Reference · how the numbers are made

Methodology.

Our authority rests on one thing: you can check our work. This page explains precisely how every number on the Economy Scoreboard is produced, what it includes, and what it deliberately leaves out. If a number can’t be measured honestly, we don’t publish it.

1. We measure at the settlement layer

Agent services advertise a payTo wallet — the address a caller pays to use them (the x402 / HTTP 402 pattern). We read the blockchain directly and sum the stablecoin (USDC, USDT, PYUSD) Transfer events into those wallets. That is the definition of “settled volume”: money that actually moved on-chain, not volume a marketplace reported about itself.

Each run scans new blocks since the last checkpoint on every supported chain, attributes each transfer to its recipient (the provider) and its sender (the payer), and appends the delta to a forward-only history. Nothing is estimated; nothing is back-filled.

2. “Real provider” vs. “listing”

Directories of the agent economy carry tens of thousands of listings, but a handful of wallets account for most of them. We report two separate things and never conflate them:

Listings — how many catalog entries exist. Context, shown small.
Real providers — distinct wallets that have received more than $0 on-chain. This is our headline count.

A service can be listed and never earn a cent. Only the second number reflects a working economy, so it’s the one we lead with.

3. Coverage — and where we’re blind

We measure EVM settlement chains directly: Base, Polygon, Arbitrum, Ethereum. We publish a coverage figure — the share of listed provider wallets that live on chains we currently read — next to the totals. Because coverage is below 100%, true economy-wide volume is higher than what we show. We would rather understate with a disclosed gap than overstate with a guess.

Not yet counted: Solana (the largest gap, a different address format and API), Stellar, Noble/Cosmos and other non-EVM rails. These are on the roadmap and will be folded in the same way — neutrally, by measurement, with no chain favored over another.

4. Concentration & price realization

Concentration — top-wallet share, top-10 share, and a Gini coefficient over per-wallet cumulative volume — tells you whether earnings are spreading out or pooling into a few wallets. A Gini near 1 means one wallet dominates; near 0 means even distribution.

Price realization compares the median quoted price (from live 402 responses) against the average amount actually paid per transaction (settled volume ÷ transaction count). They’re different populations, shown side by side, so you can judge whether advertised prices reflect real demand.

5. What we exclude, on purpose

Testnets. Test-network activity isn’t real money and is never counted.
Single-payer self-funding. When a wallet’s inflow is ≥90% from one payer on material volume, we treat it as self-dealing/wash and hold it out of the validated figure — while still showing it in the raw total. One self-payment shouldn’t masquerade as an economy.
Non-production endpoints. Services on localhost / test / example hosts are excluded from the validated figure.
Self-reported or estimated volume. If we can’t see it settle on-chain, it isn’t in the totals.
Pre-tracking history. Volume before a wallet’s first scan isn’t retroactively added.
Projections. Forecasts, if any, live outside the scoreboard and are always labeled with their method.

6. Neutrality

We are chain-, protocol-, and facilitator-agnostic. We add new rails when real agent-payment activity appears on them — the same standard for every network, including ones we have no stake in. Any paid placement, if it ever exists, will be clearly labeled and will never alter a measured number. The moment the data looks bought, the reference is worthless — so it won’t be.

Questions or corrections: hello@agentalmanac.org.