Deciding whether an observed agent payment travelled on the x402 rail. FLOOR — tens of thousands of x402 resources exist and only a slice is indexed. Every unindexed recipient makes our attributed share smaller than reality.
Measuring how concentrated agent signing authority is. COMPLETE for the wallets we have read. Wallets whose code we failed to read are excluded from both numerator and denominator, never assumed undelegated.
The only set that can make an unknown implementation an incident. DELIBERATELY EMPTY. An unknown implementation is not evidence of a god key, and until somebody has read a contract and can say why it belongs here, listing it would be exactly the unsourced claim this project exists to replace. While empty, the god_key_unknown_7702 detector stays inert by design.
Showing how much of the registry's headline count comes from very few hands. COMPLETE for Base. Other chains are not counted.
Mapping the demand side — where agent money actually lands. COMPLETE for the wallets we have scanned. It is a sample of the market, not a census of it.
Naming the contracts every launch metric here is read from, so the layer itself can be checked. The contracts we read, not all of Doppler's deployment. Taken from Doppler's own published address table rather than inferred, because we briefly mistook this layer for a single front-end's private infrastructure and the fix for that is to cite the table.
Showing who actually earns from the token-launch layer agent platforms are built on. COMPLETE for the blocks we have swept, and the sweep advances only past windows it actually read. The set includes Doppler's own protocol beneficiary, which is present on every launch by construction and is excluded from any statistic describing launchers.
Detecting whether agent money leaves the chain through Circle's own bridge. The TokenMessenger, MessageTransmitter and TokenMinter for Base and Ethereum, V2 plus the older V1 TokenMessenger. Of these, three distinct addresses can actually receive a burn from a wallet, and those are what the exit check matches against — which is why cctp_addresses_watched reads 3 against a set of this size, rather than the two numbers disagreeing. Other CCTP chains are not watched, because our payment scan only covers Base.
How to read provenance.
Every set is labelled with one of three words, because it is the only thing that matters when judging whether to trust it.
- Read from the chain — we derived it ourselves from on-chain data. Anyone with an RPC endpoint can reproduce it, and if we are wrong you can prove it.
- Harvested from a cited index — somebody else's list, taken with the source URL stored against every member. Our claim inherits their coverage, so the set is a floor rather than a census, and we say so.
- Curated by hand — a human decided. That is sometimes the only honest option, and it is the weakest kind of set, so it is labelled loudest.
What a set does not license us to say.
Knowing that a set of addresses receives agent payments does not tell us who runs them. Naming an operator is a separate claim that needs its own evidence — a published disclosure, a signed message, an explorer tag with a source — and we do not make it from a set alone. This is why the homepage carries no volume attributed to a named company: the claim would be about somebody's business, and we have not measured it.
It also does not license a total. Every set here is a sample of a larger world, and each one states its completeness in its own words. Read the definitions for what each dependent metric asserts and, more usefully, what it refuses to assert.