llms.txt — documentation index for LLMs and agentsllms-full.txt — full documentation corpusMarkdown version of the documentationskill.md — agent skill filePaddock MCP server descriptor

Methodology spec

What is in the index, why, and on what basis it is weighted.

Published Document 01 of 04Changes dated in the corrections log
Status

Published 2026-08-27. This is specification, not description: it defines how a Paddock index is constructed when one is published.

No Paddock index is currently live. The Agent Commerce Index is suspended, not retired — no new monthly print is computed, existing prints are retained unmodified, and it resumes only under a new version tag once the recipient-list truncation is fixed and the attribution basis is recomputable. Nothing here should be read as a claim that a governed index exists today.

01.1

The universe

The candidate universe is publicly traded tokens whose issuer operates a service observable on a rail Paddock measures. Sourced from CoinGecko's x402-ecosystem and ai-agents categories.

Membership in the universe is not membership in the index. A token enters the universe by being an agent-economy token; it enters the index only by clearing the tier-1 bar in §01.2.

01.2

Eligibility — tier 1 only

Constituents are drawn exclusively from tier 1. The four tiers are the output of the token→service join, and the definitions below are the corrected ones — the pre-2026-08-20 logic assigned tier 1 on a domain-name match alone and is not the basis here.

TierDefinitionEligible
T1The token's service resolves to a domain in Paddock's attribution graph and that domain recorded non-zero settled volume in the trailing 30 daysyes
T2Attribution holds but settled volume is zero — a real x402 seller whose settling wallet is not ours, or a dormant oneno
T3An unverified documentation claim of x402 support, with no attributable settlementno
T4No observed connection. Split into confirmed-absent (a live probe returned a clean non-402) and PROBE-INCONCLUSIVE (429/timeout/TLS/bot-challenge on every leg)no

Two rules that follow, and are load-bearing

Attribution alone never earns tier 1. A domain in the graph with zero settled volume is demoted to T2 and flagged, not smoothed. This is the defect corrected on 2026-08-20: three rows held tier 1 having never settled anything.

PROBE-INCONCLUSIVE is not evidence of absence. A token whose probe was rate-limited or bot-challenged on every leg is recorded as inconclusive and excluded from any “no connection” count. It is not eligible, but neither is it evidence against the issuer.

Evidence required per constituent

Each tier-1 row carries, in the published constituent table:

  • the matched domain and the matched payTo wallet(s);
  • 30-day settled USDC and transaction count;
  • a probe_statusmatched, inconclusive, or clean_negative — never a hardcoded default. inconclusive means the tier rests on registry and settlement evidence rather than on a payment challenge we decoded ourselves, and the row says so;
  • an evidence URL that resolves.

A constituent whose probe_status is not_probed is not eligible. There is no exception for a large or well-known issuer.

Standing exclusions

Excluded regardless of settled volume:

  • wallets on the circular/flagged denylist — see §04.4 for who may change it;
  • facilitator and pooled wallets — they carry pass-through volume, not the demand of a single service;
  • facilitator brand hints — a token whose name or domain matches a known facilitator brand but whose payTo was not independently confirmed. No volume is credited to these regardless. A facilitator's own token must never be credited settlement from unrelated sellers routed through it.
01.3

Weighting — settled volume, not market cap

Constituents are weighted by trailing-30-day settled volume, denominated in USDC, from the census universe.

Why not market cap. A market-cap-weighted index of this universe measures one token's price, not agent commerce. Measured 2026-08-20, the tier-1 market-cap share was 99.5685% a single issuer — it would have tracked that issuer and been almost completely insensitive to every other constituent. That is a fine statistic and a bad index.

Why settled volume. It is the thing being indexed. It is also the figure with the most disclosure attached to it: a known universe, a stated denominator, and a capture basis.

Known weakness, stated rather than hidden

Settled volume comes from a single third-party index (x402scan) with no independent second measurement. The parity check compares our stored series against that same source, so it detects our ingestion drift and not an upstream re-basing — its own verdict scaffold says the two are inseparable. An independent on-chain series is a precondition for licensing this index, not an enhancement to it.

Smoothing. Weights use a 7-day rolling average of daily medians, not a raw daily figure. Daily settlement on this rail is intraday-volatile by a factor of roughly two within a single day, and August 2026 moved between ~175,000 and ~1,230,000 daily transactions inside one calendar month.

Floors. A constituent below a stated minimum 30-day settled volume is excluded rather than carried at a negligible weight. A weight small enough to be noise is a claim of precision the data does not support.

01.4

Concentration disclosure — mandatory

Every publication of an index level carries its concentration figure, with its denominator named, in the same artifact. Not in a linked methodology page, not on request, not in a footnote.

FieldMeaning
top-constituent sharelargest constituent's share of index weight
top-3 shareconcentration of the head
effective constituent counte.g. an HHI-derived figure — how many constituents the index behaves as
ex-top-constituent levelthe index recomputed without its largest member

When the top-constituent share exceeds a stated threshold, the index level is published only alongside the ex-top-constituent level, with equal prominence.

The precedent is explicit: never quote the market-cap share without the settlement view beside it. The same rule, applied to weight.

01.5

What this index is not

  • Not a price index. It measures settlement, not token valuation.
  • Not a measure of the whole market. It measures the attributed universe. Attribution currently names the seller behind a small share of settled transactions, computed inside a capped nightly recipient fetch. Any coverage figure published with the index states that base.
  • Not investment advice, and not a recommendation. See the conflict policy.
01.6

Versioning

The methodology carries a version tag. A change to eligibility, weighting or exclusions is a new version, and a new version is a series discontinuity — recorded, dated, annotated in the series, never applied retroactively to published levels. See §02.4.

Threshold changes count. Re-tuning a number is a methodology change wearing a smaller hat.

Not appealable

A methodology change is not an error, and the appeals process is for errors. An issuer who disagrees with the basis is disagreeing with a published, dated decision — which is a fair thing to do in public, and not a thing this process reverses.

— Keep going

What to read next.

Monthly intelligence
Read the report.

The State of Agent Commerce — the citable monthly report on what AI agents actually buy.

See the report →
x402 market
What agents pay for.

Real transaction volume by category. Three sources merged into one live table.

See the market →