The classification rules of Methodology v3.2 are carried over unchanged, except as stated above. AQUIX does not state whether an address is operated by an AI agent or who operates it.
The full text of Methodology 2026.11 follows this notice and is available at aquix.ai/methodology-2026-11.
Comments on this notice may be sent to disputes@aquix.ai, which is open now. Any change to this notice or to the methodology before the effective date will be published as a second, dated notice.
AQUIX Methodology 2026.11 · Published 5 November 2026 · Effective from the first sealed release after launch (planned 18 November 2026) · Supersedes Methodology v3.2
AQUIX Methodology 2026.11
1. Summary
AQUIX publishes measurements of payments made over the x402 payment protocol on the Base network. For each address that receives such payments it publishes, at one fixed point in the chain each night, the money the address received, the money it sent, the balance it held, the part of what it received that is a proven x402 settlement, and, where the evidence is sufficient, a classification of how its money moves.
The classification describes where money went. It does not describe the operator of an address, its intentions or the quality of its business. Whether any address is operated by an AI agent is not verified.
Each night's figures are fixed at a published block, checked, sealed and released together. A sealed release is never edited. Late data, corrections and changes of method are handled by the published procedures in sections 8 and 9.
2. Scope and coverage
Network and asset. The Base network. The asset is USDC on Base, identified by its contract address, not by its symbol.
Payments covered. x402 payments settled with the EIP-3009 "exact" scheme (transferWithAuthorization). A USDC transfer is treated as an x402 settlement only if the same transaction records the payer's authorization (AuthorizationUsed) and the authorizer is the sender of the transfer.
Payments not covered. Other settlement schemes (Permit2, "upto", batch settlement, ERC-7710) and other networks are not measured. Payments made off the chain are not measured.
Facilitators. The facilitators whose transactions are read are listed publicly, each with the date from which it is recognised. Payments relayed by a facilitator before that date are not counted as settlements. The completeness of the list against external lists is not measured.
The published set. An address is published if it is admitted by the published admission rule at the night's cut-off and is not excluded. Excluded addresses (for example facilitators, routers and wallet software) are listed with the basis on which each is known. An address that is no longer admitted keeps its records but is not published.
Period. Records begin on 28 June 2026. The first published day is 7 July 2026. Days with no published figures and days with no classifications are listed.
3. Definitions
The core terms used in this document are defined below. Every published figure has a full definition in the public catalog: its identifier, definition, formula, unit, source, cut-off rule, exactness and what it does not measure. Pages print the definitions from the catalog.
- Cut-off. The last Base block before 00:00:00 UTC on the release day. Every figure is stated "through the end of the previous day (UTC)".
- Received. All USDC that arrived at an address by any route, up to the cut-off. Labelled "received in USDC, any route".
- Sent. All USDC that left the address up to the cut-off.
- Kept. Received minus sent. It equals the address's USDC balance on the chain at the cut-off and is shown with the cut-off date.
- x402 revenue. The part of what an address received that is a proven x402 settlement under section 2. It is published as "x402 revenue, verified", with the scope of that verification stated in section 7. It is never added to received.
- Headline figures. Every headline figure is on the proven x402 basis. The home page leads with the amount kept from x402 payments. Under it, x402 revenue is shown first without the two largest addresses, which are named, with the total, "x402 revenue, verified", beside it. "Received in USDC, any route" is shown beside them, labelled, and never added.
- Evidence bands. A received dollar is either "proven x402" or "not established as x402". "Not established as x402" does not mean that the payment was not an x402 payment.
- Proven share. x402 revenue divided by received.
- Exactness. Every figure carries one of four labels: exact (reproducible to the micro-USDC from the records at the cut-off), floor ("at least"), upper bound, or estimate (with the method named).
- Payer. The account that authorised an x402 payment. Seller: the recipient of the payment. Facilitator: the account that submitted the transaction. A facilitator that is also the payer is excluded from x402 revenue and shown separately; a facilitator that genuinely pays on behalf of its users is excluded as a result, and this is stated.
- Sender. Any account from which an address received USDC, by any route.
- Recency. Live: paid within the seven days before the cut-off. Stale: an older last payment. None: no x402 settlement at or before the cut-off.
- Concentration. When the largest address carries 25% or more of total x402 revenue, its share is printed beside the total. The totals are also published without the two largest addresses, which are named.
4. Data sources
- The chain. All figures are computed from Base chain data: USDC transfer events, settlement authorizations, block headers and balances at the cut-off block.
- Collection. AQUIX reads these data through its own collectors and stores them unchanged. Each night's cut-off block and its hash are recorded and checked again the following night.
- Admission and exclusion inputs. The lists that decide which addresses are published and which are excluded are copied once for each release, at its cut-off, and every step of that release uses the copy. Any later change to those lists is recorded and takes effect only in a later release.
- No other source changes a figure. Names of operators are shown only from public sources and are never inferred. Where a name is not known, it is left blank.
5. The nightly release
- One cut-off each night. All figures in a release are measured at the same cut-off block.
- Stages. Each release passes in order through building, testing and sealing. A release that fails any stage is not published, and the previous sealed release remains in place.
- Sealing. A sealed release is never edited; the database refuses changes to it. The evidence behind each classification is stored with the release and is not read again from live data.
- Checksums. A checksum of each sealed release is published in two places outside the AQUIX database: an off-site copy and a public file of releases. Both are read back before the release is declared sealed. Anyone can recompute the checksum from the published files.
- Stamp. Every page and file carries the methodology version, the catalog version, the release number, the cut-off block and time, and the version of the code that produced it.
- Release time. Published by 20:30 UTC every day, or a dated miss with its reason. No one receives a release before the public.
6. Classification
Verdicts. An address that meets the evidence requirement receives one of four verdicts: Service, Conduit, Wallet software or Circular flow. These words are used without any prefix. Internal classifier codes never appear on a page.
Limiting sentences. Every verdict is printed with its limiting sentence, the kind of evidence on which it rests, the rule that produced it, its confidence and the basis for that confidence, the release stamp, and the statement "unsolicited — the operator did not request this and did not take part". The limiting sentences are listed in the public catalog.
Circular flow. This verdict describes the money, not the operator: value returning to the same counterparties is also what a trading program, a settlement loop or a refund flow looks like. Two earlier sentences that described the operator rather than the money are withdrawn.
Conduits. An address is classified as a Conduit only on evidence of its role. The two largest addresses by x402 revenue are classified as Conduits on role evidence gathered on 27 October 2026, with the limiting sentence "role read from the contract and its payments; source code not verified." An address that forwards most of what it receives is not a Conduit for that reason alone.
"Too early to judge." An address paid by fewer than two different senders, by any route, receives no verdict. Its money is shown and remains in every total; it is counted on a separate line, outside the count by verdict, and listed in full with its evidence. Its verdict line reads "Too early to judge". A published Service address with no x402 settlement on record is shown as "too early to judge — no x402 settlement recorded".
Until 30 November 2026 the list of addresses that are too early to judge is fixed as it stood on 5 November 2026. During that period an address first published after 5 November with fewer than two senders is classified as before and marked "thin evidence". From 30 November 2026 the requirement becomes two different x402 payers, applied in each release, under the procedure for methodology changes in section 8.
One label per address. Each address carries one public label: the verdict of the night's sealed release. No other view on the site contradicts it. A check that enforces this is built after launch, before 23 November 2026; until then only the sealed verdict is shown.
Continuity. The classification rules of Methodology v3.2 are carried over unchanged, except for the changes stated in this section. Before launch, the new system must produce the same verdicts as the existing system for seven consecutive nights, apart from those stated changes.
7. Data quality and verification
Nightly checks. Every check is defined once in the public catalog, with what it reads, its severity and its consequence: the address is held, the release fails, or a warning is published. A check that cannot run is recorded as a failure. A check not run on a given night is recorded with its reason. A release is held if the number of published addresses moves by more than 10%, or received or x402 revenue by more than 25%, from one night to the next.
Implementation check. Every published figure for every address is calculated a second time, each night, by separately written code reading the same stored records. A difference greater than one micro-USDC holds the address. Because it reads the same records, this check detects errors in AQUIX's code, not errors in the data, and it is not described as independent.
Independent check. From launch, a sample of addresses is checked each night directly against the chain, outside AQUIX's stored records: the largest addresses by x402 revenue and a rotating set, so that every published address is checked about every six weeks. Each release states how many addresses were checked, how many agreed, and what share of x402 revenue they carry. An address that disagrees is held.
The basis of "x402 revenue, verified". The label rests on the following checks:
- Twenty addresses, the five largest by x402 revenue and fifteen chosen by a fixed rule, together carrying about 90% of x402 revenue: each of their 59,702 payments up to the cut-off of 24 October 2026 matched a USDC transfer on the chain exactly, in recipient, transaction, position and amount. In a sample of 2,252 of those payments, every one carried the payer's authorization on the chain.
- Separately, on 2 November 2026, an independent chain read of 75 addresses, carrying 98.98% of x402 revenue, found money received, money sent and balance equal to AQUIX's figures for every one.
- The addresses carrying the remaining 1.02% were not part of that read. They will be covered by the nightly independent check from launch.
What is not measured. A false-positive rate for each verdict would require ground truth that AQUIX does not have. The set of known infrastructure addresses is the only external measure of error, and it is reported as a lower bound.
8. Revisions, corrections and methodology changes
- Two series. The as-published series shows every figure exactly as it was published in its release, and is never changed; it exists from launch. The revised series shows the history as seen by the latest sealed release, with chain data that arrived late marked "completed later in release N", with the old value, the new value and the reason; it exists from 30 November 2026. A sealed release is never edited.
- Corrections. A numbered public correction is issued whenever a headline figure on the home page changes. Other errors are corrected without a numbered entry, and the corrected file is kept as the record. A published value that a correction shows to be wrong is marked, never changed. Correction entries are never edited.
- Presentation changes. A change is treated as presentation only if it alters no figure, no verdict, no label a reader would quote and no meaning. Any other change is a revision, a correction or a methodology change.
- Methodology changes. A material change (to a definition, a classification rule, a verdict word, the facilitator rules, the evidence requirement, or the existence of a threshold) is announced in a dated notice at least seven days before it takes effect. The previous series is kept under the previous methodology version; the new series begins under the new version; both remain available; and every changed verdict is listed by address.
- Versioning. Every release identifies the methodology version, the catalog version and the release. A public changelog records every version with its date.
- Archive. The data files of every sealed release remain available by release date.
9. Verdict lifecycle
No verdict is removed without notice. A verdict ends in one of four ways, each published with its date, its release and its reason:
- Too early to judge. The evidence falls below the requirement in section 6. The money remains shown.
- Excluded. The address is found to be infrastructure or out of scope. It leaves the published set and appears on the exclusion list with the basis on which it is known.
- Withdrawn. AQUIX can no longer support the verdict, because of a data fault, an upheld dispute, or a rule shown not to apply. The page reads "verdict withdrawn", with the date and the reason, and a correction is issued where section 8 requires one.
- Replaced by a methodology change. The previous verdict remains in the previous series and the new verdict begins in the new series. The address is listed among the changed verdicts.
In every case the past verdict remains in the sealed release that published it. The address and its money remain published unless the address is excluded. A verdict is restored only by the same rules that produced it; it cannot be restored on request.
10. Disputes and notices
- Where to write. disputes@aquix.ai, open now. Disputes should include transaction hashes.
- Proof of control. A signed message from the address; for a contract account, an EIP-1271 signature; or, for a named operator, proof of control of the domain shown.
- Factual complaints. Anyone may file a factual complaint, for example a company named on a page that does not control the address. It is answered and logged. It confers no right to control the page.
- Times. Every dispute is acknowledged within 3 days and answered within 14 days. A missed deadline is published.
- Dispute log. A public log records the date, the claim, the outcome and any correction number, without personal details. An address is named in the log only with the complainant's consent or once the outcome is published.
- Notice before a negative change. A named operator is notified 3 days before a new negative verdict about it is published. A negative verdict is a new Circular flow verdict, a withdrawal, or a move to "Too early to judge". These notices begin on 9 November 2026. Notice cannot be given to an address with no known operator; for those, the dispute procedure is the remedy.
- No effect of a claim. Claiming a page does not change a verdict.
11. Governance and independence
- Decisions on disputes. Until an independent reviewer is appointed, disputes are decided by AQUIX; this is disclosed in the comply-or-explain table.
- Comply-or-explain. AQUIX publishes a table that sets out each provision of the codes it follows, states whether it complies, and, where it does not, explains how the objective is otherwise met. The table discloses, among other things, that AQUIX has no separate compliance officer, no independent complaints reviewer and no external audit at present.
- Conflicts of interest. AQUIX publishes every address controlled by AQUIX, any x402 payment made by them, and its legal form, each with a statement of whether it bears on any verdict. Once those addresses are listed, a release fails if a published address was paid by one of them.
- Payment and access. No payment changes a published figure or verdict; buyers and the public see the same data at the same time.
12. Confidential parameters
- What is confidential. The exact threshold values used by the classification rules. The form of every rule is public.
- Why. An operator who knew an exact threshold could keep an address just below it. No such case has yet been observed.
- How the objective is still met. The measured value for each address is printed on its page, so a reader can see where the address stands. A hash of the code that holds the thresholds is published for each methodology version, so that any change can be detected. The exact threshold is disclosed in confidence to a party disputing its own verdict. A threshold is published when it is retired.
- Public by design. The evidence requirement in section 6; the concentration threshold of 25%; the facilitator list with its dates; the exclusion list with its basis; and the catalog.
- Two values already in the public domain. Two values were shown on the site before this methodology and are treated as public rather than changed, because changing them would change verdicts: a shared funder is one that covers 50% or more of an address's payers; and a trend requires at least five measured days on each side, with at least $1 per measured day.
13. Limitations
AQUIX measures one network, one asset and one settlement scheme, and follows money one step. It does not establish whether an address is operated by an AI agent, who operates it, or whether its business is sound. Payment schemes outside section 2 and payments off the chain are not seen. Payer counts before 7 August 2026 are lower bounds. Facilitators recognised late leave earlier payments outside x402 revenue. The figures describe the chain at each nightly cut-off; they are not a live record. A classification is an opinion reached by published rules, not a finding of fact about any person, and the figures are not investment advice.
14. Data access and reproducibility
- Free access. Everything published on the site is free to read, including the full list of destinations for each address.
- Files. For each address: a JSON file with identifiers, and CSV files for each address and for the register. The verification files are published with the cut-off block.
- Reproduction. A script is published that recomputes the figures from any public Base node.
15. Version history
| Version | Effective | Change |
|---|
| v3.2 | In force before 2026.11 | Revised 17 September 2026: display labels and definitions rewritten; no rule, threshold or count changed. |
| 2026.11 | The first sealed release after launch (planned 18 November 2026) | This document. Published 5 November 2026 with a notice of change. |
To be finalised before the effective date
The following points are not yet decided and are not stated above: the treatment of very large addresses in the shared-funder test; how long records are retained; the legal entity that publishes AQUIX; and the appointment of an external reviewer. Each will be published here before the effective date. Any change to this methodology is published as a second, dated notice.
Source texts: notice sha256 b46c4a4fd781b8c35953da0ce101a8926d4684f9f2023a793182a3a750d4f289 · methodology sha256 821849524b517c6788c512ad9ffc6ba31ed4759fd133c7a03d988bd3fa1dd145.