Unstable

The word volume often conceals the unit being counted. The minimum taxonomy is longer than the headline (Committee on Payments and Market Infrastructures 2016).

The asymmetry runs in one direction. A known payment can produce one or more ledger records; ledger records do not uniquely recover the payment that produced them. Inference can move from verified purpose to transfer, and it cannot move back without additional evidence.

The strongest public evidence establishes a unit-of-observation problem, not a global payment total.

Schär et al. (2026) analysed 141,310,373 Ethereum transactions involving USDT, USDC, or PYUSD during calendar 2025. Those transactions emitted 241,423,906 stablecoin transfer events inside a dataset of 593,096,940 event logs. The authors classify 31.6% of the transactions as complex; because complex transactions emit several events, 59.96% of transfer events occurred inside that class. Their inputs came from an archive node, public event signatures, and protocol factory contracts rather than proprietary end-user labels.

That is unusually strong evidence that event counts and transaction counts are different objects. The same record fixes the boundary of the finding: the study covers three tokens on Ethereum, not every token and chain. Transaction complexity identifies structure; it does not observe an invoice or prove that every simple transfer purchased a good or service.

Visa’s public dashboard addresses another layer. Its adjusted methodology counts only the largest stablecoin movement inside a transaction, uses more than three million probabilistic address labels, and applies thresholds of 1,000 transactions or $10 million over 30 days to certain unlabelled addresses. The filters are explicit and useful. Yet the retained categories include centralised-exchange deposits and withdrawals, decentralised-exchange activity, lending, investment funds, ramps, and minting and burning. Adjusted therefore means adjusted under that model, and it does not mean payments (Visa and Allium 2026).

The published payment studies proceed by adding purpose assumptions:

The proximity of the two payment estimates is suggestive. It is not independent replication. The time bases differ, the address labels are unavailable, the models make different boundary choices, and neither publication supplies a ground-truth validation set, confusion matrix, or executable pipeline.

The raw denominator is unstable for the same reason. McKinsey reports public estimates of up to $35 trillion for 2025; BCG and Allium report approximately $62 trillion; the BIS Annual Economic Report reports an estimated $28 trillion using another provider. Token coverage, chain coverage, anomaly handling, internal-event treatment, price conversion, and the distinction between transfer events and transactions differ. These figures cannot be treated as three measurements of one fully specified quantity (Higginson et al. 2026; Batra et al. 2026; Bank for International Settlements 2026).

Visa reported $14.2 trillion of payment volume for the fiscal year ended 30 September 2025. Its filing describes payment volume as activity on Visa products for purchased goods and services; nominal volume is provided by its financial-institution clients and is subject to Visa’s review. Cash activity sits in a separate total-volume measure (Visa Inc. 2025a).

That statistic is not commensurable with a calendar-year stablecoin transfer aggregate. It has a defined commercial function, a different reporting period, a network and client population, and institutional metadata that a public token log does not contain. Placing the figures beside each other can illustrate the magnitude of unlike systems, and it cannot establish relative payment adoption.

Visa’s own stablecoin programme shows why the layers must remain separate. In December 2025, Visa said its monthly USDC settlement activity had exceeded a $3.5 billion annualised run rate. Participating issuers and acquirers use USDC to discharge obligations to Visa. Visa stated that the consumer card experience does not change. A cardholder can therefore make an ordinary card payment while USDC appears later as an institutional settlement asset, and counting the settlement leg as a second consumer payment would confuse two valid records of different obligations (Visa Inc. 2025b).

USDC makes a useful case: Circle’s audited filing reveals both the issuer economics and the limits of analogy. Circle reported that reserve income was 96.0% of total revenue in 2025. At 31 December 2025 it reported $75.266 billion of USDC in circulation and approximately 88% of USDC reserves in the Circle Reserve Fund, a government money-market fund managed by BlackRock; the remainder was principally cash (Circle Internet Group 2026).

Those facts support a precise statement: Circle’s 2025 revenue was overwhelmingly sensitive to reserve balances and yields. They do not turn Circle into a money-market fund administrator, and they do not establish an equity valuation. A USDC holder holds a token subject to contractual and jurisdiction-dependent redemption rights. Ownership of a share in the Circle Reserve Fund does not follow, and Circle maintains the reserve arrangement and assumes the issuer’s redemption obligation.

Access to that obligation also depends on status. Circle’s USDC terms outside the European Economic Area distinguish Circle Mint customers from other holders. A non-customer generally cannot redeem directly with Circle unless eligible for and accepted into Circle Mint; EEA holders are governed by a separate white paper and legal regime. The terms also reserve compliance, blocking, and delay powers and state that USDC is not deposit-insured. "Fully backed," "directly redeemable," and "cash equivalent" are therefore not synonyms (Circle Internet Financial 2025).

The legal label adds another boundary. The GENIUS Act became Public Law 119-27 on 18 July 2025. It defines a payment stablecoin partly by design and an issuer’s fixed-value redemption obligation, while excluding national currency, deposits, and securities. That statutory definition does not classify each observed transfer by user purpose (U.S. Congress 2025).

The Act’s operative framework was not yet in force on 15 July 2026. Section 20 sets effectiveness at the earlier of 18 months after enactment or 120 days after primary federal regulators issue final implementing regulations. The official materials available on that date were proposed rules: the OCC’s principal proposal, the FDIC’s prudential proposal, Treasury’s AML and sanctions proposal, and a joint customer-identification proposal. The Act’s one-to-one reserve, disclosure, and redemption requirements were therefore not yet the operative federal basis for every U.S. stablecoin claim (Office of the Comptroller of the Currency 2026; Federal Deposit Insurance Corporation 2026; U.S. Department of the Treasury 2026; Federal Reserve Board 2026).

A reserve disclosure cannot establish payment purpose, and a payment classifier cannot establish reserve quality. A statutory product definition cannot do either job retroactively.

The available record does not identify a causal chain from aggregate-volume analogies to S-1 framing, acquisitions, venture funding, or market pricing.

Circle’s filing reports both high on-chain activity and a reserve-income-dominated business model. Visa reports both card payment volume and stablecoin settlement. These disclosures show that institutions track several layers at once. They do not reveal which metric investors relied on, what counterfactual valuation would have prevailed, or whether a funding decision changed because of a Visa comparison.

Goodhart’s law would require evidence that protocol operators targeted a stablecoin-volume statistic. Multiple event logs and routing legs arise from system design whether or not anyone optimises a headline. The demonstrated problem is construct validity: the measure does not uniquely represent the economic concept assigned to it.

The market inference stops there. Anyone using raw or mechanically adjusted transfer value as a proxy for payment adoption assumes a bridge that the ledger does not supply. Whether that assumption affected a price is a separate empirical question.

Four rules are enough to prevent the category collapse.

Stronger claims require a reproducible classification pipeline and an independently labelled validation sample, or records such as invoices and platform data that connect transfers to underlying obligations. A larger transfer total or agreement between proprietary models cannot supply that connection on its own.

The sources fix both population and period. Transaction-level evidence covers USDT, USDC, and PYUSD on Ethereum during calendar 2025. The payment estimates cover December 2025 activity at an annual rate in one case and calendar 2025 in the other. The three raw denominators come from different compilations: public estimates reported by McKinsey, Allium’s figure via BCG, and another provider via the BIS Annual Economic Report, each applying different coverage and handling rules. Visa’s payment volume covers the fiscal year ended 30 September 2025, and its USDC settlement figure is an annualised run rate for monthly settlement activity, stated in December 2025. Circle’s circulation and reserve figures are dated 31 December 2025, and federal regulatory status is stated as of 15 July 2026.

Each result sits on one layer and does not travel to another. A ledger record reports that balances changed; an adjustment model reports what a stated filter removed. Purpose attribution requires metadata, validated labels, counterparties, surveys, or a stated behavioural model, which a public token log does not supply by itself. Corporate filings and terms report obligations an issuer or a network has assumed, while a statute reports a legal category. No finding here may be carried from one of those layers to another without evidence that crosses the boundary explicitly.

Several questions fall outside the boundary: payment purpose for any individual transfer, a global payment total, relative payment adoption between card and stablecoin rails, continuous reserve sufficiency under stress, an unconditional claim by every holder on each reserve asset, an equity valuation for Circle, and any causal chain from aggregate-volume analogies to S-1 framing, acquisitions, venture funding, or market pricing. Whether protocol operators targeted a stablecoin-volume statistic falls outside what this record identifies, and whether a proxy assumption affected a price is a separate empirical question.

The affirmative claim is narrow: a recorded stablecoin transfer establishes that a token balance changed under a protocol, and it establishes payment purpose, settlement, redemption, reserve quality, or legal category only where separate evidence connects those layers. Two of the five observations below would defeat that claim; three would defeat findings it rests on, and those three name the sections they would unseat.

Three of the five resolve against material a reader can obtain today: Ethereum event logs behind the first and third conditions, and Circle’s published terms behind the fifth. The other two wait on a released classifier and on an aggregate built to Visa’s definition, neither of which appears in the sources read here, and that absence is part of what this note reports rather than a defect in the test. None of the five asks for a forecast, which keeps refutation conditions apart from the measurement programme set out above.

Stablecoins can be useful for payment, trading, collateral, treasury, and settlement. Precision does not require choosing one function as the real one. It requires refusing to let movement impersonate purpose.

The ledger tells us that value moved. Scholarship begins where that sentence ends.