On September 18, Visa refreshed its Onchain Analytics stablecoin dashboard. The adjusted volume line dropped. The transaction count line fell less than 2%.
That divergence is the whole story, and most of the coverage got it backwards.
I have watched enough exchange tape to know what a real demand break looks like. When payments stop, counts break first. Wallets stop initiating. Merchants stop settling. Value and count fall together, and the ratio between them stays roughly stable through the decline. That is not what happened here. Counts held flat. Value reset. When the numerator moves by a large multiple and the denominator does not move at all, you are not watching behavior. You are watching a ruler get shorter.
Visa did not publish a decline in stablecoin payments. Visa published a change in what it is willing to call a payment. That distinction is load-bearing for anyone whose model ingests that series. Ledgers don't lie. Definitions do.
Visa Onchain Analytics runs on data supplied by Allium. The product exists for one reason: the raw number — total stablecoin transfers on a chain — is worthless as a payment metric. Institutional desks have known this for years. Retail dashboards have not caught up.
The September 18 changelog documents two changes. First, the identity label set expanded from roughly 15 million tagged addresses to roughly 600 million. A 40x expansion in a single refresh. Second, Visa revised its heuristic filters, adding detection for what it calls short-term routing: the same asset bouncing through a chain of wallets inside a compressed time window.
Both changes do the same mechanical thing. They reclassify. Addresses that were anonymous become exchanges, contracts, or bots. Once labeled, their transfers get pulled out of the "adjusted" series — the line Visa presents as organic activity. Coverage spans Ethereum, Tron, Solana, and several others.
Visa named a specific case on Solana: an automated program cycling the same stablecoin through thousands of single-use wallets. Under the old rules, that flow counted as volume. Under the new rules, it does not.
There is a documentation defect sitting on top of all this. Visa's live methodology page still reads "over 3 million" labeled addresses. The changelog says 600 million. Two figures for the same product, two orders of magnitude apart, both published by a company whose core business is moving money accurately.
The adjusted stablecoin volume is a filtered flow, not a measurement. Three numbers get conflated constantly: stock, flow, and filtered flow. USDT market cap sits near $183.79 billion — that is stock, a point-in-time snapshot of outstanding tokens. Twenty-four-hour transfer volume on the same page is flow, everything that moved. Visa's adjusted number is flow after classification. Same asset, three instruments, routinely quoted interchangeably.
Deleting non-payment flow is correct methodology. But the sizing matters.
The BIS has published work that should have ended the debate already. Its researchers found that close to 60% of transfer events on Ethereum occur inside complex transactions — a single transaction containing multiple token movements and financial operations. That is not payment. That is DEX swaps, liquidations, collateral rotation, arbitrage legs. The BIS also drew a clean line between chains: Tron skews toward holding and transfer, Ethereum skews toward smart contract interaction. The same stablecoin, on two chains, doing two different economic jobs.
So the honest starting position was never "stablecoin volume equals payments." It was "we do not know what fraction of this is payments, and nobody has been incentivized to find out."
Visa is now incentivized. That is the real story.
I built an arbitrage bot in the DeFi Summer of 2020 that ran 15,000 transactions in three months between Uniswap and Sushiswap on a $500,000 base. Every one of those transactions was real. Every one paid gas. Every one moved value. Not one of them was a payment. That is precisely the class of activity Visa's routing heuristic is now designed to strip out. I know how those flows look because I generated them.
The routing heuristic is anti-evasion, and it is a moving target. Programs that manufacture volume adapt. Bots that generated clean-looking transfers in 2023 look like short-term routing in 2025. The label set is not a fixed asset. It is a live game, and Visa just made a large move.
Here is the transparency gap. Visa did not publish the size of the revision. No before-and-after total. No decomposition showing how much of the change came from label expansion versus the new routing rules. In eleven years of risk work I have learned one hard rule: when a vendor changes a methodology and does not publish the bridge, the bridge is where the information lives. Conviction without verification is just gambling.
There is a second consequence, and it is the one desks will get wrong. The series now has a regime break. Pre-September 18 and post-September 18 are different instruments measuring overlapping but non-identical things. Plot them on the same axis and you generate fiction with a clean trend line. Structure survives the storm; chaos does not.
The over-exclusion risk is real and undisclosed. Heuristic filters produce false positives. A legitimate merchant settlement routed through an intermediary wallet that sweeps and forwards looks, structurally, like short-term routing. A treasury operation consolidating balances across cold and hot wallets looks like a bot. Visa has not published an estimated error rate for the new rules. Without one, you cannot distinguish a precise filter from a blunt one, and you cannot tell whether the volume Visa removed was manufactured or merely inconvenient.
For Solana specifically, the implications cut deeper. The chain's stablecoin volume narrative has leaned on high throughput as evidence of real usage. If a meaningful share of that throughput was single-use wallet cycling, the adjusted figure will diverge from the raw figure permanently, and the gap becomes a footnote on every Solana usage chart built after September.
The retail read is that Visa cut stablecoin volume and therefore payments collapsed. That read is wrong in direction and wrong in mechanism.
Two structural facts beat it. First, the payment narrative was inflated long before September 18. The BIS 60% figure alone told you the headline number was mostly machine traffic. Second, Visa's incentive is not to make stablecoins look small. Visa's incentive is to make its own series the reference standard for what counts as a stablecoin payment. Defining the classification is the product. Every filter Visa adds increases the value of its data and decreases the comparability of everyone else's. That is not a bearish signal on stablecoins. It is a competitive move in the data market.
The genuinely bearish detail is buried, and it is not about Visa. It is the arithmetic of the label set. 600 million addresses is a large number. It almost certainly includes a long tail of one-time and low-activity wallets — the same population the routing heuristic targets. Coverage at that scale buys recall at the cost of precision. And there is no denominator: Visa's adjusted volume is a subset of an unknown universe, so nobody outside Visa can audit what was kept versus what was thrown away.
Discipline turns noise into a tradable signal. Right now the tape is noise. Nobody has a bridge.
Treat September 18 the way you would treat a stock split or an index reconstitution. Date-stamp every chart. Do not extend pre-reset trends forward. If you need a comparable series, either wait for Visa to publish a bridge or reconstruct it yourself from raw chain data — that is a two-week job and it is worth doing once.
The number that matters is not the level. It is the ratio of adjusted to raw volume. If that ratio keeps compressing while transaction counts hold flat, something is still manufacturing activity that has not been labeled yet.
Watch the labels, not the payments. Alpha hides in the friction between chains.
