Analysis of on-chain transaction history to trace the provenance of digital assets and identify exposure to sanctioned or illicit sources. Providers group addresses into clusters, attach a label to the cluster, and measure how much of a balance traces back to a labelled source. The result is an exposure assessment — a measure of proximity to labelled activity — not a finding of fact about who controlled the coins. In a crypto-funded bullion purchase the analysis sits on the payment leg, applied as an AML and sanctions control by a licensed digital-asset platform under its own licence, before the seller receives anything.
What the analysis reads, and what it cannot read
A public blockchain records amounts, addresses, timestamps and the links between transactions. It records no identity, no purpose and no agreement. Analytics exists to bridge that gap: the chain supplies a complete transaction graph, and the provider supplies the names.
Two operations do the bridging. Clustering groups addresses likely controlled by the same party, most often through the co-spending heuristic — addresses that sign the same transaction share an owner — and change-address heuristics, which identify the output that returns the remainder of a spend to the sender on UTXO chains such as Bitcoin. Attribution attaches a real-world label to the cluster: an exchange, a mixer, a marketplace, a payment processor, a designated entity.
Labels come from a narrow set of sources, and that narrowness sets the ceiling on coverage. A provider transacts with a service to capture known addresses; it scrapes open-source material; it ingests data from its own customers; it works from law-enforcement and subpoenaed records. A service that no one has interacted with, and that no one has written about, is not labelled.
How an address becomes an exposure number
- Indexing. The provider maintains its own copy of the chains it covers, transaction by transaction.
- Clustering. Heuristics collapse many addresses into one entity. Clustering is conservative by construction: where a heuristic does not hold, the provider splits an owner into several clusters rather than merging two owners into one.
- Attribution. A label is attached to the cluster from the sources above. Unlabelled clusters stay unlabelled — there is no default category of “clean”.
- Tracing. Value is followed forward or backward across hops, each hop being one transaction between clusters.
- Exposure measurement. The share of a balance reaching the address from labelled counterparties is expressed by value and by proportion, split into direct and indirect.
- Scoring. The measure is compared against the screening party’s own risk policy, which decides acceptance, review or refusal.
Steps 1 to 5 are the provider’s. Step 6 belongs to whoever is accepting the money, and two firms reading the same exposure figure can reach opposite decisions without either being wrong about the chain.
Direct exposure, indirect exposure, and the convention that splits the value
Direct exposure is value received from, or sent to, a labelled cluster one hop away. Indirect exposure runs through intermediaries — a swap, a bridge, an unlabelled wallet — and weakens with every hop, though providers differ on how many hops they count and how they weight them.
Where a transaction has several inputs and several outputs, the tainted portion has to be allocated to specific outputs by convention rather than by fact. One common convention matches inputs to outputs in order, first in first out, mirroring the way contamination is calculated in case law. The practical consequence for a reviewer: an exposure percentage is a product of the tracing convention as much as of the transaction history, and two providers applying different conventions to the same wallet will report different numbers without either having made an error.
Who runs it in a crypto-funded bullion purchase, and at what point
The screening sits with the platform. On the crypto-funded purchase route, the counterparty pays through a gateway operated by a licensed digital-asset platform; the platform screens the wallet and the transaction under its own licence, converts, and settles the seller in national currency against the order reference. The seller holds no digital assets and performs no chain analysis of its own.
The seller’s controls run in parallel and answer a different question: who the counterparty is, who owns it, and whether it is subject to sanctions. Those checks apply to the party, not to the chain. Two records from the funding leg — the screening record and the conversion record — join the counterparty’s Evidence Set, alongside the contract, the payment confirmation, the Allocation Record and the instruction log.
When it runs, and what it costs
It runs before acceptance. Automated screening of an address returns in seconds and is built to sit inside a payment flow; a manual investigation of a flagged wallet does not, and is measured in analyst hours. The cost of the control is therefore in four parts — the provider licence, per-query screening volume, analyst time on the exceptions, and the settlement delay that a review imposes on an order with a price fixation attached to it. No public rate card governs the first three; they are negotiated per firm.
The assessment is also a point in time, and it ages. Designations arrive after transactions, not before them: in April and July 2026, OFAC added addresses linked to Iran’s central bank to the SDN List, and the issuer of the stablecoin held in them froze roughly $475 million across the two actions. Where the asset is a centrally issued token, a designation can immobilise a balance at the issuer level — the address remains visible on-chain, and the holding stops moving.
What the assessment leaves on file
A screening event produces a dated record of what was assessed: the addresses, the data the provider held at that moment, the exposure measured, and the decision taken. That record establishes that a control was applied and when. It does not establish the origin of the money.
That distinction is what keeps documentary source of funds a separate requirement rather than a duplicate one. Bank statements, sale contracts, exchange records and tax filings evidence where value came from in terms a bank’s own file can hold; the chain assessment evidences that the on-chain leg was examined. A counterparty file that carries one and not the other is short on either the money or the method. In the seller’s own model the client-facing status remains a fact rather than a report: Verified, with a date and an internal reference.
Where it fails
- Missing labels are the normal failure. Unattributed clusters are common, and their absence from a result is not a finding about them.
- Time. Coverage of a service improves once it is publicly operating and has been transacted with. Addresses used early in a service’s life are the least likely to be attributed, so historic funds trace worse than recent ones.
- Nested services. A small service holding its liquidity at a larger one inherits the parent’s label, and the entity actually handling the value stays invisible until someone subpoenas the host.
- Cross-chain movement. Bridges and swaps break a single-chain trace, and coverage of the destination chain is a separate question from coverage of the origin.
- Change-address false positives. Where fee mechanics such as child-pays-for-parent make a fresh change address behave abnormally, the heuristic can misidentify a deposit as change and propagate a label to a party that never used the service. Techniques such as PayJoin are built specifically to break these heuristics.
- Provider disagreement. Coverage varies enough between providers that investigative practice is to run two tools where budget allows, and to treat a single tool’s picture as one account rather than the account.
Blockchain analytics is not wallet screening, KYC, source of funds or the Travel Rule
| Control | Question it answers | Where its data comes from |
|---|---|---|
| Blockchain analytics | Where did this value come from, and what is it close to | The chain, plus the provider’s labels |
| Wallet screening | Is this address itself listed or high-risk | Sanctions lists and provider labels, at address level, before a transfer |
| Transaction screening | Should this individual transaction proceed | Sanctions, PEP and adverse-media data at the moment of processing |
| KYC | Who is the counterparty, and who owns it | Identity documents, registries, ownership disclosure |
| Source of funds | Where did the money for this transaction come from | Documentary evidence supplied by the payer |
| Travel Rule | Who is the originator, and who is the beneficiary | Data passed between regulated providers, not derived from the chain |
The first three are frequently described as one control and are not. Analytics reasons about history; wallet screening reasons about a single address against a list; transaction screening reasons about one payment at the moment it is processed. A firm can hold all three and still have no documentary source of funds on file.
Attribution is a lower bound, and the record changes form at conversion
The most precise public measurement of how well attribution performs comes from independent research rather than from providers. At USENIX Security 2025, researchers at Delft University of Technology published Ghost Clusters, comparing commercial attribution against ground truth recovered from the seized wallet data of three illicit services. Address coverage ranged from 24.54% for a mixing service to 94.85% for one marketplace, while false positives stayed under 0.5% in all three cases and as low as 0.01%. Traced as whole value flows rather than as address counts, fewer than three in ten flows through the mixer were covered at all; for the two marketplaces it was roughly nine in ten.
Read as a compliance instrument, that is an asymmetric error profile, and the asymmetry runs in one direction. A flag is strong evidence, because the heuristics almost never invent a link. Silence is weak evidence, because they routinely miss one. A screen that returns nothing has established that nothing in one provider’s label set matched — a lower bound on exposure, not a description of origin.
The other data layer does not close the gap. In its seventh targeted update on virtual assets, published in July 2026, the FATF reported that 83% of surveyed jurisdictions — 91 of 109 — had passed Travel Rule legislation, up from 73% a year earlier, and that close to half of those that legislated had taken no supervisory or enforcement action on it. Originator and beneficiary data that is legislated but not yet enforced arrives unevenly, which returns weight to the chain layer, which is the layer that reads as a lower bound.
What follows from this is a change of form rather than a shortfall. On-chain evidence ends at conversion. From that point the counterparty’s position is carried by different instruments: a fiat settlement bound to a single order reference, and, after allocation, bars identified by serial number in a dedicated client sub-account at Brink’s Hong Kong or Singapore, named to the counterparty in the vault register, allocated by serial and segregated from the seller’s own stock. Analytics answers where value came from. The Allocation Record answers what is owned and by whom. Neither answers the other’s question, and a counterparty file assembled for a bank, an auditor or a court has to carry both, in the order they were produced.
The controls that apply this analysis to a bullion transaction — counterparty verification, source-of-funds review and sanctions screening — are set out in AML & KYC Controls in Physical Gold Transactions.