Where blockchain analysis breaks down

The common-input heuristic assumes that all inputs to a transaction share an owner. A collaborative transaction breaks that assumption by construction: several parties contribute inputs and receive outputs, and the resulting transaction has no single owner. A CoinJoin is the clearest case, and it is the reason the heuristic is described as an assumption rather than a rule. An analyst who applies it to a CoinJoin produces a cluster that contains several unrelated people.

The limit is not absolute. Collaborative transactions have their own fingerprints — the number of participants, the equality of the outputs, the fee structure — and an analyst can sometimes identify them and exclude them from clustering. But the exclusion is itself an inference, and it fails when a collaborative transaction is designed to look ordinary. The general point stands: any heuristic that assumes a property can be defeated by a transaction that does not have it, and the ledger does not label which transactions those are.

There is a second-order effect that is often missed. A user who mixes successfully has broken the link between their inputs and their outputs, but the mixed outputs are now a new set of coins with their own history. If those coins are later spent together with unmixed coins, the link is rebuilt. Mixing is not a one-time operation; it is a state that has to be maintained, and the coin control page covers the discipline that maintains it.

A custodial service holds coins for many customers and moves them in transactions that have nothing to do with any individual. When an exchange sweeps deposits into a hot wallet, the transaction spends outputs belonging to hundreds of unrelated people, and the common-input heuristic merges all of them. The resulting cluster is not a person and not even a coherent entity; it is an artefact of the service's internal accounting. An analyst who labels it as a single actor has produced a false positive at scale.

False positives are the failure mode that matters most, because they are invisible in the output. A cluster is presented as a set of addresses believed to share an owner, and the belief is not accompanied by a confidence interval. The published research is explicit that clustering is probabilistic and that precision varies with the population being examined; a heuristic that performs well on consumer wallets performs badly on exchange traffic. The correct reading of any cluster label is "the analysis believes", and the strength of that belief is not published.

The practical consequence for a user is that being merged into a large cluster is not the same as being identified, but it is not harmless either. It means that any future identification of any address in the cluster reaches the user's coins. The defence is the same as everywhere else on this site: keep holdings apart, avoid reuse, and do not spend outputs together that have no reason to be spent together.