Appeals court upholds Bitcoin Fog conviction, accepts Chainalysis tracing with no known error rate

On 25 September 2026 the US Court of Appeals for the D.C. Circuit affirmed all four convictions of Roman Sterlingov, found to have run the Bitcoin Fog mixer, and his sentence of 150 months. The case is United States v. Sterlingov, No. 24-3161. Judge Wilkins wrote for a unanimous panel.

The ruling matters beyond one defendant because of what it says about blockchain tracing as evidence.

What the court accepted

The government's case rested on Chainalysis Reactor, software that groups addresses into clusters and attributes them to a service. Reactor tied more than 900,000 addresses to Bitcoin Fog, with eight darknet markets sending it roughly 80,000 bitcoin.

The opinion states the weak point plainly: "Both experts admitted that Reactor was not peer-reviewed and that they did not know the error rate for the software." It then holds that this does not make the testimony unreliable. The usual tests for scientific evidence, the court says, "are not exhaustive", and it lists what it relied on instead:

• the experts could not recall a case in which Reactor's attribution had been wrong;
• law enforcement and industry use the tool widely;
• a government contractor confirmed that Reactor "correctly clustered and attributed 99.9% of a large number of addresses in a different case";
• Reactor found four of the five addresses Bitcoin Fog used in the government's own undercover transactions.

Sterlingov's lawyers read that last point as a 20% error rate. The court adopted the trial judge's view that the tool is "deliberatively conservative and thus underinclusive". On the missing peer review, a footnote says some technology is too new to have generated independent research, and adds: "Blockchain tracing certainly falls within this bucket."

The source code stays closed

The defence asked for Reactor's source code and did not get it. Defence counsel and their experts were shown the heuristics under a protective order. Sterlingov himself was barred from seeing them, on the reasoning that they could serve as a road map for evading tracing. The appeals court found no error.

No questions asked, held against the operator

One passage will be read closely by anyone who runs a service without accounts. Explaining why the jury could be instructed on wilful blindness, the court notes that Bitcoin Fog "didn't have an account-verification process and asked zero questions of users" and "auto-deleted its records every week". A jury, it says, could infer the service was built that way so that its operator could avoid knowing what users did.

What this changes

For users, nothing overnight. But the same clustering heuristics are applied to every mixer and every CoinJoin transaction, and a federal appeals court has now said their output can go to a jury without published error rates and without the defence reading the code. A request for rehearing or a Supreme Court petition remains possible; none had been reported at the time of writing.

This is a separate case from the Tornado Cash prosecution, whose retrial moved to April 2027. Mixers listed in this catalogue are in the mixers category.

Sources: US Court of Appeals, D.C. Circuit — opinion, No. 24-3161, Armstrong & Bradylyons — analysis, 27 Sept 2026