The D.C. Circuit Court of Appeals upheld Roman Sterlingov's conviction and 150-month sentence for operating Bitcoin Fog, a service designed to obscure cryptocurrency transaction trails. The September 25 decision affirmed the admissibility of expert testimony based on Chainalysis's proprietary Reactor software, rejecting Sterlingov's challenge to the analysis method.
Sterlingov's defense sought access to Reactor's source code, arguing it was necessary for his legal representation. The trial court provided access to the software's heuristics and underlying assumptions under a protective order but restricted direct access to the code itself due to concerns about circumvention. The appellate court found this arrangement sufficient and noted that Sterlingov's experts declined to review the materials regardless of the protective order's terms.
How the Evidence Was Assessed
Under Federal Rule of Evidence 702, courts determine whether expert testimony is reliable enough for admission by evaluating whether it rests on sufficient facts, uses reliable principles and methods, and applies those methods reliably. The appeals court applied this standard flexibly, recognizing the trial judge's discretion in assessing reliability.
FBI witness Luke Scholl and Chainalysis analyst Elizabeth Bisbee testified that Reactor attributed more than 900,000 Bitcoin addresses to Bitcoin Fog. The software employs three clustering methods: co-spend clustering, which links addresses used together in transactions; behavioral clustering, which identifies matching transaction patterns; and intelligence-based clustering, which incorporates information from data leaks and court documents.
Reliability Markers
Both Reactor witnesses acknowledged the software had not undergone peer review and that its error rate was unknown. The court nevertheless identified other indicators of reliability, including the witnesses' experience, comparisons with information obtained through legal processes, use by law enforcement and private-sector analysts, and corroboration of results.
During undercover transactions with Bitcoin Fog, the government identified five known service addresses. Reactor's cluster contained four of them. The trial court characterized Reactor as conservative and underinclusive rather than attributing the discrepancy to a 20 percent error rate as the defense suggested.
The court cited a government contractor's finding that Reactor correctly clustered and attributed 99.9 percent of addresses in a large group from another case, though the opinion noted this result was specific to that case rather than a broader product-wide error rate.
Constitutional Arguments
Sterlingov raised constitutional objections regarding access to the full software source code. The appellate court distinguished between these newly advanced claims and a previously raised objection concerning personal access to the heuristic materials. The panel rejected both Confrontation Clause arguments and an independent due-process theory, finding that Sterlingov had not demonstrated material impairment to his defense or adequately explained why the source code was necessary.
The district court had indicated willingness to provide the code if justified but found no sufficient explanation for why it was material or relevant to the defense.
Evidentiary Implications
The decision establishes that explained methods, corroboration, and protected access to proprietary methodology can support admission of blockchain tracing testimony without full source-code disclosure. Service attribution and personal ownership remain distinct evidentiary questions requiring separate treatment.
