This paper proposes a structural framework for detecting and interrupting crypto investment fraud on social platforms before financial loss occurs. Existing platform responses often treat fraud as isolated deceptive posts, malicious links, or account-level violations. This paper argues that such framing is insufficient because many real-world scams unfold as staged behavioral sequences rather than single artifacts. The proposed framework models crypto investment fraud as a multi-step conversion process involving initial contact, trust-building, migration into direct messages or off-platform messaging apps, low-friction financial solicitation, proof-of-profit theatrics, escalating transfer requests, and eventual withdrawal obstruction. On this basis, the paper introduces a pre-transaction intervention architecture composed of five components: sequence-based detection, graph-based coordination analysis, graduated friction interventions, registry-aware verification for financial solicitation, and structured victim-report intake. The paper further argues that sequence-based modeling is more robust than post-level moderation against linguistic evasion and surface-level variation, because scammers can rewrite words more easily than they can conceal the underlying conversion structure. It also identifies several deployment-oriented research challenges, including pre-migration prediction under off-platform observability loss, dynamic infrastructure-reuse analysis, cold-start transfer from adjacent scam archetypes, and adversarial sequence adaptation. More broadly, the paper suggests that platform safety should be evaluated not only by how effectively a system removes harmful content, but by how early it can interrupt harmful conversion processes before irreversible loss occurs.
Hiroko Konishi (Thu,) studied this question.