Trade-based money laundering (TBML) conceals value transfer inside ordinary trade documentation, leaving the relevant evidence fragmented across customs, financial intelligence units, banks and registries. This conceptual paper extends Kupatadze's macro-meso-micro framework into a non-exclusionary, institutionally tiered screening architecture for the customs-observable subset of TBML. Macro corridor-commodity evidence, meso declaration-level anomalies and a heterogeneous entity-graph stream run in parallel over the full declaration universe and combine through late, decomposable fusion; macro evidence reorders priority but never excludes; a stratified sentinel stream preserves score-independent observation. The framework is disciplined by an explicit typology-detectability scope condition mapping ten TBML techniques to their observables, layers and blind spots; by candidate operator specifications for each stream; by a tier vocabulary separating public-data, customs-internal and gateway-conditional evidence under a statute-level legal mapping for Bangladesh; and by a three-stage prospective evaluation pathway, from operational-baseline analysis through shadow mode to an authorised sentinel pilot, designed around the selective-labels problem in enforcement data. Published institutional evidence anchors the application: financial-intelligence reporting concentration, a development-bank pilot on trade-related suspicious reporting and external gap estimates are read as convergent screening context, never as laundering measurements, and three worked illustrations span the directional range from revenue-motivated under-invoicing to outward value transfer. A situational-prevention and crime-script reading locates where the design plausibly raises effort and risk, and where displacement remains. The framework executes no model and claims screening plausibility, not effectiveness; its evaluation pathway defines how any performance claim would have to be earned.
No takes yet. Share an insight, caveat, or question.
Sushanta Paul (2026) studied this question.