Nine independent evidence streams. One two-signal guard. A transshipment fraud finding that names the corridor, the entity network, the FTA violation — and exactly what evidence would prove it wrong.
In a recent WCO global survey, more than 60% of customs administrations reported active transshipment fraud cases. For every verified seizure, analysts estimate 20 to 50 shipments pass through undetected.
The reason is structural. A Chinese factory ships solar panels to a Vietnamese warehouse. They are repackaged. A certificate of origin is issued — notarized, correctly formatted, stamped by a legitimate authority. The bill of lading says Vietnam. The declared FOB value is plausible. The HS code is correct. Each individual document is authentic. The fraud lives in the relationship between them.
Traditional risk systems check one signal at a time. A single anomaly can be explained away. Origin verification requires seeing all nine evidence streams simultaneously — and requiring independent corroboration before a finding fires.
A manifest arrives at the destination port. The declared origin is Vietnam. BorderRev's Rules of Origin model begins its analysis — nine modules, running in parallel, blind to each other. This is what it finds.
Customs manifest data flows into 9 independent scoring modules. Their outputs are fused by a weighted engine with interaction bonuses. A two-signal guard prevents single-module over-sensitivity from creating false positives.
Each module is a blind, independent scorer. The rulebook, corridor intelligence, entity network, route forensics, and statistical baseline each produce their own EvidenceAtom — with citations and a confidence weight.
A weighted sum fuses all module scores, with interaction bonuses when complementary modules corroborate each other. The two-signal guard returns a score of zero if fewer than two independent modules fire — preventing single-module over-sensitivity.
Every EvidenceAtom includes a falsification_template: the specific evidence that would refute the finding. This is what allows findings to be used in formal proceedings without being challenged as unfalsifiable.
The roo_rulebook module doesn't know what the corridor module found. The statistical module doesn't know what the entity network module found. Each scores independently against its own data source.
When 2, 4, or 6 of those independent evaluations reach the same conclusion about the same shipment — that's not a coincidence. That's a pattern. And it's exactly the kind of multi-source corroboration that distinguishes an intelligence finding from a false positive generated by a single noisy signal.
Any single evidence module can misfire. A statistical anomaly might reflect a legitimate pricing arrangement. A flagged corridor might carry clean cargo. A short dwell time might have a legitimate explanation.
The two-signal guard is the architectural answer: if fewer than two independent modules fire, the composite score is set to zero and no finding is issued. This is not a threshold — it is a hard architectural constraint. The system cannot produce a finding on the basis of one module's assessment alone.
When six modules fire, each having evaluated a different dimension of the same shipment independently, the case is not circumstantial. It is convergent.
The roo_rulebook module embeds the Rules of Origin for major trade agreements — VJEPA, USMCA, ASEAN Free Trade Area, RCEP, and others. For each shipment, it identifies the applicable FTA corridor, retrieves the relevant rule for the declared HS code, and evaluates whether the origin claim can be satisfied given the observed trade pattern.
The four rule types — wholly_obtained, specific_process, tariff_shift, and rvc (regional value content) — each imply different minimum production requirements. A 4-day warehouse dwell satisfies none of them for solar panels.
An unfalsifiable allegation is not an intelligence finding — it is an accusation. In a formal dispute, a finding without a falsification path will not survive scrutiny. BorderRev was designed with this constraint built in from the start.
Every EvidenceAtom produced by every module includes a falsification_template: the specific, concrete evidence that would refute the flag. Not "prove your innocence" — but "here is exactly what would make this finding wrong, and we will withdraw it if you produce it."
That is the standard that holds up in a WTO dispute panel, a court proceeding, or an administrative review. That is the standard BorderRev produces.
BorderRev's Rules of Origin model connects to your customs manifest feed — or runs against a sample of historical data for a no-risk evaluation. It works with any country's manifest format. We return findings with full evidence chains and falsification templates. You decide what to do with them.