Mirror mismatches, corridor concentration, and OFAC partner anomalies, surfaced from UN Comtrade public data.
trade-finance-engine ingested UN Comtrade bilateral trade-flow records for 2023 across five HS6 commodities (refined petroleum, gold, motor cars, wheat, integrated circuits) and eight major reporters (USA, China, Germany, Japan, UK, France, Russia, Netherlands). Tier 1 ran three deterministic checks: mirror-trade mismatch (Global Financial Integrity misinvoicing methodology), high corridor concentration (Basel large exposures framework), and sanctioned-partner anomaly (OFAC SDN cross-reference). The same checks ship for customer engagements gated on lob = trade_finance_counterparty.
UN Comtrade bilateral trade-flow register, real public data.
Source. The UN Statistics Division publishes bilateral trade-flow data at comtradeapi.un.org, documented at uncomtrade.org/docs. Each row is a (year, period, reporter, partner, commodity, flow) cell with reported value in USD and net weight in kg. Free anonymous tier permits roughly 500 requests per day; production engagements hold a paid API key.
What we ingested. 2023 annual data for HS6 codes 2710 (refined petroleum), 7108 (gold unwrought), 8703 (motor cars), 1001 (wheat), and 8542 (electronic integrated circuits) across eight major reporters. Loaded into trade_finance.trade_flows with idempotent upserts on the (year, period, reporter, partner, commodity, flow) primary key.
What is mirror-paired. For every reporter-export-to-partner row, the engine self-joins on the matching partner-import-from-reporter row (same year, period, commodity_code) and computes the bilateral asymmetry. Persistent divergence above 30% is the canonical Global Financial Integrity trade-misinvoicing signal.
Why this matters. Trade-based money laundering is the single largest cross-border value-transfer channel that does not pass through correspondent-bank settlement. GFI estimates illicit financial flows from misinvoicing at $800B-$1.6T per year. Custodian banks, trade-finance lenders, and export-credit insurers each carry exposure; the UN Comtrade register is the only common public reference point for the bilateral pattern.
What ships when a trade-finance customer engages.
Three production checks gate on the customer profile lob = trade_finance_counterparty. Each runs deterministically against ingested bilateral data and produces sealed CREB output through the same orchestrator and Ava layer that powers the rest of the platform.
tf_mirror_mismatch
Bilateral asymmetry on (year, reporter, partner, commodity). Flags divergences above 30%. The classic GFI misinvoicing signal. Reference: Global Financial Integrity methodology, FATF Recommendation 32, UN Comtrade bilateral asymmetry.
tf_high_corridor_concentration
Single (reporter, partner, commodity, flow) corridor accounting for 60% or more of the reporter''s total trade in that commodity. Basel III large-exposures and BIS country-risk concern. Critical when partner is sanctioned.
tf_sanctioned_partner_anomaly
Trade flow with a country on the OFAC SDN, EU consolidated, or UN sanctions list. Cross-references the seed list and the live compliance.sanctions_list. Reference: OFAC SDN, OFAC 50 Percent Rule, EU consolidated, UN Security Council.
Pattern: USA exports refined petroleum to Turkey.
Below is a single corridor (USA exports of HS6 2710 to Turkey, 2023 period) with the kind of asymmetry the check fires on. The reporter side reported 500M USD; the mirror side reported 250M USD. The 50% gap exceeds the 30% threshold. This is illustrative of the pattern; live customer engagements run against the customer''s actual bilateral records and surface every corridor that crosses the threshold.
| Year | Reporter | Partner | HS6 | Description | Reporter export USD | Partner import USD | Delta | Tier 1 signals |
|---|---|---|---|---|---|---|---|---|
| 2023 | USA | TUR | 2710 | Petroleum oils, refined | $500,000,000 | $250,000,000 | 50.00% | MIRROR-MISMATCHGFI-MISINVOICINGCRUDE |
Pattern: BLR exports of wheat concentrated on RUS.
Below is an illustrative concentration finding. Belarus reports total wheat exports of 500M USD across four partners. RUS accounts for 425M (85%), POL for 35M (7%), CHN for 25M (5%), IND for 15M (3%). The 85% share crosses the configured 60% threshold. The financing bank carries Basel III large-exposure risk; the partner is on the OFAC SDN program (Russian Federation, Executive Order 14024); the underlying flow becomes a stacked finding: corridor concentration plus sanctioned-partner anomaly.
| Reporter | Partner | HS6 | Flow | Corridor USD | Reporter total USD | Share % | Tier 1 signals |
|---|---|---|---|---|---|---|---|
| BLR | RUS | 1001 | export | $425,000,000 | $500,000,000 | 85.00% | CONCENTRATION-HIGHSANCTIONED-PARTNERBASEL-LE |
| BLR | POL | 1001 | export | $35,000,000 | $500,000,000 | 7.00% | BASELINE |
| BLR | CHN | 1001 | export | $25,000,000 | $500,000,000 | 5.00% | BASELINE |
| BLR | IND | 1001 | export | $15,000,000 | $500,000,000 | 3.00% | BASELINE |
Eight country-level OFAC programs cross-referenced.
Seed list for tf_sanctioned_partner_anomaly. Joined to trade_flows.partner_country at run time. The live OFAC SDN feed is refreshed from compliance.sanctions_list on a separate ingestion cycle; this seed is the public-data corroboration baseline.
| ISO | Country | Program | Effective from | Notes |
|---|---|---|---|---|
| IRN | Iran | OFAC SDN | 1979-11-04 | Comprehensive embargo, ITSR. |
| PRK | Korea, Democratic Peoples Republic of | OFAC SDN | 2008-06-26 | Comprehensive embargo, NKSPEA. |
| CUB | Cuba | OFAC SDN | 1962-02-07 | Cuban Assets Control Regulations. |
| SYR | Syrian Arab Republic | OFAC SDN | 2004-05-11 | Syria Sanctions Regulations. |
| RUS | Russian Federation | OFAC SDN | 2014-03-06 | Sectoral sanctions, Executive Order 14024. |
| BLR | Belarus | OFAC SDN | 2006-06-16 | Belarus Sanctions Regulations. |
| VEN | Venezuela, Bolivarian Republic of | OFAC SDN | 2015-03-09 | Sectoral sanctions on government and PDVSA. |
| MMR | Myanmar (Burma) | OFAC SDN | 2021-02-11 | Burma Sanctions Regulations, Executive Order 14014. |
What the customer takes to a regulator.
A single mirror-mismatch finding rendered as a sealed CREB record. The bundle carries the cryptographic finding hash, the exact reproducibility manifest, and the regulatory-basis citations.
Deterministic, reproducible, court-defensible.
Deterministic
Each check is a SQL aggregate over a public bilateral dataset. Same input register, same self-join, same threshold, every run.
No external LLM
The Tier 1 verdict path is rule-based. Ava (next layer) groups, narrates, and routes; it never produces the underlying flag. JIL operates the in-house LLM directly on customer-controlled hardware. No OpenAI, Anthropic, or Vertex API.
Replay manifest
Every CREB carries the source-dataset hash, code version, schema DDL, query plan, and signal thresholds. A third party with the same inputs replays the analysis bit-identically.