Methodology and limitations
TradeGuard is a public, client-side demonstration of transparent review rules. It prioritises evidence and human judgement; it does not make autonomous compliance decisions.
Implementation boundary
The public v37 build uses deterministic browser-side rules and does not call an AI API or make an autonomous compliance decision. “RegTech Nexus AI” is the publisher name; the public workflow should be understood as explainable rule-based review support. Any future AI component would require separate documentation, testing and governance.
TBML Check
The TBML workspace structures a trade case around goods, price and valuation, HS-code alignment, quantity and unit, documents, route and ports, related parties or UBO concerns, third-party payment, repeated or unsupported shipments, business-profile mismatch and unusual payment terms.
Additional control-context fields cover jurisdiction, PEP and sanctions status, restricted or dual-use goods, payment transparency and source of funds or wealth. These are control findings and readiness information, not automatic sanctions, STR or legal conclusions.
TBML indicator weights in the public build
- Material price deviation: +25; significant price deviation: +15.
- Related-party concern: +15; third-party payment: +10.
- Route or port anomaly: +10; document inconsistency: +15.
- Duplicate or unsupported shipment concern: +20.
- Goods / HS-code mismatch: +15; quantity / unit inconsistency: +10.
- Business-profile mismatch: +10; unusual payment terms: +10.
Control-only context findings, including jurisdiction, PEP/sanctions status, restricted goods, payment transparency and source of funds, do not add numeric points in this public build.
Indicator-to-reference crosswalk
The following crosswalk explains the standards and practice families that inform the public vocabulary. It is not a legal opinion, a certification or a complete jurisdiction-specific mapping. Each institution must configure the applicable FIU, regulator, customs, sanctions and trade-finance requirements for its market.
| Public indicator | Reference families | Local configuration required |
|---|---|---|
| Price / valuation anomaly | FATF TBML typologies; Wolfsberg trade-finance principles; ICC/BAFT practice materials | Customs valuation, FIU and institutional benchmark sources |
| Goods / HS-code mismatch | FATF TBML typologies; WCO HS/customs concepts; ICC/BAFT practice materials | National tariff, customs and restricted-goods references |
| Quantity / unit inconsistency | FATF TBML typologies; WCO customs concepts; trade-document controls | National customs and product-unit conventions |
| Document inconsistency | FATF TBML typologies; FATF Recommendations 10 and 11; Wolfsberg principles | Documentary-credit, customs and record-keeping requirements |
| Route / port anomaly | FATF risk-based approach and TBML typologies; Recommendations 1, 6 and 7 | Country-risk, sanctions, shipping and customs data sources |
| Related party / UBO concern | FATF Recommendations 10, 24 and 25; Wolfsberg CDD principles | Local beneficial-ownership, CDD and company-registry rules |
| Third-party payment | FATF Recommendations 10 and 16; Wolfsberg and BAFT payment-transparency practice | Local payment, FX and wire-transfer requirements |
| Multiple / phantom shipment | FATF TBML typologies; Recommendation 11; trade and transport-document controls | Customs, shipping, carrier and internal duplicate-detection sources |
| Business-profile mismatch | FATF risk-based approach and Recommendation 10 CDD expectations | Local KYC, customer-risk and sector guidance |
| Unusual payment terms | FATF Recommendations 10 and 16; ICC/BAFT trade-practice materials | Institutional product, FX and trade-finance policy |
Price-deviation rule
When comparable positive values are available, the deviation is 0% when the declared unit price is within the reviewer-entered market range. Above the upper bound it is calculated as (declared price − upper bound) ÷ upper bound × 100; below the lower bound it is (lower bound − declared price) ÷ lower bound × 100. The current public thresholds are +15 at 50% or more and +25 at 100% or more. No live market-price feed is used; the reviewer supplies the range.
Risk bands, cap and duplicate prevention
Decision-ready bands are Low 0–24, Medium 25–49, High 50–74 and Critical 75–100. Raw scored points are capped at 100 for the TBML workspace. The six published Transaction Monitoring patterns currently total 95 points; its 100 cap is a defensive guard for future configured rules and does not activate under the current six-rule set.
Quick toggles and structured indicators use the same-concept suppression rules in the browser code, so a price, document, route, related-party, third-party-payment or goods/HS signal is not counted twice when one source already produced the corresponding flag.
Decision-readiness gate
Data completeness is the percentage of configured review fields that are non-blank. Data integrity is assessed separately for comparable currency, unit, HS-linked unit profile, price range and related context. “Unknown / not provided” is treated as not assessed, not as clean evidence. A result can show an indicative score while the decision-ready score remains withheld when integrity, evidence, confidence, route context or reviewer-note requirements are unresolved.
Transaction Monitoring
The public early-warning workspace uses configurable illustrative weights. The current pattern weights are:
- Transaction velocity shift: +20
- Value-pattern shift: +20
- New counterparty pattern: +15
- Rapid in-and-out flow: +20
- Route or corridor change: +10
- Channel or product change: +10
The configured total is capped at 100 as a defensive guard. Evidence-supported and unverified contributions are shown separately. If more than 50% of the displayed contribution is unverified, the signal is labelled Pending evidence. If an expected transaction or value baseline is missing, the relevant baseline-dependent rule is not evaluated.
PDF input modes
Full statement mode uses available transaction text and limited context indicators to pre-fill a local review. Transaction history-only mode uses anonymised transaction rows while the reviewer manually supplies an anonymised reference and any available profession, profile, baseline and evidence context. In both modes, extracted values require reviewer confirmation.
Data sources and configuration
The public build currently uses one configured national tariff reference containing 7,420 records for tariff year 2026–2027; the packaged dataset carries its source metadata. It is not a universal tariff database. The form accepts a 6–10 digit HS-code format: six digits represent the harmonised level, while exact national 8- or 10-digit verification requires the relevant country reference to be configured. An unrecognised code may continue to an indicative review, but the result remains not decision-ready until the tariff mapping is verified.
It does not connect to a live sanctions, PEP, adverse-media, market-price, SWIFT or core-banking data feed. Country-specific tariff datasets, thresholds, weights, data lineage and approval controls must be configured and validated for an institutional pilot.
Reference families and coverage boundary
The indicator vocabulary is informed by public AML/CFT and trade-finance risk guidance, including FATF risk-based standards and TBML typologies, Wolfsberg Group trade-finance principles, and BAFT/ICC trade-practice materials. These references inform the public rule vocabulary; they do not constitute regulatory approval, a complete typology library or a jurisdiction-specific legal mapping.
Additional typologies such as transshipment, sanctions-evasion routing, vessel or AIS anomalies, freight-cost anomalies and LC-amendment patterns remain roadmap items unless explicitly configured and evidenced in a deployment.
What the score means
The score is an indicative prioritisation signal. It is not a probability of money laundering, a finding of TBML, a sanctions determination, a regulatory rating or a substitute for investigation. A high score can reflect missing or unverified evidence, while a low score does not prove low risk.
Production readiness boundary
This public build has no backend, login, role-based access, saved case record, server-side audit log, live screening integration or regulatory certification. Browser rules can be inspected and altered. Institutional deployment therefore requires secure hosting, identity and access management, server-side case storage, audit logging, versioned rule governance, independent testing, privacy controls, operational resilience and applicable regulatory review.
Version and change log
- v39 · 22 September 2026: added the browser-local Document Assessment workspace with OCR-assisted extraction and credit-related document profiles.
- v37 · 22 September 2026: prints a visible report snapshot after rendering so Android Save as PDF cannot capture a blank hidden report.
- v36 · 22 September 2026: corrected Android print-preview visibility so the styled TBML and Transaction Monitoring reports render into the saved PDF.
- v35 · 22 September 2026: aligned report PDF print isolation, corrected the shared rules import version, unified navigation and footer links, and added shared build metadata and social-share tags.
- v29 · 22 September 2026: added a simple user guide, explicit analyst disclaimer, 7–365 day monitoring options, multi-PDF local analysis, downloadable text reports, risk-type labels, OCR/document-scope limitations and an indicator-to-reference crosswalk.
- v28 · 20 September 2026: added a mobile-upload fallback by placing the pinned self-hosted PDF.js files at repository root, so GitHub mobile upload does not require nested-folder preservation; updated all cache-busting references.
- v27 · 19 September 2026: removed unsupported AI-style wording, added a PDF consent gate and identifier warning, moved PDF.js to a pinned self-hosted build with evaluation disabled, aligned local-processing language, added publisher/contact details, added reference families and expanded the public security boundary.
- v26 · 19 September 2026: reconciled the public preview arithmetic, aligned the Transaction Monitoring preview and upload warning, added 6–10 digit HS-code format handling with an exact-mapping gate, published TBML weights and formulae, pinned PDF.js with integrity verification, and added the About and security-boundary page.
- Future changes to weights, thresholds, tariff references or PDF extraction behaviour should be recorded here before institutional use.