Classify, screen, and defend
every shipment you clear.

The internal reference is exactly 502 HS codes (counted in src/empires/trade/hs_code_database.py) with duty rates for US and EU corridors, plus corridor-specific FTA overrides. It is representative of the chapters that move the most volume for mid-market importers — not the full 17,000-plus WCO nomenclature. For long-tail codes we fall back to the LLM reasoning layer with ruling-citation prompting and always mark the confidence as intermediate. If a code is outside the internal reference, you see that on the response.

No — and the code says so in plain English. The screening_engine.py module carries a LEGAL_DISCLAIMER constant that is appended to every screening result: "this is a heuristic screening tool, not a substitute for a licensed denied-party screening service." It covers 506 public sanctioned entities across OFAC SDN, BIS Entity List, OFAC Consolidated, EU, UN, and Canada SEMA — enough to catch obvious hits and to stand up a workflow — but real production deployments integrate a paid feed. We surface the decision, you own the sign-off.

Five at launch, wired into classification_engine.py: USMCA, CPTPP, EU-UK TCA, RCEP, and KAFTA. Each has an origin-documentation requirement that the system will prompt for before certifying preferential treatment. If the origin documentation is missing or inconsistent, the FTA strategy is not surfaced as an option — we do not let you claim a rate you cannot defend.

Methodology #236 (Normative Reasoning) runs after every strategy suggestion and checks the output against trade regulation. The 7 strategies are each tagged low / medium / high risk in OPTIMIZATION_STRATEGIES — substantial transformation is explicitly high-risk and never appears without a human-review gate. Methodology #8 (Constitutional AI) refuses classifications that would misdeclare restricted end-uses from the EAR Part 744 keyword list.

The /audit-defense endpoint assembles the classification rationale, the duty-rate lookup, the screening records, any filing versions, and the full reasoning trace into a single dossier. Every AI decision on a shipment is logged with citations at the time of entry, so the defense is a compilation — not a reconstruction. You still brief counsel; we just hand them the evidence ordered.

No. The screening module is built on public OFAC / BIS / EU / UN / SEMA lists extracted into a representative 506-entity reference. Descartes Visual Compliance, Dow Jones Risk & Compliance, and LexisNexis Bridger Insight are on the Enterprise-tier roadmap as planned integrations — not existing partnerships. We would rather tell you that than imply a feed we do not have.

On the roadmap, not shipping today. The platform pattern used by LiquidTrade.OS — Tauri + Ollama, with Llamafile for full air-gap — is planned for Q3 2026 for import / export teams who cannot send customer and shipment data to any cloud. Waitlist signups will get early access to the local build when it is ready.

SOC 2 Type II is targeted post-launch, not audited yet. GDPR deletion workflow is in design. Every AI decision is logged with citations and tenant-scoped. Credentials are envelope-encrypted at rest. We will publish the audit report the day we have it — not describe it before we do.