File № SCN-2026-01 / A · Cross-border trade
Full file
Trade deployment dossier for defensible customs workflows.
From raw supplier spreadsheets to HS / TN VED codes with GRI reasoning and certification flags — and, for enterprise teams, search over released declaration archives. Built for a person who has to defend the output, not just receive it.
Wrong classification and inconsistent supplier data create regulatory and margin exposure — someone has to defend the file.
Qualified walkthrough on redacted samples after access request. Scoped paid pilot under NDA for enterprise teams.
The exposure
Classifying a single code is easy. Doing it at scale, defensibly, is not.
Supplier data arrives as a mess
One supplier sends Excel, another CSV, a third a PDF or a photo of a packing list. Half the descriptions are in a foreign language. Classification cannot start until someone cleans it up by hand.
Classification has to be defensible
"Looks like this code" is not enough when a customs officer, auditor, or compliance manager reviews it. The practice requires GRI reasoning, tariff references, and a confidence the signatory can argue about.
Certification rules are invisible in the spreadsheet
Whether an item needs a mandatory declaration of conformity under EAEU TR 004, TR 020, or nothing at all — that is buried in regulations, not in the supplier file. Miss it, and the shipment stops at the border.
Illustrative output
Classification the signatory can defend.
Every row ships with GRI trace, confidence, certification flags, and source references — not just the final code.
- Rule 1 → Heading 9405 (lamps and lighting fittings)
- Rule 6 → Subheading 9405.42 (LED, electric)
- · EAEU TN VED, Chapter 94
- · TR CU 004/2011, Annex 1
Measured workflow · interim
Golden-set benchmark (internal, 45 cases)
91% heading accuracy (4-digit) · median 14s per line · operator-in-the-loop. Illustrative workflow metrics on an internal golden set — easy, adversarial, and practice tiers.
- Heading accuracy (4-digit)
- Share of golden-set lines where the model’s 4-digit HS/TN VED heading matches the reference label after GRI trace review — not a guarantee on live declarations.
- Golden-set tiers
- 45 internal cases split across easy (clean supplier rows), adversarial (ambiguous descriptions, mixed languages), and practice-style (realistic noise).
- Latency
- Median 14s per line on the interim dev stack — operator-in-the-loop; production deployments are sized per engagement.
- Scoped pilot success
- Paid pilots agree KPIs on redacted samples under NDA — typically heading agreement, review time, and certification flag precision — not fixed ROI percentages.
Client pilots are measured separately on redacted data under NDA. Interim dev LLM; production deployment metrics provided per engagement.
What the stack does
Six capabilities. Four live today, two in scoped partnership.
Normalize supplier spreadsheets
Excel / CSV / PDF / photos in, a single reviewable schema out. Foreign-language descriptions translated, attributes extracted, duplicate rows merged, declaration-ready fields prepared.
Explainable HS / TN VED classification
Each row gets a code with a GRI 1–6 trace, confidence score, and source references. Today EAEU TN VED. The reasoning engine is HS-agnostic — adding another jurisdiction is a data question, not a rewrite.
Certification requirement detection
For each HS code, the system flags whether it falls under EAEU technical regulations (TR CU 004, 020, 025, etc.) and whether conformity is mandatory, voluntary, or not required. The output is a simple column — the reasoning stays traceable.
Quick classification channel
Lightweight lookup for on-the-fly checks or evaluating reasoning quality before a scoped paid pilot on operational data. Available after written request for qualified teams.
Search over released declaration archives
Turn a team's historical DT archive into a searchable precedent layer: find similar past cases by meaning, not file name. Available as a scoped deployment on redacted samples under NDA.
Declaration draft under R.055
A controlled next step, not a promise. From messy documents to a reviewable declaration draft, operator-in-the-loop. Interested teams should flag it early — this is a partnership, not a self-serve feature.
Who this is for
Built with four kinds of signatories in mind.
Importer ops lead
Spends hours cleaning supplier files before declarations can even start. Wants fewer Excel reshuffles, faster classification, and a repeatable process that does not depend on one senior expert.
Customs broker
Needs a defensible first-pass code with reasoning and precedent lookup. Treats the engine as a force multiplier for a small team, not a replacement for judgement.
Trade compliance manager
Owns the audit trail. Wants a consistent standard across the team, explainable outputs, and a clean answer on certification requirements before goods ship.
Platform / software vendor
Building a procurement, ERP, or logistics product and needs customs intelligence as an API. Interested in partnership and controlled deployment over a public signup form.
Terms before engagement
Boundaries stated before scope.
Not a replacement for a broker
Final responsibility stays with the qualified specialist. This is operator-in-the-loop tooling that prepares, explains, and accelerates — not a black-box declaration machine.
Honest about what is live
Anything in roadmap is labelled that way. No fake "already GA" on features still under co-development.
Data sensitivity
Enterprise engagements run on redacted samples under NDA. No supplier data or declaration archive is used outside the agreed scope.
Named practice — direct accountability
The principal of the practice reviews every brief. Engagements stay scoped to what can be delivered with full accountability behind each output.
FAQ