How We Built Classification Researcher: Facts, GRI and Review
What changed: Rewrote the article as a product-design account; removed CBP-ruling endorsement and mandatory architecture claims and clarified Researcher, API and GingerOS roles.
We designed Classification Researcher around a sequence: collect product facts, identify competing headings, apply the GRI and legal Notes, research relevant CROSS rulings and retain the reasoning for review. Missing composition or function should lead to a question before a suggested code. This is an account of our product choices, not a claim that a CBP ruling approved the system. Our classifications support reference, education and planning; the customer's licensed broker reviews a code before it goes on an entry.
The practical AI trade compliance for importers guide separates classification research, tariff modeling and monitoring from filing responsibilities.
We start with facts rather than a product name
The description may omit materials, construction, principal function or how components work together. Researcher asks targeted questions to distinguish the candidate headings. In GingerLite, each product is a separate case; catalog batch classification belongs to GingerOS.
GRI and Notes structure the analysis
The GRI and the relevant Section and Chapter Notes provide the legal framework [1]. Our design encodes GRI sequencing and Note checks rather than relying on a similar-looking heading alone.
For composite goods, the essential-character judgment considers the material or component's nature, bulk, quantity, weight or value, and its role in use. These are the GRI 3(b) Explanatory Note factors described in CBP HQ 950636 [2]. The report records the judgment and its basis; it does not label those factors as the Carborundum principal-use test.
CROSS research happens during the decision
Researcher reads similar rulings as decision inputs. A cited ruling must fit the relevant product facts; a citation is not a guarantee that the same result applies to a different article. Keep the product evidence alongside the reasoning.
What the products return
| Product | Output and boundary |
|---|---|
| Classification Researcher | Product-level code research and reasoning report; no numeric confidence score |
| Parts Library | Candidate-code and source-country landed-cost comparison after classification |
| Classification + Tariff Calculation API | Code and U.S. duty layers when called; no reasoning report |
| GingerOS | Customer-fitted catalog workflows, rechecks and approved ERP write-back |
Researcher's report does not include the full duty stack for every candidate or a parts-to-finished-goods code map. Use Tariff Calculator for U.S. duties by entry date. GingerOS runs on our own classification and tariff engines, with integrations fitted to the customer's process.
How we ask buyers to assess the system
Use your own reviewed products, inspect the questions and reasoning, and record what the test measured. GingerControl's Quick classifier scored 99.86% at the 10-digit level on its own internal test of 1,000+ products. That is a first-party test, not an independent benchmark or a promise for another catalog. See how a GingerOS project starts.
Frequently Asked Questions
Did a CBP ruling approve GingerControl?
This article describes our design choices and makes no claim of CBP endorsement.
Why does Researcher ask questions?
Missing materials, construction or function can change which heading applies.
Is essential character the Carborundum test?
No. GRI 3(b) essential character and the principal-use analysis are distinct.
Does the report show numeric confidence?
No. GingerControl does not show a numeric confidence score.
Does the API return the reasoning report?
No. The report comes from Classification Researcher or the relevant GingerOS workflow.
Can Researcher classify a catalog in one case?
No. GingerLite uses one product per case; batch classification belongs to GingerOS.
What does 99.86% measure?
The Quick classifier at 10 digits on GingerControl's own internal test of 1,000+ products, not an independent benchmark.
Who reviews the result before filing?
The customer's licensed customs broker reviews the code before it goes on an entry.
References
- USITC, General Rules of Interpretation
Source: USITC, General Rules of Interpretation
Checked: 2026-10-12 - CBP HQ 950636, GRI 3(b) essential-character factors
Source: CBP HQ 950636, GRI 3(b) essential-character factors
Checked: 2026-10-12
Classifying SKUs one ruling at a time?
Hand this to GingerOS.
Pick a 30-minute slot on the next page with Chen Cui, our CEO. Bring one notice or one SKU list (an ERP export, a spreadsheet or an email will do) and see what GingerOS does with it.
Reviewed by Michael Weick, LCB / CCS




