The classification workflow

Six steps. One defensible answer.

Every Kanon classification follows this sequence — in this order — with every decision recorded before any result is written.

Product capture

Describe your product

The intake flow is conversational. You describe your product — from a minimal free-text description to a complete set of physical attributes — and Kanon meets you where you are. The LLM extracts every attribute it can infer and asks targeted follow-up questions for anything that is missing and necessary for classification.

The session continues until Kanon has sufficient attribute coverage to run the classification engine with at least a Reasoned confidence level. Product records persist across sessions.

Materials, composition, and construction are extracted from your description
Function, end use, and commercial identity are confirmed
Dimensions, weight, and technical specifications are captured when relevant to classification
All attribute values are stored with value type (Nominal or Actual) and unit

GRI 1

Heading evaluation — chapter notes and section notes

GRI 1 is the first and most important rule. Under GRI 1, classification is determined by the terms of the headings and any relevant chapter or section notes. Kanon evaluates your product's attributes against the parsed condition trees for every candidate heading in the HTSUS.

Chapter notes and section notes are legal exclusions, definitions, and conditions that govern how headings apply. They can include or exclude products from a heading — and they take precedence over heading text. Kanon parses every chapter and section note in the active HTSUS corpus into structured condition trees, enabling deterministic evaluation against your product's attributes.

All candidate headings are identified from the active HTSUS corpus
Chapter notes and section notes are evaluated as structured condition trees, not as raw text
Each heading passes or fails with a specific condition recorded
If GRI 1 uniquely resolves the heading, classification proceeds to subheading differentiation

GRI 2–6

Traversal of remaining rules (if required)

If GRI 1 does not uniquely resolve a heading, Kanon applies GRI 2 through GRI 6 in the legally mandated sequence. Each rule is applied only if the preceding rule was insufficient — no rules are skipped, and no rules are applied out of order.

GRI 2a — Incomplete or unfinished goods: whether the product is classifiable as the finished article
GRI 2b — Mixtures and composite materials: each constituent is evaluated against the relevant headings
GRI 3a — Most specific description: which heading provides the more specific description
GRI 3b — Essential character: for mixtures and composite goods, which material or component gives the product its essential character
GRI 3c — Last in numerical order: if 3a and 3b are insufficient, the heading that occurs last numerically is used
GRI 4 — Most akin goods: classification under the heading for most similar goods
GRI 5 — Containers and cases specifically designed for the contents
GRI 6 — Subheading differentiation using the same rules applied at heading level

Layer 2

Additional duty and USTR exclusion evaluation

Once the 10-digit HTSUS code is established, Kanon evaluates Layer 2: the additional duty landscape. This includes applicable tariff action lists (Section 301 China tariffs, Section 232 steel and aluminum duties) and active USTR exclusion notices from the Federal Register.

USTR exclusions can eliminate additional duties for specific products and may apply retroactively, entitling importers to duty refunds for prior entries. Kanon matches all active exclusions against your classified HTS code and flags any potential applicability.

Section 301 List 1–4 applicability determined at the 10-digit HTS level
Section 232 steel and aluminum duty applicability evaluated
Active USTR product exclusions matched against classified HTS code
Retroactive exclusion applicability flagged where relevant

Classification record

Immutable record written

The classification result is written to an immutable record. Fields include the recommended HTS code, confidence level, the corpus version used, the LLM model version, the full attribute set at classification time, and the Layer 2 evaluation results.

The record cannot be edited. If a product is reclassified — due to attribute updates or a new corpus version — a new record is created and the original is preserved. The history is permanent.

Corpus version pinned: the exact HTSUS edition used is recorded
Model version pinned: the exact LLM version is recorded
Confidence level recorded: Confident, Reasoned, Split, or Insufficient
Prior records preserved: reclassification creates a new record, does not overwrite

Export

Download your Classification Support Package

Your Classification Support Package is available as a downloadable PDF or Word document. It is the human-readable form of the reasoning trace — the document a customs broker or importer presents to CBP on audit day.

Product identity and the full attribute set used for classification
Recommended HTS code with the full 10-digit breakdown
Applicable base duty rate and any additional duties
USTR exclusion status and retroactivity assessment
GRI path with a plain-language explanation of each decision point
Corpus version and USITC publication date
Classification date and confirming user

Reference guide

The General Rules of Interpretation

The GRI are the six legally mandated rules that govern how every product imported into the United States is classified under the HTSUS. Published by the U.S. International Trade Commission, they are the same rules CBP applies when auditing an import classification.

GRI 1

Heading text and notes

Classification is determined by the terms of the headings and any relative section or chapter notes. GRI 1 is applied first and resolves the majority of classifications. Chapter and section notes are legally binding — they define what is and is not included in a heading.

GRI 2

Incomplete goods and mixtures

GRI 2a: An incomplete or unfinished article is classified as the complete article if it has the essential character of the finished article. GRI 2b: Mixtures and composite materials are each evaluated against the headings.

GRI 3

Multiple applicable headings

When a product could fall under two or more headings, GRI 3a applies the most specific description first. GRI 3b evaluates essential character. GRI 3c uses numerical order as the final tiebreaker.

GRI 4

Most akin goods

Goods not classifiable under GRI 1–3 are classified under the heading for goods most akin to them. GRI 4 is applied rarely — it signals that the heading structure does not have a precise match for the product.

GRI 5

Containers and cases

Containers and cases specifically shaped or fitted to contain specific articles, suitable for long-term use and presented with those articles, are classified with those articles if they are of a kind normally sold with them.

GRI 6

Subheading differentiation

Classification at the subheading level (4-digit through 10-digit) is determined by the same GRI 1–5 rules applied at the heading level — but only among subheadings within the same heading. GRI 6 governs how the 10-digit code is resolved.

Why the sequence matters: The GRI are applied in strict order. A product cannot be classified under GRI 3 unless GRI 1 and GRI 2 have each been applied and found insufficient. Skipping rules — or applying them out of order — produces a classification that CBP can challenge. Kanon applies every rule in the mandated sequence and records the outcome of each before proceeding.

The corpus

Versioned. Pinned.
Never ambiguous.

The HTSUS is a living document. The USITC publishes revisions, adds chapter notes, and modifies heading descriptions. Every change can affect a valid classification. Kanon maintains a versioned corpus so that every classification is pinned to the exact HTSUS edition that was in effect when it was made.

Immutable corpus versions

Each USITC HTSUS publication becomes an immutable Corpus Version in Kanon. Classification records reference a specific version. The same product classified against two different corpus versions may produce different results — and both are correct, for their respective dates.

Parsed condition trees

Chapter notes and section notes are not stored as raw text. They are parsed into structured condition trees that the classification engine can evaluate deterministically against a product's attribute set. This is what makes GRI 1 evaluation reproducible rather than interpretive.

Corpus pinning on record

Every classification record includes the exact corpus version used. If a newer corpus is published after your classification, your record remains valid against the version that was active at classification time. The audit trail covers the corpus as well as the product.

The Classification Support Package

What CBP wants to see.
Already in the document.

The Classification Support Package is the exportable record of your classification. Here is what it contains.

Classification Support Package

Generated by Kanon · HTSUS Corpus v2026.1 · 2026-06-12

product_name

As recorded at classification time — immutable

product_attributes

Full attribute set (material, function, dimensions, end use) with value type (Nominal / Actual) and unit for each field

recommended_hts_code

10-digit HTSUS code with heading, subheading, and statistical suffix breakdown

gri_path

GRI rules applied in sequence, with outcome (resolved / continued) recorded for each rule

provisions_evaluated

Every heading evaluated — pass/fail — with the specific condition that determined the outcome

chapter_notes_applied

The chapter and section notes evaluated, their condition trees, and the evaluation result for your product

confidence_level

Confident · Reasoned · Split · Insufficient — with the reasoning behind the confidence assignment

base_duty_rate

Column 1 (MFN) and Column 2 rates from the active HTSUS corpus

additional_duties

Section 301 / Section 232 applicability and rates, with the tariff action list reference

ustr_exclusion_status

Active exclusions matched · retroactivity assessment · expiration date if applicable

corpus_version

Exact HTSUS corpus version and USITC publication date

classification_date

UTC timestamp at which the classification record was written

confirmed_by

The registered user who reviewed and confirmed the classification

Ready to load the cannon?

Classify any product under the HTSUS for free. Full corpus evaluation, deterministic GRI traversal, and a Classification Support Package that holds up on audit day.

Try free Why Kanon