Limits of the service
What a TariffTrust estimate is, and what it is not.
An informational estimate
TariffTrust calculates US import duties and entry fees from the Harmonized Tariff Schedule, the country of origin, the customs value and the characteristics you declare. Every amount cites its official, dated and archived source. It is not a customs determination, a ruling or legal advice; the importer of record remains responsible for the entry.
Unknown is never zero
When a measure depends on information you have not given, the estimate takes the ordinary case of that question (the goods are what the measures of their code are aimed at, and nothing particular is claimed), names each assumption and shows what another answer would give. A question without an ordinary case is asked, and the result shows the established components only. When a Chapter 99 position is still under review, or a published rate is under verification after an incident, the amounts it may modify are withheld rather than guessed. Unanswered yes/no and choice questions give a scenario for each combination of their answers, up to twelve, with bounds; beyond, a prudent range when every component can be bounded.
What we do not cover
- Classification: TariffTrust does not propose an HTS code from a product description.
- Antidumping and countervailing duties (AD/CVD) and customs valuation determinations.
- Filing entries or refund claims, and any legal or eligibility opinion.
- New IEEPA refund features; existing IEEPA research stays, and past IEEPA duties are calculated.
- Import of CBP ACE exports.
- Particular entry regimes: personal goods in a traveler's baggage, Chapter 98 entries and withdrawals from a foreign trade zone. Declared, such a regime gives no estimate, neither duties nor fees.
- Taxes and fees other than the merchandise and harbor maintenance fees.
How rules are read
Chapter 99 notes are interpreted by a language model, read again by a second model and checked by automatic controls before an operator or the automatic decision approves them. The calculation itself is deterministic: a language model never computes an amount. A weekly random sample of automatic approvals is reviewed, and a wrong published rate follows an incident procedure: suspension, notification, correction and a second notification.
Freshness
Official sources are checked every six hours. A result says when they were last verified successfully and whether the verification itself works; if it fails, changes since the last success are unknown.
Keeping your evidence
Saved estimates and proof bundles are kept as long as their workspace, unless it deletes them; a workspace without a subscription for three years is deleted, its owners told a month before, and one whose plan was never chosen 30 days after its creation. Exports let you keep your own copies; the service does not promise retention beyond its own life.