How a line is priced

Two numbers sit on every line and they are not the same kind of thing. One was counted by somebody else. The other we built by reading the schedule. This page is how the second one is built, and what happens on the lines where it cannot be built at all.

The arithmetic, in the order it is applied

Every term is read at the entry’s own month and origin, not at today’s. That is the whole difficulty: a line entered in April 2025 and the identical line entered in June 2026 are governed by different law, and a lookup tool that reads today’s schedule gets the first one wrong. The Section 232 carve-out is applied where it belongs, so a product that merely contains the metal (a “derivative article”) is not charged twice on its metal content.

What “could not be priced” means

This is the part worth arguing with, so it is stated in full rather than softened. On 389,721 of 1,401,940 lines — 27.8% — the public record does not settle the rate. Not because the data is missing: because the rule turns on a fact that only the filed entry declares. Those lines get an interval — the lowest and highest rate the rule could produce — and a named reason, and they are not dropped.

A count of lines and a share of money are different claims, so here is the second one too. In August 2026 those lines carry $58.95bn of the $91.67bn of dutiable value we can see — 64.3%. The share of money is the larger of the two, and it is the one printed wherever a page has room for only one.

Dropping them would be the easy move and it would flatter us: a smaller denominator makes the remaining agreement look tighter than it is. They stay. A count of lines we could not price cannot be wrong — which is why it is the more defensible half of what we publish, and why it is printed in the same typography as the lines that agree.

It is also the reason the check asks for your document. Your own entry is the only thing that resolves a line like this: it is the form that declares the metal share, the exclusion, or the US content.

How a rate for a whole era is put together

A rate for one line is a division. A rate for a country, a chapter or an era is an index number — and an index number is not just a bigger division. Three choices sit inside it, each defensible, each moving the answer. We publish all three rather than leave a reader to guess.

The plain average of a rate across an era uses that era’s own import mix. That is the right answer to “what was charged on what actually arrived”, and the wrong answer to “what did the rulebook cost”. When a tariff rises, imports move away from what it hits, so the later average is taken on a basket that has already dodged part of the increase. On this ledger that understates a fixed basket by roughly 16% cutting by origin and 30% cutting by origin and product.

So we publish a Fisher index beside it — the same comparison run on the old basket and on the new one, then combined. It is superlative: the price part and the quantity part separate exactly, rather than by a convention someone picked. We do not publish its usual companion, a Törnqvist index: 2024 rates sit near zero for many cells because those tariffs did not exist yet, and an index built on log changes is unstable against a near-zero base. Tested here, it fell outside the bracket its own two halves define, which disqualifies it.

Falsify it

Send one entry

The arithmetic above is what runs on a real entry, line by line, at each line’s own date and origin — including the lines it refuses to price.

Check one of your own entries

Free. No account. We do not file with CBP and take no fee on anything you recover. Your file is read on our own server and is never sold or shared.

try it on one code · the national number · the test we failed, and the conflict · what we got wrong