One Tariff Line, Sixteen Notices: What 11,789 AD/CVD Change Documents Actually Say
Save the HS codes in this guide to a free watchlist (5 codes, no card) and keep the figures you look up in one place.A free account covers this page; Pro adds 100-code watchlists and rate-change alerts.
Create a free accountShort answer: in a corpus of 11,789 US antidumping and countervailing duty change documents covering 2,581 ten- and eight-digit HTS lines, 68.0% of the documents do not state a duty number at all — they state a procedural event (a continuation, an amended determination, a review institution). Of the 10,839 trade-remedy documents, 3,465 carry a number. The single most common way to misread this file is to diff two rows on the same HTS line and call the difference a rate change: 1.9% of rows produce such a pair, and when you verify each one against its own case number, only 107 line-level moves survive.
What a "change document" is, and what it is not
Every AD/CVD action in the United States ends up as a Federal Register notice, and every notice that touches a tariff line produces a row in a change ledger: this line, this case, this program, this old value, this new value, this effective date, this FR document number. The ledger is the useful object because a notice is written for lawyers and the ledger is written for machines — but the ledger inherits the notice's habit of describing events instead of numbers.
The corpus behind this page was built on 2026-11-14T02:37:04Z from two sources: the Federal Register API (publication dates, effective dates, FR document numbers) and the USITC HTS change records (Chapter 99 rate rows). It spans 2021-07-01 to 2026-11-12 — 1,960 days — and contains 11,789 documents over 2,581 distinct HTS lines, 131 headings, 35 chapters, 159 AD/CVD case numbers and 308 distinct Federal Register documents. 10,839 rows are trade-remedy rows; 949 are HTS policy-layer rows (a Chapter 99 provision added, modified or removed); 1 is a Chapter 99 rate row.
| Event type | Documents | Share |
|---|---|---|
| admin review results | 6,603 | 56.0% |
| continuation | 1,414 | 12.0% |
| sunset review results | 1,411 | 12.0% |
| ad cvd orders | 601 | 5.1% |
| amended final | 583 | 4.9% |
| revocation | 93 | 0.8% |
| ad order | 68 | 0.6% |
| cvd order | 66 | 0.6% |
Source: Federal Register API — antidumping and countervailing duty notices — https://www.federalregister.gov/documents/search?conditions%5Bterm%5D=antidumping (fetched 2026-11-14 by re-reading the change ledger; corpus build 2026-11-14T02:37:04Z).
The order types are the whole story of why a line "changes" without its rate moving. Administrative review results are 56.0% of the corpus (6,603 documents). A review result usually announces that a review was completed and sets a deposit rate for the reviewed exporters; the other lines in the same order keep their old rate and still get a row, because the order itself was re-published. Continuations (1,414) and sunset review results (1,411) are the same shape: the order survives, the numbers mostly do not move.
The diff trap: 219 apparent changes, 107 real ones
Sort the corpus by HTS line and effective date, then compare each row's old value to the previous row's new value on the same line. You get 219 numeric pairs where the two numbers differ, with a maximum apparent swing of 116.87 percentage points. That number is wrong, and it is wrong in the most expensive direction: it looks like the schedule moved 117 points on a chemical line when nothing of the kind happened.
The correction is an identity join. A real rate move requires all three of: the same HTS line, the same case number, and the same program (AD or CVD) — plus the previous row's stated number must be the current row's own old value, i.e. the two rows must chain. Applying that filter leaves 107 line-level moves across 3 cases. 112 of the 219 apparent pairs (51.1%) are artifacts of reading two programs or two cases as one line.
| Case | Event | Lines | From → to | Change | Notice |
|---|---|---|---|---|---|
A-201-844AD | sunset review results | 60 | 20.58% → 66.70% | +46.12 pt | 2026-048832026-03-12 |
A-570-016AD | sunset review results | 32 | 76.46% → 87.99% | +11.53 pt | 2026-086352026-05-04 |
A-533-867AD | admin review results | 15 | 0.50% → 8.35% | +7.85 pt | 2026-216242026-10-22 |
Source: Federal Register API — AD/CVD determinations — https://www.federalregister.gov/documents/search?conditions%5Bterm%5D=antidumping (fetched 2026-11-14 by re-reading the change ledger; corpus build 2026-11-14T02:37:04Z).
Three cases, all antidumping, all upward: a Mexican steel order moving 20.58% to 66.70% on 60 lines at a sunset review, a Chinese case moving 76.46% to 87.99% on 32 lines, and an Indian case moving 0.50% to 8.35% on 15 lines at an administrative review. That is the entire verified movement in a corpus that spans five calendar years. Anyone who tells you AD/CVD rates "change all the time" is describing the volume of paperwork, not the frequency of rate moves.
Worked example: ten notices on one chemical line, zero rate changes
Line 2918.14.00.00 collects 10 documents in the corpus. Read in
date order with the numbers left in, the sequence looks like a rate that
oscillates between 11.25%, 19.3%, 156.87% and 40% — a 116-point ride. Add
the case number and the program and the ride disappears: the 11.25% and 19.3%
rows belong to two unrelated cases (Thailand, Belgium), the 156.87% and 40%
rows alternate between a single Chinese case's own margin and a combined
AD/CVD continuation row that restates the combined figure, and four of the ten
rows state no number at all.
The same shape appears on line 4409.10.40.10, which is the busiest line in the
corpus at 16 documents. Those 16 documents belong to 4 different case
numbers — a Chinese AD case, a Chinese CVD case, a Canadian AD case and a
Canadian CVD case — and only 2 of the 16 state a number (6.04%, 231.6%).
A buyer tracking "the rate on my line" without a case number is tracking four
unrelated orders at once.
| Effective | Event | Prog. / case | Rate | FR doc |
|---|---|---|---|---|
| 2026-02-23 | admin review results | CVD C-570-118 | — | 2026-03500 |
| 2026-02-24 | admin review results | AD A-570-117 | — | 2026-03678 |
| 2026-04-14 | admin review results | AD A-122-857 | — | 2026-07153 |
| 2026-04-14 | admin review results | CVD C-122-858 | — | 2026-07154 |
| 2026-05-05 | sunset review results | AD A-570-117 | 231.6% | 2026-08736 |
| 2026-05-05 | sunset review results | CVD C-570-118 | — | 2026-08737 |
| 2026-05-08 | admin review results | CVD C-570-118 | — | 2026-09218 |
| 2026-05-13 | admin review results | AD A-570-117 | — | 2026-09509 |
| 2026-06-24 | continuation | AD/CVD A-570-117 | — | 2026-13103 |
| 2026-06-24 | continuation | AD/CVD C-570-118 | — | 2026-13103 |
| 2026-08-06 | amended final | CVD C-122-858 | — | 2026-16003 |
| 2026-10-03 | amended final | CVD C-570-118 | — | 2026-21038 |
| 2026-10-05 | amended final | AD A-570-117 | — | 2026-21045 |
| 2026-10-19 | admin review results | CVD C-122-858 | — | 2026-21286 |
| 2026-10-19 | admin review results | AD A-122-857 | 6.04% | 2026-21296 |
| 2026-11-12 | admin review results | CVD C-570-118 | — | 2026-22950 |
Source: Federal Register API — wood mouldings and millwork determinations — https://www.federalregister.gov/documents/search?conditions%5Bterm%5D=antidumping (fetched 2026-11-14 by re-reading the change ledger; corpus build 2026-11-14T02:37:04Z).
Where the paperwork lands: 9,275 ten-digit rows, 2,514 eight-digit rows
Two thirds of the rows are ten-digit statistical lines (9,275) and one third are
eight-digit subheadings (2,514). The mix matters because a case is scoped to a
product, and a product usually spans several ten-digit lines under one
eight-digit subheading. The largest single notice in the corpus, FR document
2026-13103, carries 246 line rows — and every one of
them is a continuation row with no number attached.
Line density follows the caseload, not the schedule. The top five headings by document count are below; the fourth column is the number of distinct cases touching that heading, which is the number that tells you whether the heading is "busy" or "one order with many lines".
| Heading | Docs | Lines | Cases |
|---|---|---|---|
7304 | 716 | 110 | 6 |
4409 | 668 | 53 | 4 |
7306 | 494 | 66 | 15 |
4412 | 440 | 40 | 2 |
8708 | 423 | 56 | 7 |
Source: Federal Register API — AD/CVD determinations, grouped by HTS heading — https://www.federalregister.gov/documents/search?conditions%5Bterm%5D=antidumping (fetched 2026-11-14 by re-reading the change ledger; corpus build 2026-11-14T02:37:04Z).
Heading 7304 leads at 716 documents but touches only 6
cases over 110 lines: a handful of steel-pipe orders with very wide product
scope. By contrast 7306, fourth on the list, spreads 494 documents across
15 cases. If you import tube and pipe, the heading you care about is
probably 7306 or 7304, and the case number — not the heading —
is what you need to quote to your broker.
Most lines are shared: 68.5% of lines carry two or more cases
1,295 of the 2,581 lines in the corpus are touched by at least two
distinct case numbers, and one line (7305.11.10) is touched by
7. This is the structural reason a naive diff fails: on a shared line, two
adjacent rows in date order are routinely two different orders, and the numbers
attached to them are not comparable.
The distribution of documents per line is heavy at the bottom and thin at the top: median 3, mean 4.57, maximum 16. Only 404 lines (15.7%) reach ten or more documents.
Coverage is also uneven in a way that matters for automation: 1,612 of 2,581 lines (62.5%) have at least one document that states an actual percentage. On the other 969 lines, the ledger tells you that something happened and never tells you what the number became. That is not a data-quality bug; it is what the notices say. A monitor that assumes every change document carries a rate will report silence as stability.
Origin concentration: half the corpus is one country
| Origin scope | Documents | Share |
|---|---|---|
| China | 5,798 | 49.2% |
| India | 971 | 8.2% |
| Vietnam | 486 | 4.1% |
| Korea | 484 | 4.1% |
| Taiwan | 437 | 3.7% |
| Canada | 409 | 3.5% |
Source: Federal Register API — origin scope recorded on each AD/CVD document — https://www.federalregister.gov/documents/search?conditions%5Bterm%5D=antidumping (fetched 2026-11-14 by re-reading the change ledger; corpus build 2026-11-14T02:37:04Z).
China accounts for 49.2% of all documents in the corpus; the next five origins together account for 23.6%. For an importer, the practical consequence is asymmetric attention: a Chinese-origin line has a high prior of being under some order, so "no order" needs to be actively verified; a line from a low-frequency origin can usually be cleared with one lookup.
Tempo: 93.2% of the corpus is 2026, and 21.5% of documents land after their own effective date
10,986 of the 11,789 documents carry a 2026 effective date, across 2,363 lines. On a line touched more than once in 2026, the median gap between consecutive documents is 33 days and 48.0% of the gaps are 30 days or less — repeated paperwork on the same line is the normal case, not the exception.
The more useful tempo number is the detection lag: how long after a document's own effective date the ledger records it. Median lag is 0 days, but 2,538 of 11,783 dated rows (21.5%) are recorded after the effective date, 1,479 by a week or more and 602 by a month or more, with a maximum of 331 days. If you learn about a rate change from a monthly newsletter, you are in that tail.
The corpus is dominated by 2026 because the two upstream sources are current snapshots and the Federal Register backfill window is short: 803 documents predate 2026. Treat the pre-2026 tail as a sample, not a census — the per-year counts in this ledger cannot be used to argue that AD/CVD activity "tripled", because the earlier years were never backfilled to the same depth.
How to read a change document without getting burned
- Join on case number and program first. The line is the wrong key. 1,295 of 2,581 lines carry two or more cases, and AD and CVD on the same line are two separate rates that both apply.
- Ask whether the row states a number. 68.0% of trade-remedy documents do not. If the new value is "rate not stated", the document is procedural: it tells you a review happened, not what you will pay.
- Separate the announced margin from the cash deposit. A review result publishes a dumping margin and a per-exporter deposit rate; they are not the same number, and the deposit is what the entry actually collects.
- Check the effective date before you assume you are early. In this corpus the median detection lag is zero days but 21.5% of documents are recorded after their own effective date, so a notice that took effect before you read it is routine — the deposit applies to entries on or after the effective date, not from the day you find out.
- Re-check the order's status, not just the rate. A continuation row means the order survived a sunset review; a revocation row means it did not. Both appear in the same ledger as "changes" to the same line.
To check a specific line end to end, the fastest path on this site is the AD/CVD case browser, which lists orders and their product scope; the HS alert watchlist re-checks the lines you care about and emails each change with its old rate, new rate and source; and the duty calculator applies the resulting rate to a shipment value including MPF and HMF. The reference guide for the document types themselves is AD/CVD duty lookup; the five-year review calendar is the 2027 sunset wave.
What this census does not prove
- It is a document census, not a duty engine. It records what the notices said. A line can carry an AD rate, a CVD rate, a Section 301 overlay and a Section 232 overlay simultaneously; this page counts AD/CVD documents only.
- The pre-2026 window is shallow. 803 documents predate 2026, against 10,986 in 2026. Year-over-year comparisons from this ledger are not valid.
- "Rate not stated" is a real value. It is recorded as such rather than dropped, because dropping it would silently convert "no number announced" into "no change".
- Case scope is broader than the HTS lines listed. A notice names the lines the order covers at the time of publication; a new ten-digit line created later may fall inside the scope without appearing in the original document.
Method, in one line: take every change row, keep code,
case_number, program, old_value,
new_value, effective_date and
fr_document_number; parse a percentage only where one is printed;
chain rows only when the previous row's stated number equals the current row's
old value on the same line, case and program. Reproduction script and pinned
counts ship with the page's generator, and the generator refuses to render if
any headline number in this text drifts from the dataset.
How often do US anti-dumping and countervailing duty rates actually change?
Rarely, relative to the volume of paperwork. In a corpus of 11,789 AD/CVD change documents spanning 2021-07-01 to 2026-11-12, only 107 line-level rate moves across 3 cases survive an identity join on HTS line, case number and program. The other 112 apparent changes are artifacts of comparing rows that belong to different cases or different programs.
Why do so many AD/CVD notices not contain a duty rate?
68.0% of the trade-remedy documents in the corpus state a procedural event rather than a number. 6,603 of the 11,789 documents are administrative review results, 1,414 are continuations and 1,411 are sunset review results; those notices re-publish an order without necessarily restating every line's deposit rate.
Can I track my HTS line by looking at the line number alone?
No. 1,295 of the 2,581 lines in the corpus are touched by two or more case numbers, and one line (7305.11.10) is touched by 7. Antidumping and countervailing duties on the same line are separate programs with separate rates, so a line number plus a date is not enough to identify a rate.
What is the difference between a dumping margin and a cash deposit rate?
The margin is the calculated dumping amount as a percentage of normal value; the cash deposit rate is what the entry summary collects per exporter until the next review. A review result publishes both, and they can differ because the deposit rate is the all-others or company-specific rate applied going forward, while the margin is the finding for the reviewed period.
How current is this AD/CVD document census?
The corpus was built on 2026-11-14T02:37:04Z from the Federal Register API and USITC HTS change records, with effective dates running to 2026-11-12. 10,986 of the 11,789 documents carry a 2026 effective date; the remaining 803 predate 2026 and should be read as a sample rather than a full history.
Do higher AD/CVD rates apply to goods already in transit?
The rate applies to entries made on or after the notice's effective date. 21.5% of the dated rows in this corpus were recorded after their own effective date, and the 90th percentile of the detection lag is 10 days, so a shipment that clears after the effective date pays the new deposit even if you only learned about the change afterwards.