AD/CVD Antidumping Duty Tracker — Federal Register API

Track US antidumping and countervailing duty (AD/CVD) trade-remedy actions as a clean, structured feed. Every Commerce and ITC trade-remedy notice in the Federal Register, with the Commerce case number, the product, the country, the lifecycle stage and the precise event pulled out — keyless, official source, no scraping.

What v1.1 changed (2026-08-01 validation pass)

The headline: the case number now actually works.

• case_number was null on 99.79% of notices. It was regex-scraped out of the title and abstract, which finds it on 3 of 1,412 live ITA notices. The Federal Register publishes the Commerce case number as a structured field, docket_ids (["A-428-850"]), present on 91.6% of ITA notices — and the actor never requested it. It is now the primary source; a live run measures 88–94% coverage.
• A case-number lookup returned zero rows. caseNumber was a client-side filter over only the newest page of results, so asking for A-570-049 returned nothing — read as "no activity on this case". It is now a server-side docket lookup across the whole archive: A-570-049 returns its complete 7-document history, 2016 Initiation → 2017 Order → 2022 Sunset Review → 2023 Continuation.
• case_type was wrong on 21.9% of notices. It was inferred from keywords, so a pure-countervailing notice whose boilerplate names both statutes came back "both" (Large Diameter Graphite Electrodes, docket C-570-221), and a plain antidumping notice came back null (Steel Concrete Reinforcing Bar From Bulgaria, docket A-487-002). It is now derived from the docket prefixes, with keyword inference only as a labelled fallback.
• The truncation guard was never read. The API publishes an exact count; the actor ignored it and pinned per_page at 100, which caps a result set at 50 pages × 100 = 5,000 records. per_page is now 1,000 (reaching 10,000, the API's own hard ceiling) and every row carries upstream_result_count, upstream_count_saturated and results_truncated.
• The lifecycle vocabulary was not closed. "Continuation of Antidumping Duty Order" (62 live notices) landed in action_type: "Order" — which the code defines as the imposition of an order. Revocation, Termination and Postponement had no home at all, and "Opportunity To Request Administrative Review" — the monthly anniversary-month announcement — was classified as an actual review. action_type is unchanged (buyers filter on it); the precise event is now carried on the new action_subtype, and an undeclared value fails the run.
• Monitor mode never remembered anything. Its seen-set lived in the platform's default key-value store, which is per-run — so a scheduled monitor found an empty baseline every run and re-emitted the entire feed as unseen. It now uses a named store, keyed to the run's case number, query and agency scope.
• Nine published fields were fetched and thrown away — FR citation, volume, page range, the DATES paragraph, the raw-text URL, the ACTION paragraph, every credited agency, the document type and the abstract itself.
• Nothing was removed or renamed.

Who it's for

• Customs brokers & trade-compliance teams — know the moment a duty order is continued, revoked or amended for a product you import.
• Importers & procurement — watch a specific Commerce case number for the next determination that changes your landed cost.
• Trade lawyers & consultants — a structured docket of every notice on a case, with citations and full-text links.
• Analysts & supply-chain risk — cross-country, cross-product AD/CVD activity in one dataset.

Example input — everything AD/CVD since a date

{ "sinceDate": "2026-01-01", "maxResults": 1000 }

Example input — one Commerce case, whole history

{ "caseNumber": "A-570-049", "maxResults": 200 }

Returns every Federal Register notice on that docket, newest first. This is a server-side docket query, so it reaches the whole archive rather than the last few months.

Example input — watch for duties ending or being extended

{ "actionSubtypes": ["Revocation", "Continuation"], "sinceDate": "2025-01-01" }

Example input — scheduled new-notice alert

{ "query": "steel", "monitor": true, "monitorStoreName": "adcvd-monitor" }

Output fields — all 42 declared fields

The case

The event

Identifiers and links

document_number, fr_citation (100% coverage), fr_volume, fr_start_page, fr_end_page, agency, agency_names, html_url, pdf_url, raw_text_url (plain-text full text, 100% coverage — use it for scope language and company-specific rates), source_system, retrieved_at.

The per-source outcome contract

null means not checked. false means checked and negative.

Inputs that populate the conditionally-null fields (each verified on a live run): correction_of — {"query":"correction","maxResults":150} (2 rows). action_summary — any broad run (32 of 150 rows on that same input). action_subtype / review_stage — any broad run; {"caseNumber":"A-570-049"} returns Continuation, Correction and Initiation subtypes in seven rows.

case_number_source: "title_or_abstract"

Live drift assertions

An offline fixture cannot see a silently degraded upstream, because the fixture was shaped from the degraded data. Before any row is billed, the running Federal Register service is checked and every measured value is logged (DRIFT PROBE: {...}):

1. The echoed filter. The envelope restates the search the API actually ran ("Documents matching 'antidumping' and from International Trade Administration"). If the term or the agency is missing from that echo, the filter is not being applied and the run fails. Verified live: a bogus agency slug, a bogus condition key and a bogus date operator all return HTTP 400 — but order=zzzz returns HTTP 200 and silently reorders the whole result set, so the ordering is asserted separately.
2. Descending publication date on page 1. maxResults truncation takes the head of the list, so a silently ignored order would return an arbitrary subset rather than the newest notices.
3. A negative control. A nonsense search term must return exactly 0 documents. This is the counterpart to the FDIC echoed-filter class: a filter that has stopped being applied would return the whole corpus here.
4. A positive canary. conditions[docket_id]=A-570-049 must return the Ammonium Sulfate From the People's Republic of China case. If the server-side docket filter breaks, every caseNumber lookup would silently answer with nothing.
5. Structured docket coverage. At least 40% of the newest ITA notices must yield a case number from docket_ids (live baseline 91–94%). If the Federal Register ever drops or renames the field, case_number would silently revert to the 0.21% title-regex hit rate.
6. Closed vocabulary. An action_type or action_subtype outside the declared sets fails the run. So does an "Other" share above 35% on a run of 200+ notices (live baseline 3.1%), which would mean Commerce changed its title conventions.
7. No silent zero. An unknown actionTypes value, a malformed caseNumber, a filter combination that matches nothing, or a caseNumber with no notices in scope all fail the run with an explanatory message, rather than billing an empty success that reads as "no trade-remedy activity".

Use as an MCP tool

This actor is callable by AI agents (Claude, Cursor, etc.) via mcp.apify.com. An agent can look up a Commerce case number and get its whole documented history in one call, or filter to action_subtype: "Revocation" to find duties that have ended. The run_complete, results_truncated and case_number_source fields let an agent judge how much to trust a row without reading logs.

Pricing

Pay per result — one AD/CVD notice = one dataset item. Graduated discounts apply on paid Apify plans. A run that fails a drift assertion, or whose filters match nothing, emits nothing and bills no result events.

FAQ

How do I follow one case? Set caseNumber to the Commerce number, e.g. A-570-049. It is a server-side docket query across the whole archive.

Why does a query return notices that don't mention the word? Because the Federal Register searches the full document text, not the title. Every row carries query_in_title_or_abstract so you can filter to headline matches.

Why is case_number sometimes null? About 8–9% of ITA notices are published with no docket ID at all — usually general policy or scope-ruling notices that are not tied to one case. case_number_source tells you whether a value came from the structured field or from the weaker title regex.

Am I seeing everything? Check upstream_result_count and results_truncated on any row. The Federal Register itself caps a result set at 10,000 documents and reports count saturated at that number; narrow with sinceDate or query for an exact figure.

Does this cover the ITC? Not by default. itaOnly defaults to true (Commerce only, which is who sets the duties). Set it to false to include ITC injury determinations and sunset institutions — about 24% more notices.

Do I need an API key? No. The Federal Register API is public and keyless.

Related actors

• consolidated-screening-list-delta — OFAC / BIS / State restricted-party changes, the sanctions counterpart to this actor's tariff signal.
• nhtsa-vpic-vin-decoder and gleif-ownership-graph — entity and product identification for the importers and exporters named in these notices.