Home → Insights → Export control data

China Export Control Data: Official Sources and Court Records

A useful China export control dataset needs several kinds of evidence. Regulations explain a framework; official announcements record a particular measure; published compliance examples describe selected events; court documents answer what a court actually considered. A database that puts all four under “case law” makes it difficult to distinguish a government measure from judicial reasoning. A court corpus alone also cannot establish whether a company is currently subject to a restriction.

For overseas legal AI teams, the practical question is where to obtain each source and what a court-data supplier can contribute. This guide maps official entry points, reports three reproducible searches of SinoVerdict’s court-record index on October 11, 2026, and sets out the evidence fields to request before accepting a topic-specific delivery. It addresses data procurement and retrieval design. It is informational, not legal advice or a current sanctions screening service.

Start with the question your product must answer

“China sanctions data” can mean restrictions imposed by China, foreign restrictions affecting Chinese entities, a controlled-item classification, or a dispute about an interrupted transaction. Those searches need different authorities and source systems. Record the issuing jurisdiction before collecting documents. A mention of a foreign sanctions program in a Chinese judgment does not turn that judgment into an official entry in that foreign program.

Similarly, a user asking whether an item requires an export authorization needs current official materials and professional review. A user asking how a court discussed an export restriction needs a judicial document and the relevant passage. An AI product should make those questions explicit, rather than silently answering the first from evidence gathered for the second.

The sources below are official Chinese entry points inspected for this guide. They are a starting inventory, not an exhaustive catalogue of every export-control or countermeasure instrument. No list of currently restricted entities is reproduced here. Such a list needs its own dated collection, amendment history and reconciliation against the competent authority’s announcements.

Four source tracks that should remain distinguishable

These are suggested collection fields, not a claim that one SinoVerdict endpoint already supplies every field. A pilot should identify which fields are stored, which need extraction and which must be obtained from another source. For broader source discovery, see the official PRC court-data endpoints inventory. For tariff classification, valuation and other customs questions, use the separate customs and trade dispute guide.

There is a useful distinction between an official website and the object hosted on it. An official portal can contain regulations, speeches, guidance and republished commentary. Preserve the object’s own attribution and caveats. Domain authority does not make every paragraph a binding rule or every account an independently authenticated court decision.

Where to obtain the rules and item terminology

MOFCOM publishes the Chinese text of the Regulations on Export Control of Dual-Use Items, identified as State Council Order No. 792. The published instrument is dated September 30, 2024 and states an effective date of December 1, 2024. Treat the instrument date, website posting date and effective date as separate fields. Its text provides the regulatory source; a court record mentioning it provides a different evidence object.

The China Export Control Information portal exposes policy, FAQ and compliance sections, a dual-use item database, and links for licensing and end-user materials. That portal is a source-discovery hub. A visible entry point does not establish that an unrestricted bulk API, a downloadable history or a complete enforcement register is available. This guide has not verified those delivery capabilities.

A particularly useful source for data modelling is the portal’s FAQ on customs commodity numbers and related questions. It explains that a customs commodity number is not the basis for deciding whether an item is dual-use and distinguishes such numbers from export-control codes. Preserve both namespaces. Do not join them as interchangeable identifiers. The FAQ itself directs readers to the competent authorities’ documents and actual implementation.

For a retrieval system, this means that code matching and plain-language similarity should be separate operations. A document can discuss a technical specification without reproducing a control code. Conversely, the occurrence of a code may be inside background material rather than the operative issue. A reviewer should see the passage that supports the classification, its source type and the date of the underlying document.

The Unreliable Entity List is another source object

MOFCOM’s English reference text of the Unreliable Entity List provisions identifies Order No. 4 of 2020 and September 19, 2020 as the promulgation date. The provisions describe announcements concerning inclusion and removal. The English page expressly says that Chinese is authentic and English is for reference. Store that language relationship instead of treating two language versions as two independent measures.

The provisions are a framework document; they are not themselves a dated list of all entities subject to measures. A procurement specification should ask separately for original announcements, later changes and the effective status the supplier can actually substantiate. It should also distinguish a report of an investigation from a decision about inclusion. This guide does not determine the present status of any entity.

Entity resolution needs evidence rather than a fuzzy-name threshold alone. Preserve the name as written in the announcement, a proposed English or Chinese alias, and the basis for linking records. An AI system should be able to say that a relationship is uncertain. Automatically extending a record to similarly named companies or an entire corporate group would introduce a claim the collected source may not support.

That design also helps when a company name appears in a court record. A mention can identify a supplier, a counterparty, a quoted example or background context. It does not establish that the entity was subject to a particular measure on the transaction date. Connecting the two source tracks requires a documented match and a compatible timeline.

A published compliance example is not a court judgment

The portal hosts a compliance account concerning an unlicensed tactical-vest export. Its narrative describes a customs enforcement event in April 2025. The displayed page also includes a disclaimer about republished, compiled or excerpted material. The object inspected is a narrative compliance example, not a displayed court judgment or the underlying administrative penalty instrument.

This distinction matters even when an account contains concrete quantities and a fine. A narrated amount belongs to that account. It should not be assigned to a court holding or used to calculate an average judicial penalty without the appropriate underlying documents and denominator. We have not linked that example to a matching court record, and we make no claim that it is present in the court corpus.

For a mixed-source collection, use separate identifiers for the narrative and any later-discovered instrument. Record the relationship as confirmed, proposed or unresolved. Keep both texts available to reviewers. Replacing the original narrative with an AI-generated “judgment summary” would remove precisely the source distinction the buyer needs to assess reliability.

What our live court-record searches returned

On October 11, 2026, we queried the judgments index through its count endpoint. Each phrase was searched with both ik_smart and ik_max_word, then combined using a Boolean union. The counts below describe captured records that match those analyzed-token queries. They are not manually verified export-control cases, literal-string counts or different disputes.

All nine count responses returned zero failed shards. The downloadable query definitions and responses contain the measurement timestamps, fields, analyzers, union definitions and counts. No case-type, cause, region or year filter was applied. The body queries may overlap, so the rows must not be added. We have not measured semantic precision, recall or a distinct-case total.

The two-record pool is below 30 records: no statistical conclusion can be drawn. The title zero only describes this query against this captured index. It does not show that China has no export-control judgments, that no enforcement occurred, or that the body pool is irrelevant. Equally, the 117 body matches do not establish 117 disputes applying export-control rules.

The analyzer difference in the second row illustrates a procurement issue worth testing before accepting a “no results” answer. A query can miss records because of text analysis rather than an absence of source material. See our Chinese phrase-query analyzer audit for the underlying retrieval problem. Using a union helps with that particular mismatch; it does not certify topic relevance or remove other coverage gaps.

Build a pilot around evidence, with explicit exclusions

A reasonable pilot begins with a written question and an inclusion rubric. For example, “judicial discussion of a restriction affecting contractual performance” is a different target from “official announcements defining controlled items.” Specify whether a passing mention is excluded, whether a party’s allegation is useful, and whether a procedural ruling belongs in the delivery. These are buyer choices to settle before counting accepted records.

Keep separate labels for the authority’s action, a party’s account of it and a court’s reasoning. A proposed extraction should include a Chinese evidence passage and its role in the document. English summaries should preserve uncertainty and attribution. If the available record only reports an allegation, the summary should not promote it to an established fact.

Keep the query-level record total separate from accepted document and matter counts. A court may issue several documents under one case number, while duplicate captured records may repeat a document. Case-number grouping is useful but is not, by itself, a completed matter-level deduplication protocol. Missing identifiers should be disclosed rather than filled with invented values.

If sampling is used to assess a larger candidate pool, preserve the randomization method, seed, sample size and any strata. Show how many records were reviewable and how concentrated they were in one party or proceeding. A convenience sample of the first search results is useful for discovery, but it cannot support a population-wide accuracy claim.

Dates and missing evidence determine the limits

A research collection needs more than one clock. Preserve the instrument’s issue and effective dates, an announcement’s posting date, the event date reported in a narrative, a judgment date and your retrieval timestamp where available. A sentence describing a later measure should not automatically be assigned to an earlier transaction. Where a date is extracted rather than stored, disclose that method and retain its evidence.

Public court material is an observed publication sample. It does not supply every administrative investigation, licensing decision, settlement or unpublished dispute. The court index’s date and text coverage must be evaluated within the requested slice. Annual publication counts can change because publication changed; they do not, without further evidence, measure changes in enforcement intensity.

A missing source URL is also a real limit. SinoVerdict does not promise a working original-judgment link for every court record. Request available court identifiers and Chinese passages, and define how your team will independently verify a record when needed. Never generate a plausible government link simply to make an answer appear more authoritative.

What SinoVerdict can contribute to the source stack

SinoVerdict’s published scope is 160M+ court-data records, with REST API and MCP access and bulk licensing discussed against an agreed specification. That number describes records, not distinct cases or a complete export-control collection. The searches in this guide support only the stated candidate pools. They do not establish exhaustive current sanctions coverage or a ready-made, verified export-control topic dataset.

Use the court-record layer to evaluate what judicial material is available for your question. Keep official rules, item materials and announcement histories in their own source tracks. A coverage report should state what the requested court slice contains, what is missing, which fields are available and which transformations the proposed delivery requires. Text completeness and freshness need their own checks.

For an enterprise trial, send the research question, jurisdictions, time window, preferred document types and intended use. Ask for a trial API key and a scoped corpus coverage report before deciding production delivery. The delivery overview explains the commercial access options; the delivery acceptance guide covers checks to carry into the final specification.

Frequently asked questions