Evidence first. Conclusions no broader than the record.

Purpose and scope

What the project studies

We document episodes in which public evidence establishes, supports, alleges, or helps assess whether technology, intellectual property, research, industrial processes, samples, source code, suppliers, know-how, or competitive advantage moved to an individual or organization in China through theft, deception, unauthorized copying, insider conduct, cyber activity, commercial leverage, undisclosed conflicts, or lawful-but-strategically-harmful learning.

Cases can involve companies, universities, state bodies, military organizations, individuals, informal networks, or combinations of them. Inclusion is based on documented conduct and relevance—not ethnicity or nationality alone.

The inclusion floor

A case requires a named event, identifiable actors or institutions, a specific asset or capability, a discernible transfer or attempted-transfer mechanism, and public evidence that can be evaluated. Publication ordinarily requires a court document, signed admission, official finding or attribution, direct institutional record, or multiple independent credible sources. A government summary can anchor an adjudicated posture only when the absence of a linked dispositive record is disclosed.

We do not publish rumor-only entries, anonymous accusations without corroboration, generalized claims about Chinese people, or a guilt-by-association chain in which affiliation substitutes for evidence of conduct.

Source hierarchy

1. Primary adjudicative records: opinions, judgments, verdicts, plea agreements, signed factual statements, and final agency findings.
2. Primary procedural records: indictments, complaints, warrants, motions, sanctions notices, and official attribution statements. These establish what was alleged or attributed, not guilt.
3. Direct institutional records: regulatory filings, audited disclosures, technical reports, patents, company statements, and university records. Interested-party claims are labeled.
4. Reputable reporting and research: used to corroborate, contextualize, or identify records; material contested claims should not rest on a single secondary account when primary records are available.

Each dossier displays its evidence foundation: primary adjudicative record, primary procedural record, mixed direct records, or official summaries only. This identifies the strongest linked foundation; it does not certify every sentence as true.

Two independent labels

Every file carries an evidence status for the conduct and a separate PRC-state nexus label. A criminal conviction can coexist with “no state nexus established.” A state link can be officially attributed while the underlying charge remains unadjudicated. We do not collapse the two questions.

Claim-level sourcing

Material factual statements carry source markers linked to the file’s source record, including titles, decks, findings, legal outcomes, nexus summaries, actor roles, and actor outcomes. Automated validation checks that the source exists and is linked; a separate editorial review must determine whether the source supports the exact wording, scope, and qualifier. Quantified harm is labeled as a court finding, documented value, attributed estimate, potential harm, or unknown. Analytic leadership lessons are explicitly identified as hindsight analysis.

Inference standard

A court finding is not required before the project can describe theft or appropriation, but non-adjudicated conclusions must be visibly qualified. A supported inference means the total cited record makes the proposition more likely than not in our editorial judgment. It should ordinarily combine several specific facts, such as proven access to identified protected material, copying or concealment, a transfer path, technical correspondence between the protected work and a later product, deception, recipient communications, or independent corroboration.

Nationality, travel to China, employment by a Chinese organization, or similarity between two products is not enough by itself. Plausible alternatives—including independent development, lawful licensing, reverse engineering, public information, coincidence, or an interrupted attempt—must be stated when the record supports them. An inference is never presented as a conviction, court finding, or proven state direction.

The six-stage framework

Every dossier answers the same six questions about relationship, access, acquisition, movement, downstream use, and consequences. The structure makes cases comparable; it does not require a completed event at every stage. Each stage is labeled as adjudicated or admitted, documented, supported inference, officially alleged or attributed, disputed, interrupted, or not established. Missing evidence is recorded as missing evidence.

Visual evidence and contextual images

Case files prioritize photographs that reliably identify the named people, followed by images of the organizations, technology, or operating environment. Subject photographs must come from a court, law-enforcement or government publication, an employer or institutional profile, the subject's own public profile, or attributable editorial reporting. A court case alone does not mean a portrait or booking photograph was released, and we do not use an unverified search-result match to fill that gap.

Every visual carries a creator or institutional credit, source link, license link, descriptive alternative text, caption, and a stored file hash. Public-domain and openly licensed images are preferred. A generic “editorial use” assertion is not treated as a license. A caption states when an image is contextual rather than a photograph, exhibit, or document from the case; contextual imagery is never presented as proof of the underlying conduct.

Names and responses

We name individuals and institutions when the identity is public, relevant, and supported by the cited record. We include acquittals, dismissals, reversals, commutations, later civil orders, and meaningful denials or procedural arguments. A charging document is never silently rewritten as a finding.

Review and correction

Every file records a last-reviewed and next-review date. Developing cases receive shorter review intervals. Review includes procedural updates, claim-to-source alignment, limiting facts, source availability, and media rights. Material changes are reflected in the case status and updates log. Substantiated corrections are made visibly; the correction process is described on the corrections page.

Corpus and sampling limits

This is a curated public-source case library, not a random or comprehensive sample. It currently emphasizes English-language records and U.S. courts and agencies, so source availability, jurisdiction, enforcement priorities, and editorial selection shape the collection. The case count cannot establish prevalence, a uniquely Chinese rate, or how often comparable conduct occurs elsewhere. Cross-case patterns are hypotheses and analytical aids, not population estimates.

Limits

Public records are incomplete. Absence of a documented state nexus is not proof that none existed; a documented affiliation or funding relationship is not proof of state direction. A case-wide legal status cannot be transferred from one actor or count to another. Legal outcomes can change. Estimated harm can mix direct loss, market reaction, opportunity cost, and victim attribution. Source links can disappear; retrieved dates, locators, archive links when available, and media hashes support later auditing but do not guarantee permanent access. Each file states these boundaries rather than resolving them through implication.