Regional Industry Structure Studies

150 studies covering 141 distinct cohorts, dated 2026-10-16, compare archive composition, value-chain roles and evidence coverage.

Topic cohorts can overlap and must not be summed. Small cohorts support counts rather than stable rankings; an empty cohort is not an industry gap.

Batch labels are not a new-company time series; deduplication is not growth or recognition validity.

A separate 2026 layer contains 11,218 local list records: 7,028 new-applicant and 4,190 review records. Only 4,333 mapped company IDs overlap the public archive. Another 6,885 records resolve outside that public set; they are not unresolved and are not added to this series’ company denominators.

List status retains the stored stage. Local recommendation is not final recognition, and record counts are not new-company counts or growth rates.

Ordering describes archive coverage, not economic strength. Guangdong and Shenzhen, for example, retain separate reporting scopes; combined provincial counts appear in the relevant details.

All studies retain original URLs and filters and deduplicate stable company IDs. Parent sets and contrasts excluding the focal cohort are separate. Dimensions fixed by filters are not findings. Representation ratios describe archive label participation, not market share, output or competitiveness.

Concentration requires mutually exclusive known labels and at least 30 labelled records. The effective category count is the reciprocal of the sum of squared shares among known records: the number of equal-sized categories with that concentration. It describes archive composition, not market competition. Multi-label participation and cross-tab cells overlap and cannot be summed into distinct firms. Fewer than ten records support counts; 10–29 describe structure without stable rankings. No matches do not mean no industry.

The intersection/union ratio divides shared company IDs by the deduplicated union; focal-only IDs equal the focal count minus the intersection. Participation gaps use each full cohort as its denominator and are reported in percentage points. Cohorts may overlap and are not independent samples. A separate parent contrast excludes the focal cohort to avoid including it on both sides. These descriptive differences do not establish causes; products, archive coverage and classification sources still require verification.

Batch labels are cross-sectional: recognition, review, relocation and archive coverage can affect them, and a company can carry several. Counts are not new firms, growth or survival rates. The 2026 local lists form a separate layer; existing ID mappings do not establish final recognition.

Source coverage counts existing review configurations with bilingual excerpts and resolvable source IDs. It does not re-certify originals or entire historical texts. Non-empty fields and document-like URLs are not original evidence. Keywords establish search relevance only; broad terms, company names and copied introductions can enlarge candidate cohorts.

Provincial statistics, reporting-jurisdiction archives and combined provincial cohorts retain separate denominators. Macro statistics do not establish individual customers, deliveries, revenue or technical capability. Proximity and shared labels do not prove transactions. Classification questions receive sensitivity counts, not automatic reclassification.

Do topic counts or keyword hits establish industrial strength?

No. Topic counts describe editorial categories and keyword hits locate candidates. Comparisons require deduplicated IDs, comparable sets, primary delivery evidence and outcome measures.

This series is research material, not investment advice.