Twenty-seven numbers this directory publishes, each with its date and the page that measures it.

One address for everything measured here. Every figure below already exists on a page with its source; this page indexes them so a claim can be cited from a single URL instead of being hunted across fifty-nine.

Why this page exists. A number is only citable if the person quoting it can find it twice, and a directory that scatters its findings across process pages, cluster pages and guides is hard to quote even when every individual figure is sound. So the figures are collected here: 27 counts and ratios, grouped by what they measure, each carrying the sweep date and a link to the page where it is argued and sourced.

Two sweeps, kept apart. Most disclosure figures come from the sixteen-plant census of 2026-11-01. Figures about the directory itself and about headcount come from the 25-profile registry, dated 2026-11-03. Where the two disagree it is because they measure different things, and the older census is not recomputed when a new profile is added — a dated measurement stays dated. Each row says which sweep it belongs to.

What is not here. No prices, no rankings, no estimates. Where a plant publishes nothing, the count is of plants that publish nothing — not a guess at what the true figure might be. Nothing on this page is new: it is an index, and every row links to the page where the number is derived.

The directory itself

Figure	Scope	Dated	Source
25 plants profiled — 20 PCB fabrication, 5 wire harness and cable assembly	25 of 25	2026-11-03	/pcb, /wire-harness
15 published fields per profile, each carrying a link to its source document	25 of 25	2026-11-03	/about
3 entities measured and refused a profile: one ST-flagged listed company, one plant with production equipment in judicial auction, one declared bankrupt in 2024	3	2026-11-03	/dataset
408 field-level records released as CSV, in long format, with the exact source note for each	408 rows	2026-11-03	/dataset
124 external source URLs cited across the profiles, keyed to the record that cites each	124	2026-11-03	/dataset

Headcount, the two figures

China publishes two different employee counts for the same factory. Both are usually true; they count different populations, and only one of them describes the building. The eight plants that publish both are the only place the gap can be measured rather than guessed.

Figure	Scope	Dated	Source
19 of 25 publish a social-insurance headcount — contributors at one legal person, from the annual filing	19 / 25	2026-11-03	/headcount-check
13 of 25 publish a group headcount — the consolidated employees of the listed parent	13 / 25	2026-11-03	/headcount-check
8 of 25 publish both; group ÷ insured runs 0.8× to 26.3×, and in aggregate 4.0× — 101,069 people against 25,231	8 / 25	2026-11-03	/headcount-check
1 of 25 publishes neither — its registry filing says 0, and the dataset leaves both columns blank rather than writing zero	1 / 25	2026-11-03	/headcount-check

What plants publish about capability

The consistent finding of the sixteen-plant census: the corporate layer is published, the specification layer is not. These seven counts are the measurement of that absence.

Figure	Scope	Dated	Source
5 of 16 state a layer ceiling; the three largest plants by revenue publish none	5 / 16	2026-11-01	/high-layer-pcb
9 of 16 claim HDI; 0 of 16 state a build-up — the only notations in the directory are on an exhibition board (1+2+1) and in a plan for a plant that does not yet produce	9 / 16, 0 / 16	2026-11-01	/hdi-pcb
10 of 16 publish at least one figure in square metres; 6 of 16 publish none, including three of the four largest by revenue	10 / 16	2026-11-01	/capacity-vs-output
1 of 16 publishes an audited physical output: 2,758,117 m² produced in 2025 across 4,824 employees = 572 m² per person per year, the only auditor-signed figure here	1 / 16	2026-11-01	/capacity-vs-output
4 of the 10 publish numbers that contradict each other on their own pages	4 / 10	2026-11-01	/capacity-vs-output
2 of 16 name a laminate grade they actually buy; 0 of 16 state which mill’s plant it ships from or the date code on the pallet	2 / 16, 0 / 16	2026-11-01	/laminate-supply
0 of 16 publish an MOQ; 2 of 16 publish a lead time	0 / 16, 2 / 16	2026-11-01	/quick-turn-prototype-pcb

Certification

The one claim in a supplier deck that costs nothing to make and can be checked in ninety seconds — if the number is published.

Figure	Scope	Dated	Source
Claims across 16 plants: 14 ISO 9001, 14 IATF 16949, 12 UL, 8 ISO 14001, 3 ISO 13485, 3 GJB 9001C, 2 ISO 45001, 1 QC 080000	16 plants	2026-11-01	/certificate-check
0 of 16 publish a certificate number, an issuing body, a certification scope or an expiry date	0 / 16	2026-11-01	/certificate-check
2 of 16 carry an expired standard edition in public — ISO 9001:2008 and ISO/TS 16949:2009 in recruitment advertising, and ISO/IEC 27001:2013 on a certification page, an edition whose certificates became invalid worldwide on 2025-11-01	2 / 16	2026-11-01	/certificate-check

What a machine sees

Measured by fetching each plant’s own site the way an answer engine would, on one day from one network.

Figure	Scope	Dated	Source
0 of 7 measured Pearl River Delta plants publish structured data (JSON-LD); 2 of 7 declare hreflang for more than one language	0 / 7, 2 / 7	2026-11-01	/buyer-questions
2 of 16 publish a customs or operator code in their own name; 2 of 16 are explicitly recorded as holding none	2 / 16	2026-11-01	/trading-company-check
0 of 6 recorded industry codes say printed-circuit-board manufacturing — they say integrated circuits, vehicle repair, new materials, display devices	0 / 6	2026-11-01	/trading-company-check
6 of 13 Suining entities serve a web page, 4 of 13 over HTTPS, 3 of 13 with any English content, 2 of 13 with their own sitemap	13 entities	2026-11-01	/suining-cluster

The four clusters, measured

Figure	Scope	Dated	Source
Suining, Sichuan: 13 entities, 5,857 insured employees, 9 PCB + 4 harness — and 0 of 13 publish a certificate number; not one of the nine PCB factories publishes an ISO 9001 or IATF 16949 number, or a MOQ	13 entities	2026-11-01	/suining-cluster
Kunshan, Jiangsu: 2 PCB plants, ¥18.945bn against ¥1.18bn revenue — a 16× gap inside one county-level city	2 plants	2026-11-01	/kunshan-cluster
Pearl River Delta: 8 plants, ¥71.1bn combined revenue, 83,482 staff on a mixed basis; 7 of 8 carry a recorded industry code and not one of the seven is a PCB code	8 plants	2026-11-01	/prd-cluster
Yangtze Delta: 3 plants, ¥60.3bn group revenue of which ¥44.9bn is board; gross margin 17.59% against 36.91% on the two board lines that disclose one	3 plants	2026-11-01	/yangtze-delta

What is deliberately absent

A page of numbers is only worth something if its absences are as stated as its findings. Five things are not on this page and will not be:

The five exclusions

No prices. A unit price that is right this quarter is wrong the next, and a directory cannot maintain one. Where a quote is discussed, it is discussed as risk allocation, not as a number.
No rankings. Nothing here orders plants by quality, and no manufacturer can pay to appear. A count of what is published is not a judgement of who is good.
No estimates. Where 12 of 16 publish nothing about a field, the finding is that 12 publish nothing — not an inferred industry average.
No zeros for blanks. A missing figure stays empty. The one plant whose filing reports 0 insured is reported as 0; the plant that publishes nothing is reported as nothing. The two are not the same fact.
No silent revisions. A census keeps its date. When a new profile is added, the older sweep is not recomputed to absorb it — the next sweep is dated and published separately.

How to cite this page

Copy as-is · CC BY 4.0 · attribution required

 EMS Suppliers, Verified (2026). "Every measured number this
 directory publishes, on one page." emssuppliersverified.com/facts
 [accessed YYYY-MM-DD].

 Inline, one number:
   ... 0 of 16 plants publish a certificate number (EMS Suppliers,
   Verified, sixteen-plant census, 2026-11-01, /certificate-check).

 Inline, the dataset:
   ... 25 plants, 408 field-level records, CC BY 4.0
   (github.com/wzhi36535-code/china-pcb-supplier-data,
   release 2026.11.03).

 Rule for anything you reuse: carry the sweep date. A count without
 its date is a count without its meaning.

The machine-readable version of most of these rows is in the dataset: plants.csv for the 25 rows and their two headcount columns, fields.csv for the 408 field records, clusters.csv for the four cluster totals with their coverage ratios. The BibTeX entry is on the same page.

The dataset (25 rows, 408 field records, CC BY 4.0) · one factory, two headcounts (the 0.8×–26.3× gap explained) · stated capacity vs audited output (the 572 m² benchmark) · certificates: claims without numbers · buyer questions, answered by plant · About & methodology · all sourcing guides.