One factory. 1,495 insured staff. 39,245 employees. Both numbers are published, and both are true.

Two figures, three names, one trap

Almost every argument about supplier size starts with two people quoting different numbers that are both correct. The confusion is not about arithmetic. It is that three different fields share the English word “headcount” and only one of them describes the building you will visit.

The eight plants that publish both

Eight plants in this directory publish a figure of each kind, which makes them the only places where the gap can be measured rather than guessed. Sorted by how far apart the two numbers sit:

Read the last column as a description of the second number, not a correction of the first. Five of these eight sit between 1.1× and 6.0×, which is what an ordinary company with several sites looks like. The two ends of the table are the two things that actually go wrong: a group figure standing in for a plant, and a stale figure standing in for anything.

Why the gap reaches 26×

Four mechanisms produce the spread, and only the first is about size:

The plant whose number is zero

One record in this directory publishes neither figure, and the reason is more useful than the number would have been. Jingming registered in September 2024 and started production in May 2026. Its registry filing reports 0 insured employees — the filing year predates its workforce — while recruitment listings in the same period suggest 50–200 staff. The profile records both statements and neither as fact, and plants.csv leaves both headcount columns blank.

That distinction is the discipline of the whole directory: a blank means nobody published it, a zero means somebody published zero. Collapsing the two is how a pre-revenue plant becomes a factory with no workers, and how a directory ends up with a row that is confidently wrong instead of honestly empty. The field dictionary on the dataset page states the same rule for all 29 columns.

Which number answers which question

Neither figure is better. They answer different questions, and the usual failure is answering a site question with a company number.

The arithmetic that breaks when you mix them

Revenue per employee is the most quoted ratio in supplier evaluation and the easiest to get wrong, because revenue usually comes from a consolidated annual report while headcount can come from either level. Here is the same revenue divided both ways, for the seven plants that publish revenue and both headcounts:

The left column says DSBJ produces 34 times as much revenue per head as Yingchuangli. The right column says the eight largest companies in this table sit within 1.8× of each other. The right column is the true statement: most of these revenue lines are consolidated group revenue, so pairing them with a single-site headcount divides a company by a plant. Yingchuangli, the one single-site company here, reads almost identically down both columns — which is what a correctly matched pair looks like.

Ten minutes, five steps

 HEADCOUNT — five steps

 1. Get the entity, not the brand
    Exact legal name and unified social credit code (USCC) from the
    entity that will sign the contract. The brand on the website is
    frequently not the legal person that invoices.

 2. Pull three years of the insurance roll
    National enterprise credit registry, the company's annual report,
    the social-insurance field. Record the number AND the year for
    each. Two years of decline is a different supplier from two years
    of growth.

 3. If it is listed, pull the group figure
    Annual report, employees section: total, plus the production /
    technical / sales split. Note whether the figure is consolidated
    and how many plants it covers.

 4. Write both down before comparing
    Insured ____ (year ___) at ______________________ (legal name)
    Group   ____ (year ___) covering ___ plants in ___ countries
    A ratio without two years attached is not a measurement.

 5. Ask the two questions the filings cannot answer
    What share of people on the floor are on your own insurance roll
    rather than an agency's? How many shifts, and how many lines?
    Neither is in any public document. Both decide throughput.

Step 5 is the one that changes estimates. A plant reporting 1,000 insured employees with a 40% dispatch share has roughly 1,600 people inside it, and the labour content of your quotation was built on the smaller number. That is not deception; it is how the industry staffs a ramp, and it is invisible until someone asks.

Where the figures come from, free

The 124 source URLs behind these profiles, keyed to the record that cites each one, are published in the dataset as sources.csv; the headcounts themselves are in plants.csv, with an explicit note per row saying which figure came from which document.

What this page cannot tell you

It cannot tell you how many people are in the building today: every number here is a filing, and filings lag reality by up to eighteen months. It cannot see dispatched labour, which is precisely the labour a ramp is built on, and no public register in China discloses it by site. It does not audit, and it does not rank plants by size — a 53-person plant and a 7,052-person plant can both be the right supplier for different orders. Where a plant publishes one figure, this page leaves the other column empty instead of deriving it, so the eight rows above are a census of disclosure, not a sample. And nothing here is legal, tax or procurement advice: the headcount is one input, alongside the entity check in trading company or factory and the survival screen in is the factory still running?

Related

The dataset (all 25 rows, both headcount columns, and the field dictionary) · stated capacity vs audited output (the ratio that uses insured headcount deliberately) · Kunshan, measured (where the insured figure misleads at both ends for two different reasons) · is the factory still running? (the headcount trend as a survival signal) · trading company or factory (which legal person the roll belongs to) · verify a Chinese PCB supplier in 20 minutes · all sourcing guides.