34 “Dongguan Lighting Manufacturers”, Checked Against the Registry

What we were testing

The standard advice given to importers is: look at the supplier’s business scope — if it says manufacturing, it is a factory; if it says sales, it is a trading company.

That advice is everywhere. It appears in sourcing blogs, in YouTube explainers, in the checklist a freight forwarder emails you. It is also, as far as we can tell, largely untested. Nobody publishes how often it is right.

So we tested it on the narrowest, most boring sample we could find: one city, one product category, every company listed in it, no cherry-picking. Dongguan, lighting and lamps. We did not choose companies because they looked suspicious. We took them in the order the directory listed them, and stopped at 34.

The question was not “how many fakes are out there”. The question was: when a buyer runs the check that everyone recommends, how often does that check give the right answer?

Method, exactly as executed

Reproducibility matters more than the result, so here is the procedure without abbreviations.

1. Sampling frame. The public “Dongguan lighting and lamps” category listing on a Chinese B2B directory (Shunqi, 11467.com), which is a directory many overseas buyers reach through search. Taken in listed order, the first 34 entries with a company name.
2. Record retrieval. For each company name we retrieved the registration record and the latest published annual report from the national enterprise credit registry (via commercial API on 5 November 2026; the underlying source is the State Administration for Market Regulation and the annual-report filing system).
3. Fields captured. Registration status, year of establishment, registered capital, paid-in capital, insured headcount from the annual report, national industry code (gate and middle class), whether the business scope contains a production verb, and the registered address.
4. Definition of “production verb”. Any of: 生产 (produce), 制造 (manufacture), 加工 (process), 产销 (produce and sell), 组装 (assemble), 研发生产 (develop and produce). Matched as a substring in the scope text.
5. Matching rule. A company counted as “matched” only when the registry returned an entity with the identical registered name. Similar names were not accepted.

The sample

Founded between 2001 and 2018, median year 2011; 7 of 31 were established in 2015 or later. Median registered capital RMB 1.85 million (roughly US$26,000 at the time of writing). This is what the small and mid-sized end of Chinese manufacturing actually looks like on paper: old enough to be real, small enough that the owner still answers the phone.

Finding 1: the business scope is the weakest signal we tested

Cross-tabulating the two signals buyers are told to rely on:

The bottom-left cell is zero. Not once did we find a company the registry classifies as a manufacturer whose scope omitted production language. In the other direction, 8 of 31 companies — 26% — claim production in their scope while the registry files them under wholesale, retail or technical research.

In other words: the scope never under-claims. It only over-claims. Which makes it structurally useless as a positive test, and only mildly useful as a negative one — if a company’s scope has no production verb at all, that is worth noting. If it has one, you have learned almost nothing.

Why the scope over-claims

Three mechanisms, all mundane:

• The catalogue is broad. Standard expressions like “manufacture of lighting fixtures” are one click away from “wholesale of lighting fixtures”, and a company that intends to do both later will select both now. Scope is written for future flexibility, not present description.
• Buyers demand it. A trading company that cannot show manufacturing language in its scope loses orders and loses the right to issue certain invoices. The incentive to add it is direct.
• Nobody enforces the match. Registration authorities verify that the wording is from the catalogue. They do not verify that a production line exists behind it.

The registry’s own industry code is a better signal, and it is the field almost nobody tells buyers to read. It is assigned in the same transaction as the registration, it is less visible to the company’s marketing team, and it drives statistics that local authorities care about. It is not perfect — see where our method fails — but it disagreed with the scope in 8 of 31 cases, and in those cases it was right far more often than the scope was.

Finding 2: the median company is five people

Every Chinese company files an annual report. One field in it is 参保人数 — the number of employees enrolled in social insurance under that legal entity in that year. It is not a marketing field. It is filed with the tax and social-insurance authorities, and it is the closest thing to an audited headcount that a foreign buyer can obtain for free.

For the 28 companies with a published figure:

Median: 5. Mean: 33. The gap between those two numbers is the whole story of this dataset: a handful of genuine plants (103, 224 and 293 insured employees) pull the average up, while the typical company in this category sits at or below five.

Read the cumulative column instead of the average. Fifty-four percent of the companies with published data report five or fewer insured employees. Sixty-four percent report ten or fewer.

Now put that next to what the same companies publish about themselves. Every one of these is presented to overseas buyers as a lighting manufacturer — a category where the minimum credible production setup is a workshop, a reflow or assembly line, a testing bench and a warehouse. Five people is not a factory floor. Five people is an office with a phone and a freight forwarder on speed dial.

The four zeros

Four companies report zero insured employees. That is a legally coherent state — the entity exists, files, and pays its taxes — but it means the entity that would sign your contract employs nobody on the books. Where the work happens is somewhere else, under someone else’s name, which changes who is liable when the goods are wrong.

Finding 3: one in four has paid in under a third of its capital

Chinese companies register a subscribed capital figure and then pay it in over time. The annual report also discloses how much has actually arrived. Both numbers are public; only the first one makes it onto the supplier’s profile page.

The cleanest example in the sample: a lighting technology company with RMB 10 million subscribed and RMB 0 paid in, registered in room 804 of an office building in a Songshan Lake industrial park, two insured employees, classified by the registry as a wholesale business. Its business scope names production. A buyer comparing two quotes would see a “RMB 10 million” supplier and a “RMB 1 million” supplier and reasonably conclude the first is ten times more substantial. The registry says the first has paid in nothing.

Registered capital is a promise. Paid-in capital is a fact. Only one of them belongs in your supplier comparison.

Finding 4: 3 of 31 no longer legally exist

Two were voluntarily cancelled; one was revoked by the regulator. All three still appear on the directory page as active lighting suppliers, with product descriptions and contact details.

And the picture is worse than that, because 3 of the original 34 could not be matched to any registry record at all — not cancelled, not revoked, simply absent. Common causes are a name that changed after the directory entry was written, a sole-proprietorship registered under a different formal name, or a listing that was never a registered entity to begin with.

Combine the two: 6 of 34 directory entries — 18% — point at a company that either no longer exists or cannot be verified to exist. These are exactly the entries a buyer finds when searching for a new supplier, because directories rank well and old listings never expire.

Finding 5: where they are registered

15 of 31 (48%) are registered at a room-level address — a specific suite or floor inside a building — rather than a standalone plant address, and one more is registered through a cluster registration arrangement (集群注册), a legitimate mechanism that provides a registered address without any premises. Two of the room-level addresses sit in commercial towers rather than industrial parks.

To be fair to the sample: Dongguan’s industrial buildings commonly mix workshops and offices, so “Building 1, Room 101” can be a real production floor. Address alone proves nothing. What it does do is tell you where to look: if the registered address is a room on the eleventh floor of a commercial tower, ask which site actually makes your product, and get that site written into the contract.

Where our method fails

Two companies in this sample would have been misjudged by the rule we just recommended. Both are worth more to you than the statistics above.

Case #27 shows the industry code can understate: a 293-person operation registered as a research company, presumably because that was the category available or advantageous when it was set up. Case #23 shows the wholesale category does not exclude manufacturing — companies that both make and sell often choose the sales category because it matches how they invoice.

Two failures in 31 is a 6% error rate, which is good enough for a screening tool and nowhere near good enough for a decision. This is the entire argument for cross-checking: insured headcount caught both cases, the industry code alone would have missed both.

What this does not prove

We would rather state these limits than have a reader discover them.

• A trading company is not a scam. Many trading companies are better suppliers than factories for small orders: they consolidate, they speak English, they handle documentation, and they do not have a minimum order quantity measured in container loads. The finding here is about misrepresentation, not about fraud.
• Insured headcount is not total workforce. It excludes dispatched labour, agency workers, interns and often probationary staff. A 5-person insured entity can have 30 people on site. It also excludes anyone the company chose not to enrol — which is itself a compliance signal, but not a headcount one.
• One city, one category, 34 companies. This is a case study, not a survey. We make no claim that 29% of Chinese suppliers are trading companies, and any article that quotes us saying so is misquoting us.
• Records lag. Annual reports are filed the following year, and industry codes are updated only when a company bothers to change them. A 2026 snapshot describes 2024 and 2025.
• The directory is not the market. Companies that pay for directory listings are not a random sample of Dongguan lighting. They may be smaller, more export-oriented, or more marketing-driven than the average.

How to use this on your own supplier

Five checks, in this order, all free, all doable in about twenty minutes with the company’s registered name:

1. Confirm the entity is alive. Registration status must be “active”. Cancelled, revoked, or not found — stop here.
2. Read the industry code, not the scope. Manufacturing, or something else? If it is something else, that is a question to ask, not a disqualification.
3. Pull three years of insured headcount. The trend tells you more than the level: 40, 38, 36 is a stable workshop; 12, 5, 2 is a company winding down while still quoting.
4. Compare paid-in to subscribed capital. Under 30% paid in on a company less than five years old is normal. On a company founded in 2009 it is a question.
5. Get the production site named in the contract. The registered address is where the entity lives, not necessarily where your goods are made. Name the manufacturing site, and prohibit substitution without written approval.

If you want the two structural checks that catch what registry data cannot — whether a plant exists at a specific site, and whether the entity on the contract is the entity making the goods — that requires the pollutant discharge permit lookup and, in the end, someone standing at the gate. We cover the documentary version in Trading Company or Real Factory?

Data and reproduction

The full record for all 34 companies — status, year, capital, paid-in, insured headcount, industry codes, production-verb flag and address district — is published as a CSV so that anyone can re-run the arithmetic:

Download the dataset (CSV, 34 rows, 11 fields) →

Company names are replaced with reference codes in the public file. Every underlying record is a public document, retrievable by anyone from the national enterprise credit registry; we have withheld names because the point of the exercise is the method, not a list of companies to avoid. Journalists and researchers who need the named version can write to hello@songxinpartners.com.