Does the China company registry actually open? What our tests found.

The observations

Each row is a request to the host root, recorded with its status code. Control hosts were requested from the same machine in the same session, which is what makes the target rows mean anything.

8 August 2026: consumer connection, mainland China

Two of the three target attempts were run as a matched pair: the same request with a default command-line user agent and with a desktop browser user agent. Both returned the same code for both targets, so the user agent is not the variable here.

5 August 2026: same connection, three days earlier

The 8 August run was carried out independently, without reference to the earlier figures, and reproduced them. Two observations three days apart is not a trend, but it is better than one.

13 August 2026: twelve countries, through a third-party node network

The observations above are all from one connection inside mainland China. This run finally adds the overseas side, using a public multi-node checker so that the requests leave from hosts in twelve countries instead of our own egress. Our own egress is the failure mode that cost us the withdrawn conclusion below.

Target: 0 of 12 nodes served a page. Control: 12 of 12 served a page. Read the first two rows closely: Jakarta and Iran appear in both runs. In those cases, the target and control were requested from the same node. A node that returns 200 for one Chinese government host and 403 for another is not a node with a routing problem.

How this was measured, and why the control rows are the important ones

A single failing request tells you almost nothing. It could be the target, the network, the client, or the moment. Each control host is there to remove one of those explanations:

• www.baidu.com: a Chinese commercial host. If it also failed, the finding would be “this connection cannot reach China”, and the registry would stay unmeasured.
• www.gov.cn: a Chinese government host. This is the row that matters most. If it failed too, the finding would be “government hosts are unreachable from here”. It returned 200, so whatever is happening is specific to the registry, while the control clears .gov.cn.
• The user-agent pair: the same request sent twice, once as a plain command-line client and once as a desktop browser. Identical results remove “they are filtering non-browser clients” as the explanation.

Requests were made to the host root only. This page does not submit searches, solve or bypass CAPTCHA, automate data collection, proxy the portal, or mirror any record. Measuring whether a front door opens is a different activity from walking through it, and only the first one is described here.

The method for measuring China official source availability explains the design behind those control rows. It covers paired vantage points, paired user agents, same-session controls and the artefacts that can make a one-view test misleading. It gives readers what they need to repeat or contest these observations first-hand.

What this does not establish

• Nothing about any company. A registry that will not load is not evidence that a record is missing, that a supplier is unregistered, or that anyone is concealing anything. It is a fact about the portal on a date. It says nothing about any counterparty.
• Nothing about what a real browser abroad would see. The overseas rows come from a node network: no JavaScript, no browser fingerprint, no session: so they measure what a script sees, and only that. A portal that returns 403 to those clients could still serve a page, or a verification step: to a person with a browser. That gap is open, and until it is closed the overseas rows say that programmatic requests are declined. What a human buyer sees may differ.
• Nothing permanent. Three dates. The mainland rows are one connection. The overseas rows are one moment across twelve countries. A portal that errors this week may serve normally next week, and a different Chinese ISP may see something else entirely. Re-test before citing it; conditions move.
• Nothing about the interactive path. Even when the portal loads, it may require verification steps before returning a record. A 200 on the front page would not by itself mean a search will complete.

A conclusion we withdrew

An earlier version of our registration guide stated that Chinese government sites were unreachable from outside China. The basis was our own overseas egress failing TLS against gsxt.gov.cn, creditchina.gov.cn and www.gov.cn alike. That was wrong, and how it broke matters, because the same trap is available to anyone running this test.

The measurement lacked a control: a government host belonging to some other country. When we added one, the reasoning collapsed. www.abf.gov.au, the Australian Border Force, also failed TLS through that same egress, while serving normally on a direct connection from China. An Australian government site does not block Australian traffic and serve Chinese traffic. The failure was a property of our route. The destinations were fine.

Everything that route produced about overseas access was therefore discarded. For several days this page carried no overseas figures at all, because we would rather publish a smaller claim than a confident one we cannot support. The rows dated 13 August 2026 are the replacement, and they are a new measurement based on a different route. They do not travel our route, and they carry a control host through the same third-party network. The withdrawn conclusion stays on this page because the trap that produced it is the most useful thing here.

If you need the registration record anyway

Portal availability is a logistics problem, not a dead end. In the order we would try it:

1. Ask the supplier for the 18-character Unified Social Credit Code in writing and check it against the business licence and the contract. Our code checker validates the structure and check digit in your browser, with no portal access and nothing sent anywhere. The licence’s scope text can be read the same way in the free in-browser check.
2. Ask a counterpart inside China to run the search and send the output with the date they ran it.
3. Have the check run China-side and returned with its sources, query date and stated limits: what this desk does.

Related: how to read a registration record once you have one · where registry checks sit among the things that actually decide an order · the eight-source availability panel that extends this measurement.

What a buyer ends up with when the portal will not open

This study records which official hosts refuse which clients. It does not say what the practical fallback yields. On 21–22 August 2026 I queried nineteen registry dimensions for 45 Chinese manufacturers through a licensed commercial route. Those manufacturers are the companies on the NHTSA vehicle-manufacturer list that resolve to exactly one Chinese entity.

The refusal this study documents is about the door, not the file behind it. A buyer who cannot open the portal is not necessarily locked out of the record. The substantive fields returned for almost every company through a licensed route. This desk exists because that route needs a person on the China side, while a script pointed at the portal is refused.

Two limits worth keeping attached. This is a different kind of access, not a way around the refusal that was measured: nothing here contradicts the observations above. And the record has its own blank spots regardless of route. A branch linking the company to a production address exists for only 8 of those 45, and no access method improves that.

Citing this measurement

You are welcome to quote or reproduce these figures, including commercially, provided the observation date and the stated limits travel with them. The dates are load-bearing: a status code from August 2026 says nothing about the portal today, and citing it without its date would misrepresent it. Machine-readable evidence: the 5 August paired-control data and the 8 August replication and the 13 August twelve-country round.

Archived copies, each with its own DOI, resolving independently of this site: Zenodo · Harvard Dataverse · figshare (the 5 and 8 August rounds). An earlier Zenodo link on this page pointed at the related official-source availability panel, which holds only one of this study’s data files. That link was removed on 21 August 2026 and the correct record is the one above.

A citation with everything it needs:

Currawong, “China company registry availability: what I measured”, observations of 5 and 8 August 2026 from one consumer connection inside mainland China, and of 13 August 2026 from twelve countries through a third-party node network. https://currawongweb.com/verify/china-company-registry-availability/

@dataset{currawong_verify_china_company_registry_availability_2026,
  author    = {Bao L. Zhou},
  title     = {{China company registry availability observations}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21947623},
  url       = {https://doi.org/10.5281/zenodo.21947623}
}

If you re-run this and see something different, another Chinese ISP, another date, a real browser session from abroad. We would rather hear it than be the only source. Corrections that survive checking get published here with attribution, including ones that contradict us.

Browse all measured studies and methods in the research index.

This page reports connectivity observations. It is not legal advice, not a compliance opinion, and not a statement about any company’s registration status.

Once the registry does open, what it publishes and what a commercial database holds line up more closely than the wording suggests. See the field-by-field comparison of 17 companies.