ChecklistUpdated when we re-test

Eight Chinese official sources: three open from overseas.

Eight official Chinese sources answer different questions about a company. Measured inside mainland China and from 12 overseas nodes on 22 September 2026, three of the eight answered from overseas and five did not. A refused request says nothing about the supplier, and it does not mean a person with a browser cannot reach the site.

Read the two limits before you use this table →

Published 22 September 2026 · · Both columns measured 22 September 2026.

What each source answers, and where it responded

Eight official Chinese verification hosts. Domestic column: one consumer connection inside mainland China, 22 September 2026, with www.gov.cn and www.baidu.com answering in the same session. Overseas column: 12 public measurement nodes per host, 22 September 2026. The prober assigns a different set of nodes to each run, so the ten runs (eight hosts and two controls) used 53 distinct nodes in 32 countries. The two control hosts, measured the same way, returned 200 from all 12 of their nodes.
HostWhat it answersInside mainland ChinaFrom 12 overseas nodesReachable from overseas
wenshu.court.gov.cnCourt judgments20012 of 12 returned 200Opens
openstd.samr.gov.cnNational standard (GB) texts20011 of 12 returned 200; 1 gave no statusOpens
zxgk.court.gov.cnCourt enforcement records200 to a browser user agent; 403 to a command-line one12 of 12 returned 403Refused
sbj.cnipa.gov.cnTrademark office200 (it returned 403 on 8 August)12 of 12 returned 200Opens (it redirected on 26 August)
www.gsxt.gov.cnCompany registry (GSXT)5219 of 12 returned 403; 2 gave no status; 1 of 12 returned 521Refused
www.creditchina.gov.cnCredit China: penalties and dishonesty lists41212 of 12 returned 412Refused
credit.customs.gov.cnCustoms enterprise credit41211 of 12 returned 412; 1 gave no statusRefused
cx.cnca.cnCCC certification queries52112 of 12 returned 521No response from either side

Two limits that decide how to read this table

These status codes describe scripted requests to a host front page. They are not a measurement of whether a person with a browser can use the site. The company registry is the clearest case: it returns 521 to a script, while a person in mainland China can still reach it in a browser and work through its verification step. We have not tested a browser from overseas.

The trademark office changed between our earlier measurements and this one. On 8 August 2026 it refused scripted requests inside mainland China, and on 26 August it redirected requests from overseas. On 22 September 2026 it answered 200 from both sides. That is what a dated table is for: the answer changes, and an undated copy of it goes stale without saying so.

A source that refuses a request is not evidence about a supplier. It tells you which route to take next, nothing more.

How to re-test this yourself

The domestic column comes from one request to each host root, sent with a command-line and a browser user agent, following no redirects and making no retries, with two control hosts requested in the same session. A round where the government control host does not answer says nothing about the targets and is discarded.

The overseas column cannot be produced from a single connection. It needs independent nodes in different countries; ours came from a public multi-node prober whose per-host report anyone can open. The report links, one per host, are in the overseas dataset, and the domestic column is in the domestic probe output. Timestamps in both files are UTC; both runs fall on 22 September 2026 in China Standard Time. Because the prober picks the nodes, a node that gave no status on one host was not re-tested against the controls. Every host in the table was answered with an HTTP status by at least ten of its twelve nodes, and that is what shows the link was up. Measuring from overseas through one proxy does not work: a proxy that cannot reach any .gov.cn host will report every source as unreachable, which is a fact about the proxy.

The underlying measurements, their limits and the archived datasets are in the availability study.