Eight official Chinese sources buyers are told to check. We asked each one for its front page.

Before treating a failed lookup as supplier evidence

Separate a source refusing the request from a supplier having no record. The panel below measures the source only.

Each target host was requested at its root three times with a browser user agent on 8 August 2026. The browser-user-agent request returned the same status in all three rounds. An extra CLI request differed for the court-enforcement host. Control hosts were requested from the same machine in the same session.

Method

• Requests. HTTPS GET to each host root, three rounds, recording the status code. Round one was sent as a matched pair: a plain command-line user agent and a desktop-browser user agent, to expose client filtering. Rounds two and three used the browser user agent.
• Controls. www.gov.cn (a Chinese government host) and www.baidu.com (a Chinese commercial host) were requested in the same session. Both returned 200, so the failures are specific to these hosts — the connection itself, and China generally, both passed the control, unlike .gov.cn generally.
• English detection. For hosts that served, the first 120 KB of front-page markup was scanned for English-version markers (an English link, lang="en", an /en/ path). This is a front-page markup scan, nothing deeper.
• What we did not do. No searches were submitted, no CAPTCHA was touched, no data was collected from any of these portals, and nothing was proxied or mirrored. Measuring whether a front door opens is a different activity from walking through it.

What this does not establish

• This panel says nothing about overseas access. The eight hosts above were measured from inside mainland China only. Two earlier attempts at an overseas vantage were discarded because they failed their own control test. The route did not carry .gov.cn at all, so target and control failed together and the data proved nothing. That gap was closed on 26 August with a vantage that passes the control rule. The result is a separate panel and is reported below, not merged into this one.
• Nothing beyond the front page. A 200 on the root does not mean a search completes: the interactive path may require verification steps this measurement never touched. And an English sub-site could exist somewhere the front-page markup does not announce.
• Nothing permanent. One connection, one date. A host that refuses this week may serve next week. A different Chinese ISP may see something else. Re-test before citing it; conditions move.
• Nothing about any company. An unreachable portal is a fact about infrastructure on a date. It is no evidence that a record is missing, or that anyone is concealing anything.

What a buyer does with this

The uncomfortable reading is this. The standard advice is to check the official sources yourself. It assumes a stack of portals that, on our test date, mostly declined to serve their own front pages even from inside China, and offered no English when they did. That is not a reason to skip verification. It is the reason verification is usually executed China-side by someone for whom these portals, their search paths and their verification steps are a daily logistics problem. That is the work this desk does, with the source and query date named per line.

What is still doable while those portals are refusing. Two checks need none of them. Validate the 18-character code offline → The arithmetic runs in your browser, so no Chinese host is involved at all. Read the registered scope text the supplier already sent you → That gives a first-pass factory-or-trader signal from a document you already hold. Both are free, and neither is affected by the availability measured above.

Related: the registry-availability record this panel extends · how to read a registration record once you have one.

Re-probed two weeks later: seven of eight unchanged, one moved

This study is a snapshot, and a snapshot of client filtering can go out of date. So I ran the same probe again on 22 August 2026. Same target set, same two user agents, no redirects followed, no retries, control hosts in the round, this time from the domestic vantage point.

Refusal here is a persistent state. It never resolved on retry. Seven of eight hosts returned exactly the same code a fortnight apart. That is the part of this study that holds up. “try again later” is not a workaround for these sources.

The host that moved is the more useful finding, and it cuts against the study as much as for it. Client filtering changed within two weeks, so any statement of the form “this source cannot be opened” carries the date I checked it and needs re-probing. Written once and left standing, it rots.

The overseas round of 22 August was voided in full. A government control host returned nothing while a commercial control returned 200, which under this study’s own rule 3 means the round describes the link. The targets stay unmeasured. Both rounds and the controls: the two-week retest; raw records in china-official-source-retest-2026-08-22.json.

Re-probed a month later: all eight hosts back on the 8 August pattern

The probe ran again on 7 September 2026 from the same domestic vantage. Same eight targets, same two user agents, no redirects followed, no retries. Both control hosts answered 200.

The court enforcement portal is the host to watch. On 22 August it had stopped refusing command-line clients. On 7 September it refused them again, with the same 403 as on 8 August. Client filtering on that host is not a one-way change. It moves in both directions within weeks, so a dated observation is the only kind worth quoting.

The overseas round of 7 September was voided under rule 3, as the 22 August round was. The government control host reset the connection while the commercial control answered 200, so that round describes the egress and says nothing about the targets. Dataset: china-official-source-panel-retest-2026-09-07.csv; raw probe: official-source-probe-2026-09-07-direct.json.

26 August: the first valid overseas vantage

Every version of this page until today carried the same admission in its limits section. That it could say nothing about access from outside China, because both attempts at an overseas vantage had failed their own control test. That is now resolved. The panel below was measured from 20 locations in 20 countries and territories, in two independent rounds on 26 August 2026, using public measurement nodes whose network paths are unrelated to ours.

How we separated geography from client filtering

This page has already documented that at least one of these hosts filters on client type rather than location. On 8 August the court enforcement portal returned 403 to a command-line client and 200 to a browser user agent in the same minute. A 403 from abroad could therefore be about the client as much as the country. Two independent observations rule that out:

• The portal stopped filtering by client between 8 and 15 August. The seven-day retest recorded the change: 403/200 became 200/200. The fourteen-day retest on 22 August found it unchanged, so by the time of this measurement both client types were being served inside China.
• Hong Kong is the control that isolates origin. Every overseas node ran the same tool, sending the same request in the same shape. The Hong Kong node received 200; nodes in 19 other countries received 403. When the client is held constant and only the origin changes, the origin is the variable that is doing the work. That node answered in the second round only. The refusals are the replicated half: India, Turkey and the United States were measured in both rounds, and were refused in both.

Neither observation tells us how the distinction is enforced, and we have not tried to find out. Filtering could sit at the network edge, in a CDN rule, or in the application. This panel reports what a request receives. The mechanism behind it is out of scope.

Method, and the rule that voided two earlier attempts

• Nodes. Public measurement nodes in Austria, Bulgaria, Cyprus, Finland, Germany, Hong Kong, India, Indonesia, Iran, Israel, Kazakhstan, Moldova, the Netherlands, Romania, Russia, Serbia, Singapore, Turkey, Ukraine and the United States. Their network paths are unrelated to ours, which is the entire point, and the thing our own two attempts could not offer.
• The control rule. This study voids any round in which the control hosts do not answer, because a round where target and control fail together describes the link, and nothing else. Both earlier overseas attempts were voided under that rule, as was the overseas round of 22 August. In both rounds reported here the controls returned 200.
• Two rounds. Never one. A single snapshot cannot separate a persistent refusal from a momentary fault, so the panel was run twice. Every host’s dominant status was identical across both rounds.
• Independently checkable. Each round is retained by the measurement service at a public report URL. The court-enforcement rounds are 48dc04dek23b and 48dc4c0ck9d0 — open either and you are looking at the same raw measurement we are.

What this overseas panel still does not establish

• Nothing about a person with a browser. These are automated requests from data-centre nodes. A buyer sitting in Rotterdam with an ordinary browser and an ordinary connection may see something different, and we have not measured that.
• Nothing about intent. A 403 is a refusal. It is not a motive. Geographic access rules exist for many reasons, including licensing, load and abuse control.
• Nothing permanent. Two rounds on one day. Client filtering on this very host changed within two weeks in August; origin filtering can change just as fast. The date on this panel is load-bearing.
• Nothing beyond the front door. A 200 means the root responded. It does not mean a search completes, and it does not mean the record you want is reachable behind it.

The data. Every observation behind this panel, one row per node per host per round — 240 rows across 34 countries and territories, with the public report URL on each row: CSV · JSON. Recompute any figure above from it yourself.

Archived independently of this site, each with its own DOI: Zenodo · Harvard Dataverse · figshare. If this page ever disappears, the measurement does not.

Independent context: Vincent Brussee, “Conceptualizing the reverse great firewall”, Journal of Cybersecurity 12(1), 2026, measured 13,508 Chinese government websites from 14 countries and reports 91.3% availability from Shanghai, 66.4% from Hong Kong and 46.7–49.7% elsewhere. That study did not segment by function. This panel is eight hosts chosen for what a foreign buyer actually needs, and our Hong Kong result sits where his numbers predict it would.

What the records hold, for anyone deciding whether the obstacle is worth clearing

Five of eight sources refused. That is a statement about doors. Whether the effort of getting past them is justified depends on what is behind, which this study does not measure. On 21–22 August 2026 I queried nineteen registry dimensions for 45 Chinese manufacturers through a licensed commercial route.

The file behind the refusal is substantial. That is the honest case for arranging China-side access rather than concluding the record is not worth reaching, and it is also why this study exists. A buyer needs to know which of those doors will not open before deciding how to get the file.

This is a different access route. It is not a way around the refusals measured above, and it does not contradict any observation in this study.

Citing this panel

Quote or reproduce these results freely, including commercially, provided the test dates (8 and 22 August 2026 domestic; 26 August 2026 overseas), the vantage for each panel and the stated limits travel with them. The dates are load-bearing: a status code from August 2026 says nothing about these hosts today. Machine-readable evidence, every host and round: the domestic panel data · the overseas panel data.

Archived copies, each with its own DOI, resolving independently of this site: Zenodo · Harvard Dataverse · figshare.

Currawong, “Five of eight official Chinese sources would not open”, three-round observations of eight official hosts with same-session controls, 8 August 2026, from one consumer connection inside mainland China. https://currawongweb.com/verify/china-official-source-availability/

@dataset{currawong_verify_china_official_source_availability_china_official_source_panel_2026_08_08_2026,
  author    = {Bao L. Zhou},
  title     = {{China official verification source availability panel}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21859883},
  url       = {https://doi.org/10.5281/zenodo.21859883}
}

@dataset{currawong_verify_china_official_source_availability_china_official_source_overseas_2026_08_26_2026,
  author    = {Bao L. Zhou},
  title     = {{Which Chinese official verification portals answer from outside China: 240 observations}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22108839},
  url       = {https://doi.org/10.5281/zenodo.22108839}
}

If you re-run this from another network or date and see something different, we want to hear it. 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, and it is not a statement about any company or about the completeness of any official database.

Related: Seven days later, we re-measured all eight sources · Five client profiles against the registry