Measurement recordUpdated when we re-test

Chinese official sites: 5 of 8 returned no HTTP 200.

A failed request is not evidence about a supplier. On 8 August 2026, five of eight official hosts returned no HTTP 200 to our scripted requests from mainland China. Later sessions and an overseas panel show why each result needs its date, client and controls. See the separate GSXT study for that host.

See what to do after a failed lookup →

Published 8 August 2026 · · Domestic panel dated 8 and 22 August. The overseas panel below dated 26 August.

Short answer

On 8 August 2026, five of eight official hosts returned no HTTP 200 in three browser-header rounds from mainland China. Both controls returned 200 in their recorded first-round requests. None of the three target responses with status 200 carried an English-version marker. These scripts did not run browser searches.

26 August overseas result: the court host returned 403 in 23 requests from 19 countries and 200 in one request from Hong Kong. Node sets varied by host and round. This does not isolate geography as the cause. Read the counts and control limits →

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.

HostWhat buyers use it forResult (3 of 3)English front page
www.gsxt.gov.cnCompany registry (GSXT)521
www.creditchina.gov.cnCredit China: penalties, dishonesty lists412
wenshu.court.gov.cnCourt judgments200No marker found
zxgk.court.gov.cnCourt enforcement records200 (browser UA only)No marker found
sbj.cnipa.gov.cnTrademark office403
credit.customs.gov.cnCustoms enterprise credit412
openstd.samr.gov.cnNational standards (GB) texts200No marker found
cx.cnca.cnCCC certification queries521
www.gov.cnControl: government host200
www.baidu.comControl: commercial host200
The user-agent finding

zxgk.court.gov.cn returned 403 to the bare client and 200 to the script with a browser user-agent header in the same session. The other seven hosts returned the same codes under both profiles. Both clients were scripts; neither result establishes what a person with a browser would see.

What 521, 412 and 403 mean here

These codes record HTTP responses, but do not identify which server or intermediary returned them. IANA lists 521 as unassigned. RFC 9110 defines 412 as a failed request precondition; 403 means the request was refused. The codes alone do not establish an outage, geographic block or its cause. Sources read 9 September 2026.

Correction, 27 August 2026: what a 412 does and does not show

The saved follow-up notes report 412 with both plain and fuller browser headers. They do not isolate the cause. TLS handling or a JavaScript challenge remained possible explanations, not findings. The attempted overseas browser result failed a government-host control and was excluded.

The two hosts returned 412 to the recorded method in the tested locations. That does not establish that they were unreachable by a browser. The earlier wording, “refuse every client”, was too broad. The court counts likewise describe requests; they do not by themselves prove the cause of the different outcomes.

A status code does not prove a browser search works

The 14 August challenge write-up discusses a separate GSXT observation. This panel did not run that browser test. Its saved status codes cannot show that a challenge was passed or a company search completed.

We do not attempt to defeat challenges. Use this panel for its dated HTTP results, not as proof that all scripts fail or all browsers work. Human checks and commercial-provider routes have separate scopes.

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 compare the recorded response codes. Rounds two and three used the browser user agent.
  • Controls. The same-session requests to www.gov.cn and www.baidu.com returned 200. That distinguishes those responses from the target responses. It does not prove every route to China worked or locate the cause of a target failure.
  • 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

  • The first panel covers mainland requests only. Earlier overseas attempts failed their controls and remain excluded. The 26 August node data is a separate panel with unpaired nodes and one missing government-control status. It does not close every overseas-access question.
  • 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

Keep access and the supplier claim separate. The first 120 KB scan found no English marker in the three target responses with status 200. It did not search for English sub-sites or test interactive searches. A buyer may use the official search steps or request a dated human check with an agreed scope.

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

The 22 August 2026 retest used the same eight targets and two user agents, with no redirects or retries. It ran from a Shanghai cloud node; the baseline used a consumer connection. Both date and network changed.

Eight official hosts plus two controls, re-probed after two weeks. Domestic vantage; controls healthy.
Result after two weeksHosts
Status code identical to 8 August7 of 8, plus both controls
Changed: stopped refusing command-line clients (403/200 → 200/200)1: the court enforcement portal

Seven of eight hosts repeated their codes on 8 and 22 August. These are two sessions with no monitoring between them. They do not establish that a retry between those dates would fail.

One host had a different status pair at the two dates. The court host changed from 403/200 to 200/200. These observations do not isolate a filtering rule or a network cause. Keep both dates and networks when citing the comparison.

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.

Eight official hosts plus two controls, re-probed thirty days after the 8 August baseline. Domestic vantage; controls healthy. Queried 7 September 2026.
Result after thirty daysHosts
Status codes identical to 8 August, for both clients8 of 8, plus both controls
Still serving no front page to either client5 of 8
200 with a browser header; 403 with the bare client1 of 8: the court enforcement portal

The court host returned 200/200 on 22 August, then 403/200 on 7 September, matching the 8 August pair. The recorded response pair changed in both directions. The observations do not explain the cause or show whether a buyer could complete a search.

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: 240 overseas requests, with unpaired controls

The 26 August file holds 240 requests across 34 countries and territories: 24 requests for each of eight targets and two controls, split across two rounds. The court host alone has 20 country or territory labels. Hosts used different node sets; this is not a fully paired country comparison.

The same eight hosts and two controls, requested from overseas nodes on 26 August 2026. “Inside” is a mainland Chinese cloud host requested in the same period with a desktop-browser user agent.
HostInside mainland ChinaFrom overseas nodesReading
zxgk.court.gov.cn
Court enforcement records
20023 × 403; 1 × 20024 requests; 20 country/territory labels
sbj.cnipa.gov.cn
Trademark office
40323 × 301; 1 × 40324 requests; 18 country/territory labels
www.gsxt.gov.cn
Company registry
52121 × 403; 1 × no status; 2 × 52124 requests; 19 country/territory labels
wenshu.court.gov.cn
Court judgments
20023 × 200; 1 × no status24 requests; 16 country/territory labels
openstd.samr.gov.cn
National standards
20024 × 20024 requests; 17 country/territory labels
www.creditchina.gov.cn
Credit China
41224 × 41224 requests; 17 country/territory labels
credit.customs.gov.cn
Customs enterprise credit
41223 × 412; 1 × no status24 requests; 17 country/territory labels
cx.cnca.cn
CCC certification
52123 × 521; 1 × no status24 requests; 17 country/territory labels
www.gov.cn
Control
20023 × 200; 1 × no status24 requests; 19 country/territory labels
www.baidu.com
Control
20024 × 20024 requests; 19 country/territory labels
Read this narrowly, because the narrow reading is the one that survives

The court file contains 23 responses with status 403 and one with 200. The 200 came from Hong Kong in round two. Other hosts also had exceptions to their most common status, shown above. These counts do not establish a general geographic block or predict what a buyer would see.

What the location comparison can show

The domestic requests and overseas nodes used different network paths and client setups. Even within the node network, changing a node changes its IP, route and resolved destination. The court result is a location-associated difference; it does not isolate a country rule or exclude client handling.

  • Keep the dates separate. Domestic court codes changed from 403/200 on 8 August to 200/200 on 15 and 22 August, then back to 403/200 on 7 September. These are short sessions. They do not establish the domestic result for every client on 26 August.
  • Only one Hong Kong response. It returned 200 only in round two. India, Turkey and the United States have 403 observations in both rounds. The shared tool narrows the comparison but does not hold IP, routing or resolved server constant.

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. The complete file spans 34 countries and territories; each host spans 16 to 20. The court subset spans 20. Use each row’s node, round and host. The file does not contain 34 paired country tests.
  • Control limits. The government control returned 200 in 23 requests and no HTTP status in one. The commercial control returned 200 in all 24. This corrects the earlier claim that every control answered. Target and control nodes differ, so these results do not clear every target path. Earlier failed-control attempts remain excluded; the published file is retained for review with this limitation.
  • Two sessions. Each host’s most common status was the same in both rounds, but exceptions occurred. Two sessions do not prove a lasting refusal. Counts of all responses, including missing statuses, appear in the table.
  • Saved reports. Each row includes its original report URL. The court reports are 48dc04dek23b and 48dc4c0ck9d0. The downloads below preserve the recorded observations; report availability has not been rechecked in this review.

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. The domestic sessions also recorded different status pairs on different dates. Neither panel predicts future access or isolates a filtering cause.
  • 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. Its November 2025 tests covered 13,508 government websites using residential proxies in 14 countries, with rotating request headers and retries. We checked the published methods on 10 September 2026. They differ from this panel and do not validate our node controls or establish the cause of our Hong Kong result.

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

A separate commercial-source run on 21–22 August 2026 requested nineteen dimensions for 45 selected company codes. The selection method does not establish 45 verified manufacturer identities. These returned counts do not measure what the official hosts would return.

What a commercial route returned for 45 selected codes. Queried 21–22 August 2026.
DimensionCodes with returned records
Company-type query45 of 45
Shareholders and annual reports44 of 45
Change history and import/export credit41 of 45
Qualification certificates36 of 45

Ask which source supplied the result and what is missing. A commercial-provider hit does not prove complete records or direct official access. This table does not show that a China-side person is necessary for every check.

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, “Chinese official sites: 5 of 8 returned no HTTP 200”, 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/

BibTeX
@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.

Official source timing out? Prepare a company check using the details you hold. Review the available route and its scope; this access study does not promise a live lookup result.