derrick459/chinese-official-source-reachability

Reachability of Chinese official verification portals from outside China

If you are doing due diligence on a Chinese supplier, the advice is always "check the official registry". This dataset measures whether you can.

Eight official Chinese verification sources were measured from public vantage points outside mainland China across several rounds between August and September 2026, plus two controls (www.gov.cn and www.baidu.com).

The result is not "Chinese government sites are blocked from abroad". It is more specific and more useful than that. Across 240 HTTP observations from 34 countries in two rounds:

• Three answer normally. The national standards portal openstd.samr.gov.cn returned HTTP 200 on all 24 observations (17 countries). The court judgement database wenshu.court.gov.cn returned 200 on 23 of 24 (16 countries). The trademark office sbj.cnipa.gov.cn returned a 301 redirect, not a refusal.
• Five do not. GSXT — the National Enterprise Credit Information Publicity System, the official company register — returned HTTP 403 on 21 of 24 observations (plus two 521s and one non-response). Credit China and the customs credit portal returned 412 on every observation. The certification registry cx.cnca.cn returned 521. The judgement-defaulter search zxgk.court.gov.cn returned 403 everywhere except Hong Kong, which got a 200.

The refusals are at the application layer, not the network layer. The 50k-row RIPE Atlas file shows the same hosts are reachable at TCP/TLS: handshake completion ran 92–99% across all ten targets, GSXT included (94.0% of 1,726 sslcert measurements). The server is there, it completes the handshake, and then declines to serve you.

That distinction is the point. From the outside a 403, a 412, a JavaScript challenge and a genuinely absent record all look like "it didn't work", and an importer told to "just check the official registry" cannot tell which one they hit. These files separate the cases.

What is in here

Every file is a mirror of a dataset already published on Harvard Dataverse under its own DOI. The DOI is the citable record; this repo exists to make the data loadable without leaving the Hub.

Config	File	Rows	Source DOI
network_reachability	china-portal-network-reachability-2026-09-09.tsv	50,248	10.7910/DVN/EUOGRK
overseas_http_observations	china-official-source-overseas-2026-08-26.tsv	240	10.7910/DVN/ERALRQ
source_panel	china-official-source-panel-2026-08-08.tsv	10	10.7910/DVN/8ACXCP
source_panel_retest	china-official-source-panel-retest-2026-08-15.tsv	11	10.7910/DVN/QDED6N
registry_availability	china-company-registry-availability-2026-08-08.tsv	4	10.7910/DVN/PAMUH3
proxy_paired_control	gov-host-proxy-paired-control-2026-08-05.tsv	8	10.7910/DVN/PAMUH3
javascript_challenge	gsxt-javascript-challenge-2026-08-14.tsv	7	10.7910/DVN/AU6LMG
certificate_registry_access	certificate-registry-access-2026-08-09.tsv	10	10.7910/DVN/UAFGUT

china-official-source-overseas-2026-08-26.json is the same content as the matching TSV, in the original nested JSON form.

Row counts exclude the header line.

network_reachability

The largest file, and the only one at network layer. Measurements were run on the RIPE Atlas platform against the eight portal hostnames. Columns include measurementId, probeId, probeCountry, probeAsn, addressFamily, destinationAddress, packetsSent / packetsReceived, rttAverageMs, handshakeCompleted, responseTimeMs, serverCipher, answerCount and error.

Because measurementId and probeId are retained, every row can be traced back to the measurement on the RIPE Atlas platform rather than taken on trust.

overseas_http_observations

240 HTTP-layer observations: for each host and each node (with node_country_iso), the outcome, http_status, resolved_ip and seconds. what_buyers_use_it_for records why an overseas buyer would open that portal in the first place, which is the column that turns a network result into something a procurement reader can act on.

The small files

These are deliberately small and hand-checkable. source_panel and source_panel_retest are the eight-source panel measured once and re-measured seven days later, under two client profiles, so a reader can see what changed. proxy_paired_control pairs direct against proxied requests for the same target — the control that distinguishes "the host is down" from "the host refuses us". javascript_challenge records the challenge signature GSXT returns to a scripted client, which is the mechanism behind a large share of the failures above.

Limitations, stated plainly

• Vantage points are not a random sample of the internet. They are the public probes and hosts that were available at the time. A portal answering one node in a country does not mean it answers that whole country.
• These are point-in-time measurements, dated in every file. Chinese portals change their protections; a result from August 2026 is evidence about August 2026.
• "Refused" is not one thing. A TLS handshake failure, an HTTP challenge page and an empty result set are different failures with different causes, which is why the network-layer and HTTP-layer files are kept separate rather than collapsed into a single "up/down" flag.
• Nothing here says anything about any individual company. These datasets are about whether the sources respond, not about what they contain.

Provenance and credits

The RIPE Atlas measurements were run with research credits granted by the RIPE NCC (ticket #1094691). Publishing the resulting data was the commitment made when requesting those credits. Use of the RIPE Atlas platform and its own data is governed by the RIPE NCC's terms; this repository does not restate or reinterpret them, and the measurementId column lets you go to the source.

The CC BY 4.0 licence declared here matches the licence on the Harvard Dataverse records this mirrors.

Citation

Cite the Dataverse record for the specific file you use — those carry the DOIs. The full collection, including 17 further datasets not mirrored here (company registration fields, USCI check-digit protection, certificate registries, localisation quality), is at:

https://dataverse.harvard.edu/dataverse/currawong-verification

Methodology write-ups, with the caveats in prose, are published at:

https://currawongweb.com/research/

Who made this

Produced by Currawong while building supplier-verification tooling for overseas buyers. We also sell verification reports, so treat this as work by an interested party — which is exactly why the measurement IDs, dates and vantage points are all in the files. Check it rather than trust it.