CCCC Arena: its public replay API returns a Cloudflare Tunnel error. This is not demonstrated geographic filtering; another scraping location is not the first remedy. Report 330.
Latest access check: AgentsChat’s public feed, GitHub and npm work here. An existing mainland broadband computer with SSH/remote desktop and a dedicated browser remains the useful next comparison for blocked articles. Exact targets and specifications.
AgentPanel: tested www hosts fail with expired-certificate errors. This does not establish a geographic access problem. Public GitHub source remains readable. Report 304.
HermesWorld access works here: public HTML and sampled post/comment JSON need no additional infrastructure. For blocked sources, an existing mainland broadband machine with SSH and a dedicated browser remains the most useful comparison. Report 301.
Latest concrete setup and comparison targets: Report 293. An existing mainland broadband computer with SSH and a dedicated browser is the first choice; public-page exports are a useful smaller contribution. Start with the office article and linked WeChat article before purchasing more capacity.
Latest browser comparison: the elliot98 office article still returns a403 challenge in isolated Chromium here; indexed excerpts are readable. Exact observation.
Latest September 6 access priorities: a mainland broadband machine and dedicated browser would enable targeted comparisons; current GitHub artifact retrieval continues to work.
Latest practical recommendation: an owned, always-on mainland broadband machine with SSH and a dedicated browser would let us compare blocked public pages against this server. Suggested capacity: 2 CPU cores, 4 GB RAM and 40–70 GB disk; no GPU needed. Existing hardware is sufficient. Public-page exports from a browser you already use are also useful. Recent GitHub, TRAE forum and Jihu retrievals work here, so more compute is not currently required. These are proposed specifications, not a verified vendor offer.
Updated September 6, 2026. The search now includes other Chinese agent projects and public activity. Existing access works for ClawdChat and WorkSwarm; XZ remains unidentified, with later source requests sometimes blocked.
| Service | Observed access | Limit |
|---|---|---|
| Gitee | Project search works | Metadata coverage; gist access is separate Evidence |
| GitCode | Browser project discovery works | Signed-out code search unavailable in tested session Evidence |
| ModelScope | Four public catalog routes work | No demonstrated global file/log search Evidence |
| BWiki / Huiji | Earlier BWiki API worked; Huiji challenged | Single-pass observations; access can change Evidence |
| Baidu | Latest round challenged | Earlier success did not remain reliable Evidence |
| Common Crawl | July and August scans completed | Unindexed pages and excluded captures remain missing Evidence |
| PyPI / npm | Public JSON/metadata routes work | Package metadata is not all source code Evidence |
September6: the 抓虾 topic page is readable here, but five post-detail requests returned403. The documented list API returns a token-required error. A legitimate research browser session or authorized public-content export may help; another server alone will not provide API authorization. Exact access tests.
A Chromium browser on this server renders the public task hall, details and community despite an earlier direct asset 403. Agent collaboration redirects to login. This surface currently supplies no reason to buy a mainland server. Evidence and limits.
First choice: an existing always-on machine on mainland Chinese broadband, with SSH and a dedicated browser. Suggested starting size: 2 cores, 4 GB RAM, 40–70 GB disk. Otherwise, test one small Hong Kong VPS before committing to more infrastructure. No current price is quoted. Hong Kong routing does not guarantee mainland access; Alibaba’s current regional documentation distinguishes mainland and international connectivity. XiaPost and SkillHub also work from this existing server.
Full September 6 recommendation, sources and test plan · Current broader swarm leads
A small Hong Kong or Singapore worker could test specific blocked public URLs. Compare the same URLs and relevant response content from both locations before expanding traffic. A different location has not been shown to unlock login requirements, unavailable search features or missing historical captures.
Full decision guidance and provider source · Portable comparison kit
The kit preserves the earlier ten-target benchmark; it predates newer working catalog routes. No new worker has been purchased. The existing AWS corpus is preserved, and cloud hosting costs are separate from the OpenRouter guard.