circumvention·corpus
/

circumvention research · structured · LLM-callable

A structured corpus of how to keep the internet free.

Every paper tagged against a shared taxonomy of censors, detection techniques, and defenses — including work from Roya Ensafi, Amir Houmansadr, Eric Wustrow, Dave Levin, Nick Feamster, Nguyen Phong Hoang, David Fifield, J. Alex Halderman, Niklas Niere, and many others. An MCP server exposes the whole thing to any AI assistant — or ask the corpus directly.

Tab or → to use the example. Or install the MCP server to query from your editor.

papers
416
censors
20
techniques
23
defenses
37

§ 01 why this exists

A layer the field doesn't have yet.

The censorship-circumvention community has wonderful resources: net4people/bbs for discussion, gfw.report for original research, CensorBib as a maintained bibliography, OONI for measurement.

None of them are LLM-callable. None of them have a consistent structured-metadata schema. None of them let an AI assistant compose a corpus query with operational data in the same conversation.

This corpus adds that one missing layer.

§ 02 core papers

Hand-selected as load-bearing.

If a Lantern protocol designer hadn't read these, the team would expect them to be slowed down. Team consensus marks them as core: true; everyone using the corpus sees them surfaced first.

§ 03 self-updating

The corpus keeps crawling without us.

A pipeline polls arXiv, net4people/bbs, gfw.report, PoPETs, FOCI, USENIX Security, ermao.net, and the Paderborn upb-syssec group's publications and blog for new circumvention research, fetches each candidate via wick (browser-grade, residential-IP web access), then asks Claude to propose taxonomy tags and extract findings against the schema. Every new entry lands as an auto-ingest PR labeled for human review. Below: the most recent additions.

§ 04 connect

Plug it into your assistant.

One line. Your AI gains search_papers, get_paper, list_taxonomy, and find_related over the corpus.

How to install