Research from four US and Canadian universities shows China's Great Firewall blocked ~311K domains, with ~41K seemingly by accident, between Apr. and Dec. 2020
Context & Ripple Effects
This study is part of a growing body of measurement work that treats the Great Firewall as something that can be audited rather than guessed at. Tools like GFWeb later industrialized the approach, tracking millions of domains to recover the blocklists directly — data precise enough to pin down when China blocked ChatGPT.
What the university researchers add is scale plus error rate: ~311K domains blocked in nine months of 2020, with ~41K seemingly caught by accident. That accident figure matters because it implies the filter operates on broad matching criteria rather than a clean curated list, which later incidents would bear out.
First-order effects
- The ~41K accidentally blocked domains are mostly owned by parties with no channel to appeal or even confirm they are on the list, so misclassification translates directly into unexplained loss of reach inside China.
Second-order effects
- Opaque, error-prone blocking raises the cost of serving Chinese users for foreign site operators and CDNs, pushing them toward either self-censorship by association or accepting the risk of collateral blacklisting.
Third-order effects
- The pattern holds across later evidence: regional studies show users in Henan denied access to far more sites than a typical user, and an accidental ~74-minute blackout of all HTTPS traffic on port 443 in 2025 — suggesting enforcement is inconsistent across provinces and prone to infrastructure-level mistakes.
- If independent measurement keeps exposing both the blocklists and their errors, the firewall's effectiveness increasingly depends on opacity itself, making censorship auditing a recurring flashpoint between researchers and the state.
The trend: Independent censorship-measurement research is converting the Great Firewall from an opaque political instrument into a quantified, auditable system whose errors and regional inconsistencies are now on the record.