1. The Speech Censorship Problem
Western reporting on Chinese censorship frequently focuses on the blocking of Western platforms (Google, YouTube, Wikipedia, NYT). However, for the 1.1 billion internet users living inside China's information ecosystem, the primary mechanism of censorship is domestic speech erasure: the rapid deletion of viral social posts, the silencing of investigative essays, the suppression of trending debates, and the distribution of secret propaganda directives ordering editors not to report on specific events.
WallWatch focuses exclusively on answering: "What conversations did Chinese citizens engage in that authorities erased?"
2. 404 Purged Posts & Citizen Speech Archiving
When a citizen essay on WeChat, an answer on Zhihu, or a post on Weibo goes viral, platform content moderators (operating under Cyberspace Administration of China directives) routinely delete the content within hours, returning standard error notices such as:
WallWatch tracks and archives deleted articles by partnering with independent digital archivists (including China Digital Times' 404 Archive and community archives). For each deleted article, our pipeline preserves:
- Bilingual Documentation: The original Chinese title and translated English synopsis.
- Citizen Speech Quotes: Direct excerpts of what citizens wrote and debated before the text was erased.
- Censorship Rationale: The political, economic, or social trigger that prompted moderators to scrub the piece.
3. Statistical Detection of Vanished Trending Discussions
Trending topic algorithms on Weibo, Baidu, Zhihu, and Douyin reflect what the Chinese public is actively searching for and discussing in real time.
Natural Decay vs. Algorithmic Purge: Naturally waning topics decay according to an exponential distribution over 6 to 12 hours, gradually sliding down the ranking ladder (e.g. rank #5 → #12 → #24 → #45 → off). In contrast, politically sensitive or socially explosive topics are manually expunged:
- A topic ranking in the top 20 with high search heat abruptly disappears between two consecutive 10-minute snapshot cycles.
- Exit heat exceeds 40% of peak heat, confirming that citizen engagement did not fade naturally.
- Search queries for the purged phrase are often downgraded to return only state media commentary or show zero results.
4. Leaked Propaganda Directives (真理部指令)
The Chinese Communist Party’s Central Propaganda Department and CAC regularly issue confidential notices to domestic newsrooms and internet platform compliance teams. Known colloquially as "Directives from the Ministry of Truth" (真理部指令), these notices explicitly specify:
- Topics forbidden from independent reporting.
- Incidents where platforms must enforce "positive energy" and scrub depressing reflections (e.g. "返乡见闻渲染灰暗情绪").
- Mandatory republication of Xinhua and People's Daily wire copy.
WallWatch ingests authenticated directives documented through whistleblower leaks and researcher networks.
5. AI Enrichment & Strict Code-Level Citation Validation
Every 12 hours, WallWatch generates an executive briefing using gemini-3.8-flash via a local headless CLI pipeline on our server. To prevent AI hallucinations:
- Grounding: The LLM is supplied only with the structured catalog of verified censored items for that specific 12-hour window.
- Mandatory Citations: Every generated bullet point must explicitly cite one or more valid item IDs in JSON format.
- Programmatic Filter: Any bullet point lacking citations or citing nonexistent item IDs is rejected by our Python verification engine before publication.