What the censor is working hardest on

Ranked terms

Order is DDTI threat. Other layers only join. A blank join is a miss, not a negative.

Social differential

contained-visible: 0. suppressed-invisible: 0.

CDT campaigns

• 3 terms in NYT - Punishment and Silence: China Pulls Out the Disaster Playbook in Tibet
• 2 terms in CyberScoop - Volexity spots another China-aligned threat group exploiting Chrome and Microsoft defects
• 2 terms in Freedom House - The CCP is Trying - and Failing - to Extinguish Defiance among Chinese Citizens
• 2 terms in HRIC - HRIC Digital Rights Report - August 2026
• 2 terms in Safeguard Defenders - CGTN: See the Difference. China’s Party-State Media in Focus
• 2 terms in DW - How China buried Tibet flood images after Nepal disaster
• 2 terms in The Irrawaddy - Guangxi’s Deadly Floods and China’s Propaganda Crisis
• 2 terms in Reuters - With national holidays approaching, Chinese workers demand a proper break
• 2 terms in Vision Times - Former Chinese Journalist Accuses Jiangxi Investigators of Torture, Coercion
• 48 terms in CDT’s “404 Deleted Content Archive” Summary for August 2026
• 37 terms in CDT’s “404 Deleted Content Archive” Summary for July 2026
• 16 terms in Translation: Censored Commentary on Beijing’s Drone Storage Ban
• 17 terms in Wuhan Authorities Punish Theatre and Event Organizers After Comedian Guo Degang Riffs on “Red Tune” Without Prior Approval
• 31 terms in MiniTrue: ByteDance Censors’ List of 5,241 Sensitive Terms
• 28 terms in China’s Online Environment Gets Colder and Darker
• 25 terms in Netizen Voices, on Gyirong Floods: “Controlling Speech is One Area Where ‘China Speed’ is Actually Real.”
• 18 terms in Information Controls Follow Catastrophic Flooding in Gyirong on the China-Nepal Border
• 24 terms in Journalist and #MeToo Activist Sophia Huang Xueqin Released From Prison After Five Years
• 18 terms in Translation: “Eventually, We All Learned to Keep Quiet”
• 18 terms in Translations: Surprise Viral Hit “Niu Lai” Represents “a Revenge-Vote by Chinese Filmgoers” Amid Ban Rumors

Forecast scorecard

9 signals scored over 3040 one-step forecasts; 82% of readings fell inside the 80% band (well calibrated); adaptive bands beat the fixed-quantile baseline on 2 of 9

Abstentions

• No fused layer abstained.

Limits

• The ranking is censor attention allocation from China Digital Times, not a deletion rate.
• A missing layer is an abstention, never a zero and never calm.
• GDELT and Weibo joins are exact-term matches against the current DDTI list.
• The Generative Firewall Index is a named-panel refusal reading, not a motive claim.
• Cross-layer coincidence counts elevated layers; it does not identify a common cause.
• GitHub-as-Refuge reports pressure against persisted prior-presence baselines only.