In 2023, a comedian in China told a joke. It involved a dog. It also riffed off a common slogan of the People's Liberation Army.1
The joke went slightly viral. Online nationalists found it. They made a fuss — the kind of fuss that ripples across Chinese social media in hours. The comedian was extensively fined. He was detained for several weeks. And then, over the next few months, comedy clubs across Shanghai — China's biggest city, a rising source of cultural tourism — were shut down. Not the club where he performed. Comedy clubs. The whole scene.
Notice what just happened. A joke about a dog and an army slogan triggered a response that went through two completely different machines.
The first machine is the one Western analysts expect: state censorship. The party-state has the apparatus to silence speech, detain speakers, close venues. It used the apparatus.
The second machine is harder to see from outside. The online nationalists who made a fuss did not get instructions from the party. Nobody at the propaganda department told them to be offended. They were autonomously offended, swarmed the comedian, demanded consequences, and pushed the response upward into the state apparatus where it got executed. The mob was not the state's tool. The mob was operating on its own and the state followed where the mob's energy pointed.
Modern Chinese censorship runs in both directions at once. Top-down through the propaganda department, public security, and platform moderation. Bottom-up through nationalist Weibo mobs that take scalps without state direction. Both are real. Both have gotten worse over the last decade. Understanding the apparatus requires holding both in the same frame.
Wang frames the change he has watched by comparing China to Russia. When Putin began arresting Russia's independent bloggers at the start of the 2022 Ukraine invasion, Wang's reaction was: what sort of amateurish authoritarian regime is this?1 Not because Putin's repression was insufficient — but because Russia in 2022 still had independent bloggers. Real ones. Publishing under their own names. Some of them critical of the war.
That is the test. China squeezed out the equivalent class of voices roughly ten to fifteen years ago. There is now, in Wang's framing, no such thing as independent media in China in any bigger, more organized way.1 Every newspaper is either formally state-owned, formally CCP-owned, or extensively pre-censored by the propaganda department before publication. There is no investigative journalism that does not pass through that filter. There is no independent media on social media either, because the platforms themselves are integrated into the censorship apparatus.
The mechanism Wang reports as his strong suspicion: at companies like Tencent and ByteDance, the largest single department by headcount is probably human censors.1 Not engineers. Not designers. Not algorithm specialists. People sitting at screens reading content as it goes viral and deciding whether to remove it before it goes too viral and too critical. Tens of thousands of workers, full-time, doing nothing but pre-emptive content removal at scale.
The Voices of Spring case from 2022 captures what the apparatus is up against and what it can defeat. During the Shanghai lockdown, somebody made a six-minute video stitching together audio clips from residents trapped in the lockdown — the desperate calls to community workers, the kids without food, the elderly without medicine. It went viral fast. The censors moved to remove it. People tried to keep sharing it. They flipped it upside down. They stretched the audio. They encoded it into the blockchain so it would persist somewhere unreachable. Still the censors managed to get through a lot of the stuff.1 The technical sophistication of evasion — blockchain, format manipulation, mirror sites — was met by the operational scale of the censorship apparatus and largely defeated.
The Wang-website case shows how granular the apparatus has become. In 2022, Wang woke up one morning to discover his personal website, danwang.co, was blocked in China. Not Wikipedia. Not the New York Times. His own rinky-dink site.1 The detection threshold has moved from major Western platforms to individual analyst sites with modest traffic. There was no door to knock on, no censor to ask why, no procedure for review. Whatever algorithm flagged the site was not going to communicate with him about it. He met with the Canadian consul general in Shanghai to assess whether he needed to leave the country in a hurry — there was already context: two Canadian citizens had been taken hostage by Beijing in retaliation for Canada's arrest of the Huawei CFO at U.S. request. The personal-safety calculation was real.
This is the top-down apparatus at present scale: comprehensive, opaque, technologically aggressive, integrated across state and platform layers, indifferent to the targets it picks up.
The harder-to-see machine is the bottom-up one, and Wang names it directly: cancel culture is also very real in China.1 The Western intuition is that cancel culture is a Western phenomenon — a feature of free societies where decentralized social pressure replaces formal censorship. The intuition is wrong. The mechanism transfers across regime types. China has both kinds of pressure operating simultaneously, and the bottom-up version is in some respects more powerful than its Western analog because it has a willing state apparatus to escalate into.
The Weibo mob is the operative unit. A wrong word, a tone-deaf joke, a perceived insult to a national symbol — and the mob assembles. There is a very strong sense that if you say the wrong thing, the mobs from Weibo will come after you.1 What happens next depends on context. Sometimes the mob is just social — pile-on, doxxing, professional consequences. Often it escalates: the mob's outrage gets attention from the propaganda apparatus, which then has political cover to take action that originated as the mob's demand rather than the state's initiative.
Wang is explicit that this is not directed from above. It is not even necessarily directed by anyone from the government. It is just nationalist social media trying to take your scalp.1 The state benefits from the dynamic without having to orchestrate it. The mob does the work of identifying targets, building the case against them, and generating the political pressure for state action. The state can then either ride the wave or decline to ride it, and either way the dissident speech has been disciplined.
The dog-joke case is the canonical example. A comedian made a joke. The mob mobilized around perceived insult to the PLA. The fuss became impossible for the state to ignore. The fine, the detention, and the closure of comedy clubs across Shanghai followed. From the outside, it looks like top-down repression. From the inside, the cascade was bottom-up: the mob's demands set the policy.
The behavioral effect of the two-direction apparatus is something Wang describes as a feature of ordinary Chinese social interaction. People know things can be censored. People know police may read their social media posts. People know they can be called in for questioning based on what they have written or said.1 The result is not just public silence — it is private silence. People are pretty reluctant to really express their core opinions because they are afraid, and they're rightly afraid of the police coming after them if they write the wrong thing or say the wrong thing especially to a foreigner.1
The cost Wang flags is one most analyses of censorship under-weight: it hurts critical thinking skills. Not just speech. The cognitive operation of forming and articulating disagreement requires the experience of doing it over and over in low-stakes settings. When low-stakes settings are not safe — when even private conversations carry risk — the skill atrophies. The censorship apparatus is not just preventing dissident speech from being broadcast. It is preventing the practice of dissident thinking in the population that lives under it.
This is the long-run mechanism by which authoritarian systems hollow out their populations' capacity for the kinds of thought that would, in a different regime, generate the renewals and corrections those systems need. The propaganda apparatus delivers the line. The two-direction censorship apparatus prevents the population from practicing the cognition that would produce alternatives. Over time, the alternatives stop being thinkable.
Wang names the regime's failure mode in passing but vividly: this is just how insensitive the engineering state is. They don't understand a lot of things. They really don't want to have any jokes or mockery.1 This phrase — the insensitive engineering state — points at a structural feature of the apparatus.
The CCP under Xi has been increasingly staffed at senior levels by engineers — people who came up through technical disciplines and apply the engineer's frame to political and cultural problems. The frame treats society as a system to be optimized; humor, mockery, dissent, and ambiguity register as noise rather than signal. When a comedian's joke produces a viral fuss, the engineer's response is to remove the source of the fuss — close the comedy clubs — rather than to consider the function the fuss is performing in the social system. The optimization-frame is incapable of recognizing that some "noise" is structurally necessary to a healthy political culture.
Wang's word for this — pathetic — is sharp.1 He does not mean weak; he means failed in a specific dimension. The engineering state is technically formidable and culturally illiterate. It can build high-speed rail networks that span the country and shut down comedy clubs over a dog joke, and the same disposition produces both. The first is admirable. The second is grim.
1 Wang's account of two-direction censorship covers lines 608-680. Specific claims and their line locations: censorship has gotten "a lot worse over the last 10 years" (line 608); Russia-China comparison on independent bloggers (lines 608-612); China squeezed out the independent bloggers "10-15 years ago" (line 612); no real independent media (lines 612-614); Tencent and ByteDance largest department probably human censors (lines 620-622); Wang's website blocked 2022 (line 624); Voices of Spring 2022 Shanghai lockdown video, blockchain encoding attempt, censors largely successful (lines 618); Canadian consul meeting and two Canadian hostages context (line 628); 2023 comedian dog-joke case + comedy clubs closed in Shanghai (lines 640-644); "insensitive engineering state" framing (line 646); cancel culture is real in China (line 678); Weibo mob mechanism, "trying to take your scalp," not directed from above (lines 678-680); chilling effect on private conversation (lines 634-638); cost to critical thinking skills (line 638).
The framing has at least two productive tensions with existing vault content:
The vault's broader treatment of censorship and propaganda (Le Bon, Bernays, the propaganda-mass-persuasion-hub) treats these phenomena largely as elite-driven. Wang's bottom-up dimension complicates this — the mob is not the elite's tool, and the resulting censorship is not the elite's product even when the elite's apparatus does the executing. The framework needs a place for the autonomous mob as a censorship actor, not merely the elite.
The vault's coverage of Western "cancel culture" treats it as a feature of post-traditional liberal societies — the breakdown of consensus on speech norms, the rise of decentralized social-media-driven enforcement. Wang's claim that the same dynamic exists in modern China at scale suggests that the mechanism is not specifically about liberal post-modernity. The mob is a structural feature of any large-population, social-media-mediated society. The regime type changes what the mob escalates into; it does not change whether the mob exists.
The two-direction censorship framing converges with Le Bon's Crowd corpus on one critical point: the mob, once mobilized, is its own actor and does not need elite direction to do real damage. Le Bon's nineteenth-century crowds and Wang's twenty-first-century Weibo mobs share the structural feature of autonomy from elite control. But the divergence is sharp. Le Bon's crowd is physically present, geographically bounded, and short-lived; once the crowd disperses, the action ends. Wang's mob is virtual, geographically distributed, and persistent — the same mob can reassemble around a new target every week, with the membership partly reshuffled but the underlying disposition intact. The tension reveals: digital mobs are not just bigger Le Bon crowds. They are a different kind of entity, with different decay characteristics, different memory, and different relationships to the state apparatus that has the power to escalate their demands.
Hughes' rapid-deployment behavioral apparatus (six-minute conversational windows, BTE charts, deception detection) operates at the individual scale. Wang's apparatus operates at population scale. The convergence is on the underlying claim that behavioral surveillance shapes behavior — what gets watched gets adjusted. The divergence is on the political infrastructure required: Hughes' tools work for any actor with face-to-face access; Wang's apparatus requires a state-platform-mob architecture that only certain regime types can sustain. The tension reveals: individual surveillance capability is now widely distributed; population-scale surveillance integration is the political-technical achievement of specific authoritarian states, and other regimes that copy the individual tools without the architecture get a different and less powerful system.
Two-direction censorship is a history claim, but the mechanism reaches into psychology (mob dynamics, self-censorship under threat), behavioral mechanics (the mob as influence apparatus), and the broader question of how regime types and platform architectures interact.
Behavioral-Mechanics: Manipulation & Influence Hub — the bottom-up Weibo mob is a behavioral-influence apparatus operating at scale. The mob's mechanisms — pile-on, scapegoating, demand for visible consequences, emotional escalation, identity-mobilization through nationalist framing — are documented in the behavioral-mechanics corpus at the small-group scale. Wang's case extends them into the population-scale, platform-mediated, state-amenable form. The handshake produces an actionable refinement: when behavioral-mechanics analysis is applied to digitally-mediated contexts, the analyst should distinguish between mobs that are amplified by platforms and mobs that are amplified by platform-state coordination. The latter category — the Wang Weibo case — has different properties from either pure-platform Western mobs or pure-state-directed traditional propaganda mobs. It is a third category and needs separate analytic treatment.
Psychology: Mass Psychology Hub (Le Bon foundation cluster) — the Weibo mob is the contemporary instantiation of the mass-psychology mechanism Le Bon described, and the Wang case offers updated material on how the mechanism behaves under platform mediation rather than physical assembly. The handshake reveals a non-obvious finding: persistent nationalist mobs have memory in a way physical crowds do not. The same accounts return for the next outrage cycle; the same coordination patterns recur; the targeting templates carry over. The mass-psychology tradition assumed crowd dynamics dissipated when the crowd dispersed. Platform-mediated mobs do not fully dissipate, which means the cumulative effect on a society's speech norms is sharper than the mass-psychology tradition would predict.
Cross-Domain: Language & Semiotics Hub (or related propaganda-architecture pages) — the chilling effect on private conversation is a semiotic phenomenon: the population learns that certain words, names, and topics carry risk, and the speech repertoire is reorganized around those exclusions. Over time, the exclusions become invisible to speakers because the alternatives have not been practiced. This is the cognitive cost Wang flags — hurting critical thinking skills — operating as a slow restructuring of what is sayable, then thinkable, then imaginable. The handshake produces a long-run claim: censorship apparatuses do not just suppress speech; they reshape the cognitive infrastructure of the populations that live under them, in ways that take generations to recover from even if the apparatus is removed.
History: Propaganda as the Primary Leninist Instrument (companion concept from this same source) — the propaganda apparatus and the two-direction censorship apparatus are paired components of the same Leninist communications system. Propaganda articulates the line; censorship removes the alternatives to it. The handshake is functional: the two pages should be read as a unit. Propaganda describes what gets said and how it travels. Censorship describes what does not get said and what happens to those who try to say it. Together they describe the closed loop of the Leninist communications regime as it currently operates in China, with the bottom-up mob layer being the new feature that distinguishes the present apparatus from its pre-internet predecessors.
The four handshakes converge on a single claim: censorship is not a single phenomenon. It is a multi-layer apparatus involving state actors, platform actors, mob actors, and the populations they jointly produce. Analysis that treats it as one thing — the state suppresses speech — misses the most operationally important features.
The Sharpest Implication
If the most important censorship in modern China is not directed from the top, then the standard human-rights frame — the regime represses dissent — is missing the larger half of the apparatus. The implication is uncomfortable in two directions. For Western policy: pressure on the state apparatus does not address the bottom-up mob, because the mob is not under the state's direct control. The state benefits from the mob without orchestrating it. Even if the state apparatus were to be liberalized, the mob would still be there, operating autonomously, with potentially worse rather than better behavior because the elite buffer would have weakened. For domestic Chinese politics: the population that the propaganda apparatus has been training for decades has, in the social-media era, become a self-organizing enforcement layer that can outpace the state in identifying targets and demanding consequences. This is not the soft authoritarianism that liberalizes when economic conditions improve. It is a system in which nationalism has become its own organizing principle, separable from the state's direction, and durable across regime changes. The frightening implication is that even a hypothetical post-CCP China would inherit the Weibo mob as a feature of its political culture, and the mob is not obviously friendlier to liberal speech than the propaganda department is.
Generative Questions