The mind

What each device exploits

The exploit targets standard human cognition, not stupidity. Intelligence doesn’t patch it — on identity-charged questions, the research says intelligence makes it worse. The mechanisms are documented below, including the ones where the science is still being fought over.

Presupposition accommodation [settled]

Hearers automatically and silently add to the conversational common ground whatever a sentence presupposes, because that is the price of parsing it (Lewis 1979; Stalnaker 1973, 2002). This is the linguistic mechanism by which frames are installed without ever being defended. You cannot rebut a presupposition by answering the sentence; answering the sentence accepts it. Loftus (1975) showed the effect is not merely conversational: presupposition-loaded questions measurably alter what eyewitnesses later report remembering.

Illusory truth [settled]

Repetition breeds felt truth. Hasher, Goldstein & Toppino (1977) established that repeated statements are rated more probably true than novel ones; Fazio, Brashier, Payne & Marsh (2015) showed the effect operates even against prior knowledge — participants who knew better still drifted toward repeated falsehoods. Devices that repeat (D3, D5, D11) are running this mechanism deliberately or not; function is what the analysis maps.

Fluency and processing ease [settled]

Easily processed claims feel truer, safer, and more familiar (Alter & Oppenheimer 2009, review). Nicknames and slogans are fluency machines: “Crooked Hillary” processes faster than any rebuttal of it ever will. Fluency also explains why the two-word class label (“fake news”) outruns the particular correction every time.

Identity-protective cognition [settled, with active debate on scope]

When facts threaten group belonging, reasoning serves the group (Kahan et al. 2012; 2017). The unsettling finding: numeracy can increase polarization on identity-charged items — more capable reasoners are better at reasoning their way to the group’s answer. This is why the site insists the analysis is of architecture, not audience intelligence; the mechanism operates most powerfully on the most educated.

Motivated reasoning [settled]

Kunda (1990): people do not believe whatever they want, but they recruit evidence and standards selectively in the direction of preferred conclusions, within constraints of plausibility. Closure constructions lower the cost of that recruitment by pre-supplying the frame in which the preferred conclusion is the natural one.

Continued influence effect [settled]

Corrected misinformation keeps shaping inference after correction (Lewandowsky, Ecker, Seifert, Schwarz & Cook 2012; The Debunking Handbook 2020). A retracted claim leaves a causal hole the mind refills with the retracted content unless an alternative account is supplied. This is why D7’s retroactive “joke” labels are cheap: the content persists regardless.

The backfire-effect saga, told straight [contested]

Nyhan & Reifler (2010) found corrections could backfire, strengthening the misperception. Wood & Porter (2019) largely failed to replicate this across thousands of participants. The current view: backfire is rare, but corrections are weak against identity-fused beliefs. We report this dispute in full because a site about closed systems should model what an open one looks like.

Negation and the mud-sticks problem [settled]

Processing “X is not a crook” activates X-and-crook together; the negation tag is fragile and decays first (Gilbert 1991 on belief-by-default: understanding a claim momentarily requires representing it as true). Lakoff’s elephant is the folk version. Consequence: denials rehearse the epithet, which is why D5 is nearly cost-free to its user.

Belief by default [settled]

Gilbert (1991), “How mental systems believe”: comprehension and initial acceptance are one operation; disbelief is a second, effortful step that load, distraction, and fatigue disrupt. D10’s flood is precisely a machine for keeping the second step under-resourced.

Fear and mortality salience [contested effect sizes]

Fear appeals narrow attention and increase preference for strong, simple responses; terror-management and threat-rigidity literatures document the pattern, though effect sizes are actively debated and some findings have replicated weakly. We flag the debate honestly. What is not contested: ad baculum — argument by threat — does not need large lab effects to function at a rally.

Cognitive load and the asymmetry of bullshit [aphorism + documented model]

Brandolini’s law (2013, coinage — an aphorism, not a finding): refuting bullshit takes an order of magnitude more energy than producing it. The Gish gallop (term coined by Eugenie Scott) is the debate-stage version. Paul & Matthews (2016, RAND) formalized the “firehose of falsehood” model: high-volume, multichannel, rapid, shameless about consistency — with the counterintuitive finding that volume and repetition beat credibility.

Disgust and dehumanization [settled at mechanism level]

Disgust is the emotion that pre-empts argument: its objects are managed, not debated (Haidt on moral emotions; Bandura on moral disengagement; the infrahumanization literature on denying outgroups full human emotions). D9’s classifiers (“vermin,” “poisoning the blood”) recruit disgust to move opponents outside the circle of justification.

Sacred values and taboo trade-offs [settled]

Tetlock (2003): once a value is sacralized, cost-benefit questions about it register as offers to sin — the response is outrage and moral cleansing, not calculation. Once loyalty to the leader is sacralized (D12, D8), the auditor’s question does not compute as a question; it computes as a betrayal.

Identity fusion [settled]

Swann et al. (2012): fused individuals experience group threats as personal attacks and endorse extreme personal sacrifice for the group. Distinct from mere identification — fusion is the felt merger of self and group. D12 is a fusion-manufacturing construction; D1 is what fusion makes possible.

Anchoring [settled]

Extreme anchors drag estimates even when hearers disbelieve them. “The greatest economy in the history of the world” moves the internal estimate of a merely good economy upward in the act of being discounted. D11 exploits the fact that discounting is not deleting.

A closing note: none of these mechanisms is partisan; they are human. That is precisely why architecture, not audience, is the right unit of analysis.

Full citations for every mechanism are on the Sources page.