---

The Theater and the Whisper

Formative years under a medium that answers back — whether a generation can still form, and how to tell. An Integrative Essay with a registered close — its predictions filed in advance with margins, to be scored by a script deposited to a timestamped public archive before the data exist — in Armchair Scholar, the independent paper series in which this is No. 013: in its Memory thread (the papers on how the past keeps voting) and cross-listed to The Medium (the papers on the public square), the series' second registration after The Imprint and the Exit*. Rules in force: (1) no prophecy — windows condition on series states, never on dates; (2) no period laundering — every formative signature is registered beside two controls; (3) no monocause — four rival carriers scored at equal resolution; (4) no shelf-tightening without re-searching. The paper's machinery is lettered so it can be held to: seventeen registered rows in five families — F fragmentation, S substitution, C convergence, P period, W the theater on the witnessed events (§9); three instruments D1–D3 (§8); the item classes Fθ and Fω (§7); four registers of claims and six tripwires T1–T6 (§10). Draft v4c (archive panel revision with the cold-reader pass, September 16, 2026); the table, the specification, the instruments, and a v0 scoring script ship with the paper; the deposit — table, specification, and scoring script to a timestamped public archive before the first file that carries a registered row is public — starts the clock.*

· · ·

Youth experiencing the same concrete historical problems may be said to be part of the same actual generation; while those groups within the same actual generation which work up the material of their common experiences in different specific ways, constitute separate generation units.

— Karl Mannheim, "The Problem of Generations" (1928), in Essays on the Sociology of Knowledge (1952), p. 304 (verbatim — confirmed against the 1952 text)

Abstract

Two papers in this series conceded the same cell — the same unfilled place in the evidence — in the same month. The Broadcast and the Model, the series' defense of the minimal-effects record on mass persuasion, conceded that the record measures formed adults over days and weeks and "contains no cell for formative-age socialization"; The Living Room and the Long Grip, its account of the Vietnam cohort's long institutional grip, re-founded that account, in the words of its revision note, on "the draft's stakes at the impressionable years, with television as the shared theater of witnessing." The cohort now in its impressionable years is the first whose formative hours are shared with a medium that speaks to each person separately and answers back — a whisper beside the theater, and, for the ordinary run of a cohort's window, on the assumption the registration tests, in its place. This paper asks what an imprint mechanism built on shared witnessing does under that condition, and specifies how to tell. Part I extracts from Mannheim at passage level what an actual generation requires — "participation in the common destiny" of its moment, a bond created by being "exposed to the social and intellectual symptoms" of the same upheaval — and notes that his generation units, which disagree, are "polar forms of … response to an historical stimulus experienced by all in common," so that fragmentation must mean the loss of the common stimulus and never disagreement. It then adds one thing Mannheim did not say and states it as its own operationalization, with its antecedents named: that for units to form around a stimulus, its sharing must be known — a theater makes witnessing common knowledge; a whisper delivers private objects and withholds the knowledge. Four readings follow, with signatures stated before any series is opened: fragmentation; substitution — the imprint forming on the medium itself, at the stratum Mannheim described as absorbed "by infiltration from the environment in early youth," the ambient channel The Gate and the Field sketched; convergence — the same model's answers delivered privately, a cohort made alike without knowing it, which the published record on the medium's outputs makes the live rival and which no one-tailed dispersion row could see; and the period story, in which everyone disperses together. The prior answers bound where a signature may live: on generic ideology, divergence has not risen across generations (Jocker, van der Brug, and Rekker) and extremity rose at every age (Twenge et al.), so the signature is built on the issues of a cohort's own window, partitioned in advance into the events large enough to be witnessed in common and the issues that arrive mediated — and generic ideology is the placebo class. Part II is the registration: a dispersion statistic descended from DiMaggio, Evans, and Bryson — read with the bimodality they hold distinct from it as a co-condition, so that a cohort split into polar units is never scored as a cohort that failed to form — computed cohort-at-age against cohort-at-age on the ANES, the Cooperative Election Study, Monitoring the Future, and the GSS, on the within-composition component — within gender, race, religion, and sexual identity — so that neither the youngest cohort's gender divergence nor its greater demographic heterogeneity can drive it; a placebo restricted to theater-formed births and a second control — the feed cohort, exposed to the same medium after formation; margins set from a deposited smoke test with floors that cannot be lowered and a power basis stated once; three instruments the series lack, the substitution battery fielded concurrently on a vehicle that observes the boundary and repeated for persistence; four carriers at equal resolution; a seventeen-row table whose most informative row is registered — thinned, not replaced: the theater signature on the witnessed events beside fragmentation on the mediated issues — as a row the paper can lose once the naming test is fielded, and cannot lose before, which the paper prints; and a fifth column — the registered outcome in which no reading's row scores — with a resolution rule that includes the archive's own theater in its blast radius. The clock is indexed to public-file releases: first memo late 2027, last 2033–34. The paper states no finding about the cohort, because the series are only now admitting it, and registers the reading it most fears — a period effect mistaken for a formative one — with the same standing as the readings it prefers.

PART I — THE GRAMMAR

§ IThe Seam

Institutions are staffed by people formed somewhere else in time. The Imprint and the Exit — the series' first registration paper — registered what happens at the far end of that fact — a formed cohort leaving power on clocks it does not control — and The Living Room and the Long Grip argued the near end for one cohort: the living room that watched Vietnam, with the draft assigning stakes to what it watched, grew into the longest institutional grip in American history. Both papers assume a step neither examined, because for every cohort they could study it was safe to assume: that formation happens — that a birth-year band, passing through its impressionable years, comes out the other side as something with a shape. The persuasion thread found the same step from the other direction. The Broadcast and the Model defended the deflationary record on mass persuasion at full strength and then conceded, in August, that the record is "scope-conditioned to adult persuasion — its designs measure message potency on formed adults over days and weeks, and its outcome taxonomy contains no cell for formative-age socialization" — and, in the same note, that "nothing in the deflationary record measured that channel, so nothing in it bounds it."

The cell both threads conceded is formative-age exposure to a medium that argues, and it is not empty of people. Two-thirds of Americans aged thirteen to seventeen use an AI chatbot; about three in ten use one daily on Pew's instrument and about one in nine on another (Hinduja and Patchin, Journal of Adolescence 2026, N = 3,466: 60.2 percent ever, 11.4 percent daily, males more likely); the leading assistant reaches 59 percent of them (Pew, fielded September–October 2025, N = 1,458). About one in five use a chatbot to get news (Pew, February 2026); AI companions specifically have been used by 72 percent of teens on a separate instrument (Common Sense Media 2025, N = 1,060); and any use of AI at all, on a wider definition and a wider age band, stood at 86 percent of nine-to-seventeen-year-olds by March 2026, a quarter of them daily (Common Sense Media 2026) — a census nearly complete on presence, which is why the identification in §8 runs on the kind of use and not on whether a member has the medium. The births of 2008 through 2014 — whose entire window from fourteen to twenty-four falls after the conversational medium's release in November 2022 — are the current twelfth grade and the grades below it. They are, by the archive's own two accounts, the first cohort whose formative years are spent inside the persuasion lever the Medium thread measured on adults, and the first whose theater of witnessing shares those years with something that does not make witnessing common.

The load-bearing assumption, stated so it can be doubted. Everything in Part II assumes that generational formation is witnessed: that a cohort takes its shape from a stimulus encountered in common at the impressionable age, and encountered as common, so that a medium which privatizes the stimulus is a change to formation itself. The rival is not an outsider's; it is the reading of Mannheim most of the literature holds, and the paper states it at full strength. On the participatory reading, an actual generation is constituted by participation "in the characteristic social and intellectual currents of their society and period," by "an active or passive experience of the interactions of forces" that make up a new situation (p. 304): the peasants of 1800 are excluded because they "remain unaffected by the events," not because they did not watch. A cohort that lives through a pandemic, an AI-driven labor shock, and two presidential elections is affected by its moment whether or not it witnessed any of it in common; on the participatory reading it forms, and the medium is irrelevant to formation as such. The paper does not pretend this away. Its own claim — that the sharing must be known, not merely undergone — is an addition to Mannheim, stated as one in §3 with its antecedents, and the participatory reading is built into the table as the fifth column's structural content: if the dialogue cohort shows its predecessors' coherence on the issues of its window despite the whisper, being affected sufficed, witnessing was the mechanism of nothing, and the paper's grammar loses its object honestly.

§ IITwo Caricatures

Both are constructed ideal types — poles of the argument space, not quotations of named authors. The Obituary announces that generations are over: with no shared theater there is no shared experience, no shared experience means no generation, and the birth-year bands of the twenty-first century are marketing categories with no political existence. Its error is that it has never measured a generation's existence in the first place; it assumes what the imprint literature only ever inferred from between-cohort differences, and declares those differences gone without computing the within-cohort quantity that would show it. The Brochure runs the fallacy in reverse: Generation Alpha is coherent, digital-native, and arriving with a worldview, because every generation has arrived with one and the consultancies have already named it. Its error is the Obituary's mirror — it takes the theater's product for granted in a decade when the theater is being supplemented tap by tap, and it has no instrument either. Both assign formation a value without measuring it. The paper's position is that formation has no value until measured; that it is measurable on series that already exist; and that the series are only now admitting the cohort that would show it.

§ IIIThe Theater: What Shared Witnessing Did

Mannheim's essay is usually cited for its first distinction and read loosely on its second. Generation location is the weak term: co-presence in a historical and cultural region, the passive sharing of "the handicaps and privileges inherent" in a moment. Generation as an actuality "involves even more than mere co-presence" — "a further concrete nexus," which "may be described as participation in the common destiny of this historical and social unit" (p. 303). He says what makes the nexus: young peasants in remote districts share a location with the youth of the towns but not an actuality, because the actuality is "created between members of a generation by their being exposed to the social and intellectual symptoms of a process of dynamic de-stabilization" (p. 303). The bond is exposure to the same symptoms — being affected by the same forces. Then the passage the epigraph carries: the romantic-conservative and the liberal-rationalist youth of 1800 "belonged to the same actual generation" though nothing united their politics, because their opposed mentalities "were merely two polar forms of the intellectual and social response to an historical stimulus experienced by all in common" (p. 304). Generation units disagree. They are a generation because they disagree about the same thing.

That much is extraction, and Pilcher's reading of the essay as an undervalued legacy (BJS 1994) supports it, as does the multidisciplinary shelf Kingstone and Bristow assemble (2024): the requirement is a common stimulus, and disagreement is not fragmentation. Mannheim's 1800 cohort disagreed and was one. The failure of a generation, on his own terms, is the absence of the common object — a cohort of people who cannot be sorted into polar responses because there is nothing they were all exposed to.

What this paper adds, stated as an addition. Mannheim required that the stimulus be common. He did not require that its commonness be known. This paper claims that for generation units to form — for opposed responses to become the "concrete bond" he describes rather than a scatter of private opinions — the members must know that the stimulus was shared: each must know the others met it, and know that they know, so that arguing about it is possible. The claim has a home: Chwe's account of public ritual as the manufacture of common knowledge — a ceremony works because it lets each member of the audience know what the others know (Rational Ritual, 2001), with the Super Bowl as his American instance. It has been tested once in the field: Arias delivered the same radio program to households in Oaxaca privately, by recording, and publicly, by loudspeaker, and found that private delivery moved nothing while public delivery moved norms — the social channel, not the individual one, did the work (Political Science Research and Methods 7(3), 2019). And it has an antecedent inside the sociology of generations: Corsten's discursive-pragmatic reading of Mannheim's actuality, on which a generation is constituted in the articulation of a shared horizon and not merely in undergoing it (Time & Society 8(2–3), 1999); Edmunds and Turner's "global generations," media-borne and, in their account, enabled by interactivity rather than dissolved by it (BJS 56(4), 2005), is the same idea run the other way, and is carried as the substitution reading's kin. A theater, in this paper's sense, is a common-knowledge generator; a whisper is its negation. That is the paper's operationalization — anticipated, tested, and not previously registered — and not Mannheim's; it is what the fifth column can falsify, and §1's rival is the reading on which it is false. The claim is not idle. Munger's account of the Boomers founds their cohort consciousness on television as a shared lens (Generation Gap, 2022) — the theater thesis in print — and The Living Room and the Long Grip found its American instance and stopped one step short of this paper's claim: television not as content but as the theater in which the war's stakes and the official deception about them "were witnessed simultaneously by an entire age band," with the draft assigning stakes so that watching was not optional. Simultaneity and visibility are that paper's claims; that everyone knew that everyone watched is this paper's, and the border is drawn here so the parent is not made to say it.

The empirical literature built on Mannheim found the requirement's fingerprint wherever it looked. Schuman and Scott, asking a national sample which "national or world events or changes over the past 50 years" had been especially important, found the answers clustering in each respondent's late teens and early twenties — the Depression for one cohort, the war for the next, Vietnam and the assassinations for the next — as regular as a tide table (ASR 1989), and the replications by Schuman and Corning — nine U.S. surveys from 1985 to 2010 and six countries (POQ 2012; Corning and Schuman 2015) — carried the pattern forward through the cohorts that name the Kennedy assassination, the moon landing, the fall of the Wall, and September 11, with two amendments the paper adopts: the critical period runs wider than Mannheim's seventeen — they call his bracket "probably too circumscribed," and find the Kennedy peak among those aged eight to twenty-two at the time — and the replications stop at the births of about 1990, so no feed-formed cohort has ever been asked the question. Krosnick and Alwin found the attitudes formed in those years the most persistent of the life course (JPSP 1989; AJS 1991); Sears and Valentino found the political events of a campaign season catalyzing preadult socialization in exactly those years (APSR 91(1), 1997); and Erikson and Stoker's draft-lottery natural experiment — the evidence The Living Room and the Long Grip rests on — found the stakes a birthday assigned still legible decades later (APSR 105(2), 2011). And the modern formative-event model — Ghitza, Gelman, and Auerbach's cohort account of presidential voting, in which the cumulative impression of political events weighs most in the teenage and early-adult years and persists for life (AJPS 67(3), 2023) — is the quantitative form of the same finding. Every cohort these studies could reach had a theater.

§ IVThe Whisper: What the Medium Is

The medium that now shares the cohort's formative hours is documented from the vendors' own deployment statements in The Gate and the Field, the series' specification paper on measuring machine persuasion at population scale, and this paper borrows that census rather than rebuilding it: answers rendered atop most searches, an assistant inside the world's largest messenger, a conversational model replacing the default assistant on Android devices on a schedule announced for September 4, 2026 and rolling over several weeks — its execution attested at user-report and trade grade, with no first-party completion statement located, as of the first entry of The Gate and the Field's flip log (September 14, 2026). Among the cohort itself the figures are Pew's: 64 percent ever, three in ten daily, about one in five for news; information-seeking (57 percent) and schoolwork (54 percent) the leading uses, casual conversation 16 percent, emotional support 12 percent. What the census cannot show, and what this paper is about, is the shape of the encounter. A conversation with the medium is individual — prompted by one person's question, in one person's words. It is private — no one else sees it. And it is unrecoverable — the same question asked by two people, or by one person twice, does not return the same text, and there is no archive of what anyone was told. Two members of the cohort who spent the same evening asking the medium about the same event were not exposed to the same object, and neither knows what the other was told. The text differs; whether the content differs is a question the paper must not assume, because the published record on the medium's outputs, none of it on a cohort's political questions, points the other way: the same model answers different people alike (Bommasani et al. 2022 on outcome homogenization; Anderson, Shah, and Kreminski 2024 on homogenized ideation; Doshi and Hauser 2024 on the collective diversity of ideas falling as individual output rises; Wenger and Kenett 2026), private exposure moves views toward the model's (Jakesch et al. 2023), and the medium's search narrows what people go looking for (Sharma, Liao, and Xiao 2024). That record is on creative ideation and algorithmic homogenization, not on the answers a cohort receives about its window, and it is carried as the reason for a reading and not as its measurement. A whisper may deliver private objects that are the same object — sharing without the knowledge of it — and §5 registers that as a reading of its own.

This is where the whisper differs from the feed, and the difference has to be stated exactly, because the feed too individualizes. The feed of 2012–2022 individualized the sequence: each person's stream was their own, but its items were public objects — the same video seen by forty million people, the same post, the same clip — and the feed's own machinery of trending and virality restored common knowledge for anything large enough. The feed is a private sequence of public objects, and a viral event on it is a theater, small or vast. The dose theory has a shelf, and the paper stands on it rather than beside it: Prior showed that the end of the captive broadcast audience — choice, not content — redistributed exposure and its political consequences (Post-Broadcast Democracy, 2007); Bennett and Iyengar diagnosed the erosion of the shared audience that The Broadcast and the Model takes as its structural premise (Journal of Communication 58(4), 2008); Katz and Scannell asked whether television as a shared object had ended (ANNALS 625, 2009); and Webster and Ksiazek measured, against the fragmentation thesis, how much audience overlap survived it (Journal of Communication 62(1), 2012) — the same reservation this paper enters for the feed's largest events. What that shelf did not have was a medium that privatizes the object rather than the sequence. The dialogue individualizes the object: the text one person receives exists for no one else. There is nothing to go viral, because nothing is shared to begin with. This is the dose theory behind the paper's two steps — the feed step, which privatized selection, and the dialogue step, which privatizes the object — and it is why the paper does not claim the theater is gone. The feed still makes the largest events common; the cohort still has a Super Bowl, an election night, an assassination on a college lawn. The media-events literature reached the same shape a generation ago: Dayan and Katz's ceremonial "high holidays of mass communication" (1992, p. 1) have suffered an "apparent decline … in both frequency and centrality" under segmentation, while live disruptive events survive (Katz and Liebes, International Journal of Communication 1, 2007) — thinned, not replaced, in the vocabulary of that field. And the dose is modest at the news end: weekly use of a chatbot for news is 16 percent of under-thirty-fives globally and rose nowhere among the United States, Britain, France, and Germany in 2026, with trust in chatbot answers at 20 percent against 37 for the news itself (Reuters Institute, Digital News Report 2026). The claim is narrower for both reasons: for the ordinary run of a cohort's window — the issues that arrive through argument rather than spectacle — the object of formation is, on the assumption Part II tests, now private.

Mannheim, again, had already described where such a medium would work. Below the reflective level that opens "round about the age of 17, sometimes a little earlier and sometimes a little later" (p. 300) lies the "inventory of experience which is absorbed by infiltration from the environment in early youth" — the stratum that "tends to stabilize itself as the natural view of the world," transmitted so that it "seep[s] in without either the teacher or pupil knowing anything about it" (pp. 299–300). A theater works at the reflective level: an event, witnessed, argued about. A whisper works at the a-problematical one: not an event but an environment, not witnessed but absorbed. This is the ambient channel The Gate and the Field could only sketch for adults — "climate rather than weather" — described, for youth, in 1928; and it is the mechanism of the second reading below. The paper is not the first to treat a medium as the substance of a generation: the media-generations school (Gumpert and Cathcart; Bolin and Westlund; Vittadini, Siibak, Reifová, and Bilandzic; Bolin's Media Generations) has argued for two decades that the medium of one's youth is itself the generational experience, and Munger predicts rising generational identification for younger cohorts on that logic — a prediction the paper carries as the substitution reading's kin, noting that the two existing instruments for generational identification point the other way in cross-section: younger adults identify less strongly with their age group than those over fifty-five (Trachtman, Anzia, and Hill, Research & Politics 2023), and label self-identification is common but flat across parties and demographics (Lindner, Stelboum, and Hakim, Socius 2024). Identification is not the paper's quantity; the sign is recorded so the kin is carried honestly. The paper's step beyond the school — and beyond the post-broadcast shelf above — is the object: from which medium a cohort formed on, and from a medium that fragments the audience, to a medium whose objects are not shared at all.

One property of the whisper distinguishes it from every prior formative medium and supplies the paper's identification, with a caveat printed beside it. Television was universal in the Vietnam cohort's living rooms; the feed was near-universal after 2012. The dialogue is used by two-thirds of the cohort, daily by three in ten, for news by one in five — and AI in some form by 86 percent of nine-to-seventeen-year-olds (Common Sense Media 2026), so the split that identifies is not whether a member has the whisper but whether the member asks it what is true. The cohort is split, from the inside, by the epistemic use of the medium — a source of fragmentation in its own right, and what allows a medium effect to be told from a period effect without a platform, a panel, or a survey house. The caveat: exposure is self-selected, and Pew's own tables say by whom — use is higher among Black and Hispanic teens, higher-income households, and older teens — the covariates that also predict attitude dispersion. A gradient is therefore evidence consistent with a medium effect until it is instrumented, and §8 names the instruments the archive can see, with their weaknesses: the September 2026 default-assistant substitution — the flip The Gate and the Field priced — and the patchwork of school-district policies on classroom AI use, which is closing as districts adopt rules.

§ VFour Readings, and the Prior Answers

Four readings follow from §§3–4, each predicting a different signature on series that already exist. They are stated here, before any series is opened, and they are not exclusive.

Fragmentation. Without a known-common stimulus, no cohort-level imprint forms on the issues that arrive mediated. The birth-year band holds formative-window attitudes that are not responses to one thing, and their within-cohort dispersion rises toward the between-cohort differences the imprint literature has always rested on. The signature: at fixed ages, dispersion on the mediated issues of the window steps up for cohorts formed after the feed and again after the dialogue, beyond the movement in theater-formed cohorts over the same years and beyond the movement in the feed cohort exposed to the same medium after formation; the naming test returns fewer shared events and less concentration; and the step tracks the exposure gradient within the dialogue cohort.

Substitution. The imprint forms anyway, with the medium itself as its object. What the cohort shares is not an event but an environment, absorbed at the a-problematical stratum, and its coherence reappears on an axis the imprint literature never measured: trust in machine answers, epistemic deference, the habit of asking the medium what is true, and — the whisper's own test — the belief that one's peers would get the same answer. The signature: a discontinuity on that battery at the dialogue band's birth-year boundary, measured concurrently against the feed cohort, persisting when the same birth cohorts are asked again at twenty-two to twenty-four; and the naming test returning the medium itself as a formative event at a rate no earlier technology reached.

Convergence. The whisper delivers private objects that are the same object. One model, or a few, answers the cohort's questions alike, and the cohort comes out of its window more alike on the mediated issues than any theater-formed cohort — without any member knowing it, because nothing was witnessed. This is the reading the homogenization record makes live (§4), and it is opposite in sign to fragmentation. The signature: dispersion on the mediated issues falling for the dialogue cohort at fixed age, beyond both controls, lower among the epistemic users than the rest of the band — and, on the concurrent battery, agreement across respondents about what the medium told them exceeding their belief that their peers were told the same. Convergence the members know about is a theater by this paper's own definition and is printed as one; convergence they do not know about is the fourth reading, and it is what a one-tailed table would have mistaken for the theater's survival.

Period. Nothing formative has happened. Dispersion rises in every cohort together, because the decade is dispersing everyone; the young look fragmented because the old do too. The signature: the placebo moves with the treatment.

The prior answers bound where the first signature may live, and the paper found them before it wrote its outline, which is the archive's standing discipline against tightening a claim without re-searching the shelf. Jocker, van der Brug, and Rekker, decomposing the General Social Survey and the ANES by age, period, and cohort, separated partisan sorting from ideological divergence and found that sorting rose — within generations — while divergence did not, and that newer generations arrive less sorted than those they replace (Political Behavior, 2024). Twenge, Honeycutt, Prislin, and Sherman, on three surveys and ten million respondents from 1970 to 2015, found more extreme self-placement and fewer moderates by the early 2010s — among adults as much as among twelfth-graders (PSPB, 2016). On generic ideology, then, the fragmentation reading's feed-era step has been looked for and not found, and the extremity that was found is a period effect visible at every age. Two more prior answers bound the field. Boxell, Gentzkow, and Shapiro found polarization since 1996 growing fastest among the Americans least likely to be online — the old — the period reading's standing exhibit and the strongest published case that a medium's arrival did not drive the young apart (PNAS 114(40), 2017). And one feed-era measurement of within-generation spread exists, at the grade of a think-tank report on a YouGov sample: Heys found "less of a spread of values for the Millennial/Gen Z generation than there was for older generations in relationship to liberal and conservative ideological orientations" (Berkeley Institute for Young Americans, 2024; N = 2,270, fielded 2020) — not fixed-age and not on window items, but pointing, as far as it points, against fragmentation and toward convergence. The nearest kin to this paper's own question is the POSTGEN program's mixed-methods study of how social media reshape the ideological structuring of political attitudes across generations, in Italy around the 2024 European elections (De Sio et al., "Assessing public sphere influence on political attitudes across generations," Frontiers in Political Science 8, 2026); the narrower question — whether within-cohort variance on a cohort's own window issues, at fixed age, is lower for theater-formed than for feed- or dialogue-formed cohorts — remains unasked, and the paper says so at the grade of a search, not a proof.

Three consequences follow and are binding. The signature is built on the issues of a cohort's own window, partitioned in advance into two classes: the witnessed events — those large enough to have been met in common by the whole cohort, which a theater would have produced whatever the medium — and the mediated issues, which arrived through argument, search, and conversation rather than spectacle. The partition is made before the data by a registered criterion (§7), and the fragmentation reading lives on the second class only. Generic ideology is the placebo class, and the paper's first computation on any file is to reproduce the Jocker–van der Brug–Rekker finding on it, so that a window-issue signature is not a generic-ideology artifact. And the extremity trend is registered as the period reading's evidence, not the fragmentation reading's.

Four carriers, not one, could produce a step in the dialogue cohort, and rule (3) of the headnote — no monocause — requires them scored at equal resolution. The feed before the dialogue, on the 2012 dating that Twenge (iGen, 2017) and Haidt (2024) argue and Odgers disputes (Nature, 2024) — carried with the dispute, not settled by the paper — predicts that the feed cohort already shows the step and the dialogue adds nothing beyond the privatized sequence. The pandemic predicts a trust scar for whoever was in the impressionable years in 2020–21: the epidemic-exposure literature puts it at 7.2 points of confidence in elections, 5.1 in government, 6.2 in leaders at maximal exposure between eighteen and twenty-five (Aksoy, Eichengreen, and Saka, Economic Journal 134(660), 2024; Bottasso, Cerruti, and Conti find the scar on young Europeans' political trust, Journal of Regional Science 62(4), 2022) — a window that licenses no differential prediction for cohorts who met the pandemic at fourteen or at ten, so the carrier is priced by that literature's magnitude and discriminated by the paper's own comparison: the 2006–2007 births, who met the pandemic at fourteen and the dialogue at seventeen, against the 2010–2012 births, who met the pandemic in childhood. The exposure window straddles the feed–dialogue boundary, so the scar mimics neither reading cleanly. The pandemic is also the carrier the public conversation has already chosen: the trade literature's "Gen Z 1.0 and 2.0" — Janfaza's coinage, carried by Edelman (2025), split at whether high school ended before or during the pandemic — is the fragmentation claim with the pandemic and not the medium as its cause. And the collective-memory literature adds a discipline the paper adopts: sharp events produce critical-period effects and prolonged ones produce lifetime effects across every cohort (Schuman and Corning 2012, on Iraq against the Gulf War; Osborne, Sears, and Valentino, Political Psychology 32(1), 2011) — so a prolonged stimulus like a pandemic, or a medium's arrival, is by that literature's own finding a candidate for the period reading before it is a candidate for formation. Economic formation predicts a directional shift — beliefs about luck and effort set by growth in the window (Giuliano and Spilimbergo) — which moves means, not variances. And the witnessed events of the window themselves are the paper's discriminator, now registered rather than suspected: if the dialogue cohort shows the theater signature on the witnessed events while fragmenting on the mediated issues, the medium thinned the theater and did not replace it, and a generation still forms on the events large enough to be witnessed in common.

§ VIThe Exhibit, Placed — and Removed From the Statistic

The widely reported divergence between young men's and young women's politics is the one exhibit the public conversation already treats as this paper's finding, and the paper's obligation is to say precisely what it is, what it is not, and what it does to the paper's own numbers. At source: in the United States, Gallup finds women aged eighteen to twenty-nine identifying as liberal at a 2017–24 average of 40 percent, twelve points above 2001–07 and fifteen points above young men, while young men sit near 25 percent — 24 in 1999, a peak of 29 in 2016 — with no long-run movement at all. The divergence is one-sided: women moved, men did not. Across thirty-two European countries, Nennstiel and Hudde's decomposition of 466,089 Eurobarometer respondents aged twenty to twenty-nine from 1990 to 2023 finds fourteen countries with near-equal positions, seven with a stable modern gap, and eleven where a gap emerged or widened, larger where gender equality is higher, and "mainly small to moderate" throughout — a pattern that, in the authors' words, contradicts "the narrative of a strong and uniform shift" (European Sociological Review, 2025). And in its behavioral form the gap is real: across twenty-seven countries in the 2024 European Election Studies, more than 21 percent of young men voted for the far right against about 14 percent of young women, and an age–period–cohort analysis over eight elections since 1989 finds the gap greatest for Millennials and Generation Z (Milosav, Dickson, Hobolt, Klüver, Kuhn, and Rodon, Journal of European Public Policy, 33(2)).

The behavioral gap is itself contested across the Union: on a pooled European sample Off, Alexander, and Charron find no youth gender gap in far-right voting at all, with Poland, Finland, and Romania driving what appears in the aggregate (European Journal of Politics and Gender, 2025) — and the paper carries both findings, since they disagree about the same data. What the exhibit is not is the paper's quantity, though it is not "not dispersion" either, by the paper's own statistic. Within-cohort variance decomposes into a within-group component and a between-group component, and a one-sided shift by young women raises the second and therefore the whole. Left in, the exhibit would score the fragmentation rows by arithmetic. The point generalizes: successive cohorts are markedly more heterogeneous by race, religion, and sexual identity, which is why Pew now refuses generational comparisons that are not age-matched and warns that apparent generational differences "may be driven by their demographic traits" (Parker, 2023) — so within-cohort variance rises by composition alone, and neither of the paper's controls, being differently composed cohorts, removes it. The rows are therefore registered on the within-composition component — the variance remaining within cells of gender, race, religion, and sexual identity — and the between-composition components are reported beside them. The between-gender component is the exhibit's own registered quantity, reported in this section at every memo, with one more fact printed: exposure to the medium already differs by sex (Hinduja and Patchin), so a gender-tracking gradient is expected on adoption grounds before it is evidence of anything. Placement then follows by two tests. A one-sided directional shift inside one sub-cohort is exactly what a theater with polar units produces — it is Mannheim's 1800, the romantic-conservatives and the liberal-rationalists responding in opposed ways to a stimulus all of them met — and it is what a whisper could produce only through exposure that differs by sex. So: is the between-gender component's movement one-sided, as Gallup finds for the United States? And does it track the exposure gradient by sex? One-sided and gradient-flat, the exhibit is a theater-with-units phenomenon or a period story, and the section says which; symmetric or gradient-tracking, it is fragmentation-consistent. Either way it confirms nothing — a discipline the paper needs, because the exhibit is the one part of this argument the reader arrives already believing.

The same logic now binds the rows and not only the exhibit. Mannheim's units are polar responses to one stimulus, which is a claim about the shape of a cohort's distribution and not its width: DiMaggio, Evans, and Bryson hold dispersion and bimodality "analytically and empirically distinct: one may find much bimodality within a relatively narrow range of opinion, or a flat distribution of persons across a very wide range of opinion" (AJS 102(3):694), and a theater with two units can raise a variance exactly as a scatter of private opinions does. A variance ratio alone therefore scores 1800 — and the 1960s — as fragmentation, and the table answers that in the rows: every fragmentation row carries bimodality as a co-condition (§9), so that dispersion rising with bimodality rising prints as units, the theater's own product, and only dispersion rising without it prints as fragmentation. The within-composition statistic has a cost of the same shape, and it is printed: what it removes is every component of dispersion aligned with the cells, and exposure to the medium differs by sex, income, and race, so medium-driven dispersion that runs along those lines is removed with the composition — a bias toward the null the rows carry, with the raw and the within-composition statistics reported side by side.

PART II — THE REGISTRATION

§ VIIThe Series and the Signature

The series, and what each can carry. Monitoring the Future has surveyed a national sample of twelfth-graders every year since 1975, and is the first series the dialogue cohort enters: the 2008 births reach twelfth grade in 2026, the 2014 births in 2032. Its 2024 file — 6,895 respondents across six forms of 1,138 to 1,159 each, response rate 76 percent, released to ICPSR in October 2025 — fixes the paper's design facts and its limits, and so does its recent history: a randomized paper-versus-tablet split in 2019 (Miech et al., Addiction 2021, about 41,900 students across grades — little difference in prevalence, less missing data on tablets); all tablets in 2020, halted on March 15 at 3,770 respondents, a quarter of a normal year; web from 2021. The 2020 file can anchor no pooled row, and lower tablet missingness moves the non-placement share, so the F3 row carries the mode flag on its own. The series carries generic self-placement with a large non-placement category; institutional ratings; social-issue and social-media items; and no items on the events of any window. It therefore carries the placebo class and the non-placement row, and, once the exposure items are fielded, the gradient — but it cannot carry the live wire — the row on which the fragmentation reading is decided, F1 in §9. The Cooperative Election Study — sixty thousand in even years and roughly twenty-five thousand in odd ones (25,700 in 2021; 24,500 in 2023), with an eighteen-to-twenty-four cell of five to six thousand in even years and about two thousand in odd — carries the window issues and the live wire, scored after the series' attention and consistency screens with the screened share printed at every memo, because Pew's audits locate the opt-in artifact in exactly this cell at magnitudes far above any margin here — twenty percent of under-thirties in an opt-in sample agreeing that the Holocaust is a myth against three in a probability sample (Pew, March 2024) — and a drift in that share would move a registered ratio with no formation at all; the ANES, which admits the cohort in 2028 with an eighteen-to-twenty-four cell of several hundred, reports beside the live wire and scores no row until its own cell has a floor of its own; the General Social Survey admits the cohort across 2026–2032; the Harvard Youth Poll runs semiannually; the Eurobarometer series Nennstiel and Hudde used supplies a cross-national arm for the placebo class only, since it carries left–right and nothing keyed to a window.

Item classes, and the partition. Three classes, and a signature claimed on the wrong class fails its row. Window issues (F), partitioned before the data into witnessed events (Fθ) and mediated issues (Fω) by a registered criterion: an item is θ if the event it keys to is named by at least 15 percent of the theater-formed placebo in the naming test or, before the naming test is fielded, ranks in the top ten of the year's most-followed stories in a probability series named at deposit (Pew's news-interest index was retired in 2012 and cannot serve the window) — the pandemic, the 2020 and 2024 elections, the October 7 attacks and the Gaza war, the September 2025 assassination on a college lawn are θ by any reading; an item is ω otherwise — the policy questions of the window that arrived through argument rather than spectacle, and AI itself as an object of attitude. What the deposit freezes is the criterion and the candidate list; classification by the naming test executes at the first memo after the test is fielded and governs from then on. Until then the second leg classifies what it can: items it marks θ carry the theater's witnessed-events row provisionally, items it cannot classify carry the ω default, and anything scored provisionally is re-scored when the naming test's classification arrives. An item that could not be classified by the criterion is excluded and listed. And because the window issues of one cohort are not the window issues of another — the feed band's Fω items and the dialogue band's are different questions asked in different years — no dispersion ratio is computed across unmatched items: Fω items are scored across windows only in matched pairs, frozen at deposit, with the same response format, the same number of scale points, the same polarity, and both members classified ω; an item that spans both windows is scored within-item; an unmatched item is reported and not scored. Within a pair the raw variances on the identical scale are compared: standardizing an item on its own pooled distribution equalizes nothing across items and injects the pool's composition into the ratio — and standardized on a placebo's distribution in the same years it absorbs the period effect the controls exist to measure, which the v0 script's synthetic run showed as a period effect reading 0.67 on the live wire — so standardization serves cross-format reporting only and never a scored ratio. Generic ideology (G): the placebo class. Meta-epistemic items (M): none exist in any series; §8 supplies them.

Bands and windows. Cohorts are dated by the year a birth-year turns seventeen. Theater-formed, births through 1994, subdivided for the baseline into the Vietnam-window cohort of 1943–56, the cable cohort, and the early-internet cohort; feed-formed, 1995–2005; dialogue-formed, 2006 onward, with a core band of 2008–2014. Every comparison is cohort-at-age against cohort-at-age — never a cohort against a different age, which is the collinearity The Imprint and the Exit was caught on — and every boundary is moved two years in each direction, with a signature that survives only one placement reported as boundary-dependent.

The two controls. An age-band placebo — "theater bands at thirty-five to fifty over the same years" — would fail twice: by 2030 that age band includes feed-formed births, and a difference against any older age band identifies only where no shock hits the young harder than the old, which is what the impressionable-years hypothesis says shocks do. So: the placebo is the theater-formed births of 1994 and earlier, whatever their age in the file year, over the same calendar years; and the post-formation control is the feed cohort — the births of 1995 through 2005 — at whatever ages it holds in 2026–2032, never an age window, so that no dialogue-formed birth-year can enter the control as the years pass (an age window of twenty-one to thirty would admit the 2006–2011 births by 2032): a cohort exposed to the same medium after its window closed. A movement the placebo shares is the period's. A movement the post-formation control shares is the medium's but not formative — adults dispersing under the same climate — and it is scored under the period reading too. Only a movement in the dialogue band that neither control shares is a candidate for formation — and because the comparison across age at fixed period identifies only under age-invariant period effects, which is an untestable constraint of exactly the kind Fosse and Winship show every age–period–cohort design to need (Annual Review of Sociology 2019), the deposited script prints bounds under a registered family of departures — the period effect allowed to differ across the age gap by up to five and up to ten percent of its size — and the period row's threshold is read on the bound nearest the null, never on the point. The placebo is the one comparison in the table that is not cohort-at-age — it is the same births aging across the files — which is why it identifies only under the constraint just stated.

The statistic. It descends from DiMaggio, Evans, and Bryson's operationalization (AJS, 1996), in the measurement lineage Baldassarri and Gelman carried forward (AJS 114(2), 2008), and takes dispersion and bimodality. Dispersion is the variance of each item — on the raw scale within a matched pair, standardized on its pooled distribution for reporting across formats — within band by age-window by year under the series weight, computed on the within-composition component (§6) — and the design has a precedent: variance-function regression inside a hierarchical age–period–cohort model, which Zheng, Yang, and Land used to decompose within-group variance in self-rated health into age, period, and cohort components and found cohort variance falling to mid-century births and rising for recent ones (ASR 2011); the paper's quantity is that method's, on attitudes; bimodality — kurtosis, in DiMaggio, Evans, and Bryson's operationalization, computed on the raw distribution and on the within-composition component — is a co-condition of every fragmentation row (§9) rather than a report beside it; bounds come from a two-thousand-draw bootstrap over respondents, inflated on the youth series by the design effect the series itself publishes for its clustered school sample (Monitoring the Future, Volume I, Appendix C) — which this paper reads off those tables as a factor of roughly 1.15 to 1.6 on a variance, retrieved rather than assumed. The youth series' non-placement category — "none of the above, or don't know," 36.6 percent of twelfth-graders in 2005–2018 and above 40 percent since (Gimbrone, Bates, Prins, and Keyes, SSM – Mental Health 2, 2022) — is handled by a rule and a coupling both stated: dispersion is computed among the placed, the non-placement share is a separate registered quantity, and because removing the unsure mechanically raises variance among the placed, every dispersion row on that series is computed twice — placed-only, and with the unsure imputed to the scale midpoint — and must clear its margin both ways. The rule is the youth series' alone: the Cooperative Election Study's "not sure" is a placement category on its ideology item and its issue items carry no midpoint, so the rows on that series are cleared once, on the placed. A coherence ratio — within-band dispersion over the between-band difference of means — is not used: its denominator is what the record says is near zero, and with three bands it is not even unique. In its place, the between-band share — the share of an item's variance accounted for by band membership, η², at fixed age — which the fragmentation reading predicts falls for the dialogue-versus-feed comparison relative to the feed-versus-theater one.

Dilution, printed. The exposed subgroup is a minority — one in five for news — and a cohort-level ratio is a mixture: a band-wide ratio of 1.15 with a fifth exposed requires the exposed subgroup's variance to be about 1.75 times the unexposed, and at the gradient's identifying level — epistemic use, a smaller share than news use — larger still: 2.5 at a tenth exposed. The table therefore carries the band-wide rows conditional on the exposure share observed at each file, at the level of the gradient the row uses, and prints the exposed-subgroup ratio the band-wide margin implies.

Power, on one basis. For normal data with 1,500 a side, the standard error of the log variance ratio is near 0.052 and a ratio of 1.16 is detectable at eighty-percent power; a figure of 1.20 to 1.25 for a single youth-series year, 1.11 to 1.14 for three pooled years, and 1.08 for the Cooperative Election Study's cell is a bare-significance figure on the same inputs, and the two must never stand on different bases in one paragraph. On one basis — eighty-percent power at α = .05, forms of 1,138 to 1,159, the placed at 60 percent, and the kurtosis of peaked five-point items inflating the standard error by a factor of 1.2 to 1.6 — a single youth-series form detects a ratio of 1.29 to 1.41; three pooled years 1.16 to 1.22, and 1.17 to 1.28 once the series' published design effect is applied; the Cooperative Election Study's even-year cell 1.09 to 1.13 and its odd-year cell 1.15 to 1.21; the ANES 2028 cell 1.30 to 1.42. The floors are therefore floors and not detection thresholds: 1.15 on the youth series pooled — reachable at eighty-percent power only at the low end of the kurtosis range and not under the design effect — and 1.10 on the Cooperative Election Study, reachable in even years at the low end of the kurtosis range. The deposited script prints, for every row, the ratio detectable at eighty-percent power and at bare significance from the actual cell in each file, with the item's kurtosis computed from the file rather than assumed, and the first memo at which each row's cell reaches its floor at the stated power; until the smoke test on the 2012–2024 public files has run, no figure in this paragraph is a margin, and the margins above the floors are set from that test at deposit. The youth-series rows carry a mode flag where they straddle 2019–2020.

§ VIIIThe Instruments the Series Lack

Three, at fielding grade, in the youth series' own formats.

The exposure items (D1), and their instruments. Frequency of back-and-forth use of an AI assistant in the last thirty days, on the six-point days scale; the first-source item — "when you want to know what is true about something in the news or politics, which of these do you usually try first?" with an AI assistant among six options; and whether, in the last year, the respondent has asked an assistant what it thinks about a political question or to explain one. The estimand is the within-cohort exposure gradient, three-valued: none, instrumental, epistemic; the first item counts homework, so the gradient that identifies is the second and third. The gradient is self-selected, and the paper says by whom, so it is scored two ways: conditional on race, household income, parental education, and political interest, and labeled consistent with; and, where the file allows, instrumented — by device platform at the September 2026 default-assistant substitution, the natural experiment The Gate and the Field registered for adults, and by school-district policy on classroom AI use. Neither instrument is strong, and the weakness is stated in the instrument itself: the substitution extends a migration that had made the conversational model the default on newer Android devices since 2024–25 — its execution attested at user-report and trade grade as of the flip log's first entry, September 14, 2026, with no first-party completion statement — and the control arm is not unexposed, since iOS has carried a chatbot integration since December 2024, so the platform instrument is a difference in depth of default, not presence, and, as The Gate and the Field states the estimand for the same flip, a treatment change on invokers; the policy instrument's variation is collapsing, with schools reporting AI rules rising from 51 to 74 percent in a year (Gallup, Voices of Gen Z, 2026), and its exclusion restriction fails on the paper's own facts — districts that adopt rules differ in resources and politics that predict dispersion directly — so it is carried as a robustness check and not as identification. Each instrument carries a registered minimum first stage (an F-statistic of ten on the exposure item); below it, F6 is scored on the conditioned gradient alone, labeled consistent with, and no row attributes anything to the medium. The gradient's "none" cell is printed at every file, because AI use of some kind already stood at 86 percent of nine-to-seventeen-year-olds in March 2026 (Common Sense Media 2026) and the identifying contrast is the epistemic level, not presence.

The meta-epistemic battery (D2), fielded concurrently. Four items on the five-point agree–disagree format: whether an assistant's answer about a public question is usually more reliable than what most people say; whether, if an assistant and a trusted person disagreed about a fact, the respondent would go with the assistant; whether an assistant is the first place the respondent goes to understand something new; and whether most people the respondent's age would get the same answer if they asked. A comparison of the dialogue cohort to "the feed band at the same age" cannot be made, since the feed band was seventeen in 2012–2022 and the battery did not exist. The design is therefore concurrent, on a vehicle that observes the boundary: the battery is fielded in the same year on the youth series at seventeen to eighteen and on the Harvard Youth Poll, whose eighteen-to-twenty-nine sample holds the 2006–2007 births at nineteen to twenty in 2026 and, above them, the feed cohort's 1997–2005 births (the 1995–1996 births fall outside its frame), and the estimand is a discontinuity at the birth-year boundary — a jump in coherence at the 2006 births beyond the smooth age trend that adoption produces in everyone, estimated with respondents on both sides of the boundary. Arms at seventeen to eighteen and twenty-one to thirty would leave the boundary births unobserved by either and make the jump an extrapolation across a two-year gap; the Harvard Youth Poll closes it. It is an online panel with the same opt-in exposure as the Cooperative Election Study, so the same screens apply to it and the screened share is printed for it too. Two items are added for the convergence reading: for two registered window prompts, what the assistant told the respondent, coded in a frozen frame for cross-respondent agreement — the fourth item asks whether peers would get the same answer; these ask what they did get. The repeat at twenty-two to twenty-four on the same birth cohorts tests persistence against novelty; it does not separate cohort from age, and the paper says so — the discontinuity does that, the repeat does the other, and substitution needs both. The novelty confound also has a signature of its own, registered: unfamiliarity shows up as uncertainty, not distrust — adults over sixty-five are the age group most often "not sure" about AI, while adults eighteen to twenty-nine, who use the medium at the same rates as teens (66 percent ever, 31 daily, 15 for news), are the most negative about its effects, with concern among them rising from 31 to 55 percent between 2021 and 2026 (Pew, June and August 2026). A coherence discontinuity that coincides with a fall in the not-sure share across the boundary is adoption; one that does not is a candidate for formation, and the battery carries the not-sure share as a registered quantity for that reason.

The naming test (D3). Schuman and Scott's open item, in the code frame Schuman and Corning published for their replications so that the paper's concentration figures are comparable to theirs rather than to a frame of its own devising: which national or world events or changes of the past fifty years seem especially important, up to two answers. Two quantities: the naming rate — the share naming any event inside their own window — and the concentration of named events within the band. The theater signature is a high rate and high concentration, window-peaked, as every cohort Schuman and his co-authors asked displayed; the fragmentation signature is a rate ten points below the theater bands' at the same age, or a concentration a quarter lower; the substitution signature is a cohort that names the medium itself — "AI," "when ChatGPT arrived" — at a rate above what the published replications record for any technology: computers were named by 3.5 percent in 1985, 9.9 in 2000–01, and 7.7 in 2009–10, communication and transport by under 6, with the mentions youth-skewed and form-sensitive (Schuman and Corning 2012, Table 1 and Appendix A) — so the 15 percent the table registers sits above every published technology baseline, and the data behind them are archived at ICPSR, where the age-matched theater baselines the naming rows need already sit. The concentration figure carries two caveats. It tracks the window's event landscape as much as its medium — a September 11 cohort concentrates — and the one feed-era measurement points against the fragmentation reading on this quantity: in Pew's 2016 survey of the most significant events of a lifetime, Millennials showed the highest consensus of any generation (86 percent naming September 11) while five of their top ten appeared on no other generation's list. Concentration is confounded by lifetime length and recency, which is why the naming rows are matched on age at survey and read beside the θ/ω partition. One more caveat is carried both ways: Koppel and Berntsen argue that dating important public events to one's youth is a belief that important things happen when one is young rather than a signature of formation (Quarterly Journal of Experimental Psychology 67(3), 2014), and Janssen finds no support for that bias (Applied Cognitive Psychology 29(1), 2015); the naming rows are read with the dispute open, which is one more reason no naming row confirms a reading on its own.

§ IXThe Table, the Carriers, and the Clock

The table's rows are printed in full in the registration artifact that ships with the paper; what follows is the shape, with every row's class, series, margin, controls, and decision rule fixed there. The rows are lettered by reading and numbered within it — F1–F6 fragmentation, S1–S4 substitution, C1–C2 convergence, P1–P3 period, W1–W2 the theater on the witnessed events — and the instruments they lean on are §8's D1 (the exposure items), D2 (the concurrent battery), and D3 (the naming test).

Fragmentation, six rows, on the mediated issues. F1, the live wire: dispersion on the matched Fω items for the dialogue core band at eighteen to twenty-four on the Cooperative Election Study exceeds the feed band's at the same age by the deposited margin (floor 1.10 on that series), on the within-composition component with the raw statistic beside it, with neither control sharing the rise — and with the bimodality co-condition: kurtosis on the same items does not fall, which is to say bimodality does not rise, past a registered kurtosis-ratio margin — 0.75 in the v0 script, calibrated at deposit on the theater cohorts' own ratios, because a bounded five-point scale lowers kurtosis mechanically as dispersion rises and a margin set at the bootstrap interval would print most fragmentation as units; dispersion rising with bimodality rising past the margin prints units, not fragmentation. F1 is two-tailed: the same computation read for a fall is C1. The ANES 2028 reports beside it and scores no row until its cell has a floor of its own. F2: the dialogue's step exceeds the feed's own step over the theater bands, on the matched items and under the same co-condition — scored only if F1 scores; if F1 scores and F2 fails, the feed did the work. F3: the non-placement share on the youth series for the dialogue band at seventeen to eighteen exceeds the feed band's at seventeen to eighteen (the 2012–2022 files) by eight points, with the 2019 randomized split-half's mode difference subtracted and the placebo's non-placement reported beside rather than netted, since an adult series' don't-know is not the same quantity — its own row, and coupled to the youth-series dispersion rows by the imputation rule. F4: the naming test's rate or concentration falls below the theater bands' by the registered margins. F5: the between-band share of variance on the matched Fω items falls for the dialogue-versus-feed comparison to no more than half its value for feed-versus-theater, under the same co-condition — and only where the feed-versus-theater share itself clears a registered minimum, since a share near zero has nothing to halve and the ratio is noise, the denominator problem that retired the coherence ratio. And F6, the identification row: within the dialogue band, dispersion on the Fω items rises with epistemic exposure — the top level's variance against the bottom's by a registered ratio of at least 1.10 with its bootstrap interval above one — conditional on the named covariates, and — where an instrument clears its registered first stage — under it; a dialogue-band effect flat across the gradient does not score for the medium under any row above, and F1 through F5 then transfer to the period or feed carriers.

Substitution, four rows, on the concurrent design. S1: the battery shows a coherence discontinuity at the 2006 birth-year boundary — dispersion at least 15 percent below and a mean at least half a point above the trend fitted on the feed cohort across the Harvard Youth Poll's ages, with respondents on both sides of the boundary — provisional by rule. S2: the coherence persists when the same birth cohorts are asked at twenty-two to twenty-four; S1 without S2 is novelty and substitution does not score. S3: the naming test returns the medium itself as a window event at 15 percent or more, above the published baselines for prior technologies. S4: a majority of the dialogue band names an assistant as its first source for what is true about a public question, against under a fifth of the concurrent feed cohort — a live bet on both direction and threshold, since the only deference-type items yet asked of teens point the other way: they expect today's AI to do worse than the people whose job it is at hiring, diagnosis, and driving, and better only at teaching a skill (Pew, February 2026).

Convergence, two rows. C1: dispersion on the matched Fω items for the dialogue core band at eighteen to twenty-four falls below the feed band's at the same age by the deposited margin (the reciprocal of the 1.10 floor on the Cooperative Election Study), neither control sharing the fall, and lower among the epistemic users than the rest of the band — F1's computation read for the opposite sign, a live bet. C2: on the concurrent battery's two window prompts, the cross-respondent agreement rate on what the medium told the dialogue band exceeds a majority while the fourth item's belief that peers were told the same does not — convergence the members do not know about; both above a majority is convergence known, a theater by this paper's definition, and is printed as one; provisional by the same rule as S1, a live bet.

Period, three rows, given standing. P1, the shared rise: dispersion rises in the dialogue band and in the theater-formed placebo over the same years, so that the placebo-adjusted ratio falls under 1.05 — the same computation as F1, read for the opposite sign, and the reading the paper most fears; if it scores, F1 through F5 are scored against fragmentation whatever their raw values. P2: the Jocker–van der Brug–Rekker finding reproduced on the public files before deposit and re-scored at every memo — a row the paper expects to score and counts for no one. P3, the post-formation control: the feed cohort — the 1995–2005 births at their ages — disperses on the matched Fω items by the same ratio as the dialogue band — the medium moved adults and adolescents alike, which is climate, not formation, and is scored for the period reading.

The theater, two rows. W1: on the Fθ items — the witnessed events of the window — the dialogue band shows the theater signature: dispersion within 1.05 of the theater bands' at the same age, naming within ten points, window-peaked; scored once the naming test has classified the items, and not before. W2, thinned: W1 scores and F1 scores — the theater signature on the witnessed events beside fragmentation on the mediated issues, both at their registered margins. An outcome merely suspected and not registered would make a table that could not lose; W2 is a row, with margins, graded a live bet — and the scoreability disclosure below says on which files it can be lost, because on the public files alone it cannot.

The fifth column, with its resolution rule. None of the readings' scoring rows — F1 through F6, S1 through S3, C1 and C2, W1 and W2 — scores under its stated rule across the 2026–2032 files. The consequence has two branches, and the memo does not choose between them; a registered criterion does. If the generic placebo shows no step beyond the record's, the non-placement row does not score, the convergence rows do not score, and the naming test returns theater-level rates for the dialogue band, the cohort formed as cohorts always have — being affected sufficed, witnessing was the mechanism of nothing, and the paper's grammar lost its object. If the naming test does not return theater-level rates, formation could not be read from these series at the resolution claimed, and the paper says so. Either branch includes the archive's own theater in its blast radius: if the first scores, the shared witnessing The Living Room and the Long Grip built its mechanism on was not necessary for formation either, and the Memory thread says so.

The live-bet census, disclosed rather than balanced: fragmentation three (F1, F2, F6), substitution three (S1, S2, S4), convergence two (C1, C2), period one (P1), theater one (W2), with F1, P1, and C1 one computation. Scoreability, disclosed beside it: from the public files and the deposited script alone, a stranger can score F1, F2, F3, F5, P1, P2, P3, C1, and W1's dispersion half; F4, F6 in its instrumented form, S1 through S4, C2, W1's naming half, W2, and the fifth column's resolution need instruments D1 through D3 fielded by someone, and until they are, the theater can neither win nor lose on this table and the fifth column cannot resolve. The rows the protocol closes at 2029 are the ones it names below; the naming-test rows are not among them, and the paper prints that asymmetry rather than papering it.

The carriers. Each row is scored beside the four carriers of §5 by the discriminating tests fixed in the specification: the exposure gradient, conditioned and instrumented, for the medium; exposure-matching on the feed items for the feed; the 2006–07 against the 2010–12 births for the pandemic; means against variances for economic formation; and the θ/ω partition for the witnessed events themselves.

The clock, honestly indexed — and closed against deferral. The table, the specification, and a scoring script — the dispersion, non-placement, between-band-share, and naming computations with the partition, the boundary rule, the two controls, and the imputation rule wired in, smoke-tested on the 2012–2024 public files, and the source of the deposited margins — are deposited to a timestamped public archive before the first file that carries any registered row is public — expected, on the two series' release calendars to date, to be the Cooperative Election Study's 2026 common content rather than the youth series' 2026 file — with the generic baseline reproduced in the same deposit; until the deposit this is a draft that can still be loosened, and after it a table that can only be struck. Memos follow the public microdata releases: the youth series' files reach ICPSR roughly eighteen months after fielding — the 2024 file, fielded February to June, arrived October 30, 2025 — so the first memo scores the 2026 file in late 2027 and the 2032 file in 2033 or 2034; the Cooperative Election Study and the ANES 2028 enter on their own release dates. Six memos. Anyone may score what the public files carry. A registration that could not lose for eight years would not be one, and the protocol answers that charge in four rules. Shock classes are registered in advance — a war involving U.S. forces, a domestic mass-casualty attack, a financial crisis, a contested presidential outcome, a pandemic declaration, a platform withdrawal of the medium — and only a registered class can flag a row, and only a class listed on that row in the table, exhaustively and exclusively as in The Imprint and the Exit; a flagged row's window extends by one file, once in the row's life, not once per shock; instrument rows that have not been fielded by the 2029 memo — F6 in its instrumented form, S1, S2, and S4, the rows the protocol names — are scored not fielded and closed, which is a score against the reading that needed them; and the fifth column's branches are resolved by the criterion above, not by the memo. The boundary sweep is scored as the intersection it is, with the script printing joint power over the placements; across rows there is no family-wise correction, and the table is read row by row under the asymmetry statement, never as "some row scored." After deposit the table is strike-only: a row may be withdrawn with a dated reason, never loosened, and never after the first file that carries it is public; no margin and no boundary may move. Every memo's summary line reports the F1/P1 conjunction, so that "nothing happened" can never be read as "the theater confirmed." And two consecutive memos missed declares the registration lapsed as of its last memo.

§ XThe Ledger of Claims and Non-Claims

The four registers — C for what is carried on cited scholarship, E for what is measured at a stated grade, D for design, P for what is predicted — are the archive's; the Notes define them. Register C. Formation as the step both threads assumed and neither examined; the shared-stimulus requirement extracted from Mannheim at passage level, with generation units as its own proof that disagreement is not fragmentation; the known-sharing requirement as this paper's addition, owned as one, with the participatory reading as its stated rival; the theater as a property of media — common knowledge of witnessing — rather than of content; the whisper as private object rather than private sequence, which is what distinguishes it from the feed; the a-problematical stratum Mannheim described as the substitution reading's mechanism; the four readings with their signatures stated in advance; the prior answers as boundaries on where a signature may live; the θ/ω partition; the matched-item rule; bimodality as the units' signature and the fragmentation rows' co-condition; the convergence reading with its content-agreement test; the exposure gradient as identification, conditioned and instrumented; the four carriers; the exhibit, and demographic composition with it, as variance components removed from the statistic by construction, with the cost printed. Register E, at stated grade: the two concessions, verbatim, in-archive; Mannheim's passages at pp. 299–300, 303, and 304, verified against the 1952 text; Jocker, van der Brug, and Rekker 2024 and Twenge et al. 2016 as the prior answers, with Boxell, Gentzkow, and Shapiro 2017 and Heys 2024 beside them at their grades; De Sio et al. 2026 as nearest kin; Arias 2019 as the field test of the known-sharing claim, Corsten 1999 and Edmunds and Turner 2005 as its antecedents; Prior 2007, Bennett and Iyengar 2008, Katz and Scannell 2009, and Webster and Ksiazek 2012 as the dose theory's shelf; the homogenization record (Bommasani et al. 2022; Anderson, Shah, and Kreminski 2024; Doshi and Hauser 2024; Jakesch et al. 2023; Sharma, Liao, and Xiao 2024; Wenger and Kenett 2026); the Cooperative Election Study's even- and odd-year samples from its cumulative guide, and Pew's 2024 opt-in audit; Common Sense Media's 2026 census; Gimbrone et al. 2022 for the non-placement share; Hinduja and Patchin, Miech et al., and Jocker et al. at their published venues; the Monitoring the Future item distribution and the 2024 file's design facts, with the 2019 tablet split-half and 2020 web transition from the series' own overview; Gallup's 40 and 25; Nennstiel and Hudde's 32, 466,089, and 14/7/11; Milosav et al.'s 21 and 14 across 27 countries beside Off, Alexander, and Charron's pooled null; Pew's 64, three in ten, one in five for news, and 59/23/20 with N = 1,458 (December 2025), with the use categories and the expectation items from its February 2026 report, and the 18–29 figures (66/31/15; concern 31→55) from June and August 2026; Hinduja and Patchin's 60.2 and 11.4; Schuman and Corning's technology baselines and Pew's 2016 consensus figures; the youth series' 2019 split, 2020 quarter-sample, and 2021 web transition; the epidemic-scar figures; the series calendar and the eighteen-month microdata lag. Register D: the signature specification with its partition, its matched-item rule, and two controls; the exposure items with their instruments and their first-stage floor; the concurrent battery on the Harvard Youth Poll with its repeat and its two content prompts; the naming test in the published code frame — each with instrument, estimand, population, and failure mode. Register P: the seventeen rows and the fifth column. Nothing else predicts anything.

Non-claims. No finding about the cohort — the series are only now admitting it. No forecast of the cohort's politics: direction, party, or turnout. No claim that the theater is gone: the paper's claim is scoped to the mediated issues of a window, and the witnessed events are registered to show the theater alive. No claim that the whisper's objects differ in content — the convergence reading is registered because they may not. No monocause. No verdict on the mental-health dispute; Haidt and Odgers are carried as the feed carrier's contested evidence. No persuasion claim — what the medium does to a formed adult is The Gate and the Field's question. No prophecy: windows condition on series states, and the clock is the release calendar.

Tripwires. T1, the controls: every formative signature is registered beside the placebo and the post-formation control; a rise either shares is printed as the period's. T2, item-class discipline: a signature claimed on generic ideology, on the witnessed events, on unmatched items, or on the meta-epistemic axis without the discontinuity and the repeat, fails its row; a fragmentation signature claimed without its bimodality co-condition fails its row. T3, exposure carried: the share of the cohort without the medium is in every table, never averaged away; a cohort effect that does not track the gradient — conditioned, and instrumented where the file allows — is reported as one that does not. T4, the exhibit cannot decide: the between-gender component is the exhibit's quantity and no row's, and every between-composition component is likewise no row's; the section may place it and may not confirm a reading. T5, the clock: memos on the public-file calendar; registered shock classes only; one extension per row per life; unfielded instruments closed at the 2029 memo; the lapse rule in force. T6, the shelf: no claim is tightened without a re-search; Jocker, van der Brug, and Rekker, the media-generations school, and the POSTGEN program are the standing examples of what a tightened claim would have missed.

§ XIThe Close

The imprint literature has always inferred that generations form from the differences between them; on attitudes, it never had to compute the quantity that would show a generation failing to form — the method exists, and has been run on health — because for every cohort it could study the theater was there. For the first cohort in the broadcast century's long shadow, the theater is there only for the events large enough to force themselves into common view, and the ordinary issues of the window arrive — on the assumption this paper registers rather than asserts — as private objects that no one else received. This paper does not say what that has done. It says how to tell: on which items, partitioned and matched how, in which series, at which ages, by which statistic and its co-condition, against which two controls, with which margins set from a deposited test and floored before the files exist, and on what clock — with the outcome it expects registered as a row it can lose once the naming test is fielded, and disclosed as unlosable until then.

The archive closes on the falsifier it most fears, and it is not the death of the theater. It is P1 — dispersion rising in the young and in the theater-formed old over the same years, a period effect the paper mistook for a formative one, with every fragmentation row scored against on a computation the paper wrote itself. That row has the same standing as the rows the paper prefers, because a registration that cannot lose is not one. The invitation is the same as The Imprint and the Exit's, aimed one stage earlier: the cohort has not entered power; it is entering the surveys. Anyone with the public files and the deposited script can score the nine rows the public files carry, and the archive binds itself to score the table six times, on the release calendar, until the last of the 2014 births has been asked what it witnessed and whether anyone else was there.

Notes

Registers. C marks conceptual claims carried on cited scholarship; E marks empirical claims re-pullable from named public sources at their stated grade; D marks design claims — the specification, the instruments — each with instrument, estimand, population, and failure mode; P marks the seventeen registered rows and the fifth column, and nothing else in the paper predicts anything.

Rules. (1) No prophecy (windows condition on series states; the clock is the public-file calendar). (2) No period laundering (every formative signature is registered beside the theater-formed placebo and the post-formation control; a rise either shares is printed as the period's). (3) No monocause (four carriers at equal resolution). (4) No shelf-tightening without re-searching (the prior answers and the nearest kin were found before the outline was written, and the coverage round and the archive panel found more).

Prior work. The Living Room and the Long Grip and The Broadcast and the Model (this archive, 2026) conceded the cell this paper occupies; The Imprint and the Exit supplied the registration grammar; The Gate and the Field the ambient channel and the meta-epistemic outcomes. Each is cited once as what it is; the evidence is cited at its primary sources.

What this paper adds to Mannheim, stated once more. He required that a generation's stimulus be common. This paper claims its commonness must be known — a theater manufactures common knowledge, a whisper withholds it — and states the claim as its own operationalization, anticipated in Corsten's discursive-pragmatic reading and tested in Arias's field experiment, with the participatory reading (Mannheim's) as the fifth column's content.

Review record. One adversarial round (two personas, thirteen majors — the live wire re-seated on series that carry window items; the window issues partitioned into witnessed events and mediated issues by a registered criterion; the age-band placebo replaced by two controls; the coherence ratio withdrawn for the between-band share; the substitution design made concurrent with a birth-year discontinuity; the exposure gradient conditioned and instrumented; the gender exhibit removed from the statistic; the "thinned" outcome registered as a row; the protocol closed against deferral) and the standing coverage round (twenty-nine verified items — the composition confound closed by construction; Chwe as the home of the common-knowledge claim; the variance-APC precedent; the naming-test baselines pinned; the youth series' 2020 quarter-sample corrected; the first-source row regraded; the Android instrument's weaknesses stated; the European gender-gap evidence carried both ways). The archive panel (September 2026; four referees, blind — verification, methods, adversary, coverage; reports and the decision memo under reviews/013): the bimodality co-condition on the fragmentation rows; the convergence reading and its two rows; the power paragraph put on one basis and the floors called floors; the matched-item rule; the battery re-vehicled onto the Harvard Youth Poll; the post-formation control defined by births; the Fosse–Winship departure family; the deposit gate keyed to the first row-carrying file; the instruments' first-stage floor and the district instrument demoted; the Cooperative Election Study's data-quality screen and its odd-year samples; the parent papers' claims re-bordered (the earlier paper established simultaneity; the known-sharing claim is this paper's operationalization); Arias, Corsten, Edmunds and Turner, Prior, Bennett and Iyengar, Katz and Scannell, Webster and Ksiazek, Boxell, Heys, Koppel and Berntsen, Sears and Valentino, Erikson and Stoker, and the homogenization record integrated; and a v0 scoring script filed with the companions, its smoke test pending the public files — its synthetic run adding four clauses: raw scale within matched pairs, the calibrated kurtosis-ratio margin, F5's denominator guard, F6's margin. The cold-reader pass (September 16, 2026): the series named at first use, the sister papers glossed where first cited, the rules numbered in the headnote, a legend for the lettered rows and the instruments, "deployment grade" unpacked, the "says so" tic trimmed; the rhetoric findings and the 25-percent cut recorded as an outward-copy option. Referee reports and response letters are filed with the archive.

Verification record. The Mannheim epigraph and passages verified against the 1952 text (pp. 299–300, 303, 304). Venues pinned at imprint for Aksoy, Eichengreen & Saka; Ghitza, Gelman & Auerbach; Osborne, Sears & Valentino; Katz & Liebes; Bottasso, Cerruti & Conti; De Sio et al.; Zheng, Yang & Land; Fosse & Winship. The Gallup, Pew, Milosav, Nennstiel–Hudde, Hinduja–Patchin, Reuters Institute, Common Sense, and Gallup Voices of Gen Z figures re-pulled at the coverage round. Two of the paper's own attributions were wrong in early drafts and corrected: the Political Behavior paper is Jocker, van der Brug & Rekker; the European Sociological Review paper is Nennstiel & Hudde. At the archive panel: Hinduja & Patchin, Miech et al., and Jocker et al. pinned at their published venues; the Cooperative Election Study's even- and odd-year samples taken from its cumulative guide; Dayan & Katz's "high holidays of mass communication" pinned at p. 1; Katz & Liebes quoted exactly; the September 2026 substitution regraded at v4c to the flip log's first entry (execution attested at user-report and trade grade, September 14, 2026; no first-party completion statement), having stood at v4 at The Gate and the Field's "announced; execution unobserved"; the 004 quotation labeled as its revision note. Author-side pins assigned to the deposit: the Monitoring the Future cross-time index pull; the frozen Fθ/Fω candidate list and the matched pairs; the generic-baseline reproduction on the public GSS and ANES files; the Schuman–Corning code frame and ICPSR baselines at page level; the probability series that serves the partition's second leg.

Companion artifacts. The execution brief; the Phase 0 corpus index with its Round 1 and Round 2 shelves and the imprint record; the Phase 1 claim outline; the Phase 2 signature specification with its v2 and v3 amendments; the Phase 3 registration table v3; the v0 scoring script with its synthetic-data validation; the referee reports and response letters — filed with the archive.

References

Aksoy, C. G., B. Eichengreen & O. Saka (2024). "The Political Scar of Epidemics." Economic Journal 134(660):1683–1700. · Alwin, D. F. & J. A. Krosnick (1991). "Aging, Cohorts, and the Stability of Sociopolitical Orientations over the Life Span." American Journal of Sociology 97(1):169–195. · Anderson, B. R., J. H. Shah & M. Kreminski (2024). "Homogenization Effects of Large Language Models on Human Creative Ideation." Proceedings of the 16th Conference on Creativity & Cognition (C&C '24). · Arias, E. (2019). "How Does Media Influence Social Norms? Experimental Evidence on the Role of Common Knowledge." Political Science Research and Methods 7(3):561–578. · Baldassarri, D. & A. Gelman (2008). "Partisans without Constraint: Political Polarization and Trends in American Public Opinion." American Journal of Sociology 114(2):408–446. · Bennett, W. L. & S. Iyengar (2008). "A New Era of Minimal Effects? The Changing Foundations of Political Communication." Journal of Communication 58(4):707–731. · Bolin, G. (2017). Media Generations: Experience, Identity and Mediatised Social Change. London: Routledge. · Bolin, G. & O. Westlund (2009). "Mobile Generations: The Role of Mobile Technology in the Shaping of Swedish Media Generations." International Journal of Communication 3. · Bommasani, R., K. A. Creel, A. Kumar, D. Jurafsky & P. Liang (2022). "Picking on the Same Person: Does Algorithmic Monoculture Lead to Outcome Homogenization?" Advances in Neural Information Processing Systems 35 (NeurIPS 2022). · Bottasso, A., G. Cerruti & M. Conti (2022). "Institutions matter: The impact of the covid-19 pandemic on the political trust of young Europeans." Journal of Regional Science 62(4):1122–1148. · Boxell, L., M. Gentzkow & J. M. Shapiro (2017). "Greater Internet use is not associated with faster growth in political polarization among US demographic groups." Proceedings of the National Academy of Sciences 114(40):10612–10617. · Chwe, M. S.-Y. (2001). Rational Ritual: Culture, Coordination, and Common Knowledge. Princeton: Princeton University Press. · Common Sense Media (2025). Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions. July 16, 2025. · Common Sense Media (2026). The Common Sense Media Census: AI Use by Tweens and Teens, 2026. June 8, 2026 (SSRS Opinion Panel; N = 1,204; ages 9–17; fielded March 18–26, 2026). · Corning, A. & H. Schuman (2015). Generations and Collective Memory. Chicago: University of Chicago Press. · Corsten, M. (1999). "The Time of Generations." Time & Society 8(2–3):249–272. · Dayan, D. & E. Katz (1992). Media Events: The Live Broadcasting of History. Cambridge, MA: Harvard University Press. · De Sio, L. et al. (2026). "Assessing public sphere influence on political attitudes across generations: a mixed-methods study of generational political structuring in Italy." Frontiers in Political Science 8. · DiMaggio, P., J. Evans & B. Bryson (1996). "Have Americans' Social Attitudes Become More Polarized?" American Journal of Sociology 102(3):690–755. · Doshi, A. R. & O. P. Hauser (2024). "Generative AI enhances individual creativity but reduces the collective diversity of novel content." Science Advances 10(28):eadn5290. · Edelman, R. (2025). "The Great Gen Z Divide." Edelman, September 15, 2025 (crediting the "Gen Z 1.0 / 2.0" coinage to Rachel Janfaza). · Edmunds, J. & B. S. Turner (2005). "Global generations: social change in the twentieth century." British Journal of Sociology 56(4):559–577. · Erikson, R. S. & L. Stoker (2011). "Caught in the Draft: The Effects of Vietnam Draft Lottery Status on Political Attitudes." American Political Science Review 105(2):221–237. · Fosse, E. & C. Winship (2019). "Analyzing Age-Period-Cohort Data: A Review and Critique." Annual Review of Sociology 45:467–492. · Gallup (2024). "U.S. Women Have Become More Liberal; Men Mostly Stable"; "Exploring Young Women's Leftward Expansion." · Gallup, Walton Family Foundation & GSV (2026). Voices of Gen Z: AI. April 8, 2026. · Ghitza, Y., A. Gelman & J. Auerbach (2023). "The Great Society, Reagan's Revolution, and Generations of Presidential Voting." American Journal of Political Science 67(3):520–537. · Gimbrone, C., L. M. Bates, S. Prins & K. M. Keyes (2022). "The politics of depression: Diverging trends in internalizing symptoms among US adolescents by political beliefs." SSM – Mental Health 2:100043. · Giuliano, P. & A. Spilimbergo (2014). "Growing Up in a Recession." Review of Economic Studies 81(2):787–817. · Gumpert, G. & R. Cathcart (1985). "Media grammars, generations, and media gaps." Critical Studies in Mass Communication 2(1):23–35. · Haidt, J. (2024). The Anxious Generation. New York: Penguin. · Heys, E. (2024). Cultural Evolution: Measuring Differences in Generational Values. Berkeley Institute for Young Americans, June 2024. · Hinduja, S. & J. W. Patchin (2026). "Risks and Harms of Conversational Artificial Intelligence (CAI) Chatbot Use Among US Youth." Journal of Adolescence 98(5):1597–1606. · Jakesch, M., A. Bhat, D. Buschek, L. Zalmanson & M. Naaman (2023). "Co-Writing with Opinionated Language Models Affects Users' Views." Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. · Janssen, S. M. J. (2015). "Also No Support for the Youth Bias: Reply to Koppel and Berntsen." Applied Cognitive Psychology 29(1):71–72. · Jocker, T., W. van der Brug & R. Rekker (2024). "Growing Up in a Polarized Party System: Ideological Divergence and Partisan Sorting Across Generations." Political Behavior 46(4):2263–2286. · Johnston, L. D., P. M. O'Malley, J. G. Bachman & J. E. Schulenberg (2012). Monitoring the Future National Survey Results on Drug Use, 1975–2011. Volume I: Secondary School Students. Ann Arbor: Institute for Social Research (Appendix C, "Estimation of Sampling Errors"). · Katz, E. & T. Liebes (2007). "'No More Peace!': How Disaster, Terror and War Have Upstaged Media Events." International Journal of Communication 1:157–166. · Katz, E. & P. Scannell, eds. (2009). The End of Television? Its Impact on the World (So Far). The ANNALS of the American Academy of Political and Social Science 625. · Kingstone, H. & J. Bristow, eds. (2024). Studying Generations: Multidisciplinary Perspectives. Bristol: Bristol University Press. · Koppel, J. & D. Berntsen (2014). "Does Everything Happen When You Are Young? Introducing the Youth Bias." Quarterly Journal of Experimental Psychology 67(3):417–423. · Krosnick, J. A. & D. F. Alwin (1989). "Aging and Susceptibility to Attitude Change." Journal of Personality and Social Psychology 57(3):416–425. · Lindner, A., A. Stelboum & Y. Hakim (2024). "Embracing Generational Labels." Socius 10. · Mannheim, K. (1952 [1928]). "The Problem of Generations." In P. Kecskemeti (ed.), Essays on the Sociology of Knowledge, 276–322. London: Routledge & Kegan Paul. · Miech, R. A., M. P. Couper, S. G. Heeringa & M. E. Patrick (2021). "The impact of survey mode on US national estimates of adolescent drug prevalence: results from a randomized controlled study." Addiction 116(5):1144–1151. · Miech, R. A., L. D. Johnston, M. E. Patrick & P. M. O'Malley (2025). Monitoring the Future: A Continuing Study of American Youth (12th-Grade Survey), 2024. ICPSR 39444, v2. · Milosav, Đ., Z. Dickson, S. B. Hobolt, H. Klüver, T. Kuhn & T. Rodon (2026). "The youth gender gap in support for the far right." Journal of European Public Policy 33(2):444–468. · Munger, K. (2022). Generation Gap: Why the Baby Boomers Still Dominate American Politics and Culture. New York: Columbia University Press. · Nennstiel, R. & A. Hudde (2025). "Is there a growing gender divide among young adults in regard to ideological left–right self-placement? Evidence from 32 European countries." European Sociological Review 41(6):862–883. · Odgers, C. L. (2024). "The great rewiring: is social media really behind an epidemic of teenage mental illness?" Nature 628:29–30. · Off, G., A. Alexander & N. Charron (2025). "Is there a gender youth gap in far-right voting and cultural attitudes?" European Journal of Politics and Gender. · Osborne, D., D. O. Sears & N. A. Valentino (2011). "The End of the Solidly Democratic South: The Impressionable-Years Hypothesis." Political Psychology 32(1):81–108. · Parker, K. (2023). "How Pew Research Center will report on generations moving forward." Pew Research Center, May 22, 2023. · Pew Research Center (2016). Americans Name the 10 Most Significant Historic Events of Their Lifetimes. December 15, 2016. · Pew Research Center (2024). "Online opt-in polls can produce misleading results, especially for young people and Hispanic adults." March 5, 2024. · Pew Research Center (2025). Teens, Social Media and AI Chatbots 2025. December 9, 2025. · Pew Research Center (2026a). How Teens Use and View AI. February 24, 2026. · Pew Research Center (2026b). Americans and AI 2026. June 17, 2026; and the short read of August 18, 2026 (N = 3,488) on concern about AI by age. · Pilcher, J. (1994). "Mannheim's Sociology of Generations: An Undervalued Legacy." British Journal of Sociology 45(3):481–495. · Prior, M. (2007). Post-Broadcast Democracy: How Media Choice Increases Inequality in Political Involvement and Polarizes Elections. Cambridge: Cambridge University Press. · Reuters Institute for the Study of Journalism (2026). Digital News Report 2026. Oxford. · Schuman, H. & A. Corning (2012). "Generational Memory and the Critical Period: Evidence for National and World Events." Public Opinion Quarterly 76(1):1–31. · Schuman, H. & J. Scott (1989). "Generations and Collective Memories." American Sociological Review 54(3):359–381. · Sears, D. O. & N. A. Valentino (1997). "Politics Matters: Political Events as Catalysts for Preadult Socialization." American Political Science Review 91(1):45–65. · Sharma, N., Q. V. Liao & Z. Xiao (2024). "Generative Echo Chamber? Effect of LLM-Powered Search Systems on Diverse Information Seeking." Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems. · Trachtman, S., S. F. Anzia & C. Hill (2023). "Age-group identity and political participation." Research & Politics 10(2). · Twenge, J. M., N. Honeycutt, R. Prislin & R. A. Sherman (2016). "More Polarized but More Independent: Political Party Identification and Ideological Self-Categorization Among U.S. Adults, College Students, and Late Adolescents, 1970–2015." Personality and Social Psychology Bulletin 42(10):1364–1383. · Twenge, J. M. (2017). iGen. New York: Atria. · Vittadini, N., A. Siibak, I. Reifová & H. Bilandzic (2014). "Generations and media: the social construction of generational identity and differences." In N. Carpentier, K. C. Schrøder & L. Hallett (eds.), Audience Transformations. London: Routledge. · Webster, J. G. & T. B. Ksiazek (2012). "The Dynamics of Audience Fragmentation: Public Attention in an Age of Digital Media." Journal of Communication 62(1):39–56. · Wenger, E. & Y. N. Kenett (2026). "Large language models are homogeneously creative." PNAS Nexus 5(3):pgag042. · Zheng, H., Y. Yang & K. C. Land (2011). "Variance Function Regression in Hierarchical Age-Period-Cohort Models." American Sociological Review 76(6):955–983. · This archive (armchair.marshallcahill.com), 2026: The Living Room and the Long Grip (No. 004); The Broadcast and the Model (No. 005); The Imprint and the Exit (No. 011); The Gate and the Field (No. 012).

The companion deck — The Theater and the Whisper — sits beside this paper, with the signature specification, the three instruments, the registration table, the v0 scoring script, the execution brief, the corpus index, both internal rounds, and the archive panel's record filed with the archive. The clock runs from the deposit — table, specification, and scoring script to a timestamped public archive before the first file that carries a registered row is public. Six memos follow the public-file calendar, the first in late 2027 and the last in 2033 or 2034, and the falsifier the paper most fears — a period effect mistaken for a formative one — has the same standing in every one of them as the readings the paper prefers.